Specimens from the following herbaria were analysed: AAU, ALCB, B, BA, BAF, BHCB, BHZB, BLH, BM, BOL, BOTU, BR, BRIT, C, CAS, CEPEC, CESJ, COL, CORD, CTES, CVRD, DS, E, EA, ESA, F, FCAB, FLOR, FURB, G, GH, GMUF, GOET, GUA, HAL, HAMAB, HAS, HB, HBR, HERBAM, HNMN, HRB, HRCB, HSTM, HUEFS, HUFSJ, HURB, IAC, IBE, ICN, INPA, IPA, K, KANU, L, LE, LG, LIL, LL, LP, M, MA, MBM, MBML, MG, MO, MVM, MY, NBYC, NY, OS, P, PH, PMSP, PR, PRC, PRE, R, RB, RFA, RFFP, S, SMU, SP, SPF, SRGH, TEX, UEC, UMO, UNA, UPCB, US, USF, VDB, VIC, W and WAG (herbaria acronyms according to Thiers, cont. updated). Fresh specimens, field notes, photographs and specimens for cultivation were gathered by the authors during several field trips across North, Central and South America, between 1980 and 2017. The indumentum and shape terminology follow Radford et al. (1974); the inflorescence terminology and morphology follow Weberling (1965, 1989) and Panigo et al. (2011), as implemented by Pellegrini and Horn (2017); fruit terminology follows Spjut (1994); and seed terminology follows Faden (1991). Species distribution is based on literature, herbarium specimens and fieldwork data.

Characters were scored mainly from living specimens in the field and specimens in cultivation and later complemented by spirit and herbarium samples from the aforementioned herbaria. When no living or herborised specimens were available for examination, information was taken from published literature. We have studied at least five specimens for each taxon, with the most representative specimen chosen as the voucher for the morphological matrix (Table 1). Some characters were chosen based on previous studies (i.e. Eckenwalder and Barrett 1986; Simpson 1987; Barrett and Graham 1997; Simpson and Burton 2006), with most characters being scored for the present study. Character coding followed the recommendations of Sereno (2007) for morphological phylogenies. Primary homology hypotheses (De Pinna 1991) were proposed for root, stem, leaf, inflorescence architecture, floral, fruit, seed, palynological and anatomical characters. A total of 96 discrete micro- and macromorphological characters were scored, being treated as unordered and equally weighted (Suppl. material 1).

Data were entered into a matrix of characters per taxa using the software Mesquite 3.20 (Maddison and Maddison 2017; Suppl. material 2). All characters were treated as unweighted and unordered. Maximum Parsimony (MP) analysis was performed using PAUP* 4 (Swofford 2003), with a heuristic search with 1000 random taxon additions and tree bisection-reconnection (TBR) branch swapping. Consistency index (CI) and retention index (RI) were used to assess the degree of homoplasy in the dataset and ACCTRAN (accelerated transformation optimisation; Swofford and Maddison 1987) was used for character optimisation. Statistical support for each branch of the cladogram was evaluated with Bootstrap Support (BS) analyses with 1000 random addition replication. The search parameters used to estimate the bootstrap values were the same as the initial heuristic search. The Bremer Index (BI) was also used to evaluate clade reliability based on the presence of secondary homologies (Bremer 1994). The Bremer Index was calculated by increasing the number of the optimal tree steps until all clades collapsed. Mesquite 3.20 was used to reconstruct the ancestral character states, while WinClada ver. 1.0000 (Nixon 2002) was used to trace the synapomorphic characters on the strict consensus tree.

Sequences of the genes ndhF and rbcL were retrieved from GenBank for 26 taxa representing all currently accepted genera in Pontederiaceae, including outgroups Anigozanthos Labill. and Xiphidium Aubl. (Haemodoraceae) and the tree was rooted with Philydraceae. All sequences were aligned using Muscle (Edgar 2004) implemented on Geneious software (Kearse et al. 2012), with subsequent adjustments in the preliminary matrices made by eye.

Combined analyses of the plastid regions and plastid+morphology datasets were performed. Prior to combining our data, we performed the incongruence length difference (ILD) test (Farris et al. 1994) to investigate the incongruence between DNA data sets. Analyses, using maximum parsimony (MP) on both matrices, were conducted with PAUP* 4 (Swofford 2003). A heuristic search was performed using TBR swapping (tree-bisection reconnection) and 1000 random taxon-addition sequence replicates with TBR swapping limited to 15 trees per replicate in order to prevent extensive searches (swapping) in suboptimal islands, followed by TBR in the resulting trees with a limit of 1000 trees. In all analyses, the characters were equally weighted and unordered (Fitch 1971). Relative support for individual nodes was assessed using non-parametric bootstrapping (Felsenstein 1985), with 1000 bootstrap pseudo-replicates, TBR swapping, simple taxon addition and a limit of 15 trees per replicate.

For the DNA partitions of the model-based approach, we selected the model using hierarchical likelihood ratio tests (HLRT) on J Modeltest 2 (Darriba et al. 2012. For the morphological partition, the standard discrete Markov model (Mkv) was used, following Lewis (2001) with rates set to equal. A Bayesian analysis (BA) was conducted with mixed models and unlinked parameters, using MrBayes 3.1.2 (Ronquist and Huelsenbeck 2003). The Markov Chain Monte Carlo (MCMC) was performed using two simultaneous independent runs with four chains each (one cold and three heated), saving one tree every 1000 generations, for a total of ten million generations. We excluded as ‘burn in’ trees from the first two million generations and tree distributions were checked for a stationary phase of likelihood. The posterior probabilities (PP) of clades were based on the majority-rule consensus, using the remaining trees, calculated with MrBayes 3.1.2 (Ronquist and Huelsenbeck 2003).

The cladistic analysis retrieved 228 equally parsimonious trees with 209 steps, Consistency Index (CI) of 0.5913, Homoplasy Index (HI) of 0.4087, Retention Index (RI) of 0.8618 and Rescaled Consistency Index (RC) of 0.5096. All 96 coded characters were parsimony-informative. The strict consensus (Fig. 1) and the majority-rule trees (Fig. 2) are presented and discussed below.

The Haemodoraceae+Pontederiaceae clade is supported by seven characters: the presence of septal nectaries (Character 44), perianth 6-lobed (Character 58, plesiomorphic), perianth with 3+3 arrangement (Character 59, plesiomorphic), epipetalous stamens (Character 66, homoplastic), stamens dimorphic (Character 69), endothecium with a basal thickening (Character 72) and non-tectate-columellate exine (Character 76). Pontederiaceae is recovered as monophyletic with high statistical support (BS=100; BI=7; Fig. 2), being supported by: dimorphic leaves (Character 12), leaf-blades late bifacial (Character 13), involute ptyxis where the blade of the new leaf encloses the petiole of the preceding leaf (Character 14), leaf-blades with xylem and phloem alternate in the central portion of the blade and xylem abaxial and phloem adaxial at the margins (Character 15), the presence of a ligule (Character 16), non-equitant leaves (Character 18, reversion), sessile leaves early-deciduous (Character 18), inflorescence deflexed at post-anthesis and in fruit (Character 37), sessile flowers (Character 39), absence of fibrillar tannin cells in the perianth (Character 47), presence of aerenchymatous tissue in the receptacle (Character 48) and in the perianth (Character 49), perianth connate producing a conspicuous tube (Character 56, homoplastic), perianth ranging from lilac to purple or blue (Character 57, homoplastic), posterior lobe(s) with a nectar guide (Character 63, homoplastic), pollen grains bisulcate (Character 75), presence of aerenchymatous tissue in the ovary walls (Character 79) and the presence of an anthocarp (Character 91).

HeterantherasensuPellegrini (2017a), is recovered as monophyletic with high statistical support (BS=99; BI=3; Fig. 2). It is supported by: plants mostly to completely submersed (Character 3, homoplastic), indefinite base (Character 4), water-binding/mucilaginous roots (Character 6), rhizome absent (Character 7), stems freely branching and elongated (Character 9 and 10, homoplastic), ligules 2–several parted (Character 17), spirally-alternate sessile leaves (Character 18), sessile leaves evenly distributed along the stem (Character 20, homoplastic), basal bract conduplicate (Character 30), main florescence reduced to a solitary cincinnus (Character 32), sparse aerenchymatous tissue in the perianth (Character 49), perianth tubular (Character 50), filaments obliquely inserted (Character 65) and unevenly trilobate stigma (Character 87). Within Heterantheras.l., we recover two main clades in the majority rule (Fig. 2), with only one of these being also recovered in the strict consensus (Fig. 1). The H. limosa group is composed by H. limosa (Sw.) Willd., H. oblongifolia Mart. ex Schult. & Schult.f. and H. rotundifolia (Kunth) Griseb., being characterised by: the absence of clonal reproduction (Character 2, homoplastic), sessile leaves late-deciduous (Character 19, homoplastic), petiolate leaves with elliptic to ovate blades (Character 27, homoplastic), the posterior perianth lobe with flanged base (Character 62) and a nectar guide consisting of a sole spot or dark band (Character 63, homoplastic), sigmoid filaments (Character 67), ovary hemiseptalous (Character 80, homoplastic), axile-parietal placentation (Character 83) and placentation 2-flanged (Character 84, homoplastic). The second clade, named by us as the H. dubia group, is composed of H. dubia (Jacq.) MacMill., H. gardneri (Hook.f.) M.Pell., H. seubertiana Solms and H. zosterifolia Mart. This group is characterised by: the presence of cleistogamous flowers (Character 43), inflated filaments (Character 68), gynoecium 1-locular (Character 77, homoplastic), ovary aposeptalous (Character 80, homoplastic), intrusive-parietal placentation (Character 83, homoplastic) and placentation slightly 2-flanged (Character 84).

Pontederias.l. is also recovered as monophyletic with high statistical support (BS=93; BI=6; Fig. 2), being supported by: distichously-alternate sessile leaves (Character 18), petiolate leaves pulvinate (Character 25), tristylous flowers (Character 42), dense aerenchymatous tissue in the perianth (Character 49), perianth campanulate or infundibuliform or hypocrateriform (Character 50, homoplastic), perianth coiled and tightly enclosing the fruit at post-anthesis (Characters 53 and 55), perianth lobes equal in shape in the same whorl (Character 60) and with obtuse apex (Character 61, homoplastic), stamens 6 (Character 64, reversion), filaments J-shaped or recurved-decurved (Character 67), anthers dorsifixed (Character 71), style J-shaped (Character 85), stigmas evenly trilobate to trifid or capitate (Characters 87), stigma wet (Characters 88), anthocarp tightly enveloping the fruit (Character 92) and anthocarp hardened and ornamented (Characters 93 and 94). Pontederias.l. is recovered by us arranged in five clades in the strict consensus (Fig. 1) and in the majority rule (Fig. 2). The E. paniculata group is highly supported (BS=95; BI=1; Fig. 2), being composed by E. paniculata (Spreng.) Solms and E. paradoxa (Mart. ex Schult. & Schult.f.) Solms. It is characterised by: its annual life cycle (Character 1, homoplastic), the lack of clonal reproduction (Character 2, homoplastic), inflated sheath of the leaf subtending the inflorescence (Character 29, homoplastic), flat basal bract (Character 30, homoplastic) with a caudate apex (Character 31, homoplastic), main florescence with a fistulose main axis (Character 34, homoplastic), inflorescence erect at post-anthesis and in fruit (Character 37, reversion), floral organs lacking tannin cells of the homogeneous type (Character 45), perianth with a moderate amount of granular tannin cells (Characters 51 and 52), perianth spirally-coiled at post-anthesis (Character 54, homoplastic), ovary walls lacking tannin cells (Character 78, homoplastic), ovary hemiseptalous (Character 80, homoplastic) and septae lacking tannin cells (Character 82, homoplastic). Based on morphology, E. meyeri A.G.Schulz should also be placed in the E. paniculata group. Monochoria is recovered as monophyletic with high statistical support (BS=96; BI=2; Fig. 2), being characterised by eight non-homoplastic synapomorphies: pedicellate flowers (Character 39, reversion), perianth only basally connate (Character 56, reversion), absence of a nectar guide (Character 63, reversion), presence of a petalo-staminal tube (Character 66), stamens unequal (Character 69), presence of a filament appendage (Character 70), enantiostylous flowers (Character 71, reversion) and poricidal anthers (Character 72). Eichhornia crassipes is recovered as a sole species with high statistical support (BS=94; BI=1; Fig. 2), being characterised by: its free-floating habit (Character 5), the production of new rosette through stolons (Character 8), flabellate ligules (Character 17), spirally-alternate petiolate leaves (Character 22, homoplastic), perianth loosely enveloping the fruit (Character 55, homoplastic) and nectar guide consisting of a sole spot (Character 63, homoplastic). Eichhornias.s. was recovered with low statistical support (BS=56; BI=2; Fig. 2), being composed by E. azurea (Sw.) Kunth, E. diversifolia (Vahl) Urb. and E. heterosperma Alexander. It is characterised by: growing as mostly submerged plants (Character 3, homoplastic), stems freely branching and elongated (Character 9 and 10, homoplastic), sessile leaves late-deciduous (Character 19, homoplastic), petiolate leaves evenly distributed along the stem (Character 23, homoplastic), flowers self-compatible (Character 38, homoplastic), floral tissues lacking granular tannin cells (Character 46, homoplastic) and presenting fibrillar tannin cells (Character 47, reversion), nectar guide consisting of a sole spot or dark band (Character 63, homoplastic) and ovary walls lacking aerenchymatous tissue (Character 79, reversion). Finally, PontederiasensuLowden (1973) was recovered by us as monophyletic with high statistical support (BS=97; BI=3; Fig. 2). It is characterised by: flowers self-compatible (Character 38, homoplastic), nectar guide consisting of two spots (Character 63, homoplastic), pseudomonomerous ovary (Character 77), the presence of epithelial cells in the septae (Character 81, homoplastic), pendulous and unflanged placentation (Characters 83 and 84), fruit an achene (Character 89), seeds one per locule (Character 90) and smooth testa (Character 95). Nonetheless, the subgenera proposed by Lowden (1973) cannot be maintained, due to P. rotundifolia L.f. (i.e. P. subg. Reussia) being nested within P. subg. Pontederia (sensuLowden 1973).

The ndhF characters represented 503 characters of the plastid dataset, with GTR+G as the nucleotide model selected. The rbcL characters represented 1355 characters of the plastid dataset, with HKY+G+I as the nucleotide model selected. The plastid dataset represented 1858 characters, of which 241 characters were variable and 119 characters were parsimony-informative. The plastid Bayesian analysis recovered a mostly resolved tree with 23 well-supported clades (>PP 95%) (Fig. 2). The congruence between the plastid and morphological datasets is illustrated in Figure 2. In both analyses, Pontederias.l. and HeterantherasensuPellegrini (2017a) are strongly supported, but the relationship between the species is greatly different. In Heteranthera, the morphologically based topology is better resolved and recovers two clades, while the plastid dataset recovers two clades plus H. gardneri in a polytomy (Fig. 2). In Pontederias.l., both datasets recover the genus arranged in five clades, but the relationship between them is different. In the morphological dataset, Eichhornias.s. is the first lineage to diverge, followed by E. crassipes, Pontederias.s. and Monochoria, sister to the E. paniculata group. Alternatively, in the plastid dataset, the E. paniculata group is undoubtedly recovered as the first lineage, followed by E. crassipes, Monochoria and Pontederias.s., sister to Eichhornias.s.

Topologies produced by MP and BI analyses, based on the combined plastid + morphology datasets, were highly congruent and provided higher support for more clades than the results based on independent datasets (Fig. 3). Thus, based on the combined plastid + morphological datasets (1858 analysed characters, of which 353 were variable and 140 parsimony-informative), the maximum parsimony analysis found 24 trees (CI=0.6471, RI=0.7858) whose MRC presented 23 highly supported clades (BSP 75%). The combined Bayesian analysis recovered a fully resolved tree with 25 mostly well-supported clades (>PP 95%) (Fig. 3). The topology recovered for the Bayesian combined analysis (Fig. 3) is almost identical to the one recovered for the plastic dataset (Fig. 2), differing in only very small details. On the other hand, the Parsimony combined analysis recovers E. crassipes, Pontederias.s. and Eichhornias.s. in a well-supported clade, with this clade being recovered in a polytomy together with the E. paniculata group and Monochoria.

The topologies recovered from the combined plastid and the total evidence datasets strongly corroborate the bi-generic circumscription of Pontederiaceae suggested by Pellegrini (2017a). They are also congruent with previous phylogenetic studies using molecular and/or combined datasets (Graham and Barrett 1995; Barrett and Graham 1997; Graham et al. 1998, 2002; Ness et al. 2011) and partially congruent with the morphologically based phylogenetic tree of Eckenwalder and Barrett (1986). The phylogenetic tree recovered by Kohn et al. (1996) differs greatly from our results and all previous studies due to part of the polyphyletic Eichhornia being recovered as sister to Heterantheras.l. Most molecular studies in the family (Graham and Barrett 1995; Barrett and Graham 1997; Graham et al. 1998, 2002; Ness et al. 2011) recover a well-supported Pontederiaceae, divided into two main lineages, corresponding to a well-supported Heterantheras.l. (sensuPellegrini 2017a) and poorly-supported Pontederias.l.; using ndhF, rbcL, plus a restriction-site in the chloroplast genome in Graham et al. (1998, 2002) and five nuclear gene families recovered employing an expressed sequence tag (EST) study by Ness et al. (2011). As in previous studies (Graham and Barrett 1995; Barrett and Graham 1997; Graham et al. 1998, 2002; Ness et al. 2011), we recover Pontederias.l. arranged in five main lineages, each representing a well-supported morphological group (i.e. Eichhornia paniculata group, Monochoria, E. crassipes group, Eichhornias.s. and Pontederias.s.). The monophyly of HeterantherasensuPellegrini (2017a) is indisputable and the inclusion of Hydrothrix and Scholleropsis in Heteranthera was strongly corroborated.

The monophyly of Pontederiaceae was rarely, if ever, questioned by previous authors. Perhaps for this reason, little attention was ever given to the family’s putative morphological synapomorphies. Amongst the 18 morphological synapomorphies recovered for Pontederiaceae, one was previously suggested by Arber (1925; i.e. with xylem and phloem alternate near the centre of the blades, plus xylem abaxial and phloem adaxial near the margins), three were suggested by Simpson (1987, 1990; i.e. late bifacial and ligulate leaves and bisulcate pollen grains) and four were suggested by Simpson and Burton (2006; absence of fibrillar tannin cells in the perianth and presence of aerenchymatous tissue in the receptacle, perianth and ovary walls). Nonetheless, the peculiar involute ptyxis where the blade of the new leaf encloses the petiole of the preceding leaf, non-equitant leaves, sessile leaves early-deciduous, inflorescence deflexed at post-anthesis and in fruit, sessile flowers, perianth connate producing a conspicuous tube and the presence of an anthocarp, are suggested here for the first time as synapomorphies for Pontederiaceae.

Almost, if not all, leaf synapomorphies recovered for Pontederiaceae seem to be directly correlated. These characters seem to be related to the adaptive shift to a completely aquatic lifestyle in the family and an adaptation to changes in water level. The leaves of Pontederiaceae are characteristically dimorphic, being morphologically divided into sessile and petiolate leaves (Horn 1988). Leaf dimorphism is widely distributed across the Embryopsida, being generally related to changes in function (e.g. reproductive leaves in ferns), growth form (e.g. juvenile and mature leaves of Monstera spp.) or environmental changes (Allsopp 1965). The dimorphic leaves of Pontederiaceae seem to fit the latter situation, since the petiolate leaves are always floating or aerial, while the ribbon-like or acicular sessile leaves are the first type produced by the germinating plantlet and seen to be an adaptation to the aquatic environment. Furthermore, the presence of a petiole greatly helps to keep the leaves at or above the water level, through cell elongation in the petiolar region. This strategy can be easily observed in several distantly related aquatic plant families (e.g. Alismataceae, Asteraceae, Cabombaceae, Haloragaceae, Nymphaeaceae, Onagraceae, Ranunculaceae etc.; Allsopp 1965; Sculthorpe 1967; Cook 1996). The peculiar vascular bundle arrangement observed in Pontederiaceae is exclusive to the family and few other monocots (Arber 1925). This feature seems to be a result of the reversion from abaxialised unifacial leaves to bifacial leaves, which, according to Simpson (1990), might be related to the adaptive shift and radiation to an aquatic lifestyle in the family. The remaining closely related families (i.e. Haemodoraceae and Philydraceae) possess consistently abaxialised unifacial leaves, with blades ranging from cylindrical, terete, laterally compressed to more rarely plicate (Simpson 1990, 1998; Hamann 1998). Nonetheless, the evolutionary relevance of bifacial leaves is significantly harder to infer, since unifacial leaves are noticeably common in several aquatic plants. The reversal from equitant to alternate leaves seems to be a by-product from the reversion from unifacial to bifacial leaves. As aforementioned, the involute ptyxis in Pontederiaceae is extremely unusual, since the blade of the new leaf encloses the petiole of the preceding leaf. This feature is also unique in the Angiosperms and is easily observed in most species in the family but is especially obvious in E. crassipes (Fig. 7C). This feature might also be related to the adaptive shift and radiation to a completely aquatic lifestyle in Pontederiaceae, being most likely a result of the reversion to bifacial leaves. Developmental studies focusing on the ontogeny of the leaves in Pontederiaceae, in comparison to some members of Haemodoraceae and Philydraceae, might help us better understand the mechanics of the reversal from unifacial to bifacial leaves in the family and how this shift might have affected general leaf morphology and the appearance of novel structures such as the ligule.

As aforementioned, the leaves of Pontederiaceae are dimorphic, with both sessile and petiolate leaves being produced at different times in the plants’ life. Sessile leaves represent the plesiomorphic state and are the first ones produced after seed germination. They vary in number from 5–many per plant and allow plants to become established in a submersed habitat (Horn 1988). The sessile leaves can range from early-deciduous to persistent in mature plants, while in some species of Heterantheras.l., petiolate leaves are never or very rarely produced (Horn 1985, 1988; Eckenwalder and Barrett 1986). The petiolate leaves are produced at posteriori and are considered the mature leaf type in the family. The initial petiolate leaves are morphologically plastic, allowing for a transition from a submersed to an immersed environment. This plasticity, coupled with the elongation of the stem, allows Pontederiaceae plants to successfully develop to and at the water surface (Horn 1988). In Heterantheras.l., the sessile leaves suffer a reversion from distichously to spirally arranged, producing the characteristic basal rosettes in the juvenile phase of many Heteranthera species (Horn 1988). Thus, early-deciduous sessile leaves and early production of petiolate leaves give a clear adaptive advantage to the Pontederiaceae, enabling them to tolerate a wide variation in water depth during their development, also allowing juvenile plants to successfully reach mature emergent or floating growth-forms (Horn 1988). This might have ultimately allowed the diversification of Pontederiaceae and their complete invasion of the aquatic environment.

The presence of a leaf sheath projection is striking in Pontederiaceae, with its morphology being relevant to the systematics of the family. Ligules and ligule-like structures are recorded for several members of Embryopsida, being especially common in some lycophytes (i.e. Selaginellales and Isoëtales) and several monocots (i.e. Alismatales, Arecales, Asparagales, Commelinales, Dioscoriales, Poales and Zingiberales) (Kubitzki 1998; Rudall and Buzgo 2002; Kellogg 2015). Despite possessing the same name, there is no evidence supporting the homology of these structures between lycophytes and monocots and not even between different groups within the monocots (Rudall and Buzgo 2002). The definition and characterisation of ligules in monocots has varied greatly depending on the author, having Poaceae as their main focus. These authors have proposed three distinct definitions for ligules: (1) a subtype of stipule (Bischoff 1834; Regel 1843; Lubbock 1891, 1895; Arber 1925); (2) a structure of mixed origin between stipules and petioles (Glück 1901; Majumbdar 1956); and (3) an avascular projection of the leaf-sheath, situated between the leaf-sheath and the blade (Colomb 1887; Philipson 1935; Dahlgren et al. 1985; Chaffey 1994; Rudall and Buzgo 2002). In Commelinid monocots, ligules and ligule-like structures are recorded for Arecales (i.e. the hastulae present is some Arecaceae leaves), several families of Poales (e.g. Cyperaceae, Joinvilleaceae, Juncaceae, Poaceae, Restionaceae), Commelinales (exclusively in Pontederiaceae) and Zingiberaceae (i.e. Costaceae and Zingiberaceae) (Kubitzki 1998; Rudall and Buzgo 2002; Kellogg 2015). As aforementioned, ligules and ligule-like structures in Commelinales seem to be restricted to Pontederiaceae and are unknown to any of the other four families of the order (Kubitzki 1998; Rudall and Buzgo 2002; Pellegrini pers. obs.). These structures might also be a result of the reversion from unifacial leaves to bifacial leaves or even an independent adaptation to the aquatic lifeform in the family. In the unifacial-leaved clade, composed by Philydraceae (Haemodoraceae+Pontederiaceae), Pontederiaceae is the only exclusively aquatic family and also the only one to possess ligule-like structures (Figs 4F, 6C, 7C, 9E), dimorphic leaves, petiolate leaves and bifacial leaves. Nonetheless, ontogenetic studies are necessary to understand the origin of these structures in the family. In Pontederiaceae, these ligule-like structures have been treated under different names according to the authors, having been named stipules (Schwartz 1926), ligules (Castellanos 1958; Pellegrini and Horn 2017), ochreas (Rutishauser 1999) or simply as leaf-sheath projections (Pellegrini 2017a). Different names have also been applied by the same author, depending on the development and shape of these structures (i.e. Cook 1998). Regardless of the name adopted for these ligule-like structures in Pontederiaceae, their systematic and taxonomic relevance is undeniable. As aforementioned, this structure is recovered as synapomorphic for the family. Alternatively, within Pontederiaceae, the morphology of this structure can be easily used to define the two clades recovered in phylogenetic studies. Pontederias.l. can be easily characterised by it mainly truncate ligules, being rarely flabellate (i.e. E. crassipes); while Heterantheras.l. can be characterised by its 2–several-parted ligules.

Out of the reproductive synapomorphies recovered by us for Pontederiaceae, some of them seem to be related to pollination, while the others seem to be related to fruit dispersal. Sessile flowers are recovered by us as a synapomorphy of Pontederiaceae, with the sole reversion occurring in Monochoria. This character seems to be directly related to another reproductive synapomorphy for the family (i.e. perianth connate to part of the receptacle and the filaments producing a conspicuous tube). Pedicel and floral tube length seem to be inversely correlated, with tube elongation helping with the floral display by elevating the perianth lobes. Added to that, the contraction of the pedicel might also provide extra stability for heavier floral visitors that require landing platforms in order to properly visit flowers (e.g. butterflies). Alternatively, the reversion from sessile to pedicellate flowers in Monochoria might have played a key role, by giving flowers the needed mobility in order to avoid floral damage during buzz pollination (Wang et al. 1995). Bisulcate pollen grains are rather rare in the monocots, being recorded for only a handful of families, such as: Araceae (Grayum 1992), Arecaceae (Harley and Baker 2001), Dioscoreaceae (Caddick et al. 1998), Iridaceae (Rudall and Wheeler 1988) and Velloziaceae (Halbritter and Hesse 1993). Of the aforementioned families, only Arecaceae (Arecales) is a member of the Commelinid monocots and it is but distantly related to Pontederiaceae (Saarela et al. 2008; Hertweck et al. 2015; APG IV 2016). In Haemodoraceae, Simpson (1983) recorded the occurrence of biporate pollen grains in some genera from subfamily Conostylidoideae. Nonetheless, Simpson (1987, 1990) considers the biporate pollen grains in Haemodoraceae not homologous to the bisulcate pollen grains in Pontederiaceae. This view is also shared by us in the present study.

The first synapomorphy related to diaspore dispersal is the deflexed position of the inflorescence at post-anthesis and in fruit. This shift in the inflorescence position during fruit development will almost certainly allow the mature fruits to reach the water after their maturity. The deflexed inflorescences also elongate in length, which ultimately places the maturing fruits at or under the water surface. This seems to be the first step in diaspore dispersion in most species of Pontederiaceae. The following adaptations are related to increasing the floatation period of the diaspores. The first and most obvious seems to be the presence of an anthocarp. According to Spjut (1994), an anthocarp is a type of fruit which possesses attached and developed floral parts that aid in its dispersal. It is more commonly recorded for plants with inferior ovaries, but it is not exclusive to them (Spjut 1994). In Commelinales, all fruits have persistent perianth parts, but only in Pontederiaceae does an enlarged perianth actively aid in the dispersal of the diaspores (Pellegrini, pers. observ.), with Tradescantia zanonia (L.) Sw. (Commelinaceae) being an exception (Pellegrini 2017b; Pellegrini and Faden 2017). In Pontederiaceae, the anthocarp seems to be related to hydrochoric dispersion, which is also supported by the remaining synapomorphies for the family (i.e. presence of aerenchymatous tissue in the receptacle, perianth and ovary walls). The anthocarp is especially developed with thick aerenchymatous tissue in Monochoria, Pontederias.s. and in the E. paniculata group (Lowden 1973; Cook 1989, 1998; Simpson and Burton 2006; Pellegrini, pers. observ.; Figs 5F, 6K & 9K), that provides long flotation periods for the diaspores (i.e. around 15 days; Barrett 1988). In the remaining lineages of Pontederiaceae (i.e. Heterantheras.l., E. crassipes group and Eichhornias.s.), the anthocarp is thin, probably resulting in a much shorter flotation period (i.e. probably around 24h), with seeds being secondarily dispersed by other biotic and/or abiotic means (Barrett 1978; Pellegrini and Horn, pers. observ.). In the closely-related Haemodoraceae and Philydraceae, the perianth is also connate, producing a characteristic hypanthium and partially to completely persistent in fruit (Hamann 1998; Simpson 1998). Nonetheless, they do not aid in the dispersal of diaspores, since in all species, the persistent perianth is only marcescent and does not develop during fruit development, being ultimately torn open by the mature fruit (Pellegrini, pers. observ.). These observations are also supported by the complete lack of aerenchymatous tissues in floral organs of both families, with aerenchyma being recorded only in the septae of the hydrochoric Philydraceae (Simpson and Burton 2006). In Commelinaceae and Hanguanaceae, the persistent perianth also does not develop during fruit maturation; with the exception of Buforrestia C.B.Clarke (Commelinaceae), where the persistent sepals are as long as, or longer than, the mature capsule (Bayer et al. 1998; Faden 1998). Nonetheless, the perianth of Buforrestia does not seem to aid in the dispersion of the diaspores, since the perianth only loosely involves the capsules, which remain attached to the pedicel and dehisce at maturity (Pellegrini, pers. observ.). In Hanguanaceae, the fruits consist of variously coloured berries that detach from the persistent sepaloid perianth and are most probably zoochoric (Bayer et al. 1998). On the other hand, in Commelinaceae, the fruits are primarily dehiscent capsules (rarely indehiscent capsules or berries), that do not rely on the persistent sepals for dispersion, with fruits or seeds being autochoric or more rarely zoochoric (Pellegrini and Faden 2017).

All 18 synapomorphies recovered by us for Pontederias.l. are suggested here for the first time. Sand-binding roots were recovered by Smith et al. (2011) as plesiomorphic for Haemodoraceae and probably for all Commelinales, despite the authors’ not sampling Hanguanaceae in their analysis. These sand-binding roots produce specialised hairs that bind soil, especially larger sand crystals, creating a protective layer that envelops the roots (Smith et al. 2011). These authors also state that all studied specimens of Philydraceae and Pontederiaceae had non-sand-binding roots, in contrast to Haemodoraceae. On the other hand, sand-binding roots are commonly observed in several lineages of Commelinaceae, but especially in species growing in dry environments (Smith et al. 2011; Pellegrini, pers. observ.). After several field studies and cultivation of several species of Pontederiaceae, we have observed that all species of Heterantheras.l. possess water-binding (i.e. mucilaginous) roots, while the absence of an external mucilage layer on the roots was characteristic of Pontederias.l. The water-binding roots of Heterantheras.l. are most probably not homologous to the sand-binding roots in the order, since they do not seem to have specialised hairs, like those described for Haemodoraceae (Smith et al. 2011). The mucilage layer seems to be produced by the secretion of chemical compounds near the root apex which polymerises in contact with water (Pellegrini, pers. observ.). Nonetheless, further anatomical and histochemical studies are needed to better understand this feature.

The presence of leaves with pulvinate petioles in Pontederias.l. is easily observed in the field, since most pulvini are lighter or darker than the rest of the petiole. On the other hand, in dried specimens, this difference in colouration is only sometimes maintained, making this character not always obvious to untrained eyes. Added to that, the pulvini in Pontederias.l. are seldom swollen, as would be expected in most eudicot plants with articulated leaves. Nonetheless, this feature seems to be key for the emergent and floating species, especially the perennial ones, since they are subjected to the greatest amount of environmental variation. Floating species like E. crassipes are easily dragged by water currents, forcing all leaf-blades to change their position in order to better absorb sunlight. Perianth-coiling at post-anthesis seems to be poorly documented in the literature for most Angiosperm families and more so in the monocots. It is known to occur in the monocots only in the distantly related Bromeliaceae (Poales), being characteristic to some genera of subfamilies Pitcairnioideae and Puyoideae (Smith et al. 1998; Hornung-Leoni and Sosa 2008). In Commelinales, the persistent perianth is marcescent in Philydraceae, Haemodoraceae and Hanguanaceae, while in Commelinaceae, the sepals are marcescent and the petals are deliquescent (Pellegrini, pers. observ.). In Pontederiaceae, the perianth in Heterantheras.l. is also marcescent at post-anthesis, only loosely enclosing the developing capsule. In Pontederias.l., the perianth is either spirally-coiled or revolute at post-anthesis, tightly enclosing the developing fruit, with two independent shifts to deliquescent perianths loosely enclosing the developing fruit (i.e. E. crassipes and Eichhornias.s.). This might be related with increasing long-distance diaspore dispersal in the rooted species, with the anthocarp ridges possessing aerenchymatous tissue in most species. This character seems to greatly increase the dispersion range of most Pontederias.l. lineages that, unlike E. crassipes and Eichhornias.s., are not easily vegetatively dispersed by the fragmentation of floating stems. In E. crassipes, the plants are free-floating and can easily disperse in waterbodies with moving waters, while in Eichhornias.s., the plants have elongated stems, which possibly help diaspores to disperse further away from the mother plant’s base, thus decreasing parental/offspring competition.

Tristyly is an extremely rare type of heterostyly, recorded for a handful of families, only two being monocots (i.e. Amaryllidaceae and Pontederiaceae; Barrett 1993). According to Kohn et al. (1996), tristyly evolved only once in Pontederiaceae. As aforementioned, in Kohn et al. (1996), they recover part of the polyphyletic Eichhornia as sister to Heterantheras.l. and tristyly as a synapomorphy for Pontederiaceae as a whole, with four reversions to homostyly. However, we recover tristyly as a synapomorphy of Pontederias.l. alone, with only two reversions to homostyly. In E. diversifolia (Vahl) Urb. and E. natans (P.Beauv.) Solms, the flowers seem to be consistently pseudo-homostylous, which could be related to miniaturisation connected with these species’ floating growth-form (Barrett 1988). In Monochoria, there is a shift from tristyly to enantiostyly (i.e. two different types of heterostyly; Barrett 1993), that could be easily explained by the shift in the group’s pollination syndrome. Monochoria species are enantiostylous, lack septal nectaries and exclusively offer pollen as a floral reward (Wang et al. 1995) and this, most likely, is connected with the buzz pollination syndrome of their flowers. Furthermore, poricidal, basifixed, polymorphic anthers are typical to buzz-pollinated flowers (Cook 1989; Wang et al. 1995). This shift from nectar-flowers to pollen-flowers seems to be the main cause of the peculiar floral morphology and loss of tristyly in Monochoria.

In Pontederiaceae, three different patterns in perianth-lobe shape can be observed: (1) perianth lobes all equal, thus producing an actinomorphic perianth (e.g. H. dubia); (2) equal to subequal in the same whorl, producing either actinomorphic or zygomorphic perianths, depending on the presence of a nectar guide [e.g. actinomorphic in M. hastata (L.) Solms and zygomorphic in E. crassipes]; and (3) unequal perianth lobes, with more than one morph in the same whorl, producing strongly zygomorphic perianths (e.g. H. gardneri). In Commelinales, the perianth lobes pattern seems to be extremely variable, being equal in the same whorl in Hanguanaceae, unequal in Philydraceae (due to the fusion of three posterior lobes) and variable in Commelinaceae and Haemodoraceae (Pellegrini, pers. observ.). In Commelinaceae, sepals are almost invariably different from the petals, except in Palisota Rchb. ex Endl. in which the sepals are characteristically petaloid (Faden 1998). Furthermore, both sepals and petals can range from equal to unequal, producing strongly zygomorphic flowers (e.g. Aneilema R.Br., Commelina L., Polyspatha Benth.; Faden 1998). In Haemodoraceae, there is much variation in the shape of the perianth lobes (Simpson 1990, 1998). Nonetheless, equal perianth lobes seem to be plesiomorphic in the monocots (Sauquet et al. 2017; Stevens 2001–onwards) and dominant in the family, being recorded for 11 out of 14 genera (Pellegrini, pers. observ.). Thus, equal to subequal lobes in one perianth whorl (the apices are obtuse to round) is recovered by us as a homoplastic synapomorphy for Pontederias.l. (Fig. 1). The perianth in Pontederias.l. ranges from campanulate to infundibuliform to hypocrateriform, while in Heterantheras.l., it is almost exclusively tubular, a distinctive synapomorphy for the latter genus. The only exception is H. gardneri, which possesses an infundibuliform perianth and which might be explained by miniaturisation. In Philydraceae, the perianth is consistently infundibuliform, while the perianth in Haemodoraceae shows great plasticity, depending on the genus, ranging from flat to hypocrateriform to tubular to the peculiar split and falcate perianth of Anigozanthos (Simpson 1990, 1998).

Out of the four synapomorphies recovered for the E. paniculata group, two had been previously proposed by Eckenwalder and Barrett (1986; annual life cycle) and Barrett and Graham (1997; annual life cycle and the absence of clonal reproduction). All currently accepted species in this group are known to inhabit seasonal and, generally, short-lived waterbodies. Thus, the annual life cycle and the absence of clonal reproduction are more than expected. However, all previous studies in the family failed to notice the peculiarly inflated sheath of the leaf subtending the inflorescence and the flat basal bract (Fig. 5B). These characters are easily observed in E. paniculata and E. meyeri, due to their elongated inflorescences, while in E. paradoxa, the inflorescence has its internodes greatly contracted, thus making the flat basal bract extremely hard to observe, especially in dried specimens.

Monochoria comprises species with extremely autapomorphic morphology, being traditionally grouped based on their: pedicellate, actinomorphic and enantiostylous flowers, basally connate perianth and its basifixed and poricidal anthers (Cook 1989, 1998). Due to its enantiostylous flowers and basifixed anthers, Monochoria has traditionally been considered closely related to Heteranthera (Eckenwalder and Barrett 1986; Cook 1998). Nonetheless, molecular data provide strong support that Monochoria is instead sister to the clade composed of E. crassipes, Eichhornias.s. and Pontederias.s. (Graham and Barrett 1995; Kohn et al. 1996; Barrett and Graham 1997; Graham et al. 1998, 2002; Ness et al. 2011; this study). Aside from the six aforementioned synapomorphies, Monochoria is also supported in our present analysis by other six characters. Out of these characters, only the basal bract with a caudate apex was previously described as characteristic of Monochoria by Cook (1989). The presence of an inflated sheath in the leaf subtending the inflorescence, flat basal bract and fistulose main axis are shared between the E. paniculata group and Monochoria and are most likely plesiomorphic for Pontederias.l. The caudate apex in the basal bract is observed in all species of Monochoria. Nonetheless, M. korsakowii can also present a leaf-like basal bract (Cook 1989). The actinomorphic perianth is a result of the loss of the nectar guide in this lineage which, as aforementioned, is directly related to the shift in pollination syndrome in the group. Additionally, other four floral modifications in Monochoria seem to be associated with this shift in the group’s pollination syndrome: (1) pedicellate, actinomorphic and enantiostylous flowers; (2) basally connate perianth (which helps to expose the stamens and allows the bees to properly visit the flowers); (3) unequal, basifixed and poricidal anthers; and (4) the loss of septal nectaries. The presence of a petalo-staminal tube is also unique in the family and most probably is the result of the reduction of the length of the hypanthium. Finally, the thickened and ridged anthocarps are also observed in the E. paniculata group and Pontederias.s., being directly related to the fruits primary hydrochoric dispersal syndrome (see comment above).

Despite being well-known, E. crassipes possesses the most peculiar vegetative morphology in the polyphyletic Eichhornia and one of the most peculiar in the family as a whole. It is so peculiar that specimens are easily identified, even when lacking any reproductive structures (Pellegrini and Horn, pers. observ.). It is the only species in the family to possess a free-floating growth form, the only one to produce stolons and the only one to possess inflated petioles. Nonetheless, one of the most peculiar characters in E. crassipes has been greatly disregarded by most specialists in the family. Castellanos (1958) was one of the first to properly describe and illustrate the flabellate ligules of E. crassipes. All synapomorphies recovered for E. crassipes seem to be directly related to its peculiar free-floating growth form, which also enabled it to become the most troublesome weed of the world (Gopal and Sharma 1981). The morphology of Eichhornias.s. is clearly a result of its floating growth form and the tendency of these plants to grow in deeper water bodies. The late-deciduous sessile leaves (sometimes persistent for most of the plant’s adult life) are characteristic of this group, but especially striking in E. diversifolia, hence its name. This protraction of the submerged phase seems to give the species in this clade a clear developmental advantage by helping them to reach the water surface and produce enough petiolate leaves to allow them to float properly. Furthermore, the even arrangement of the petiolate leaves along the mature stem might help provide the needed stability to the elongated floating stem.

From all the recovered clades in Pontederias.l., Pontederias.s. goes hand-in-hand with Monochoria in the number of reproductive synapomorphies. Out of the eight recovered synapomorphies for this clade, six are reproductive, with only the presence of epithelial cells in the septae, which are shared with Monochoria, being homoplastic. All the remaining five reproductive synapomorphies are directly correlated, but their evolutionary chronology is much harder to infer. The most parsimonious view is probably that all characters were triggered concomitantly by the appearance of the pseudomonomerous ovary, which caused the change in placentation morphology and ovule number. The abortion of most of the gynoecium might have caused a key shift in the reproductive strategy in this lineage from investing in a great number of small seeds with little chance of reaching maturity, to investing into a single big seed with a good amount of provision and guaranteeing that it has a greater chance of reaching maturity. The smooth testa seems to be a simple byproduct of negative selection of ornamentation, since the seeds stopped being individually dispersed with the change of reproductive strategy. Finally, the achene gives this lineage a great evolutionary advantage since it is easily dispersed by water, with a long floatation period due to its thick parenchymatous walls. Furthermore, many species also possess complex ornate achenes, with teeth and spikes that efficiently stick to fur, feathers, fabric etc., most likely having animals as their primary dispersers (Pellegrini, pers. observ.).

With the present recircumscription of Pontederia, Pontederiaceae now is organised in two monophyletic genera (i.e. Heteranthera and Pontederia). As stated by Pellegrini (2017a) and corroborated by nine phylogenetic studies (Eckenwalder and Barrett 1986; Graham and Barrett 1995; Kohn et al. 1996; Barrett and Graham 1997; Graham et al. 1998, 2002; Ness et al. 2011; this study), the recognition of two genera seems to be the best and most taxonomically conservative option available, since it avoids the description of new genera and the reestablishment of names that were rarely, if ever, used in any relevant taxonomic or floristic study. Finally, this option makes the differentiation of the two accepted genera easy, using either fresh, liquid or herbarium samples. Thus, the genera of Pontederiaceae can be differentiated using the key below: