Aquatic plants have evolved repeatedly from terrestrial ancestors, resulting in a broad phylogenetic dispersion and remarkable morphological diversity (Cook 1999). One conspicuous example of this diversity is the floating-leaved growth form, in which plants rooted underwater produce leaves that float on the water surface (Sculthorpe 1967; Hutchinson 1975; Schuyler 1984). Several diverse plant lineages have evolved this habit, including the genus Nymphoides (Menyanthaceae), one of the most diverse and widespread floating-leaved groups (Cook 1996; Kadereit 2007). There are approximately 50 Nymphoides species worldwide (Tippery et al. 2008). Some (e.g. N. indica) have rather broad distributions, but most are restricted to a single continent or even to a local geographic region (Aston 1973, 1982; Raynal 1974b; Sivarajan and Joseph 1993; Li et al. 2002). Several regional taxonomic treatments exist for Nymphoides, but there has been no recent comprehensive treatment of the genus worldwide.

With two exceptions (N. cambodiana and N. exiliflora), Nymphoides species produce distinctive floating leaves that subtend umbellate flower clusters or lax racemes (Aston 1973; Sivarajan and Joseph 1993; Tippery et al. 2009). Flowers are borne above water, but the capsular fruits develop underwater (Aston 1973; Raynal 1974a). The flowers of most species are pentamerous with yellow or white petals that can be glabrous or ornamented with hairs or laciniate wings. Seeds differ among species in their shape (globose or elliptical, some compressed laterally), size (from 0.4 mm to over 5 mm), surface ornamentation (smooth or tuberculate, some carunculate), and number (from one or two to over 100 per fruit). Although floral and seed characters vary among species, other features like inflorescence architecture and fruit structure are more consistent across taxa (Raynal 1974a; Aston 1982, 2003; Chuang and Ornduff 1992; Sivarajan and Joseph 1993).

Two major types of inflorescence architecture exist in Nymphoides, and these differ by their relative internode elongation (Tippery et al. unpublished data). Floral axes are either expanded (having internodes that elongate between pairs of flowers) or condensed (with each cluster of several flowers supported by a single floating leaf) (Aston 1982; Sivarajan and Joseph 1993). The temperate Eurasian species N. peltata produces a unique third inflorescence type, in which two leaves support each of several successive flower clusters on an axis with elongated internodes (Raynal 1974a). The condensed inflorescence morphology characterizes the majority of Nymphoides species throughout the range of the genus, whereas expanded-inflorescence species occur only in Australia and tropical Asia. A recent phylogenetic study of Nymphoides, which sampled 12 species, inferred a basal grade of expanded-inflorescence species (and N. peltata) toward a clade of species with condensed inflorescences (Tippery et al. 2009). The evolutionary interpretation from this result is that the condensed inflorescence morphology arose once in Nymphoides from an ancestor having an expanded inflorescence. However, because Tippery et al. (2009) sampled only about half of the species with expanded inflorescences and one-tenth of condensed-inflorescence species, more robust conclusions have awaited further taxon sampling, particularly of Nymphoides species with expanded inflorescences.

One of the more intriguing floral characteristics of Nymphoides is the presence of dimorphic heterostyly, a sexual condition characterized by the spatial separation of anthers and stigma occurring reciprocally in different plants (i.e. the anthers of one morph type are at the height of stigmas in the other morph type). Heterostyly promotes outcrossing by presenting pollen at a different height than the stigma of the same flower, but corresponding to the height of receptive stigmas on compatible flowers of the opposite morph type (Barrett and Shore 2008). In addition, heterostylous plants frequently are genetically self-incompatible, which prevents interbreeding between plants of the same morph type (Ganders 1979). Heterostyly occurs in a number of distantly related plant lineages, and its phylogenetic distribution supports the inference of multiple, independent origins (Ganders 1979; Barrett 1992; Barrett and Shore 2008; Cohen 2010). Heterostyly is the ancestral condition in Menyanthaceae, with the syndrome lost independently in several nonheterostylous lineages (Tippery et al. 2008). In addition, four Nymphoides species (N. aquatica, N. cordata, N. krishnakesara, N. macrosperma) are dioecious, having both flowers and individuals that are unisexual (Ornduff 1966; Sivarajan and Joseph 1993). Still another condition characterizes the gynodioecious N. cristata, a species in which individuals produce either pistillate (female) or hermaphroditic flowers (Vasudevan Nair 1973). Gynodioecy and dioecy also promote outcrossing, with the latter representing the most effective mechanism to avoid self-pollination (Charlesworth and Charlesworth 1978). Furthermore, gynodioecy often constitutes a transitional step in the evolution of dioecy (Barrett 2010). The existence of a similar ‘gynodioecy pathway’ to dioecy in Nymphoides is supported by the previously inferred sister relationship between N. cristata and N. cordata (Tippery et al. 2008). The combination of ancestral heterostyly and derived dioecy and gynodioecy in Nymphoides (Tippery et al. 2008) provides an opportunity to compare the evolutionary patterns of these different sexual conditions and to elucidate why dioecy and gynodioecy might have evolved in a group already equipped with the heterostylous outcrossing mechanism.

Phylogenetic methods provide an appropriate means of evaluating evolutionary hypotheses pertaining to a variety of factors, including sexual condition (heterostyly, gynodioecy, and dioecy) and inflorescence architecture (condensed or expanded) in Nymphoides. Accordingly, we conducted a phylogenetic analysis of Nymphoides to determine relationships among species and to evaluate patterns of character evolution. Specifically, we used more extensive taxon sampling to test the results of previous analyses, namely that the condensed inflorescence morphology had a single evolutionary origin, and that dioecious and gynodioecious species comprise sister lineages. Our study represents the first complete sampling of Australian Nymphoides species in a molecular phylogenetic context and includes a nearly complete set of species having the expanded inflorescence morphology. In total, we analyzed data for over half of the genus worldwide. Using nuclear (ITS) and plastid (matK/trnK) data, we obtained a molecular appraisal of interspecific relationships in Nymphoides and the most comprehensive phylogenetic evaluation to date of inflorescence architecture and sexual condition evolution in the genus.

MATERIALS AND METHODS

Morphological Data—Morphological data for 54 Nymphoides taxa (including two subspecies of N. indica; Appendix 1) were obtained from literature (Aston 1973, 1982, 1984, 1986, 1987, 1997, 2002, 2003; Raynal 1974b; Sivarajan and Joseph 1993) or directly from observations of preserved specimens (in ethanol or dried) or live plants. We omitted two species from consideration that until recently were circumscribed within Nymphoides: N. exigua (F. Muell.) Kuntze, which Tippery and Les (2009) transferred to Liparophyllum (as L. exiguum), and N. stygia (J. M. Black) H. Eichler, a rare and possibly extinct species with uncertain taxonomic affinity (Aston 2009). Other species synonymized by some authors (Nymphoides cristata/N. hydrophylla, N. elegans/N. moratiana, N. humboldtiana/N. indica/N. thunbergiana; Ornduff 1969; Klackenberg 1990; Sivarajan and Joseph 1993) were coded separately and retained provisionally as distinct.

Nymphoides species were scored for 28 vegetative and reproductive features used to distinguish species (Appendices 1–2). Morphological data were verified against specimens where possible; otherwise they were obtained only from literature sources. Eighteen species were scored using only published information, as specimens were unavailable or in poor condition (Appendix 1). Six species (N. flaccida, N. hastata, N. lungtanensis, N. microphylla, N. minor, and N. verrucosa) were excluded from the morphological phylogenetic analysis (see below) because published morphology data were insufficient to distinguish them. Specifically, the excluded species had uncorrected p distances (determined in PAUP*; Swofford 2002) equal to zero in pairwise comparisons with more than four other taxa. After exclusion of these species, there were no zerolength pairwise distances. Dioecious and gynodioecious species with unisexual flowers (character 5) were coded as ambiguous for heterostyly (character 4). Inflorescence morphology was coded according to whether expanded internodes separate pairs of flowers (expanded inflorescence type) or whether flowers occur in dense clusters (condensed inflorescence type and N. peltata; character 7). Two species that have been reported or illustrated with condensed inflorescences, N. beaglensis (Aston 1987) and N. crenata (Aston 1973), but which have the ability to elongate their floral axis internodes (N. P. Tippery, pers. obs.), were scored as having expanded internodes following observations of field and herbarium specimens. Quantitative characters (characters 6, 8, 9, 10, 11, 20, 21, and 22) were coded using break points established by visual inspection of ranges plotted for individual species (Fig. S1). Phylogenetic analyses of morphological data (see below) were conducted with and without quantitative characters included.

Molecular Data—Nucleotide sequence data were obtained from 143 specimens collected and preserved in the field using a saturated NaCl/CTAB solution (Rogstad 1992) or from dried herbarium specimens. Specimens were identified using relevant literature sources (see above). In addition, two specimens (labeled ‘Nymphoides sp. 1’ and ‘Nymphoides sp. 2’, Appendix 3) were considered unidentified species. Genomic DNA was extracted using a standard method (Doyle and Doyle 1987) and amplified for the selected gene regions following Les et al. (2008). Primers used for amplification and sequencing were as follows: ITS: ITS2, ITS3, ITS4, and ITS 5 (Baldwin 1992); matK and trnK introns: trnK-3914 (dicot) and trnK-2R (Johnson and Soltis 1995), 1F (Bremer et al. 2002), and 0445F, 0503R, 1011R, 1556R, 1749F, 1848F, and 1966R (Tippery et al. 2008). For accessions that were difficult to sequence or for taxa that were represented by multiple accessions, matK/trnK sequence data were obtained preferentially from the more variable 5′ trnK intron region. Amplified DNA fragments were purified using 0.1 µL ExoSAP-IT® enzyme mixture (Affymetrix, Inc., Santa Clara, California), 0.9 µL water, and 1.0 µL amplification product in a 2.0 µL reaction. Sequencing reactions were conducted using 2.0 µL, of cleaned amplicon, 1.0 µL of Big Dye® (Applied Biosystems, Foster City, California), 2.0 µL of 5 × ABI buffer, and 3.2 pmol of sequencing primer in a 10 µL reaction. Cycle sequencing and cleanup followed Les et al. (2008); sequencing was performed on an ABI PRISM^® 3100 genetic analyzer (Applied Biosystems).

Chromatograms were edited using the program 4Peaks ver. 1.7 (Griekspoor and Groothuis 2005) and assembled into contigs using CodonCode Aligner ver. 3.0.3 (CodonCode Corporation, Dedham, Massachusetts). Nucleotide sequences were aligned manually under the similarity criterion (Simmons 2004) using MacClade ver. 4.06 (Maddison and Maddison 2000). Sequences newly obtained for this study were combined with those previously reported (Tippery et al. 2008, 2009; Tippery and Les 2009, 2011) for 13 Nymphoides species and eight Liparophyllum species (sensu Tippery and Les 2009) that were retrieved from GenBank (Appendix 3). Insertions and deletions (indels) were scored for the aligned nucleotide matrices using simple indel coding (Simmons and Ochoterena 2000) implemented with the program SeqState ver. 1.4.1 (Müller 2005). Aligned morphological and molecular data matrices were uploaded to TreeBASE (study number 11079)

Phylogenetic Analyses—Data were analyzed separately (morphology, ITS [nucleotide and indel], matK/trnK [nucleotide and indel]) using both equally-weighted maximum parsimony (MP; Fitch 1971) and maximum likelihood (ML; Felsenstein 1973) methods. Partition-homogeneity / incongruence-length difference (ILD) tests were conducted using PAUP* ver. 4.0b10 (heuristic search, 1,000 replicates, maxtrees = 1,000; Farris et al. 1994; Swofford 2002) with a significance threshold of p < 0.01, to evaluate the relative congruency of the nuclear (ITS DNA and indels) and plastid (matK/trnK DNA and indels) data partitions. After initial results indicated that the partitions were significantly incongruent (see Results), we employed taxon jackknifing (Lecointre and Deleporte 2005) to determine the species responsible for the incongruence, a procedure that proved effective in a previous study of N. montana (Tippery and Les 2011). To keep jackknif e analyses computationally tractable, clades (containing up to three species) that resolved consistently on both ITS and matK/trnK phylogenies (see Results) were treated as stable lineages (i.e. they were removed or replaced in jackknife analyses as a single unit). Ingroup species or clades were pruned singly or in pairs, with outgroup taxa excluded from ILD analyses. After determining that congruence could not be achieved even by removing up to two ingroup species or clades (see Results), the nuclear and plastid data partitions were evaluated independently.

Heuristic tree searches were performed under parsimony in PAUP* (Swofford 2002) with 100 replicates of random stepwise addition and branch swapping by tree bisection and reconnection (TBR), using maxtrees = 100,000. Data matrices for which heuristic searches recovered the maximum number of trees were subjected to ten independent searches with maxtrees = 100,000, after which duplicate trees were removed. Summary tree sets from these analyses thus contained up to 1,000,000 trees, minus any duplicate topologies. To assess the completeness of tree searching, ten further searches were conducted for trees incompatible with the strict consensus tree and having equal or lower tree lengths than the minimum already found, using search parameters as above. If no such trees were found, the consensus tree was considered to represent all most-parsimonious trees. Support for internal branches was evaluated using 1,000 bootstrap (Felsenstein 1985) replicates in PAUP* with the following options: heuristic search, one random stepwise addition per replicate, swapping by TBR, and maxtrees = 10,000. Trees were depicted as one of many most-parsimonious phylograms or strictconsensus cladograms of all most-parsimonious topologies. The outgroup comprised species of Liparophyllum, which previous studies resolved as the sister genus to Nymphoides (Tippery et al. 2008; Tippery and Les 2009).

After model selection with jModeltest ver. 0.1.1 (Posada 2008) under the AIC criterion (Akaike 1974), likelihood analysis was implemented using GARLI ver. 0.97.r737 (Zwickl 2006) with default settings except as noted. Morphological and indel data were evaluated using the Mkv model (Lewis 2001), and models of DNA evolution were applied to ITS (GTR + G) and matK/trnK (TVM + G) aligned nucleotide data. Ten separate likelihood runs were performed using different random starting seeds, and the tree with the maximum likelihood score was compared with the parsimony consensus tree. Bootstrap analysis was conducted in GARLI using 1,000 replicates.

Alternative Topology Testing—Nodes that differed in topology between the ITS and matK/trnK data analyses (see Results) were subjected to statistical topology testing using parsimony and likelihood methods. Incongruence was evaluated under parsimony with the Templeton/ Wilcoxon signed-rank test (Templeton 1983), conducted in PAUP*. Nodes that were specific to the ITS or matK/trnK strict consensus tree were used as topological constraints on the other data set, and the effect of the constraint was evaluated as the degree of tree length increase relative to the unconstrained analysis. Constrained and unconstrained trees also were constructed and site-specific likelihoods obtained under the likelihood criterion in GARLI, using the parameters described above. Congruence for these trees was evaluated using the Shimodaira-Hasegawa (SH) test (Shimodaira and Hasegawa 1999) and the approximately unbiased (AU) test (Shimodaira 2002), both implemented in the package scaleboot ver. 0.3–3 (Shimodaira 2008) in the program R ver. 2.12.0 (R Development Core Team 2009). We judged nodes to be significantly incongruent if the three statistical tests all yielded significant p values (p < 0.05).

Character Evolution—State transitions for two features, heterostyly (character 4, Appendix 2) and inflorescence architecture (character 7), were mapped onto the ITS and matK/trnK trees under the parsimony criterion in PAUP* (Swofford 2002), using both accelerated (acctran) and delayed (deltran) transition optimization. States for outgroup species were assigned following previously published data (Tippery et al. 2008).

RESULTS

Morphological Data—Of the 54 Nymphoides taxa that were coded for morphology (Appendix 1), six lacked data for more than half of the characters, and all of these except N. siamensis were excluded from the phylogenetic analysis of morphological data (see Materials and Methods). Another excluded species (N. lungtanensis) lacked data for 12 characters. Species that were excluded from the morphological data analysis also lacked molecular data. All 28 coded morphological characters were parsimony-informative (Appendix 2), and 12.4% of cells lacked data for the matrix of 48 taxa. With quantitative characters excluded, the matrix of 20 characters had 16.4% missing data. Except for chromosome number (character 1), which was scored for only 14 species, all morphological characters were scored for over half of the included taxa (Appendix 1).

Molecular Data—Using molecular data, we analyzed 31 Nymphoides species and two provisionally unidentified specimens (Appendix 3). The ITS data matrix (955/167 characters [nucleotide/indel], 338/135 parsimony-informative, 4.6% missing data excluding gaps) comprised 142 accessions, and the matK/trnK matrix (2,654/44 characters, 218/25 parsimony-informative, 32.4% missing data overall, 5.8% missing in the 5' trnK intron region) had 126 accessions. The ILD test on a combined ITS + matK/trnK data matrix containing 123 accessions indicated significant incongruence between the nuclear and plastid data partitions (p = 0.001). Successive pruning of taxa failed to yield a nonsignificant incongruence (p > 0.01) after removing all possible pairs of species or consistently resolved clades.

Phylogenetic Analyses—Phylogenetic analysis of morphological data with no characters excluded obtained 566,183 most-parsimonious trees (150 steps, CI = 0.27, RI = 0.65; not shown). Several clades were resolved on the strict consensus tree, but none with high bootstrap support. Relationships that were resolved included a clade of the dioecious North American species N. aquatica and N. cordata plus the gynodioecious Indian species N. cristata, and a separate clade of dioecious species native to India (N. krishnakesara + N. macrosperma). Another resolved clade contained Nymphoides species with relatively small leaves and flowers (N. furculifolia + N. minima + N. planosperma + N. simulans; Appendices 1–2). With quantitative characters excluded, 781,205 most-parsimonious trees were obtained (161 steps, CI = 0.25, RI = 0.61; Fig. 1). The strict consensus tree from this analysis was considerably more resolved than the tree constructed using all characters, but with branches still receiving poor (< 50%) or marginal bootstrap support. Most of the relationships described above were recovered also in the second analysis, and additionally some deeper nodes were resolved. Among these were two large clades of species with condensed inflorescences (Fig. 1, arrows) and a clade containing N. herzogii (also condensed inflorescence) and N. peltata (unique inflorescence type). Heterostyly and homostyly did not exhibit any clear phylogenetic signal, although homostylous species (N. hydrophylla, N. parvifolia, N. sivarajanii) were implicated as closely related to the dioecious/gynodioecious species mentioned above.

Analysis of the ITS data matrix obtained 648,168 most-parsimonious trees (1,047 steps, CI_exc = 0.65, RI = 0.94), and analysis of the matK/trnK data matrix yielded 999,788 trees (412 steps, CI_exc = 0.88, RI = 0.97; Fig. 2). The topologies of likelihood trees (not shown) were fully congruent with the corresponding strict consensus trees recovered under parsimony. The majority of species for which multiple accessions were analyzed resolved to single, well-supported clades on both the ITS and matK/trnK trees, sometimes with no branch lengths between accessions. Clades with notable sequence variation within a species included the geographically widespread N. indica, for which the primarily Australian accessions comprised several clades (ITS data) or were paraphyletic to N. ezannoi and N. fallax + N. humboldtiana (matK/trnK data). Nymphoides aurantiaca accessions formed a diverse assemblage that either was unresolved relative to N. cambodiana (ITS data) or represented a distinct clade (matK/trnK data). In both analyses, N. crenata accessions resolved to two subclades separated by substantial genetic distance between them, and a single accession of N. exiliflora from the Northern Territory was distinct from four other accessions of this species from Queensland.

FIG. 1.

Strict consensus (maximum parsimony - MP) phylogenetic tree of relationships among Nymphoides species, constructed using morphological character data (Appendix 1). The tree was rooted at N. crenata, following results from molecular data analyses. Bootstrap support values are given as MP above and maximum likelihood below each node; dashes (-) indicate support below 50%. Arrows indicate two large clades containing only species with condensed inflorescences.

Separate analyses of nuclear and plastid data produced largely incongruent trees, although they did resolve some of the same relationships (Fig. 3, circled nodes). Both data matrices strongly supported a monophyletic Nymphoides relative to Eiparophyllum. The only other major point of agreement involved the placement of N. minima, which resolved as the sister species to a clade containing about half of the species analyzed. The nodes that resolved the position of N. minima were recovered by both nuclear and plastid data, mostly with moderate support. Elsewhere on the trees, both data sets resolved clades containing N. aquatica + N. cordata, N. aurantiaca + N. cambodiana, N. fallax + N. humboldtiana, N. furculifolia + N. parvifolia + N. quadriloba + ‘Nymphoides sp. 1’ + ‘Nymphoides sp. 2’, and N. planosperma + N. simulons + N. spongiosa.

Alternative Topology Testing—The ITS and matK/trnK consensus trees differed at 18 nodes on the ITS tree and eight on the matK/trnK tree (Table 1; Fig. 3, nodes with letter designations). Statistical tests of topological congruency revealed that nine nodes on the ITS phylogeny were significantly incongruent with the matK/trnK tree, and five nodes on the matK/ trnK tree were incongruent with the ITS tree (Fig. 3, arrows). Some nodes that were significantly incongruent had marginal or moderate bootstrap support values, although several wellsupported nodes (bootstrap > 90%) also were incongruent. Significantly incongruent nodes were primarily located in the basal grade of lineages toward N. minima, with only one significantly incongruent node in the clade sister to N. minima.

Character Evolution—The evolution of two characters was reconstructed on the nuclear and plastid trees (Fig. 3). The condition of heterostyly (character 4, Appendix 2) resolved as the ancestral state for the root node of Nymphoides and for approximately half of the internal nodes on the ITS and matK/trnK trees. Ancestral nodes were reconstructed to be heterostylous primarily in the grade of species leading to N. minima and its sister clade, whereas most internal nodes in the latter were reconstructed to be homostylous. Despite their topological incongruence, the ITS and matK/trnK phylogenies both reconstructed four transitions from heterostyly to homostyly and four reversions. The dioecious N. aquatica and N. cordata were largely unresolved relative to other species, although both their most recent ancestor and the precursor of the gynodioecious N. cristata were reconstructed to be homostylous on both trees. Inflorescence morphology (character 7) mapped to four transitions within Nymphoides on the ITS tree (all from expanded to condensed inflorescence type) and four transitions on the matK/trnK tree (two transitions in each direction; Fig. 3). Reconstructions optimized for accelerated or delayed transitions returned the same ancestral states within Nymphoides for both heterostyly and inflorescence morphology.

DISCUSSION

Causes of Tree Incongruence—Phylogenetic study of Nymphoides using two genomic regions, the nuclear ITS and the plastid matK/trnK, failed to recover congruent tree topologies. Alternative topology tests indicated that the placements of many species on the nuclear and plastid trees were significantly incongruent (Fig. 3, arrows). In a previous study that analyzed fewer Nymphoides, the source of incongruence was traced to a single species, N. montana, which was deemed to be of hybrid origin because its nuclear (ITS) sequence data were most similar to N. geminata but its plastid (rbcL and trnK) sequence data closely matched N. spinulosperma (Tippery and Les 2011). In the current study, however, which included 31 ingroup taxa, the disagreement between nuclear and plastid data trees was too extensive to be attributed to only one or even several species. Systematic removal of taxa failed to produce nonsignificant ILD p values in any scenario where one or two species (or consistently resolved clades) were removed, and systematic removal of three or more species or clades was computationally intractable (25 ingroup species or clades arranged into inclusion groups of 22 yielded 2,300 combinations).

FIG. 2.

Nymphoides phylogenetic trees, constructed using ITS (left) or matK/trnK data (right). Each tree represents one of many most-parsimonious trees (ITS: one of 648,168, matK/trnK: one of 999,788; branch lengths indicated). Two species that were resolved as paraphyletic on the matK/trnK tree (N. indica and N. montana) are indicated with asterisks (*). Branch support values are depicted on the strict consensus tree (Fig. 3).

FIG. 3.

Strict consensus (MP - maximum parsimony) phylogenetic trees constructed using ITS (left) or matK/trnK data (right) for Nymphoides. Multiple accessions (see Fig. 2) are condensed into one terminal branch for each species. The paraphyletic resolutions of N. indica and N. montana on the matK/trnK tree are indicated by asterisks (*). Bootstrap support values are given as MP above and maximum likelihood below each node; dashes (-) indicate support below 50%. Central lines are provided as a visual aid to locate identical species between trees and are not meant to indicate similarly resolved phylogenetic relationships. Nodes resolved by both ITS and matK/trnK are circled, whereas incongruent nodes are labeled with letter designations (Table 1). Significantly incongruent nodes (see text) are indicated with arrows. Ancestral state reconstructions using parsimony are depicted for two characters. Hashed lines indicate the presence of heterostyly (character 4, Appendix 2), whereas homostylous lineages have smooth lines. Lineages ancestral to dioecious (N. aquatica, N. cordata) or gynodioecious species (N. cristata) and species with unknown reproductive systems (Appendix 1) are shown with sparsely hashed lines. Inflorescence morphology, coded as the presence or absence of expanded floral axis internodes (character 7), is depicted using black lines for expanded inflorescence morphology, white lines for condensed inflorescence morphology, and grey lines for Liparophyllum gunnii (solitary inflorescence).

There are several possible explanations for the incongruence between nuclear and plastid data. The different signal could be due to incomplete lineage sorting, gene duplication, or hybridization (Maddison 1997), the latter possibly coincident with speciation. Incomplete lineage sorting characterizes relatively recent divergences (Maddison 1997) and would not be expected to produce most of the reticulate patterns we observed across large phylogenetic distances in Nymphoides. It could, however, potentially explain the incongruence among more recently diverged Nymphoides species (e.g. in the clade sister to N. minima). Gene duplications are unlikely to have produced phylogenetic incongruence in the genetic loci we studied, one of which (ITS) routinely homogenizes to a single copy through the process of concerted evolution (Álvarez and Wendel 2003) and the other, a haploid plastid locus (matK/trnK), presumably lacks substantially divergent allelic variants within a species. Hybridization, in contrast to other sources of incongruence, would allow for spontaneous genetic exchange between species that already have diverged phylogenetically and this possibility is most consistent with our results in Nymphoides.

TABLE 1.

Evaluation of nodes that differ between ITS and matK/trnK strict consensus trees (Fig. 3). Node letters identify constraints from one tree that were applied to the other data set (e.g. node A from the ITS tree was used as a constraint on the matK/trnK data analysis). Maximum parsimony (MP) data are given as the length increase of the constrained tree relative to the unconstrained analysis and the Templeton test p value (see text). Maximum likelihood (ML) values display the score decrease of constrained trees relative to unconstrained and p values from the SH and AU tests (see text). Values significant at p < 0.05 are highlighted with an asterisk (*). Nodes for which all three tests yielded significant results are highlighted on Fig. 3.

The ITS sequence chromatograms obtained during this study notably did not show any polymorphic nucleotide sites, which in other groups (e.g. Les et al. 2009, 2010) have provided straightforward evidence of hybridization (i.e. both alleles of extant parental species or recombinants thereof are separable using molecular cloning methods). On the other hand, under a more ancient hybridization scenario (e.g. hybrid speciation; Rieseberg 1997) the genetic background of one or both parental taxa could have been lost subsequently, along with any corresponding phylogenetic signal. Studies in other groups have shown that nuclear DNA can be lost preferentially from one parental genome over another following hybridization (Buggs et al. 2009). Furthermore, the ITS region is prone to homogenization through concerted evolution (Álvarez and Wendel 2003), which could lead to the loss or dilution of signal from the maternal nuclear genome, thereby rendering evidence of the maternal parent only in the plastid genome. In our study, approximately half of the Nymphoides species resolved differently on the ITS and matK/trnK trees, which could be interpreted to represent an unreasonable amount of hybridization. However, a similar proportion of hybrid species has been demonstrated in the genus Persicaria Mill. (Kim and Donoghue 2008). In that study, the authors were able to correlate hybridization with allopolyploid speciation, and thus determine that polyploids were the result of hybridization events between diploid parental lineages.

In Nymphoides (x = 9), chromosome numbers are known for relatively few species, although multiple diploid (2n = 18), tetraploid (2n = 36), and hexaploid (2n = 54) species have been reported for the genus (Ornduff 1970; Li et al. 2002). Without being able to ascribe ploidy levels to most of the Nymphoides species in this study, it is impossible to identify allopolyploid lineages or assign them to putative diploid parents. Of the Nymphoides species with published chromosome counts, three included species (N. crenata, N. cristata, and N. indica) reportedly are diploid, and seven are tetraploid (Appendix 1). Nymphoides peltata represents the only species reported to be solely hexaploid, although both tetraploid and hexaploid counts have been obtained for N. geminata and N. montana. The anomalous N. lungtanensis (not included in the molecular analysis) is triploid (2n = 27) and does not produce fruit (Li et al. 2002). Conceivably, this species could represent a sterile hybrid between diploid and tetraploid species, but genetic data have not been evaluated to address this possibility or to identify putative parental taxa. If N. lungtanensis did arise by hybridization, it could be a contemporary example of a hybrid species origin and a system in which to examine post-hybridization effects on genome organization. With regard to other sources of phylogenetic incongruence, however, chromosome numbers are known for too few species to conclude which are derived from a single ancestral species (i.e. diploid, autopolyploid, or descended from a polyploid ancestor) and which may have originated through hybridization (i.e. allopolyploid).

Several anomalous Nymphoides species are polyploid, and hybrid origins (allopolyploidy) or spontaneous genome doublings (autopolyploidy) could have been factors in the evolution of their distinct morphologies. Nymphoides peltata is the most morphologically divergent species in the genus, having a unique inflorescence architecture, large, compressed fruits, and flattened seeds with a marginal ring of stiff projections. In our study, nuclear and plastid data each resolved N. peltata within a clade of other Nymphoides species, thus potentially indicating a hybrid origin, which also would be consistent with its hexaploid chromosome number. Another interesting chromosomal pattern occurs in N. aquatica and N. cordata, two tetraploid species native to temperate North America and the only dioecious species analyzed thus far using molecular data. Our phylogenetic data were unable to determine the closest relative of these sister species, but genetically similar species include the gynodioecious and diploid N. cristata (matK/trnK data) and the homostylous N. ezannoi (ITS data; Fig. 3). Interestingly, more extensive sampling of the largely sympatric N. aquatica and N. cordata has revealed the existence of interspecific hybrids having polymorphic ITS sequences (N. P. Tippery, unpublished data). This result not only confirms that contemporary Nymphoides species hybridize, it also presents a system in which to study patterns of morphological evolution and genome rearrangement in hybrid lineages.

Taxonomic Implications—Included in the molecular data analyses were two accessions, provisionally labeled ‘Nymphoides sp. 1’ and ‘Nymphoides sp. 2’ (Appendix 3), that differed from their nearest relatives by considerable branch lengths (Fig. 2). Both specimens were collected in Western Australia and resolved consistently within a clade that also contained N. furculifolia, N. parvifolia, and N. quadriloba (Figs. 2–3). The specimen of ‘Nymphoides sp. 1’ (Cowie 4390) was cited by Aston (2003) as an extreme example (due to tubercle ornamentation) of the “Kimberley” seed morphology of N. quadriloba. The specimens identified as N. quadriloba in this study also were collected in the Kimberley region of Western Australia, and although their seeds have a similar caruncular projection, they differ from ‘Nymphoides sp. 1’ in having a smooth surface. Molecular data were not obtained for specimens of N. quadriloba having the “typical” or “Carpentaria” seed morphologies (Aston 2003), and further research will be required to determine the extent of molecular and morphological variation in this species before any taxonomic recommendations can be made. Our rather preliminary results, however, suggest that N. quadriloba may need to be circumscribed more narrowly, with additional species potentially to be uncovered after further geographic sampling.

The second unidentified species, ‘Nymphoides sp. 2’ (Barrett & Barrett 2640), resembles N. planosperma in the morphology of its spongy leaves with open sinuses (cf. Aston 1982). The seeds, however, differ from any other Nymphoides by having a flattened teardrop shape and a mostly smooth surface that is roughened toward the broader funicular end. ‘Nymphoides sp. 2’ likely represents an undescribed species (H. I. Aston, pers. comm.), although formal description and naming is deferred pending further research.

Molecular data also were able to inform the issue of whether the neotropical N. humboldtiana represents a distinct species or whether it should be considered synonymous with the paleotropical N. indica, as some authors have suggested (e.g. Ornduff 1969). Although the two taxa allegedly are indistinguishable morphologically (Ornduff 1969), they differ at least chromosomally with the former being tetraploid and the latter diploid (Ornduff 1970; Appendix 1). Moreover, the species concept for the widespread N. indica is considerably broad, encompassing plants with diverse floral and seed morphologies that grow in Africa (N. indica subsp. occidentalis), Asia, and Australia (Raynal 1974b; Sivarajan and Joseph 1993; Li et al. 2002; Aston 2003). Specimens of N. indica analyzed for this study resolved to a genetically diverse clade (ITS data) or assemblage (matK/trnK data) that was not sister to N. humboldtiana (Fig. 2). Instead, specimens of N. humboldtiana resolved consistently as sister to N. fallax, another neotropical species. These results support the maintenance of N. humboldtiana and N. indica as distinct species, despite their extreme morphological similarity. Moreover, the genetic and morphological diversity of N. indica and its potentially paraphyletic relationship to N. ezannoi (Fig. 2) indicate that N. indica s. l. might more prudently be parsed among lineages that can be diagnosed using molecular and/or morphological evidence. Such resolution will depend on a dedicated study of specimens throughout the range of N. indica, including other species that might be more closely related to it than what we report here.

Heterostyly and Dioecy—Differences in sexual condition represent important evolutionary distinctions among Nymphoides species. A clear phylogenetic pattern of heterostyly, homostyly gynodioecy, and dioecy was not obtained through analysis of either ITS or matK/trnK data (Fig. 3). The three homostylous species in the grade toward N. minima (N. disperma, N. geminata, N. simulans) did not form a monophyletic group, indicating that each lineage evolved homostyly independently from a heterostylous progenitor. An exception to this pattern was observed for the homostylous species N. minima and members of its sister clade (e.g. N. furculifolia, N. parvifolia, N. rautanenii), whose phylogenetic relationships indicate that homostyly was maintained in their ancestral lineage, from which heterostyly re-evolved subsequently. Several authors have proposed a low likelihood for the independent acquisition of heterostyly, preferring instead to reconstruct multiple losses of the syndrome using weighted character transitions (e.g. Kohn et al. 1996; Schoen et al. 1997). Other studies, however, have reconstructed multiple origins for heterostyly in families where the condition is phylogenetically diffuse (e.g. Graham and Barrett 2004; McDill et al. 2009; Cohen 2011).

The ancestral state reconstructions presented herein for Nymphoides are subject to limited taxon sampling and phylogenetic uncertainty, and thus it would be premature do draw conclusions about evolutionary patterns. In addition, it should be noted that our characterizations of heterostyly in Nymphoides species were based on published reports, and except for several species in which crossing experiments have been conducted (N. geminata and N. montana, Haddadchi 2008; N. humboldtiana, Barrett 1980 [as N. indica]; N. indica, Shibayama and Kadono 2003; N. peltata, Wang et al. 2005), these likely reflected only the observation of reciprocal herkogamy and not necessarily the presence of associated self-incompatibility Several of the species that are reported to be homostylous have rather small flowers (< 1 cm; character 10, Appendix 1), and these could either have lost the style/filament polymorphism as their size reduced or simply be too small for researchers to have observed heterostyly. Furthermore, the apparent homostyly of species does not necessarily implicate the loss of self-incompatibility. Homostyly could correlate with smaller flowers and increased self-compatibility as Haddadchi (2008) reported for N. geminata, or these traits could be unrelated for other Nymphoides. Data currently are unavailable, however, regarding the presence or effectiveness of self-incompatibility in most Nymphoides species. More comprehensive reproductive system data will be required before we can properly understand the evolutionary history of homostyly and heterostyly in Nymphoides.

The ancestral condition of the dioecious species N. aquatica and N. cordata similarly is difficult to assess, because of the poor resolution of their related species. The ITS phylogeny placed the homostylous African species N. ezannoi as their closest relative, whereas the matK/trnK data ambiguously resolved five species or clades as the most closely related (Fig. 3). Consequently, the reconstruction of homostyly or heterostyly in the most immediate ancestor of the dioecious species requires more confident determination of phylogenetic relationships, including increased taxon sampling, to be conclusive. Phylogenetic analyses also resolved the gynodioecious N. cristata in an uncertain position relative to other species, although it also was reconstructed to have a homostylous ancestor in both analyses, and the matK/trnK data resolved it as a potential sister species to N. aquatica + N. cordata. Further phylogenetic resolution and taxon sampling will be required to determine whether N. cristata represents the sister lineage to dioecious species (including the two dioecious Indian species not yet sampled) and whether its ancestor was homostylous or heterostylous. Even after more confidently resolving the relationships among species with different sexual conditions, careful research (e.g. on self-incompatibility and developmental genetic factors) will be required to reconstruct the evolution of dioecy in Nymphoides.

Inflorescence Architecture—Parsimony reconstruction of ancestral states revealed that there were at least two transitions in Nymphoides from an expanded inflorescence morphology to a condensed type (Fig. 3), with four transitions (including to the unique N. peltata inflorescence type) mapped onto the more thoroughly resolved ITS tree. The matK/trnK tree indicated two reversions from condensed back to expanded inflorescence morphology, but this reconstruction involved poorly supported nodes (nodes V–X, Fig. 3). The repeated evolution of condensed inflorescence morphology and its prevalence in the genus indicates that it may have a strong selective advantage, at least under certain conditions, and thus may have contributed to the diversification of some Nymphoides groups, particularly the clade of condensed-inflorescence species comprising N. minima and its sister lineage. Notably, all extra-Australian species with condensed inflorescences in our study resolved within the sister clade of N. minima, possibly implicating a single major ancestral dispersal event out of Australia for condensed-inflorescence Nymphoides species.

In the basal grade of species leading to the N. minima branch, all species are exclusively Australian except for the Australian and tropical Asian species N. aurantiaca, its partially sympatic sister species N. cambodiana, and the wide-ranging Eurasian species N. peltata (Fig. 3). Species in the clade sister to N. minima are predominantly extra-Australian, except N. indica and the consistently resolved clade of N. furculifolia + N. parvifolia + N. quadriloba + ‘Nymphoides sp. 1’ + ‘Nymphoides sp. 2.’ Apart from the major clade of condensed-inflorescence species discussed above, the other two evolutionary origins of condensed-inflorescence morphology (as reconstructed on the ITS phylogeny; Fig. 3) involve exclusively Australian species. These species also differ morphologically from most species in the major condensed-inflorescence clade. Species in one clade (N. elliptica + N. triangularis) have glabrous, mauvewhite petals with lateral wings and relatively long petioles on the leaves subtending their flower clusters (Appendix 1; Aston 1984). The other clade (N. planosperma + N. simulans + N. spongiosa) contains species with relatively small proportions and spongy leaves (Appendix 1; Aston 1982, 2002). Notably, in contrast to the overall pattern for Nymphoides, the latter clade resolved congruently on ITS and matK/trnK trees (Fig. 3). Future research should be directed toward understanding the multiple independent origins of condensed inflorescence morphology and determining, for example, whether the developmental mechanisms that produce this morphology are the same in every lineage.

The elucidation of phylogenetic relationships in Nymphoides clearly requires further work to examine the incongruence between nuclear and plastid sequence data, which hampers the reconstruction of ancestral character states and evolutionary patterns. Moreover, putative hybrid species origins must be verified using chromosomal data and potentially additional genetic markers that can provide further supporting evidence. Nonetheless, several patterns have emerged from the current study, most notably that inflorescence architecture evolved at least two times from an ancestrally expanded morphology into a derived, condensed morphology. Reticulate inheritance patterns throughout the evolutionary history of Nymphoides may have provided genetic variation that facilitated the morphological diversity and extensive geographic range of the genus today.