Aliso: A Journal of Systematic and Floristic Botany Aliso: A Journal of Systematic and Floristic Botany

The diverse, aquatic Hydrocharitaceae have defied stable classification for nearly two centuries. Anatomical and morphological convergence characterize many aquatic plants and undoubtedly have hindered the ability of researchers to ascertain accurately those features representing reliable phylo genetic markers within Hydrocharitaceae. Most prior classifications of the family have emphasized few characters to define major taxonomic subdivisions (i.e., they were highly artificial). Previous studies using molecular data have shown that DNA sequences provide novel indications of phylogeny not indicated previously by morphologically based classifications; however, they have not yet rec ommended alterations to the classification for the family. We conducted a more comprehensive phy logenetic study of Hydrocharitaceae to better elucidate evolutionary relationships among the genera that in turn could be used to provide insight for improvements in classification. We analyzed different data sets (55 morphological characters; chloroplast rbcL, matK, trnK intron sequences; nuclear ribo somal ITS region sequences) singly and in various combinations using maximum parsimony and maximum likelihood methods of phylogenetic reconstruction. Phylogenetic analysis of combined data yielded a fully resolved tree depicting four well-supported, major clades within Hydrocharitaceae. We use these results to propose a phylogenetic classification of Hydrocharitaceae recognizing four sub families that correspond to these clades: Anacharidoideae, Hydrilloideae, Hydrocharitoideae, and Stra tioideae. Phylogenetic analysis also indicated the pattern of derivation with respect to submersed life forms, hydrophilous pollination, and marine habitation in the family. Character reconstructions indi cated that several features, (e.g., ovule type; occurrence of detaching male flowers), once thought to provide strong phylogenetic markers in Hydrocharitaceae, actually are highly homoplasious and have acutely mislead past attempts at classification of the family.


INTRODUCTION
Hydrocharitaceae Juss.("hydrocharits") are aquatic monocotyledons currently circumscribed as comprising 17 genera and approximately 75 species (Cook 1996).Despite their relatively small size, Hydrocharitaceae exhibit some of the greatest diversity of any aquatic angiosperm family, including freshwater and marine species; annual and perennial life histories; amphibious, free-floating, and submersed lifeforms; broad to narrowly linear leaves in rosettes or caulescent arrangements; showy to highly reduced flowers; wind, insect, and water pollination, and male flowers that detach and float on the water surface as some examples.The latter character is unique to this family among angiosperms.
As in many other aquatic plants, the combination of morphological reduction and convergent aquatic adaptation has made it difficult to establish phylogenetic relationships with certainty, especially using characters derived from comparative morphology and anatomy (Les and Haynes 1995).The inability of traditional morphological characters to provide data adequate for resolving credible and consistent infrafamilial relationships in Hydrocharitaceae is particularly evident in past studies that have yielded extremely volatile classifi-cations.Notably, various authors have proposed to subdivide this small family even further into as many as nine different families including Blyxaceae (Aschers.& Giirke) Nakai, Elodeaceae Dumort., Enhalaceae Nakai, Halophilaceae J. Agardh, Hydrillaceae Prantl, Otteliaceae Chatin, Stratiotaceae Link, Thalassiaceae Nakai, and Vallisneriaceae Link (Dumortier 1829;Link and Willdenow 1829;Chatin 1855e;Agardh 1858;Prantl 1880;Nakai 1943Nakai , 1949) ) in a mosaic of circumscriptions.There also has been little agreement with respect to the grouping of subordinate genera among different classifications proposed (Table 1).Other anomalous Hydrocharitaceae classifications are compared in Tomlinson (1982), , and Tanaka et al. (1997).Despite these many previous classifications, none has yet been based on a phylogenetic analysis, thus they provide only weak hypotheses of relationships.
Another complication concerns the correct placement of Najas (Najadaceae Juss.) that has been shown to possess a close affinity to Hydrocharitaceae (Shaffer-Fehre 1991a, b; Les et al. 1993Les et al. , 1997;;Les and Haynes 1995;Tanaka 1997), but until recently had been assumed to be related quite distantly to the family (e.g., Dahlgren et al. 1985).Prior molecular studies incorporating rbcL and matK data (Les et al. eral invasive and notoriously weedy species (e.g., Egeria densa, Hydrilla verticillata) whose phyletic relationships may provide information useful for studying the evolution of invasive characteristics in aquatic plants.
The fundamental importance of securing accurate phylogenetic information as a basis for further study cannot be overstated.Systematically, the diverse Hydrocharitaceae provide an excellent model system for demonstrating the evolutionary transition from terrestrial to aquatic habitats in flowering plants.Sculthorpe (1967) specifically identified three "biological trends" in the family: hermaphroditism to unisexuality and dioecy; entomophily and anemophily to hydrophily; and freshwater to marine habitation.Yet, to achieve a valid interpretation of these trends, the phylogenetic relationships within the family must be resolved thoroughly and confidently.Hydrocharitaceae also contain sev- The objective of this study is to provide a more comprehensive phylogenetic analysis of intergeneric relationships within Hydrocharitaceae that is based on information compiled from various sources including morphological and molecular (both chloroplast DNA and nuclear DNA) data.We use these resulting indications of phylogeny to explore some of the evolutionary trends within the family and to propose modifications to the subfamilial classification of this unusual group of aquatic angiosperms that long has defied a satisfactory systematic treatment.
Table 2. Characters and states used in morphological phylogenetic analysis of Hydrocharitaceae genera (see text for references).

DNA Sequence Data
Chloroplast DNA (cpDNA) data.-Wecompiled cpDNA sequences for 18 genera including Butomus (used as the outgroup), Najas, and 16 of the 17 genera recognized within Hydrocharitaceae s.s.The cpDNA data consisted of 1183 base pairs (bp) of rbcL data with 17 of the sequences retrieved from GenBank as deposited by Les et al. ( 1997) and one sequence newly generated for Maidenia rubra.The rbcL sequence for M. rubra was produced using the same methods as described in Les et al. (2002a).Our matK data set consisted of 1582 bp with 12 sequences retrieved from GenBank as deposited by Tanaka et al. ( 1997) and sequences newly generated for Apalanthe granatensis, Butomus umbellatus, Lagarosiphon major (Ridl.)Moss, Maidenia rubra, and Vallisneria americana that we obtained using the methods described in Les et al. (2002b).An unusually long insertion/deletion (indel) and high variability near the 3 '-end of matK impaired sequencing of the complete matK region, resulting in shorter sequences in the latter two genera.As a result, our data set for Vallisneria americana and Maidenia rubra contained only 616 bp of matK and Nechamandra lacked matK data entirely.Following the methods reported in Les et a!. ( 1997) we obtained trnK intron data for 13 genera consisting of 1546 aligned sites for the 5' region and 207 aligned sites for the 3' region.The 5' intron sequence for Nechamandra alternifolia was somewhat shorter at 1335 bp.We were unable to obtain 3' trnK intron region sequences for Halophila stipulacea (Forssk.)Acherson, Maidenia rubra, Nechamandra alternifolia, Najas marina, and Vallisneria americana.
Nuclear ribosomal DNA (nrDNA).-Weobtained DNA sequences of the nriTS region (ITS-1, ITS-2, 5.8S) for all 18 genera included in the study following the methods described in Les et a!. (2002b).All new cpDNA and nriTS sequences have been deposited in GenBank (Table 4).

Data Analyses
Phylogenetic analyses of morphological data were carried out using maximum parsimony as implemented by the program PAUP* vers.4.0 beta 4a (Swofford 1998).Character states were treated as unordered and heuristic searches were conducted using a simple addition sequence referenced to Butomus which retained one tree at each step, and tree-bisection-reconnection (TBR) branch swapping restricted to best trees using stepwise addition for starting trees while saving multiple trees (MULPARS option).Trees were rooted using Butomus as the outgroup following the rbcL analysis of Les et a!. (1997).Strict consensus was used to depict results yielding multiple, equally parsimonious trees.Internal support was determined from 1 000 bootstrap replicates using a "full" heuristic search with search options as described above.
Indels were treated as missing data.For matK, we scored the presence or absence of distinctive indel motifs to produce a separate binary character set (gaps in other regions were too variable to score confidently).The resulting indel matrix for matK consisted of 23 characters and was included only in maximum parsimony analyses (see below).Data from all cpDNA regions were combined and analyzed as a block partitioned separately from the morphological and nrDNA data.
In some cases (Blyxa, Halophila, Hydrocharis, Limnobium, Najas, Thalassia) we combined different molecular loci sequenced from different but congeneric species under the same genus for analysis.Molecular data were analyzed using maximum parsimony (MP) and maximum likelihood (ML) methods.The MP analyses of cpDNA and nrDNA essentially followed the same approach used to analyze the morphological data (see above).Results yielding equally minimal length trees were depicted as strict consensus trees with internal nodal support calculated from 1000 replicates as described above.Sequences from Butomus were used to root the trees.
The cpDNA data matrix for ML analyses used the same data set included in the parsimony analyses except for the removal of large indel regions (> 10 base pairs) in the matK and trnK regions.After aligning the complete set of ITS sequences, it was apparent that many regions were too divergent to reasonably ensure the maintenance of nucleotide site homologies.We identified those regions where excessive divergence occurred and removed questionably aligned regions prior to our analyses of the nriTS data.Although preliminary analyses indicated that removal of the highly divergent regions did not influence the resulting tree topologies to any great extent, we analyzed only the nriTS data from the less divergent regions.
For each molecular data set (cpDNA and nrDNA), we employed an iterative search strategy (Swofford eta!. 1996;Sullivan et a!. 1997) to evaluate different models of molecular evolution using likelihood ratio tests.The data sets were examined initially under 24 models of substitution to deter-  Model parameters were optimized on sets of the ten most parsimonious trees found in searches using equally weighted parsimony for the nriTS and cpDNA data sets, respectively.ML searches (heuristic searches with TBR branch-swapping) subsequently were conducted using the best fit substitution model (GTR + I + f for both data sets) fully defined.Model parameters were as follows: nriTS data (AC-0.90678,AG-1.8306,AT-1.2144,CG-0.8699,CT-4.8948,I = 0.1368, a = 0.5080); cpDNA data (AC-1.8934,AG-2.1376,AT-0.2620,CG-0.9488,CT-2.5945,I = 0.2426, a = 0.9424).Nodal support was estimated by bootstrap analyses (200 replicates, heuristic searches and TBR branch-swapping) under the best-fit model.
Data set congruency was estimated by implementing the ILD/partition-homogeneity test of PAUP* (Swofford 1998).The ILD test was used in the six pairwise comparisons of nriTS, matK, rbcL, and trnK intron data sets as well as in an overall comparison of cpDNA and nriTS data.In addition, the morphological data set was compared to each molecular data set.Considering the precautions stated by Yoder et al. (200 1) regarding the ILD test, we selected a value of P < 0.005 as the level where potential data incongruence might be indicated.
Observing no major incongruence among trees generated from cpDNA or nrDNA (see Results), we combined the molecular data for analysis.Because the pattern of molecular evolution for these data sets differed conspicuously as demonstrated by their distinct model parameters (see above), we restricted the analysis of the combined molecular data to maximum parsimony methods.Where both approaches were used on individual data sets, we observed that ML and MP analyses produced very similar results.
Although the topology resulting from analysis of morphological data alone differed somewhat from the molecular analyses (see Results), it varied only at nodes that were sup- ported poorly.Thus, the overall agreement between the morphological and molecular trees warranted their combined analysis.We first conducted separate combined analyses of the morphological data with either the cpDNA or nrDNA sequences, using the MP methods described above.With identical topologies resulting from those analyses, we combined the morphological data with all molecular data and analyzed the combined data set using the MP methods described above.Bootstrap support was obtained in all instances from 1000 replicates and a full heuristic search.
Because of its unique occurrence within Hydrocharitaceae, we mapped the character of detached male flowers on the combined data phylogeny using both ACCTRAN and DELTRAN optimizations to determine the number of origins for the trait within the family.We used the same optimizations to explore the number of origins for orthotropous ovules, given their importance in early classifications of the family.Codings for pollination types and life-forms were provided for the terminal taxa on the final combined data tree.

Morphological Analyses
Character state distributions for 55 characters scored for 19 study genera and analyzed using maximum parsimony produced four equally minimal length trees of 164 steps with a consistency index (CI) = 0.482, a consistency index excluding uninformative characters (Ciexc) = 0.465, and retention index (RI) = 0.587 (Fig. 1).Internal bootstrap support exceeded 50% for only a few nodes, notably for clades consisting of Hydrocharis and Limnobium (100%), Vallisneria and Maidenia (71 %), Vallisneria, Maidenia, and Nechamandra (62%), and genera with a submersed life-form (63%).Lower internal support ( 41%) was obtained for clades consisting of the marine genera (Enhalus, Halophila, Thalassia), for Apalanthe, Egeria, and Elodea (34% ), and for genera with emergent or floating-leaved life-forms (34% ).Najas resolved in a position somewhat distant from Hydrilla in a clade comprising the marine genera and Vallisneria clade; however, it was placed in this part of the tree with poor internal support (12%).
Although material of the rare Appertiella hexandra C. D. K. Cook & Triest was unavailable for molecular analysis, morphological data placed the genus within a clade comprising the marine genera, Vallisneria clade, Najas, and Lagarosiphon.Placement within this clade was not supported strongly, and its relationship to other members of the clade was otherwise unresolved.
Both likelihood and parsimony methods recovered a clade of the marine genera (Enhalus, Halophila, Thalassia) and a monophyletic group comprising Maidenia, Nechamandra, and Vallisneria with strong support (100%).Hydrilla and Najas resolved as a clade (MP) or as a basal grade (ML) within a larger clade containing the groups mentioned previously (Fig. 3).The association of Najas within this hydrocharit clade was supported strongly (96-98%) regardless of the method of analysis.Other clades supported strongly by both methods of analysis consisted of Hydrocharis and Limnobium (100%), Apalanthe, Egeria, and Elodea (100%), Blyxa and Ottelia ( l 00% ), and the placement of Lagarosiphon sister to the two previously mentioned clades (100%).The precise pos1t10n of Stratiotes and interrelationships of the larger clades remained unresolved or weakly supported (MP = 45%) in the cpDNA analyses (Fig. 3).
Nuclear ribosomal DNA (nrDNA).-ThenriTS data produced an alignment of 947 sites for the 18 genera analyzed.
After removing from the analysis those regions of questionable alignment (see above), the resulting data set consisted of 414 sites.Parsimony analysis of this reduced data set generated a single minimal length tree of 573 steps with a Cl = 0.716, Clexc = 0.602, and RI = 0.577.Maximum likelihood analysis employing a GTR + I + r model generated one optimal tree having a lnL score of -2552.7346(Fig. 4).
Both ML and MP analyses of the ITS data produced similar topologies and comparable levels of internal support (Fig. 4).Each analysis resolved the marine genera (Enhalus, Halophila, Thalassia) as a relatively well-supported clade (72-86% ).A clade consisting of Apalanthe, Egeria, and Elodea also was well supported (94-99%) and positioned sister to a relatively well-supported clade of Blyxa and Ottelia (79-83%) and in succession, sister to Lagarosiphon (68-70% ).Clades consisting of Maidenia, Nechamandra, and Vallisneria, and of Hydrocharis and Limnobium also were strongly supported (100%) in both analyses.Maximum parsimony reversed the position of Maidenia and Nechamandra as resolved by ML analysis and cpDNA analyses; however, that minor discrepancy was supported only moderately (38% ).The positions of Halophila and Thalassia were reversed by the ML analysis compared to the parsimony analysis that agreed with the cpDNA analyses; again, this relatively minor inconsistency had only moderate support (58%).Both analyses placed Najas near Hydrilla and the Vallisneria clade (Maidenia, Nechamandra, Vallisneria), but somewhat distant (however, with weak support) from the clade of marine genera (Enhalus, Halophila, Thalassia).Each analysis resolved Hydrocharis and Limnobium as the basal clade in the family.Combined molecular data.-TheILD test indicated no significant incongruency among the molecular data partitions when using P < 0.005 as a threshold value.The following comparisons: nriTS-matK, nriTS-trnK, rbcL-trnK, trnK-matK yielded P-values in the range of 0.288-0.721.Although rbcL data were less congruent with nriTS (P = 0.028) and matK (P = 0.010), the different data sets produced very similar (even identical) topologies and the overall agreement between the cpDNA and nriTS data was high (P = 0.461).Most discrepancy among molecular data sets presumably was due to the inclusion of the divergent genus Najas.Excluding Najas from the ILD tests raised all P-values above 0.116.Yet, the resolution of Najas embedded well within the family (a major question of this study) was indicated in every data set analyzed.From these results, we concluded that the combined analysis of our molecular data was warranted.Maximum parsimony analysis of the combined 4955 molecular characters produced three equally minimal length trees of 2984 steps with a Cl = 0.779, Clexc = 0.664, and RI = 0.690.The resulting topology of the strict consensus tree (Fig. 5) retained the same basic groups as those resolved by the previous molecular analyses (Fig. 2-4).Most resolved nodes had strong support (11 nodes >98%).Support for the association of Hydrilla and Najas was elevated to 98% and the inclusion of that clade with the marine genera and Vallisneria clade received comparable support (98%).
Combined molecular and morphological data.-TheILD test showed significant incongruency between the morphological data set and every molecular data partition (P = 0.001 in all instances).However, the inclusion of morphological data only slightly altered the tree topologies produced using each single molecular data set (either cpDNA or nriTS data), and in each case, the same topology was produced in the combined analysis (see below).Furthermore, the addition of the morphological data to the combined molecular data resulted only in the resolution of two additional nodes and did not otherwise alter the tree topology produced by the analysis of combined molecular data.Because of these results, and the understanding that even a P-value of 0.001 does not necessarily preclude the combinability of data (Yoder et al. 2001), we concluded that the combination of our morphological and molecular data was warranted.
Parsimony analysis of combined morphological and cpDNA data (total of 4596 characters) yielded a single minimal length tree of 2242 steps with a CI = 0.737, Clexc = 0.637, and RI = 0.696 (Fig. 6).Parsimony analysis of combined morphological and nriTS data (total of 469 characters) yielded a single minimal length tree of 627 steps with a CI = 0.584, Clexc = 0.514, and RI = 0.550 (Fig. 6).The combination of morphological data either with cpDNA data or nrDNA data produced an identical topology that differed only by the degree of internal support provided by each different molecular data set (Fig. 6).Parsimony analysis of all data combined (total of 5010 characters) yielded a single minimal length tree (identical in topology to those shown in Fig. 6) of 2695 steps with a CI = 0.720, Clexc = 0.621, and RI = 0.675 (Fig. 7).Overall, the combined molecular/morphological data showed results similar to previous analyses; however, these trees were completely resolved with moderate support (72%) for the position of Stratiotes between the Hydrocharis-Limnobium clade and remaining genera, and low support (56%) associating the Lagarosiphon-Egeria clade with the clade containing Hydrilla, Najas, Vallisneria and the marine genera.Eleven of the 15 nodes (73%) were supported with bootstrap values exceeding 96%.2, 3).Numbers above branches represent internal support of the tree as provided by bootstrap analysis (1000 replicates).
(ACCTRAN/DELTRAN = 4 gains) in the family.If the occurrence within some species of Elodea (which is polymorphic for the latter trait) also is considered, then five independent origins would be indicated for the characteristic within the family (Fig. 7).
Emergent and floating life-forms occurred in the outgroup (Butomus) and basal genera (Hydrocharis, Limnobium, Stratiotes) when mapped on the combined data cladogram, with submersed life-forms representing a conspicuously derived condition in Hydrocharitaceae (Fig. 7).When pollination types (as classified by Cook 1982) were mapped on the same cladogram, entomophily was plesiomorphic with hydrophily being derived and homoplasious (Fig. 7).

Morphological Studies and Inferences
Botanists long have strived to produce a classification of Hydrocharitaceae that would arrange the highly diverse genera into natural subordinate groups.One of the earliest taxonomic studies of Hydrocharitaceae was made by Richard ( 1811) who divided the family into two groups according to stigma number (3 or 6) with further subdivision based on conspicuous vegetative characters (Table 1).However, Rich-ard's highly artificial scheme produced unlikely associations such as the grouping of the marine Enhalus with the freshwater Stratiotes, and the association of the highly specialized Vallisneria with the relatively unspecialized Blyxa.Endlicher (1836-40) modified Richard's treatment by subdividing his first group (which he recognized as tribe Vallisnerieae Dumort.) to yield yet another tribe Anacharideae Endl.
The French botanist Adolf Chatin evaluated anatomical evidence to clarify taxonomic relationships within Hydrocharitaceae (Chatin 1855a-g).Chatin's survey of ovule types and other anatomical features persuaded him to follow Richard's subdivision of the family, and he recognized two tribes: the "true" Hydrochariteae (comparable to Richard's second group) and Vallisnerieae (which reincorporated Endlicher's Anacharideae), defined primarily by the presence of vascular elements in the former and their absence in the latter (Chatin 1855c, d).By that circumscription, tribe Hydrochariteae included Enhalus, Hydrocharis, Limnobium, Ottelia, and Stratiotes; whereas, tribe Vallisnerieae contained Apalanthe, Elodea, Hydrilla, and Vallisneria (Chatin 1855c, d).
Because Chatin determined previously that Vallisneria possessed orthotropous ovules (Chatin 1855e), he concluded that Ottelia, with anatropous ovules, was worthy of transfer to a distinct tribe, a distinct family, or possibly even to some closely related family (Chatin 1855d).Subsequently, a more thorough survey convinced Chatin to recognize Otteliaceae as a distinct family (Chatin 1855e) to accommodate those genera with anatropous ovules (Enhalus, Ottelia, Stratiotes).This realignment required Chatin to refine his earlier classification, which was based on the presence or absence of vascular elements.He used vasculature differences and integument number (which he perceived to differ in the family) to define tribes within each of the newly circumscribed families (Chatin 1855e).
Chatin's classification was criticized by Caspary (1857a, b) who challenged the accuracy of his anatomical observations and corrected several misconceptions regarding ovule anatomy such as Parlatore's (1855) misinterpretation of Stratiotes as having orthotropous ovules and Chatin's misconceptions concerning integument numbers.Caspary argued convincingly that various anatomical discrepancies did not warrant the acceptance of Chatin's proposed classification.He published a revised treatment of the tribe Hydrilleae that included Elodea, Hydrilla, and Lagarosiphon by virtue of their similarly reduced anatomy (Caspary 1858).This tribe (often recognized as Anacharideae) of anatomically reduced, submersed Hydrocharitaceae has been retained by contem-porary authors such as Tomlinson (1982) who tentatively also included Nechamandra; noting, however, that instead it may belong with Vallisneria.
Ovule morphology remains misunderstood in Hydrocharitaceae and deserves a more thorough examination.Cronquist (1981) and Dahlgren et al. (1985) remarked that ovules of Hydrocharitaceae usually were anatropous, and Schmidt-Mumm (1996) simply described the family as having anatropous ovules.However, our survey found that orthotropous ovules actually predominate, occurring in 9 of the 17 hydrocharit genera where ovule type has been reported (it remains unknown in Appertiella; Tables 2, 3).
From these examples, it is apparent that much of the previous taxonomic history of Hydrocharitaceae has been influenced by highly artificial approaches where modifications to classification have been made on the basis of the distributions of a small number of characters as well as misinformation regarding their character states.Even fairly recent studies (e.g., Shaffer-Fehre 1991a, b) have attempted to define infrafamilial taxa using relatively few characters.It is understandable that this approach has not yet produced a stable, more natural classification.
Over the years, many morphological studies have provid- ed further insight into structural similarities and differences among hydrocharit genera.However, the extensive level of morphological and anatomical diversity found within Hydrocharitaceae does not readily disclose a conspicuous pattern of relationships within the family.Furthermore, most aquatic taxa are well known for reductions in vegetative and floral structures leading to convergence in form (Sculthorpe 1967) that generally makes it difficult to interpret morphological characters for phylogenetic analysis.The extent to which parallel reduction and convergent adaptations have influenced interpretations of character state homologies in Hydrocharitaceae has not yet been evaluated adequately.Only recently, some progress in clarifying this issue has been made using more empirical approaches to evaluate distributions of larger numbers of morphological characters in the family.
Several years ago, Dr. C. D. K. Cook (Botanic Gardens, Zurich) sent to us his unpublished results of various morphological cluster analyses that he had used to analyze phe-netic relationships among Hydrocharitaceae genera (C.D. K. Cook pers.comm.).Although some of those analyses produced several consistent clusters (e.g., Hydrocharis and Limnobium; Egeria and Elodea; Maidenia and Vallisneria) other associations of genera varied among the analyses or were too diffuse to provide much insight.Although this approach was not phylogenetic, it was the first analytical attempt to obtain a more natural perspective of relationships in the family that was based on more than just a small subset of characters.
Our cladistic analysis of 55 morphological and anatomical characters for 18 hydrocharit genera represents the first comprehensive phylogenetic analysis of morphological data conducted at the generic level for Hydrocharitaceae.Parsimony analysis (Fig. 1) indicated that some of the same phenetic groupings obtained by Cook also were resolved cladistically.We obtained high bootstrap support (100%) for a clade containing Hydrocharis and Limnobium, and moderate support (66%) for a clade uniting Vallisneria with Maidenia.The Left: Best cladogram recovered (lnL = -2552.7346)using maximum likelihood analysis (GTR + I + r model).Numbers above branches represent bootstrap values obtained from 200 replicates.Right: single, minimal length tree (573 steps) recovered from maximum parsimony analysis.Numbers above branches represent bootstrap values obtained from I 000 replicates.Both methods of analysis recovered similar clades with comparable levels of internal (bootstrap) support.Minor discrepancies occurred in the positions of Halophila (likelihood) and Vallisneria (parsimony), deviations that were not supported strongly.latter result is particularly informative given that other work (Les et al. unpubl. data) shows clearly that Maidenia is nested within Vallisneria; yet strong cladistic support for these genera still is not forthcoming from these morphological data alone.
The vegetatively similar Hydrocharis and Limnobium are believed to represent the most primitive elements within Hydrocharitaceae (Kaul 1969(Kaul , 1970;;Wilder 1975;Cook and Urmi-Konig 1983b) and resolve as a sister group to theremainder of the family (Fig. 1).Richard (1811) remarked on the difficulty of distinguishing these genera in absence of their fairly distinctive flowers.Shaffer-Fehre ( 1991a, b) found the testa structure of Hydrocharis dubia to be so similar to Limnobium that she merged the species with that genus.Cook and Urmi-Konig (1983b) also found it difficult to separate Hydrocharis and Limnobium vegetatively stating that the "affinities of Limnobium sensu lato are clearly with the genus Hydrocharis."The high degree of internal support for the clade containing Hydrocharis and Limnobium (Fig. 1) indicates that these genera are indeed closely related using morphological criteria.
As in Cook's phenetic analyses, our morphological phylogenetic analysis resolved Egeria and Elodea as sister genera; we also resolved Apalanthe in a position adjacent to that clade (Fig. 1).Although weakly supported (35-47% bootstrap), the association of these three similar genera is not surprising given their taxonomic history.St. John (1962, 1965) reduced Apalanthe to a subgenus of Elodea; whereas, Hauman-Merck (1912, 1915) merged both Apalanthe and Egeria with Elodea on the basis of their pollination system.Cook (1985) remarked that Apalanthe shared many features in common with Egeria or Elodea and was "patristically related" to Egeria.Cook and Urmi-Konig (1985) concluded that Elodea probably was related most closely to Apalanthe.Despite their similarities, the distinctness of all three genera is evidenced by a number of features unique to Apalanthe and the fact that Elodea is unable to hybridize with Egeria (Cook and Urmi-Konig 1985).A close phylogenetic relationship between Maidenia and Vallisneria always has been quite evident morphologically.Aston (1973) viewed the two genera to be very closely related because of their similar floral morphology.Hutchinson (1959) merged Maidenia with Vallisneria, and Cook (pers. comm.)essentially regards Maidenia to be a modified Vallisneria.Shaffer-Fehre (1991a) also placed Maidenia with Vallisneria on the basis of similarities in their seed coat structure.However, Tomlinson (1982) suggested that the vegetative morphology of Maidenia might indicate its relationship to Anacharideae.As mentioned above, this possibility is untenable given the results of a recent molecular study of Vallisneria indicating that Maidenia is embedded within the genus phylogenetically (Les et al. unpubl. data).
Our morphological analysis shows weak support for an alliance of Nechamandra with Vallisneria and Maidenia (Fig. 1).Yet, this result is not surprising given that Nechamandra was once included in Vallisneria (

Butomus (OUTGROUP)
Fig. 6.-Cladogram resulting from maximum parsimony (MP) analysis of combined morphological and molecular data.The same tree was recovered whether morphological data were combined either with cpDNA data (I tree @ 2242 steps), or with nriTS data (I tree @ 627 steps).Numbers above branches represent bootstrap values obtained from 1000 replicates.Levels of bootstrap support provided by combined morphology and cpDNA data are shown by the upper numbers above each branch; those from combined morphology and nriTS data are shown by the lower numbers.Triest 1983).Cook also has suggested that Nechamandra is related closely to Vallisneria (Tomlinson 1982) but b) concluded that the seed coat structure of Nechamandra placed it closer to Blyxa, Najas, and Stratiotes.However, our analysis (Fig. 1) indicates that these four genera show no close association whatsoever based on phylogenetic inferences using parsimony analysis of morphological characters that include seed coat features.
Among other Hydrocharitaceae, the endemic Madagascar genus Appertiella has been placed closest phylogenetically to Lagarosiphon (Cook and Triest 1982).Although not definitive, our morphological cladogram (Fig. 1) resolved both genera in a topology consistent with that interpretation.
A major difference between our phylogenetic analyses and Cook's earlier phenetic results concerns the interrelationships of the marine genera Enhalus, Halophila, and Thalassia.Cook's phenetic analyses consistently split these three genera into 2-3 different groups, but never grouped them together, a result reminiscent of taxonomic schemes (e.g., Hartog 1970) that segregated these genera among different subfamilies of Hydrocharitaceae.Conversely, our phylogenetic analysis resolved the marine Hydrocharitaceae as a clade, albeit with weak support (Fig. 1).Tomlinson's (1982) remark that Thalassia was "much more specialized" than most Hydrocharitaceae is supported by our morphological cladistic analysis (Fig. 1) that shows the genus to occupy a position quite derived in the family.
Few specific relationships have been postulated between and among other hydrocharit genera.Triest (1982) suggested that Lagarosiphon was closely related to Elodea and Hydrilla, a result inconsistent with relationships depicted in the morphological cladogram (Fig. 1) as well as results from analysis of combined data (Fig. 7).Shaffer-Fehre (1991a) noted a similarity in the seed coat anatomy of Blyxa and Lagarosiphon, taxa that also did not resolve closely in our morphological analysis.Aston ( 1973) remarked that Blyxa resembled juvenile plants of Ottelia ovalifolia (R. Br.) Rich.
where their degree of similarity could lead to confused identifications.Although Blyxa and Ottelia did not associate as a clade in the morphological cladogram (Fig. 1), they were not far removed in that analysis.Most nodes were not supported strongly in the morphological cladogram, yet our phylogenetic analysis did indicate that hydrocharit taxa having large, showy flowers and more complex anatomy and morphology (e.g., Hydrocharis, Ottelia, Stratiotes) tended to occur more basally than did those taxa with highly reduced floral and vegetative morphologies.All genera represented by submersed life-forms occurred more distally in the tree than did those having either emergent or floating-leaved life-forms (Fig. I).The marine genera (Enhalus, Halophila, Thalassia) resolved in a relatively specialized phylogenetic position with respect to most of the freshwater genera (Fig. 1).These results support the common perception that evolution in Hydrocharitaceae has proceeded generally via transitions from emergent to submersed life-forms and from freshwater to marine taxa.
The position of Najas.-Believing that morphological homologies generally were poorly understood in Alismatidae, Sculthorpe ( 1967) regarded any discussion on relationships of Najas as "phylogenetic speculation."However, he did conclude that the genus was not primitively simple as earlier authors had believed.Although Najadaceae long have been regarded as allied phylogenetically to Potamogetonaceae, some morphological data have indicated a possible association with Hydrocharitaceae.Rendle (1901) and Singh (1965) interpreted the outer floral "envelope" of Najas as similar to the spathe found in Hydrocharitaceae.Although Rendle (1901) believed that Najas was most closely allied to Zannichellia L. (Zannichelliaceae ), he also remarked on the similarity of the genus to Elodea, Hydrilla, and Lagarosiphon of Hydrocharitaceae.Miki (1937) suggested that "an intimate affinity" existed between Najas and Hydrocharitaceae based upon his evaluation of various morphological characters.He believed that Najas was derived from Hydrocharitaceae and was remote phylogenetically from Potamogetonaceae.Wilder (1975) remarked that unlike most Alismatidae, Najas and Hydrocharitaceae similarly lack the ability to produce nonprecocious buds.Compelling morphological evidence of a close relationship between Najas and Hydrocharitaceae was provided by Shaffer-Fehre (1991a, b) who discovered unique seed coat features that linked together these taxa.Les et al. ( 1 993) and Les and Haynes (1995) showed that Najas and Hydrocharitaceae could be resolved as a sister group by phylogenetic analysis of cpDNA data as well as morphological data.
Our morphological analysis supports the inclusion of Najas within Hydrocharitaceae (Fig. 1 ), with a moderate degree of internal support (64% ).Parsimony analysis embedded Najas rather deeply within the family, close to the marine genera.Although not definitive, this result shows overall that Najas is not so distinct morphologically as to preclude its placement within Hydrocharitaceae.However, because Najas is modified for hydrophilous pollination and is highly reduced otherwise, it is difficult to ascertain the influence of convergent character states attributable to morphological reduction.

Molecular Studies
Chloroplast DNA ( cpDNA).-Hydrocharitaceaehave been fairly well studied taxonomically, yet few efforts focused specifically on the elucidation of phylogenetic relationships within the family until the relatively recent advent of molecular data.Les et al. (1993) and Les and Haynes (1995) evaluated higher level relationships within subclass Alismatidae, using cladistic analysis of morphological and cpDNA sequence data, which indicated a close relationship between Najadaceae and Hydrocharitaceae (as noted above).Soon afterward, two studies appeared using cpDNA sequences that evaluated intergeneric relationships in Hydrocharitaceae.Tanaka et al. (1997) examined relationships among Najas and 13 hydrocharit genera using rbcL and matK sequence data analyzed by neighbor-joining methods.Les et al. (1997) included two Najas species and 20 species from 15 hydrocharit genera in a survey of relationships in subclass Alismatidae using a weighted parsimony analysis of rbcL data.The results of these analyses produced topologies that differed in several details including the precise placement of Najas within the family (Fig. 2).However, the topologies generated by the two studies were extremely similar overall considering that different outgroups, taxon sampling, data sets, and analytical methods were used.Most nodes were supported quite well in both trees indicating that DNA sequence data showed potential for resolving at least some questions of relationships within the group.However, a number of branches (notably those leading to Najas, Hydrilla, and the marine genera) were long, a situation where it would have been desirable to use a likelihood analysis that typically performs better than either neighbor-joining or weighted parsimony methods in such instances (Page and Holmes 1998).
Our present cpDNA sequence analyses improved on these earlier studies in several ways.We increased the sample of hydrocharit genera (including Najas) to 18, lacking only the rare Appertiella, which has not been relocated in the field in recent years.We have added rbcL and matK sequences for those genera not surveyed for these loci in the prior studies.We also have added to the analysis two additional cpDNA loci, namely the 3' and 5' trnK intron regions.These modifications increased the extent of cpDNA sequence data nearly twofold over previous studies.Furthermore, we have analyzed the cpDNA data using maximum likelihood as well as maximum parsimony to better assess the presence and influence of long internal branches in the phylogenetic trees.Another refinement was the use of Butomus (Butomaceae) as the outgroup.This genus is closest phylogenetically to Hydrocharitaceae (Les et al. 1997) and thus better suited than Hydrocleys Rich., the outgroup used by Tanaka et al. (1997).
It is satisfying that the trees resulting from our expanded cpDNA data (Fig. 3) retain topologies very similar to those recovered in the earlier studies (Fig. 2).In particular, all associations strongly supported (>90%) in the earlier cp-DNA analyses were retained in the results from our expanded analyses.A notable improvement was the increased level of internal support (96-98%) for the placement of Najas within the clade including Hydrilla, the three marine genera, and Vallisneria.This result inspires confidence in accepting the merger of Najas within Hydrocharitaceae.
Nuclear ribosomal DNA (nrDNA).-ThenriTS sequences of hydrocharit genera display a highly mosaic pattern of evolution.In some regions there is relatively high similarity among certain groups (arguably the most closely related genera) yet extreme divergence among others; whereas, in other regions, there is fairly high similarity across all genera, or in some cases, extreme divergence among all genera.Much variation in the nriTS region was expressed as indels that made alignment quite difficult.As expected, the 5.8S region was strongly conserved and aligned easily.Several other regions conserved across genera also existed in the spacers (ITS-1; ITS-2) and could be aligned readily.Exclusion of the highly variable sites in our analyses improved the suitability of the region for phylogenetic analysis, but the smaller number of characters reduced the degree of internal support that could be provided by the ITS data.However, the inclusion of a nuclear DNA marker not only provides additional characters for phylogenetic reconstruction, but also serves to evaluate the possibility of different histories for maternally inherited (e.g., cpDNA) vs. biparentally inherited nuclear DNA that can arise through hybridization, lineage sorting, etc. (Page and Holmes 1998).The extent of even our reduced nriTS data was adequate to serve both purposes.
A common cpDNA and nrDNA history for hydrocharit taxa was indicated by the highly congruent topologies of cladograms obtained for each data set (Fig. 3, 4).Notably, these different data sets produced several of the same associations such as a large clade consisting of Apalanthe, Blyxa, Egeria, Elodea, Lagarosiphon, and Ottelia that resolved in the same topology.A monophyletic marine clade and clades consisting of Hydrocharis and Limnobium and also Maidenia, Nechamandra, and Vallisneria also were resolved by both sets of sequences.The nriTS data also provided additional internal support for the phylogenetic analysis of the family (68-100% for the clades mentioned).We observed no major inconsistencies that would warrant against combining the nriTS data with the other data sets for parsimony analysis.
Phylogenetic insights from the combined data analysis.-Weview the cladogram generated from our combined data analysis (Fig. 7) to reasonably represent the best currently available estimate of phylogenetic relationships among Hydrocharitaceae genera.This cladogram is based on data from several sources representing both maternally and nuclear encoded characters, and shows relatively high internal support for most clades.Neither the different data sets nor the method of analysis influenced the resolution of any strongly supported groups except for Hydrilla and Najas, which were not always supported as a clade.
Relationships depicted in the combined data cladogram provide insight into the futility of previous attempts at classification for the family.In no instance did any of the classifications, based on a small subset of morphological characters, provide an appraisal of relationships that resembled those depicted in the combined data tree.Classifications such as those proposed by Richard (1811), Chatin (1855e, g), and Caspary (1857a) suggested generic associations that are at odds with the combined cladogram because they relied extensively on few, highly homoplasious characters.Anatomical characters such as those emphasized by these earlier authors are problematic in aquatic plants, which typically show strongly convergent patterns of reduction (Sculthorpe 1967;Dahlgren et al. 1985).Although this complicating factor now is generally recognized, it was poorly understood in the early to mid-nineteenth century when these classifications appeared.
Even those anatomical features such as ovule type (anatropous vs. orthotropous), which typically are regarded as having strong phylogenetic utility (e.g., Dahlgren et al. 1985), are highly homoplasious in the family (3-4 separate origins of orthotropous ovules were indicated by the combined analysis) and singularly are unsuitable for determining natural clades.Perhaps the most striking example of a misleading convergent character in Hydrocharitaceae is illustrated by some taxa having highly unusual male flowers that are released from submersed plants in bud and float to the surface where they open and drift as independent units.Although this bizarre floral mechanism occurs nowhere else in the flowering plants, the trait has evolved as many as five separate times within Hydrocharitaceae as indicated by the combined data tree (Fig. 7).By its uniqueness, it is understandable why this unusual feature has been considered to represent a strong indication of infrafamilial relationships in the past; however, it is now apparent that the feature is extremely homoplasious.Some indication of the repeated evolution of this condition is given by Cook (1982) who recognized different pollination subcategories among the taxa having detached male flowers.Cook's unique designations for Elodea ("III-A:' where pollen floats on the water surface) and Hydrilla ("III-C" where pollen is discharged through the air) correspond to independent origins of the trait as indicated by our phylogenetic analysis (Fig. 7).A third group (III-B) where anthers directly contact stigmas, unites Maidenia, Nechamandra, and Vallisneria (a strongly supported clade), yet is convergent for Enhalus and Lagarosiphon.Interestingly, the occurrence of similar pollination systems in Enhalus and Vallisneria is responsible for their placement together in many previous classifications (e.g., Hartog 1970).
Even fairly recent classifications of Hydrocharitaceae (e.g., Shaffer-Fehre 1991a), have relied principally on the distribution of few character states and portray groups that are inconsistent with the phyletic relationships depicted in our combined data cladogram.Dahlgren et al. (1985) considered distributions of larger numbers of characters, but their classification of Hydrocharitaceae ultimately was biased by their emphasis on relatively few features such as perianth structure.Although their classification (Table I) is fairly compatible with the results of our combined analysis, it misplaced various genera such as Enhalus, Hydrilla, Ottelia, and Thalassia.
We believe that results of our combined data analysis inspire much greater confidence by minimizing effects of small numbers of convergent characters in constructing our phylogeny.Analyses of larger morphological data sets, either by phenetic or phylogenetic approaches have consistently indicated the same associations of certain genera as those also resolved by phylogenetic analysis of molecular data.Examples of clades recovered consistently by either approach include Hydrocharis and Limnobium (100% bootstrap support in all analyses), Egeria and Elodea (47-100% in all analyses), and Maidenia and Vallisneria (66-100% in all analyses except MP analysis of nriTS data).Other clades resolved by all phylogenetic analyses included Maidenia, Nechamandra, and Vallisneria (57-100%), the marine genera (41-100%), and Apalanthe, Egeria, and Elodea (35-1 00% ).These results are difficult to question given the relationships indicated are consistently mirrored by various data sets and analyses.In particular, we view the relatively isolated, well-supported, basal clade comprising Hydrocharis and Limnobium to represent a distinct subfamily to which the name Hydrocharitoideae Eaton should be applied.
Although the clade containing Apalanthe, Blyxa, Egeria, Elodea, Lagarosiphon, and Ottelia was not resolved fully using morphological data alone, it was recovered using all other data sets and is depicted as a strongly supported clade (100%) in our combined analysis.Consequently, we now regard this group to represent a fundamental phyletic subdivision of Hydrocharitaceae warranting taxonomic recognition at the rank of subfamily.This clade contains most genera once placed together in tribe Anacharideae (also known as Hydrilleae Horan.);however, it also includes genera (Blyxa, Ottelia) not associated previously with the group and also excludes Hydrilla, which was placed formerly in the tribe.This well-supported clade has not been recognized by any prior classification system.The oldest available name at the rank of subfamily appears to be Anacharidoideae Thome.Although based on the genus Anacharis (a later synonym of Elodea), this name retains priority as a subfamily in accordance with article 11.3 of the International Code of Botanical Nomenclature (Greuter eta!. 2000).Presently, we recognize subfamily Anacharidoideae to include Apalanthe, Appertiella, Blyxa, Egeria, Elodea, Lagarosiphon, and Ottelia.
Our results indicate that Lagarosiphon is not most closely related to Elodea and Hydrilla as Triest (1982) concluded, but is relatively close to Blyxa as Shaffer-Fehre (1991a) observed.The vegetative similarity of Blyxa and Ottelia observed by Aston (1973) indicates their close relationship as depicted by the combined data cladogram (Fig. 7) despite our inability to resolve this clade in our morphological analysis (Fig. 1).
Another fundamental clade within Hydrocharitaceae includes the three marine genera along with Hydrilla, Maidenia, Najas, Nechamandra, and Vallisneria that associate with high internal support (99%) in the combined analysis (Fig. 7).Although the morphological (Fig. 1) and nrDNA trees (Fig. 4) excluded some taxa (Hydrilla in the former, the marine clade in the latter), the exclusions were not supported strongly.This clade is of particular significance because it includes the genus Najas, once placed in a distinct and quite distantly related family by some.Although several previous analyses (Les et a!. 1997;Tanaka et a!. 1997) have placed Najas within Hydrocharitaceae, internal support for its inclusion was only moderate (51-66%) leaving the merger of these taxa to be questionable.The existence of long branches (Fig. 7) characterizing Najas (and also Hydrilla, with which it is sometimes associated) also raises the issue whether their relationship may be spurious due to the effects of longbranch attraction.Using ML approaches that provide some correction for this problem (Page and Holmes 1998), we still recovered the placement of Najas within the family with strong support (96%) for cpDNA data and with moderate support (74%) for nrDNA data.The combined (MP) analysis also shows Najas to associate strongly (98%) with Hydrilla, a result also recovered in MP analysis (92%) of cpDNA data and nriTS data (56%).Although the phylogenetic association of these genera is well supported using MP, the long branch-lengths of both taxa show them to be quite divergent from one another at the molecular level.We recommend that this clade of eight genera be recognized as a separate subfamily to which we assign the previously published name of Hydrilloideae Luerss.
Our results have compelled us to endorse the merger of Najadaceae and Hydrocharitaceae that must take into account the nomenclatural issue of priority to apply the correct family name.Both names originate from the same publication date: (Hydrocharitaceae Juss., Genera Plantarum, vol. 67. 4 Aug 1789;Najadaceae Juss., Genera Plantarum, vol. 18. 4 Aug 1789).According to the most recent International Code of Botanical Nomenclature (Greuter et a!. 2000), they both are legitimate names of equal priority.In such instances, article 11.5 establishes priority based on the first choice to be effectively published (Greuter et a!. 2000).As far as we can ascertain, Shaffer-Fehre (1991b) was first to merge the families formally under the name Hydrocharitaceae; thus, we accept her decision to retain this name and accept it with priority over the name Najadaceae.
All data support the monophyly of the marine "seagrasses" (Enhalus, Halophila, and Thalassia), which have not been classified together previously (Table 1).The monophyly of these genera indicates that a single evolutionary colonization of the marine habitat involved taxa having different pollination mechanisms, i.e., surface-pollination in Enhalus vs. hydrophily in Halophila and Thalassia.The marine clade is derived within Hydrocharitaceae, supporting the major biological trend ascribed previously by Sculthorpe (1967).
Our phylogeny of Hydrocharitaceae shows hydrophily to be derived relative to entomophily, which occurs in the outgroup and most genera resolved basally (Fig. 7).This result supports another of Sculthorpe's (1967) biological trends attributed to Hydrocharitaceae.Hydrophily in Najas and its position distant from the two hydrophilous marine genera indicates several derivations of underwater pollination within the family (see also Les et a!. 1997).
The genus Stratiotes has an extremely unusual life-history, spending the vegetative portion of its life-cycle as a submersed plant at the bottom of ponds, but then rising up to the surface where it produces a floating rosette of emergent leaves and showy, aerial flowers (Sculthorpe 1967).Morphologically, this genus could be regarded as an evolutionary intermediate between the floating-leaved life-form that characterizes Hydrocharis and Limnobium, and the submersed life-form that occurs in all other hydrocharit genera.It is intriguing to hypothesize that the biphasic life-form of Stratiotes may have enabled plants to gradually acquire adaptations to a submersed existence over evolutionary time, without necessitating an abrupt abandonment of terrestrial adaptations such as entomophilous flowers.The intermediate phylogenetic placement of Stratiotes between the floatingleaved and submersed genera of Hydrocharitaceae (Fig. 7) is consistent with this interpretation.We propose that Stratiotes should be placed within a separate subfamily (published previously as Stratioideae Luerss.) to reflect its distinct position in the family.
One remaining question concerns the precise phylogenetic position of Appertiella, which could not be obtained for molecular analyses.Our best estimate at present is to tentatively regard it as being related most closely to Lagarosiphon (Cook and Triest 1982), a conclusion consistent with, but not supported unambiguously by the results of our morphological analysis (Fig. 1).This proposed relationship should be tested once material becomes available for DNA analysis and comparison with the molecular data compiled for other genera surveyed in this study.
Combined phylogenetic data analysis has significantly improved our understanding of intergeneric relationships in Hydrocharitaceae by producing a well-resolved and wellsupported cladogram that lends credibility to our suggested improvements in the classification of the family.We recommend the taxonomic division of Hydrocharitaceae into four subfamilies, which correspond to the major clades depicted in our combined data analysis (Fig. 7).A synopsis of our proposed classification is presented in Table 5.

Fig. 2 .
Fig. 2.-Comparison of cladograms from previous studies using cpDNA to study phylogenetic relationships in Hydrocharitaceae.Left: Cladogram redrawn from Tanaka et a!.(1997) from matK and rbcL sequence data analyzed by neighbor-joining (NJ) and rooted using Hydrocleys as the outgroup.Right: Cladogram adapted from Les et a!.(1997) generated with rbcL data analyzed by weighted maximum parsimony (wMP) and rooted using multiple Alismatidae outgroups.The trees are highly congruent in topology and show similar levels of internal support.Numbers above branches in both trees represent bootstrap values.

Fig. 3 .
Fig. 3.-Cladograms depicting results of combined cpDNA analysis (present study) using maximum likelihood (ML) and maximum parsimony (MP) methods.Left: Best cladogram recovered (lnL = -14,964.04352)using maximum likelihood analysis (GTR + I + r model).Numbers above branches represent bootstrap values obtained from 200 replicates.Right: Strict consensus of three trees (2405 steps) generated from maximum parsimony analysis.Numbers above branches represent bootstrap values obtained from I 000 replicates.Both methods of analysis recovered similar clades with comparable levels of internal (bootstrap) support.
Fig. 4.-Cladograms showing results of nriTS data analysis using maximum likelihood (ML) and maximum parsimony (MP) methods.

Fig. 5 .
Fig. 5.-Maximum parsimony (MP) analysis of combined cpDNA and nriTS data.The strict consensus of three trees (2984 steps) generated from maximum parsimony analysis is shown.Numbers above branches represent bootstrap values obtained from 1000 replicates.All but two resolved nodes show strong internal support (2:98%) from bootstrap analysis.
Fig. 7.-Single most parsimonious tree (2695 steps) recovered from MP analysis of all data combined.The tree is shown as a phylogram depicting relative branch lengths (scale bar = 20 steps).Numbers above branches represent bootstrap values obtained from 1000 replicates.Life-forms of genera are indicated by symbols in parentheses: (E) = emergent; (F) = floating-leaved; (S) = submersed.Boxes indicate genera where male flowers detach from plants and float on the water surface (category III pollination types).Pollination types (from Cook 1982) are indicated as: ENT (entomophilous); H (hypohydrophilous); III-A (pollen released on water surface); III-B (anthers making direct contact with stigma); III-C (pollen shed ballistically).Closed circles indicate four independent origins of type III pollination as shown by character state reconstructions using either ACCTRAN or DELTRAN optimizations.The open circle indicates a fifth origin of type III pollination in Elodea (which is polymorphic, possessing species having both attached and detached male flowers).Ovule type also is highly convergent on this tree, with ACCTRAN and DELTRAN optimized reconstructions indicating 3-4 separate derivations of orthotropous ovules (indicated by stars), respectively.

Table 3 .
Matrix of characters and character states (see Table2) used in morphological phylogenetic analysis of Hydrocharitaceae genera and Butomus (outgroup).Polymorphisms are indicated in parentheses.?= data missing; -= data not applicable.