The present study samples 60 taxa from the Tradescantia alliance (as circumscribed by Hertweck and Pires 2014), including all four genera currently accepted for subtribe Tradescantiinae and two genera from subtribe Thyrsantheminae (i.e. Elasis and Tinantia Scheidw., out of six genera). The ingroup includes 42 species of Tradescantia (ca. 90 species total in the genus) and five species of Gibasis (11 spp.). The outgroup is represented by five species of Callisia (20 spp., with representatives from sections Leptocallisia and Callisia), five species of Tripogandra (ca. 22 spp.) and the monospecific Elasis. The analysis is rooted in two species of Tinantia (ca. 13 spp.), since the genus is consistently recovered as the first lineage to diverge in the Tradescantia alliance (Wade et al. 2006; Burns et al. 2011; Hertweck and Pires 2014). Type species from all sections and series currently accepted for Tradescantia and Gibasis were sampled, with the exception of the type species of Gibasis [i.e. G. pulchella (Kunth) Raf., not available for analysis when this study was carried out]. Aside from this, the sampling of this study aimed to represent the morphological and diversity centers in each genus, especially Tradescantia. I have studied at least five specimens for each species, with the most representative specimen chosen as the voucher (Table 1). I was also unable to sample Weldenia Schult.f. and Thyrsanthemum Pichon due to both being endemic to Mexico and represented by few specimens in most herbaria, especially in South American collections. Since both genera are phylogenetically well-placed and supported based on molecular data, and are morphologically quite distinctive from the studied genera, including them in the present analysis is dispensable, since it would only increase the degree of homoplasy and phylogenetic noise in the analysis. For the same reason, Callisia warszewicziana (Kunth & C.D.Bouché) D.R.Hunt, which was originally sampled, was excluded from the final analysis.

Characters were scored mainly from living specimens at the field and specimens kept at the Jardim Botânico do Rio de Janeiro greenhouses, and later complemented by spirit and herbarium samples from the following herbaria: ALCB, B, BA, BHCB, BHZB, BM, BOTU, BRIT, C, CAL, CEPEC, CESJ, CGE, CGMS, CNMT, COR, CORD, CVRD, EAC, ESA, F, FCAB, FCQ, FLOR, FUEL, FURB, GUA, HAMAB, HAS, HB, HBR, HDCF, HRB, HRCB, HSTM, HUCS, HUEFS, HUFSJ, HURB, IAC, ICN, INPA, JOI, K, L, MBM, MBML, MG, MO, MY, NY, P, PACA, PMSP, R, RB, RFA, RFFP, SCP, SP, SPF, SPSF, U, UEC, UFRN, UPCB, US, W, WAG, and WU (herbaria acronyms according to Thiers, continuously updated). Many additional specimens were examined during collections made on expeditions in Brazil and in the USA, between 2010–2016. When living or herborized specimens were not available for examination, information was taken from published literature (Table 2). A limited number of characters used here have been analyzed in previous studies (i.e. Evans et al. 2000; Panigo et al. 2011), with most characters being coded and scored for the first time in 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 (partly illustrated in Fig. 1), palynological, anatomical characters (illustrated in Fig. 2), phytochemical and cytological characters. A total of 114 discrete characters were scored, being: 97 macromorphological, three micromorphological, two palynological, three anatomical, four cytological, and five phytochemical (Suppl. material 1). All the characters were treated as unordered and were equally weighted. The terminology for indumentum and shapes follows Radford et al. (1974); inflorescence terminology follows Weberling (1965, 1989) and Panigo et al. (2011); stigmatic micromorphology terminology follows Owens and Kimmins (1981) and Owens et al. (1984); fruit terminology follows Spjut (1994); seed terminology follows Faden (1991); phytochemical terminology follows Martínez and Martínez (1993); cytology terminology follows Jones and Jopling (1972) and Martínez and Ginzo (1985); pollen terminology follows Poole and Hunt (1980); anatomical terminology follows Tomlinson (1966, 1969); and general macromorphological terminology follows Pellegrini (2015).

Figure 1. 

Some macromorphological characters used in the phylogenetic analysis. A subpetiolate leaf (Character 8) and asymmetrical base (Character 16), in Tradescantia tenella Kunth B complicate leaves (Character 8), in Tradescantia crassula Link & Otto. C impressed secondary veins (Character 19), in Tradescantia fluminensis Vell D predominantly axillar to thyrse-like synflorescence (Character 24), in Callisia repens (Jacq.) L. E synflorescence with two paraclades (Character 26), in Tradescantia zanonia (L.) Sw. F contracted cincinni (Character 34), fused back to back (Character 35), vestigial cincinni bracts (Character 38), flower display of 60° (Character 48), shorter antesepalous stamens (Character 72), sigmoid filaments (Character 73), and zygomorphic androecium (Character 76), in Tripogandra diuretica (Mart.) Handlos G supernumerary cincinni bracts (Character 37), in Tradescantia praetermissa M.Pell H cincinni bracts saccate at base (Character 43), tubular flower (Character 47), fused petals (Character 60), clawed petals (Character 62), shorter antesepalous stamens (Character 72), connective expanded and transversally linear (Characters 77–80), round anther sacs (Characters 81–82), pollen white in vivo (Character 83), and trilobate stigma (Character 91), in Tradescantia zebrina Heynh. ex Bosse. I tubular flower (Character 47), pedicels geniculate at anthesis and pre-anthesis (Character 51), fused sepals (Character 53), filaments bearded with sparse and short hairs at mid-length (Characters 66–71), shorter antesepalous stamens (Character 72), connective expanded and transversally linear (Characters 77–80), round anther sacs (Characters 81–82), pollen white in vivo (Character 83), and trilobate stigma (Character 91), in T. zanonia J sepals all keeled (Character 56), in T. fluminensis K filaments basally bearded with dense and long hairs (Characters 66–71), connective expanded and rhomboid (Characters 77–80), anther sacs ellipsoid (Characters 81–82), and pollen yellow in vivo (Character 83), in T. fluminensis L pistil longer than the androecium (Character 86) and punctate (Character 91), in Tradescantia cerinthoides Kunth. All photos by M.O.O. Pellegrini, except G by H. Huaylla.

Figure 2. 

Anatomical characters used in the phylogenetic analysis. A leaf epidermis with silica crystals in specialized cells with thickened cell walls (Character 103–104), in Callisia multiflora (M.Martens & Galeotti) Standl B leaf epidermis with silica crystals in specialized cells without thickened cell walls (Character 103–104), bundle sheath in the mesophyll with longitudinal sclerenchymatic extensions (Character 105), in Gibasis pellucida (M.Martens & Galeotti) D.R.Hunt. C detail of the silica crystals in the leaf epidermis, in Tripogandra aff. glandulosa D raphides inside the raphide canals, evidencing the different morphology and position from the silica crystals in the leaf epidermis, in G. pellucida. All photos by S. Yankowski & F.B. Faden; A based on Spencer 92-308 (US), B, D based on Rosen 4645 (US), C based on Bogner 2381 (US).

Data was entered into a matrix of characters per taxa using the software Mesquite 3.20 (Maddison and Maddison 2017; Suppl. material 2). 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 using character optimization of ACCTRAN (accelerated transformation optimization; Swofford and Maddison 1987). 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. Bremer Index (BI) was also used to evaluate clade reliability based on the presence of secondary homologies (Bremer 1994). 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 majority-rule (50% values) and strict consensus trees. The complete data matrix and trees are available at TreeBase (http://purl.org/phylo/treebase/phylows/study/TB2:S21372).

Gibasis is recovered as polyphyletic, with all sampled species recovered in a polytomy in the strict consensus (Fig. 3), or placed in a basal grade in the majority-rule (Figs 4A, 5). In the strict consensus, Tripogandra (sensuHandlos 1975) was recovered as paraphyletic (BS= 80; BI= 3), due to the inclusion of C. filiformis (M.Martens & Galeotti) D.R.Hunt (Fig. 3). However, in the majority-rule, Tripogandra is recovered as monophyletic, with C. filiformis sister to it, but without statistical support (Figs 4A, 5 clade G). The Tripogandras.l. clade (i.e. including C. filiformis) is supported by a 60° torsion in the floral display (Character 48), petals ranging from pink to lilac to purple (Character 64, homoplastic), antesepalous filaments shorter than the antepetalous (Character 72, homoplastic), filaments sigmoid at anthesis (Character 73, homoplastic), filaments sigmoid at post-anthesis (Character 74, homoplastic), chromosome count of n= 8 (Character 106), and medium-sized chromosomes (Character 107, homoplastic) (Figs 3, 4A, 5 clade G). Consequently, Callisia is also recovered as paraphyletic (BS= 85; BI= 3), due to the inclusion of C. filiformis in Tripogandra. Callisias.s. is supported in the present analysis by: leaves congested at the apex of the stems (Character 7, homoplastic), inflorescences mainly axillary, producing a raceme-like synflorescence (Character 24, homoplastic), bracteoles conspicuous (Character 45, homoplastic), pedicels apically non-gibbous (Character 49, homoplastic), penicilliform stigma (Character 91), stigmatic papillae longer than 1μm (Character 93, homoplastic), and by the absence of sulphated phenolic acids (Character 114, homoplastic). In the strict consensus, the clade composed by Tripogandras.l.+Callisia(BS= 90; BI= 8) is supported by a combination of 20 characters, the non-homoplastic ones being: semicircular to flabellate connectives (Characters 78 and 80); seeds with reticulate to foveolate testa (Character 100); and leaf epidermis possessing specialized cells with thickened walls carrying silica bodies (Character 104) (Figs 3, 4A, 5 clade F). This clade is recovered inside a polytomy, together with Tradescantias.s. and the polyphyletic Gibasis (BS= 66; BI=5). The polytomy composed of Tradescantias.s.+Gibasis grade+(Tripogandras.l.+Callisia) is sister to Elasiss.l. (Fig. 3).

Tradescantias.s. was recovered arranged in five well-supported clades, with the innermost clades herein called Core Tradescantia (Figs 3, 4A, 5 clade L), which possesses most of the genus morphological diversity and includes most of its species richness. The first lineage to diverge in Tradescantias.s. represents T. sect. Austrotradescantia (sensuPellegrini 2015). The section is recovered as monophyletic (BS= 94; BI= 5), being supported by: sepals elliptic to broadly elliptic (Character 55), all keeled (Character 56); filaments basally, densely bearded (Characters 66–69 and 67–70, homoplastic) with long moniliform hairs (Characters 68 and 71); style obconic at base (Character 89) and conic at apex (Character 90), stigma punctate (Character 91) with type D papillae (Character 92); seeds with costate testa (Character 100); chromosome count of n=10–numerous (Character 106), bimodal chromosomes (Character 108, homoplastic); and the presence of 1 C-glycosides (Character 111, homoplastic) (Figs 3, 4A & 5 clade I). The section is further divided into two fairly-supported clades (Figs 3 & 4A). The first one has medium support (BS= 66; BI= 3) and was named by Pellegrini (2015) as the T. fluminensis group, being characterized by: prostrate (Character 3, homoplastic), herbaceous stems (Character 4, homoplastic), chartaceous or membranous leaf-blades (Character 10, homoplastic), saccate cincinni bracts (Character 43, homoplastic), the presence of 6-hydroxy-luteine (Character 113, homoplastic), and by the absence of sulphated phenolic acids (Character 114, homoplastic). The majority-rule recovers the T. fluminensis group arranged in two morphological complexes (Fig. 4A), that can be interpreted as: (1) the T. tenella species complex (BS= 57), being characterized by its rugose seeds (Character 100, homoplastic), and hilum shorter than ½ the length of the seed (Character 102, homoplastic); (2) and as the T. fluminensis species complex, being characterized by plants with indefinite base (Character 2, homoplastic), and equal cincinni bracts (Character 40, homoplastic). The second clade has high statistical support (BS= 90; BI= 4) and was named by Pellegrini (2015, 2016) as the T. crassula group, being characterized by its: stems unbranched to branched only at base (Character 5, homoplastic), conduplicate and/or falcate leaf-blades (Character 11, homoplastic), inconspicuous secondary veins (Character 21, homoplastic), pistil longer than the stamens (Character 86, homoplastic), seeds cleft towards the embryotega (Character 96, homoplastic), and hilum longer than ½ the length of the seed (Character 102, homoplastic).

Figure 4. 

Congruence between morphological and molecular datasets. A, majority-rule tree showing the sections and series proposed by Hunt (1975, 1980, 1986b) color-coded; the five newly proposed subgenera are represented by the black bars; Tradescantia guatemalensis C.B.Clarke ex Donn.Sm. is depicted in red, to highlight its placement as sister to Elasis hirsuta (Kunth) D.R.Hunt; the ● represents Tradescantias.s.; Bremer Index support values are depicted over the branches, while bootstrap support values are depicted under the branches B simplification of the current hypothesis for phylogenetic relationships in tribe Tradescantieae, based on molecular data. Thyrsanthemum Pichon and Weldenia Schult.f. are depicted in red since they were not sampled in the present study. Modified from Hertweck and Pires (2014).

The second lineage in Tradescantia is highly supported and here named the Campelia clade (BS= 91; BI= 6), being composed by T. sect. Campelia, T. sect. Corinna, T. sect. Cymbispatha, T. sect. Rhoeo, and T. sect. Zebrina (Figs 3, 4A, 5 clade K). The clade is supported by: synflorescence with one or more coflorescences (Character 26), presence of peduncle bracts (Character 29), presence of supernumerary bracts (Character 37, homoplastic), spathaceous cincinni bracts (Character 39, homoplastic); flowers with pedicels geniculate at anthesis and pre-anthesis (Character 51), unequal sepals (Character 54), androecium with filaments from outer series shorter than the inner (Character 72, homoplastic), pollen white in vivo (Character 83, homoplastic), pistil longer than the stamens (Character 86, homoplastic), and semilateral embryotega (Character 98, homoplastic) (Figs 3, 4A). The monospecific T. sect. Rhoeo is recovered as an independent lineage at the base of the Campelia clade, sister to two clades. One of these clades represents a monophyletic T. sect. Cymbispatha (sensuPellegrini et al. 2016), herein called the T. commelinoides group, being well-supported (BS= 76; BI= 5) and characterized by: herbaceous stems (Character 4, homoplastic); distichously-alternate leaves (Character 6, homoplastic), with membranous leaf-blades (Character 10, homoplastic), and acute apex (Character 20, homoplastic); equal (Character 40, homoplastic), basally fused (Character 42), not overlapping cincinni bracts (Character 44, homoplastic); keeled dorsal sepal (Character 56, homoplastic), chartaceous sepals (Character 57, homoplastic), rotund to rhomboid petals (Character 61, homoplastic), and pistil shorter than the stamens (Character 86, homoplastic) (Figs 3, 4A). Tradescantia sect. Cymbispatha is recovered as sister to a small, well-supported (BS= 79; BI= 4) clade, herein called the T. zebrina group, composed by T. sect. Campelia, T. sect. Corinna, and T. sect. Zebrina. Tradescantia sect. Campelia and T. sect. Corinna are monospecific, being recovered in the strict consensus in a polytomy with both sampled species of T. sect. Zebrina (Fig. 3), thus making the section paraphyletic. In the majority-rule (Fig. 4A), T. sect. Zebrina is also recovered as paraphyletic, due to the inclusion of T. soconuscana Matuda, the sole species of T. sect. Corinna. The T. zebrina group is supported by: leaf-blades abaxially velutine (Character 16, homoplastic); flat cincinni bracts (Character 40, homoplastic), with saccate base (Character 42, homoplastic); tubular flowers (Character 46, homoplastic), sepals basally to completely fused (Character 52), stigmatic papillae longer than 1μm (Character 92, homoplastic); and chromosomes with asymmetric complements (Character 107, homoplastic) (Figs 3, 4A). The sister relation between the T. zebrina and the T. commelinoides group is fairly well-supported (BS= 61; BI= 4), being sustained by: the presence of pubescence in the abaxial side of the leaves (Character 15, homoplastic), leaf-blades cuneate at base (Character 18, homoplastic); sessile to subsessile flowers (Character 49, homoplastic); connectives ranging from cordate to sagittate to linearly-tapered (Characters 77 and 79, homoplastic), rounded anther sacs (Characters 80 and 81, homoplastic); seeds with hilum equal to ½ the length of the seed (Character 100, homoplastic); and chromosome count n= 7 and 8, n= 5, 6, 10 and 11, due to Robertosnian Changes (Character 104) (Figs 3, 4A). The Campelia clade is recovered as sister to Core Tradescantia with high statistical support (BS= 92; BI= 8), and sustained by: spirally-alternate leaves (Character 6, homoplastic), leaf-blades with truncate to amplexicaulous (Character 18, homoplastic), symmetric base (Character 19, homoplastic), acuminate apex (Character 20, homoplastic); overlapping cincinni bracts (Character 43, homoplastic), conspicuous bracteoles sometimes completely involving the cincinnus (Character 44, homoplastic); membranous sepals (Character 56), stigmatic papillae equal or shorter than 1μm (Character 92, homoplastic); conspicuous embryotega (Character 97); equal or larger than 10µm (Character 105), symmetric chromosomes (Character 107) (Figs 3, 4A, 5 clade J).

As aforementioned, Core Tradescantia consists of three smaller clades (herein called the Mandonia, Setcreasea, and Tradescantia clades), restricted to drier environments (i.e. Seasonally Dry Tropical and Subtropical Forests) in the American continent. Core Tradescantia is well-supported (BS= 88; BI= 4; 5 clade L), and defined by: tuberous roots (Character 1, homoplastic); stems unbranched to branched only at base (Character 5, homoplastic); conduplicate leaf-blades (Character 11, homoplastic); petals ranging from lilac to purple or pink (Character 63, homoplastic), connectives quadrangular to rectangular to slightly curved (Characters 77 and 79), anther sacs C-shaped (Characters 80 and 81), and pistil the same length as the stamens (Character 85, homoplastic). In the strict consensus, all the three clades are recovered inside a polytomy (Fig. 3), while in the majority-rule, the Tradescantia clade is the first to diverge, being sister to the Mandonia and Setcreasea clades (Figs 4A, 5 clade N). The Tradescantia clade is well-supported (BS= 96; BI= 3), and composed by most species of T. sect. Tradescantia (i.e. series Virginianae and Tuberosae), thus including the type-species of the genus (i.e. T. virginiana L.). Nonetheless, T. sect. TradescantiasensuHunt (1980) is still recovered as paraphyletic, due to series Sillamontanae and Orchidophyllae being nested within the Setcreasea clade (Figs 3, 4A). The Tradescantia clade is supported by: linear leaf-blades (Character 9, homoplastic); pedicels apically non-gibbous (Character 49, homoplastic), filaments densely bearded with moniliform hairs (Characters 65 and 68, homoplastic), and stigmatic papillae restricted to the margins of the stigma (Characters 93) (Figs 3, 4A, 5 clade M). The sister relation between the Mandonia clade and Setcreasea clade has low statistical support (BS= 65; BI= 1), as evidenced by the polytomy recovered in the strict consensus tree. However, it is morphologically supported by: pubescent ovary with eglandular hairs (Character 84); hilum shorter than ½ the length of the seed (Character 100, homoplastic); production of hydroxy-luteolin (Character 108, homoplastic), the absence of 6-hydroxy-lutein (Character 111, homoplastic), and the absence of sulphated phenolic acids (Character 112, homoplastic) (Figs 3, 4A, 5 clade N). The Mandonia clade is a well-supported clade (BS= 91; BI= 4), composed by T. sect. Mandonia and T. sect. Parasetcreasea. It is supported by: leaf-blades abaxially pubescent (Character 15, homoplastic); inflorescences mainly axillary, producing a raceme-like synflorescence (Character 24, homoplastic), sessile main florescences (Character 28, homoplastic), the presence of supernumerary bracts (Character 36, homoplastic), reduced cincinni bracts (Character 37, reversion), cincinni bracts not overlapping (Character 43, homoplastic); chartaceous sepals (Character 56, homoplastic), filaments apically spirally-coiled at post-anthesis (Character 73), style ½ time longer than the stamens (Character 85), and style spirally-coiled at post-anthesis (Character 87) (Figs 3, 4A, 5 clade O). Finally, the Setcreasea clade is statistically well-supported (BS=92; BI=4), despite being defined exclusively by homoplastic characters. It is characterized by the combination of: leaf-blades with cuneate base (Character 18), acute apex (Character 20), inconspicuous secondary veins (Character 21); saccate cincinni bracts (Character 42); tubular flowers (Character 46), pedicel the same length as the floral buds (Character 49), hyaline sepals (Character 58), fused petals (Character 59), clawed petals (Character 61), epipetalous stamens (Character 74); and chromosomes medium-sided (i.e. bigger than 5µm and smaller than 10µm; Character 105) (Figs 3, 4A, 5 clade P). The Setcreasea clade is composed by T. sect. Separotheca, T. sect. Setcreasea, and the remaining species of T. sect. Tradescantia (i.e. series Sillamontanae and Orchidophyllae).

The present study features the most extensive sampling of Tradescantia and its relatives, in a phylogenetic analysis (almost 50% of the species currently accepted in the genus), and most of the morphological diversity in subtribe Tradescantiinae (sensuFaden and Hunt 1991). It is also the first phylogenetic study to sample all sections and series proposed by Hunt (1975, 1980, 1986b), including all type species for each infrageneric rank. This study is also the first morphologically based phylogeny for Tradescantia, and the first in the family to include macromorphological, anatomical, palynological, cytological, and phytochemical characters. A high degree of homoplasy is observable in the present dataset, based on the CI, RI, and RC indexes, being congruent with the scenario expected for Commelinaceae (Evans and Faden 1998; Evans et al. 2000). Nonetheless, contrary to results for the whole family of Evans and Faden (1998) and Evans et al. (2003) in which their morphologically derived topology was highly incongruent with the molecular dataset (Evans and Faden 1998; Evans et al. 2003), the herein presented relationships inside Tradescantia, are congruent with the ones previously recovered for the genus, based on molecular data (Burns et al. 2011; Hertweck and Pires 2014; Figs 4A,B, 5). The Bremer Index, which is used for the first time in phylogenetic analysis in Commelinales, also gives strong statistical support for most clades, despite the homoplasy in the present dataset. The incongruence between the morphological and molecular datasets observed by Evans and Faden (1998) and Evans et al. (2000, 2003), might be a reflection of the differentiated coding of some key morphological characters. This is mainly due to the dramatically different sampling between the present study (i.e. subtribal and generic level) and the studies by Evans and Faden (1998) and Evans et al. (2000, 2003; i.e. family and infrafamily level). Characters related to inflorescence architecture and androecium morphology, were key in resolving the backbone of the present analysis. Furthermore, micromorphological, anatomical, palynological, cytological, and phytochemical characters were also essential in giving support to the backbone of the herein presented topology. From a wider perspective and despite the high degree of homoplasy, it is not unexpected for well-coded morphological characters to be highly congruent to molecular datasets. In Commelinales, Haemodoraceae and Pontederiaceae have yielded similar results regarding the congruence between different datasets. In Haemodoraceae, the morphological phylogeny by Simpson (1990) is widely corroborated by the molecular phylogeny by Hopper et al. (2009), with recent anatomical data Aerne-Hains and Simpson (2017) further increasing the congruence between morphology and molecular data. In Pontederiaceae the congruence is even clearer, where all phylogenies for the family published so far (i.e. Eckenwalder and Barrett 1986; Graham and Barrett 1995; Kohn et al. 1996; Barret and Graham 1997; Graham et al. 1998; Ness et al. 2011) recover the same evolutionary history, regardless of the dataset (Pellegrini 2017).

As aforementioned, the results of the present study are highly congruent with previous phylogenetic studies. Nonetheless, they still differ significantly from the previous ones, regarding generic limits and relationships. Similar to Hertweck and Pires (2014), the herein presented analysis recovers Tradescantia as paraphyletic in its current circumscription. In Hertweck and Pires (2014), the paraphyly of Tradescantia is caused by two species of Gibasis. In the present analyses, the paraphyly of Tradescantia is due to the position of T. guatemalensis as sister to E. hirsuta. This is the first time T. guatemalensis is sampled in any phylogenetic study, but based on the protologue of T. sect. Coholomia (Hunt 1980) describing the cincinni as arranged side by side, it was clear that this species was placed in the wrong genus. Elasis, as described by Hunt (1978), is characterized by possessing free, elongated, non-geniculate and fasciculate cincinni (which can also be interpreted as several cincinni emerging side by side from the same node), six equal stamens, inconspicuous connectives, and truncate stigma. This description matches perfectly with the one of T. guatemalensis, with both species being differentiated solely on pubescence, inflorescence position, and flower color. Thus, it is clear that in order to maintain a monophyletic Tradescantia and a coherent Elasis, T. guatemalensis needs to be transferred to the later. Elasiss.l. can be differentiated from Tradescantias.s. by the following characters: 1–several, shortly-pedunculate, free, elongate cincinni (Characters 30, 31, 32, 34, 35), bracteose cincinni bracts (Character 38), anther connectives not expanded (Characters 77 & 79), silica crystals in specialized cells with thickened walls (Character 103 & 104), bundle sheath in the mesophyll with longitudinal sclerenchymatic extensions (Character 105) and chromosome number n= 4–5 (Character 106).

The present analysis was also unable to recover a monophyletic Gibasis. Nonetheless, in the present analyses Gibasis is not partially nested within Tradescantia, as recovered by Hertweck and Pires (2014), being actually distantly related to Tradescantia. In fact, the herein presented results suggest a closer relationship between Tradescantia and the Callisia+Tripogandra clade, based on inflorescence architecture. Tradescantia and Gibasis are morphologically, anatomically, cytologically, and phytochemically very distinct, and should be kept as distinct genera (Fig. 5). I believe that the present analysis was unable to recover a monophyletic Gibasis due to the limited sampling of this morphologically diverse genus in the present study. The present sampling of Gibasis had solely the intention of testing its relationship to Tradescantia, and not its monophyly. According to the simulations of Hillis (1996), the greater sampling for a certain group, the higher its statistical support and phylogenetic resolution. Furthermore, I was unable to sample the type species of Gibasis (i.e. G. pulchella) in the present study. Thus, I believe that further analysis focused on Gibasis and its systematics, might yield different results. Aside from recovering the genus as monophyletic, it would possibly recover the geniculate cincinni as a synapomorphy for Gibasis, as hypothesized by Hunt (1986a).

Figure 5. 

Majority-rule tree showing the relation between the genera in the Tradescantia alliance, highlighting the reproductive synapomorphies recovered for each lineage. Synapomorphies for Tinantia (clade A): cincinni verticillate, flowers zygomorphic, filaments sigmoid, anthers dimorphic with inconspicuous connectives, style sigmoid, stigma truncate, and hilum C-shaped. Synapomorphy for clade B: basal bract inconspicuous and tubular. Synapomorphies for Elasiss.l. (clade C): cincinni fasciculate, flowers actinomorphic, filaments sigmoid, anthers dimorphic, style sigmoid, and stigma truncate. Synapomorphies for clade D: double-cincinni fused back to back. Synapomorphies for clade E: seeds triangular to round triangular or tetrahedral, ventrally ridged, hilum elliptic or punctate, and leaf epidermis with silica crystals in specialized cells with thickened cell walls. Synapomorphies for Callisias.s. (clade F): inflorescences mainly axillary, bracteoles conspicuous, pedicels apically non-gibbous, and penicilliform stigma. Synapomorphies for Tripogandras.l. (clade G): floral display with a 60° torsion, petals ranging from pink to lilac to purple, antesepalous filaments shorter than the antepetalous, and filaments sigmoid at anthesis and post-anthesis. Synapomorphies for Tradescantias.s. (clade H): double-cincinni fused back to back; frondose cincinni bracts, pedicels deflexed at post-anthesis, seeds elliptic to oblong in outline, ventrally flattened, hilum linear and longer than ½ the length of the seed, leaf epidermis lacking silica bodies in specialized cells, and diffuse bundle sheath in the mesophyll. Synapomorphies for T. subg. Austrotradescantia (clade I): sepals elliptic to broadly elliptic, all keeled, filaments basally and densely bearded with long moniliform hairs, style obconic at base and conic at apex, and stigma punctate with type D papillae. Synapomorphies for clade J: overlapping cincinni bracts, conspicuous bracteoles sometimes completely involving the cincinnus, membranous sepals, stigmatic papillae equal or shorter than 1μm, and conspicuous embryotega. Synapomorphies for T. subg. Campelia (clade K): synflorescence with one or more coflorescences, presence of peduncle bracts, presence of supernumerary bracts, spathaceous cincinni bracts, flowers with pedicels geniculate at anthesis and pre-anthesis, unequal sepals, androecium with filaments from external series shorter than the internal, pollen white in vivo, pistil longer than the stamens, and semilateral embryotega. Synapomorphies for Core Tradescantia (clade L): petals ranging from lilac to purple or pink, connectives quadrangular to rectangular to slightly curved, anther sacs C-shaped, and pistil the same length as the stamens. Synapomorphies for T. subg. Tradescantia (clade M): pedicels apically non-gibbous, filaments densely bearded with moniliform hairs, and stigmatic papillae restricted to the margins of the stigma. Synapomorphies for clade N: ovary pubescent with eglandular hairs; hilum shorter than ½ the length of the seed. Synapomorphies for T. subg. Mandonia (clade O): inflorescences mainly axillary, sessile main florescences, the presence of supernumerary bracts, reduced cincinni bracts, cincinni bracts not overlapping, chartaceous sepals, filaments apically spirally-coiled at post-anthesis, style ½ time longer than the stamens, and style spirally-coiled at post-anthesis. Synapomorphies for T. subg. Setcreasea (clade P): saccate cincinni bracts, tubular flowers, pedicel the same length as the floral buds, hyaline sepals, fused petals, clawed petals, and epipetalous stamens.

Callisia is a historically challenging group in Commelinaceae, especially regarding its taxonomy (Hunt 1986c). Once again, the genus is recovered as non-monophyletic, with species being recovered together with Tripogandras.s. and with the different sections proposed by Hunt (1986c) being recovered as distinct lineages. The present topology is highly congruent with the molecular-based hypothesis presented by Bergamo (2003). In the Callisias.s. lineage recovered in the present study (i.e. Fig. 5 clade F), two morphological groups are clearly distinguishable. The first is represented in the present dataset exclusively by C. monandra (Sw.) Schult. & Schult.f., and can be characterized by its pedunculate double-cincinni, long-pedicellate flowers, glandular-pubescent pedicels and sepals, androecium reduced to 1–3, antesepalous stamens, anthers with elongate anther sacs, inconspicuous connectives, and sessile stigmas, being equivalent to C. sect. Leptocallisias.s. [i.e. Aploleia Raf. sensuMoore Jr. 1961, with the addition of C. cordifolia (Sw.) E.S. Anderson & Woodson]. The remaining species sampled in the present study represent C. sect. Callisia, due to the inclusion of the type species [i.e. C. repens (Jacq.) L.]. This lineage is characterized by its sessile double-cincinni, short-pedicellate flowers, sepals glabrous or with eglandular hairs, androecium with (3–)6 stamens, when reduced to 3 stamens than antepetalous, round anther sacs, expanded flabellate or sagittate connectives, and elongated styles. Based on the molecular-based results by Bergamo (2003), combined with the herein presented morphological results, I suggest that both groups should probably be recognized as distinct genera in the near future (Pellegrini in prep.). Aside from that, other lineages of Callisias.l. recovered by Bergamo (2003) and Hertweck and Pires (2014) possess specific morphologies (Fig. 6D, G), but were not sampled in the present study. In the same way as Callisias.s. and Aploleia, these lineages also need to be recognized at the generic level, in order to maintain a morphologically cohesive and monophyletic Callisia (Pellegrini in prep.).

As aforementioned, C. warszewicziana was excluded from this analysis due to the great noise this extremely autapomorphic species created. The exclusion of C. warszewicziana caused no changes in the backbone of the analysis, but considerably increased the statistical support for all branches. As recovered by Bergamo (2003) and Hertweck and Pires (2014), C. warszewicziana is nested within the Callisia/Tripogandra generic complex, but morphologically deviant from all remaining species (Fig. 6F). Ongoing phylogenetic studies in the group seem to reveal the need to reestablish Hadrodemas H.E.Moore and other satellite genera lumped in Callisias.l. by Hunt (1986c), in order to retain morphologically cohesive and monophyletic genera in subtribe Tradescantiinae (Pellegrini in prep.). Tripogandra as circumscribed by Handlos (1975) is also recovered as paraphyletic. This result is not surprising, since Bergamo (2003) had already recovered part of C. sect. Leptocallisia [i.e. C. gracilis (Kunth) D.R.Hunt] nested within Tripogandra. Callisia gracilis and C. filiformis both share the typical Tripogandra-type inflorescence pattern, with the cincinni bracts reduced to only a membranous crest (i.e. vestigial), the 60°-degree torsion in floral display, the outer whorl of stamens shorter than the inner whorl, and basic chromosome count of n= 8 (Fig. 5 clade G; pers. observ.). Aside from these species, C. ciliata Kunth is also morphologically similar, being a putative candidate for being placed in Tripogandras.l. Pollen morphology should play an important role into solving this problem, since Tripogandra is known to be the only genus in the family with granular-verrucose tectum (Poole and Hunt 1980). I believe that with an improved sampling of Callisia and Tripogandra, it would be possible to shed some light on this intricate group and potentially solve this taxonomic impasse. Nonetheless, I also believe it would be precocious to implement taxonomic changes in the Callisia/Tripogandra complex in the present study, without further studies focusing in the group.

Figure 6. 

Floral morphology of subtribe Tradescantiinaes.l.A Tinantia Scheidw B Thyrsanthemum Pichon C Weldenia Schult.f. D–I Callisia Loefl. s.l.: D Cuthbertia Small E Aploleia Raf. (i.e. C. sect. Leptocallisias.s.) F Callisias.s. (i.e. C. sect. Callisia) GCallisia sect. Brachyphylla D.R.Hunt H Hadrodemas H.E.Moore (i.e. C. sect. Hadrodemas) I–J Tripogandra Raf. s.l.: ICallisia sect. Leptocallisiapro parteJ Tripogandras.s.K–L Elasis D.R.Hunt: K E. hirsuta (Kunth) D.R.Hunt L E. guatemalensis C.B.Clarke ex Donn.Sm.) M.Pell M Matudanthus D.R.Hunt N Gibasis Raf. O–T Tradescantia L. emend. M.Pell.: OT. subg. AustrotradescantiaP–QT. subg. CampeliaRT. subg. MandoniaST. subg. SetcreaseaTT. subg. Tradescantia. A by E. Barbier, B, L, P by P. Acevedo-Rodriguez, C by S. Cross, D by D. Rankin, E by J. Amith, F, J, N–O, Q–T by M.O.O. Pellegrini, G by S. Eduardo, H by C. Willemsen, I by B.E. Hammel, K by A. Kay, and M by A. Garcia Mendoza.

Subtribe Thyrsantheminae has been consistently recovered as polyphyletic by all morphological and molecular phylogenies so far (Evans et al. 2000, 2003; Wade et al. 2006; Burns et al. 2011; Zuiderveen et al. 2011; Hertweck and Pires 2014; Pellegrini et al. unpublished data). Additionally, the subtribe completely lacks any kind of micro- or macromorphological synapomorphies. This should not be unexpected, since Faden and Hunt (1991) when formally proposing the subtribe, comment on the marked heterogeneity of the group. The Tradescantia alliance, as proposed by Hertweck and Pires (2014), represents the exclusively Neotropical crown group in tribe Tradescantieae, being composed by all genera of the non-monophyletic subtribes Thyrsantheminae and Tradescantiinaes.s. This clade can be uniquely defined by the presence of pollen grains with rugose to rugose-insulate tectum as its synapomorphy, and originating the specialized tectum ornamentation of genera such as Callisia, Tinantia and Tripogandra (Poole and Hunt 1980; pers. observ.). As originally envisioned by Hunt (1983), subtribe Thyrsantheminae was characterized as possessing: free, stipitate and non-geniculate cincinni, sometimes reduced to 1–3-flowered and sessile cincinni; actinomorphic flowers; and six, fertile and equal stamens with expanded connectives. This diagnosis only needed to be slightly expanded by Faden and Hunt (1991) in order to include Weldenia, but due to the inclusion of Tinantia this diagnosis was expanded to also include zygomorphic flowers, and unequal stamens with inconspicuous connectives (Faden and Hunt 1991; Hunt 2015a,b,c). Even in its original circumscription (i.e. Hunt 1983), subtribe Thyrsantheminae still represented a polyphyletic and heterogeneous assemblage, with different genera possessing either alternate or fasciculate cincinni, inconspicuous or expanded connectives (the connectives of Elasis and Matudanthus D.R.Hunt are not expanded, and the connectives of Thyrsanthemum are very seldom so), and dorsal or semilateral embryotega (pers. observ.). Furthermore, based on the diagnosis provided by Faden and Hunt (1991), it is unclear why Gibasis was not included in this subtribe, instead of being included in Tradescantiinae. Gibasis lacks the diagnostic double-cincinni of TradescantiinaesensuFaden and Hunt (1991), matching perfectly the diagnosis of Thyrsantheminae. Based on morphological and molecular data (Evans et al. 2000, 2003; Wade et al. 2006; Burns et al. 2011; Zuiderveen et al. 2011; Hertweck and Pires 2014; Pellegrini et al. unpublished data; this study), Gibasis is clearly more closely related to Elasis than it is to Tradescantia, as shown by Hertweck and Pires (2014). Furthermore, Matudanthus is unambiguously closely related to Elasis, being differentiated exclusively by the presence of tuberous roots, sessile cincinni, and tannin cells in the petals (Hunt 1978; pers. observ.). When finally sampled in a phylogenetic study, it is expected for Matudanthus to emerge as part of the Gibasis+Elasis clade, due to its inflorescence architecture, floral morphology, seed morphology, pollen features and anatomical pattern. Another monospecific genus not sampled in any phylogenetic study so far, Gibasoides D.R.Hunt, is strikingly similar to Thyrsanthemum. Both genera can only be differentiated based on the development of the main axis of the inflorescence, which produces a perfect thyrse with elongate main axis and alternate cincinni in Thyrsanthemum, and an umbellate-thyrse with inconspicuous main axis and verticillate cincinni in Gibasoides (Hunt 1978). Whether it will be recovered as sister to Thyrsanthemum or nested within the latter when included in a phylogenetic study, remains to be seen. For the time being, it is better to recognize them as distinctive genera. Finally, the last poorly-known genus in the Tradescantia alliance, Sauvallea C.Wright ex Hassk., received recent attention by Hunt and Arroyo-Leuenberger (2015). The authors clarify several morphological features for the genus, but due to the probable extinction of the its sole species, S. blainii C.Wright ex Hassk., our knowledge on this taxon is still limited. The genus was treated as incertae sedis by Faden and Hunt (1991) and Faden (1998), being a putative member of either subtribe Tradescantiinae or Thyrsantheminae. Its inflorescence morphology is unique in the Tradescantia alliance due to its spathaceous basal bract, being reminiscent of Cyanotis D.Don s.l. (including Belosynapsis Hassk.; Hunt and Arroyo-Leuenberger 2015). However, Cyanotiss.l. possesses an exclusively Palaeotropical distribution, inflated filaments and styles, generally basally poricidal anthers, and apical embryotega. Thus, it is unlikely that Sauvallea proves to be a member of subtribe Cyanotinae, if it is ever recollected and/or sampled in a phylogenetic study (pers. observ.).

Inflorescence morphology has been widely used in the taxonomy of Commelinaceae, throughout the years (e.g. Clarke 1881; Brückner 1930; Pichon 1946; Hunt 1975, 1980, 1986a, 1986b, 1986c; Panigo et al. 2011) for the delimitation of different taxonomical ranks. The herein presented analysis corroborates that different inflorescence patterns are indeed recovered as synapomorphies for different lineages. Tinantia can be easily differentiated from ingroup genera by its leaf-like basal bract [that becomes somewhat spathaceous and conduplicate in species like T. anomala (Torr.) C.B.Clarke], pedunculate inflorescence with verticillate, free, and non-geniculate cincinni, and conspicuous and persistent bracteoles (Figs 3, 4A, 5 clade A). Based on the molecular results published so far (Bergamo 2003; Evans et al. 2003; Wade et al. 2006; Burns et al. 2011; Hertweck and Pires 2014), tubular and hyaline basal bracts seem to have evolved only once in Commelinaceae, being a putative synapomorphy for the clade composed by Callisias.l., Elasis, Gibasis, Tradescantia, and Tripogandras.l. (i.e. clade B, Fig. 5). Despite the limited sampling of the Tradescantia alliance in the present study, the recovered topology also supports this hypothesis (Figs 3–5). All remaining genera in the family have either leaf-like, bracteose, or spathaceous basal bracts (pers. observ.). In the clade composed by Thyrsanthemum+Weldenia (and most likely also including Gibasoides), the inflorescences are characterized by their alternate cincinni, subtended by linear bracts, and greatly reduced, generally absent or deciduous bracteoles (Hunt 1978, 2015b, c). If confirmed to be part of this clade, the inflorescence architecture of Gibasoides could be easily explained by the suppression of the main axis of the inflorescence, causing the alternate cincinni to acquire as pseudo-verticillate arrangement. This feature could easily derived from a inflorescence very similar to the one of Thyrsanthemum goldianum D.R.Hunt, which is the only species in the genus with long-stipitate cincinni, equivalent to the ones of Gibasoides laxiflora (C.B.Clarke) D.R.Hunt (Hunt 1978, 2015c; pers. observ.).

The double-cincinni fused back to back, seems to have evolved only once in Commelinaceae, recovered in the present study as a synapomorphy for the clade composed by Callisias.l., Tradescantia, and Tripogandras.l. (i.e. clade D, Fig. 5). Nonetheless, molecular based phylogenetic studies (i.e. Bergamo 2003; Evans et al. 2003; Wade et al. 2006; Burns et al. 2011; Zuiderveen et al. 2011; Hertweck and Pires 2014; Pellegrini et al. in prep.) recover the double-cincinni as a synapomorphy for subtribe Tradescantiinae (sensuFaden and Hunt 1991) plus Elasis (i.e. clade B, Fig. 5). In this scenario, the reversion from the double-cincinni to verticillate or fasciculate cincinni is synapomorphic for the clade formed by Gibasis+Elasis (and most likely also including Matudanthus). Despite being recovered in the present study as polyphyletic, Gibasis can be recognized by its main florescences with free, verticillate and geniculate cincinni (Hunt 1986a; Fig. 5). The double-cincinni is very conserved in this clade, however in C. warszewicziana the cincinni are commonly only basally fused, and can range from 1–2(–3). Some variation is also observed in Tradescantias.s., with some species sometimes producing abnormal inflorescences with alternate looking cincinni (e.g. T. cymbispatha, T. fluminensis, and T. zanonia), others commonly producing axillary inflorescences with a solitary cincinnus (e.g. T. crassula and all species in the Mandonia clade), and others producing inflorescences with 2–3(–5) cincinni, fused back to back (i.e. T. valida) (pers. observ.). As aforementioned, Tripogandras.l. can be easily defined by the extreme reduction of the cincinni bracts, coupled with its double-cincinni fused back to back, and the 60° torsion in floral display. The polyphyletic Callisias.l. also possesses several inflorescence characters that give morphological support to the different lineages recovered by molecular datasets (Bergamo 2003; Evans et al. 2003; Wade et al. 2006; Burns et al. 2011; Zuiderveen et al. 2011; Hertweck and Pires 2014; Pellegrini et al., in prep.), and on a smaller scale the results herein presented. Callisias.s. (i.e. C. sect. CallisiasensuHunt 1986c) could be easily characterized by its synflorescences with mainly axillary main florescences, sessile main florescences, cincinni bracts similar to the bracteoles, double-cincinni fused back to back, tubular flowers, white to hyaline petals, glabrous filaments, flabellate connectives, and penicilliform stigmas.

Sauvallea blainii must once again be considered a taxon of uncertain phylogenetic affinity regarding inflorescence morphology. As explained by Hunt and Arroyo-Leuenberger (2015), due to the paucity and poor preservation of available specimens, our knowledge on the genus will remain limited until it is, if ever, recollected. Nonetheless, Hunt and Arroyo-Leuenberger (2015) were able to describe the main florescence as being greatly reduced, apparently to a solitary cincinnus, with conspicuous and persistent bracteoles (which can be observed protruding from the spathaceous basal bract), and with each cincinnus producing most probably only one flower. Based on this description, Sauvallea could be more closely related to Tinantia, with which it shares the frondose (i.e. spathaceous or leaf-like) basal bract, and expanded and persistent bracteoles. The number of cincinni in Tinantia is variable within and between species, with several species known to produce solitary cincinni (Faden 1998). Furthermore, Sauvallea and Tinantia also share an annual life cycle, subequal to unequal petals, and six stamens bearded with moniliform hairs (pers. observ.). Alternatively, it is also possible for Sauvallea to represent a further reduction of the inflorescence architecture exhibited by Callisias.l. This scenario is supported by the description of Hasskarl (1870), in which he describes the androecium as similar to Tradescantia due to its six equal stamens, barbate with moniliform hairs, expanded connectives, and gynoecium similar to Callisia due to its bilocular and pilose ovary and elongate style. This description brings S. blainii considerably closer to C. cordifolia (i.e. AploleiasensuMoore Jr. 1961, or C. sect. LeptocallisiasensuHunt 1986c) due to floral morphology, which would explain the common misidentifications between the two species in Cuba (Hunt and Arroyo-Leuenberger 2015). Nonetheless, this relationship would require major changes in the inflorescence architecture (e.g. shortening of the peduncle, reduction of the double-cincinni to a solitary cincinnus accompanied by the expansion of the cincinni bract, and shortening of the pedicels). Both hypotheses remain to be tested once both genera are sampled in a comprehensive phylogenetic study.

When Hunt (1975) reduced Setcreasea to a section of Tradescantia, the author commented on the strong emphasis given by previous authors to sympetaly and epipetaly, questioning its systematic significance. As aforementioned, these characters can be found in other genera of Commelinaceae, as well as in three lineages of Tradescantia (i.e. the Campelia, Setcreasea, and Mandonia clades). Nonetheless, these two characters played a significant part in Hunt’s sectional treatment of Tradescantia. Furthermore, sympetaly has also been regarded by other Commelinaceae specialists as a key morphological character in the evolution and systematics of the group. It has been used as the defining character of tribes, subtribes, genera, and sections (e.g. Clarke 1881; Brückner 1930; Pichon 1946; Hunt 1975, 1980, 1986a, 1986b, 1986c). Despite his own critical view on sympetaly, Hunt (1975, 1980, 1986b) also gave great weight to this character, making it decisive in differentiating some of the sections in Tradescantia, such as: (1) T. sect. Mandonia vs. T. sect. Parasetcreasea; and (2) T. sect. Campelia + T. sect. Corinna vs. T. sect. Zebrina. Nonetheless, systematic studies have shown that this character probably evolved several times in the family (Bergamo 2003; Evans et al. 2003; Wade et al. 2006; Burns et al. 2011; Zuiderveen et al. 2011; Hertweck and Pires 2014). Furthermore, my observations have revealed that sympetaly does not seem to be constant in the same evolutionary lineage, suffering several reversions along the way (Figs 3, 5). Finally, T. pallida produces flowers either with conate or free petals, sometimes in the same individual, which gives further support to my interpretation.

Epipetaly is by far one of the least common and less studied characters in Commelinaceae. It is found exclusively in some species of Tradescantia and Weldenia. In Tradescantia, epipetaly is only found in 12 species from six sections (i.e. corresponding to three clades in our analysis): T. andrieuxii C.B.Clarke, T. brevifolia (Torr.) Rose, T. buckleyi (I.M.Johnst.) D.R.Hunt, T. hirta D.R.Hunt, T. leiandra Torr., T. mirandae Matuda, T. orchidophylla Rose & Hemsl., T. pallida (Rose) D.R.Hunt, T. pygmaea D.R.Hunt, T. rozynskii Matuda, T. sillamontana Matuda, T. schippii D.R.Hunt, and T. soconuscana Matuda. In most species, the stamens are fused throughout or most of the claws length. However, in the species that do not possess clawed petals (i.e. T. mirandae, T. orchidophylla, T. rozynskii, and T. sillamontana), the stamens are clearly basally fused to the petals and recorded by the first time in the present study. According the herein presented results, epipetaly does not seem to be directly connected either with sympetaly or clawed petals, but might actually represent an independent character.

Clawed petals are found in several genera from tribe Commelineae and Tradescantieae. In tribe Commelineae, clawed petals are restricted to genera from the Commelina clade (i.e. Aneilema R.Br., Commelina L., Dictyospermum Wight, Pollia Thunb., Rhopalephora Hassk., and Tapheocarpa Conran), that together with the presence of hook-hairs are putative synapomorphies for the group (Evans et al. 2003). In all these genera, the petals are invariably free from each other. In tribe Tradescantieae clawed petals are uncommon, found in Coleotrype C.B.Clarke, Cyanotiss.l., some species of Tradescantia, and Weldenia. These genera are in general distantly related, suggesting that this character evolved several times and has no obvious phylogenetic pattern. In Tradescantia, clawed petals have been previously observed in: T. andrieuxii, T. brevifolia, T. buckleyi, T. hirta, T. huehueteca (Standl. & Steyerm.) D.R.Hunt, T. leiandra, T. pallida, T. pygmaea, T. schippii, T. soconuscana, and T. zebrina Heynh. ex Bosse. Differently from the pattern observed in tribe Commelineae, in Tradescantia many, but not all, species with clawed petals also present connate petals.

Tubular flowers are generally associated by most taxonomists with sympetaly by traditional taxonomy. However, as indicated by Harris and Harris (2001), the shape of the perianth has no relation with the connation of the perianth segments. Thus, in the present study, I have considered flowers of species such as T. zanonia (L.) Sw. as possessing shallow, wide and funnelform to cupuliform tubes, instead of being truly flat, as in T. fluminensis. Flat flowers seem to be ancestral state in Commelinaceae, with most genera in the four major lineages of the family possessing primarily flat flowers. Non-flat flowers are found in Callisia, Coleotrype, Cyanotiss.l., some species of Tradescantia, and Weldenia, being far more common than most of the characters described so far. Sympetaly and epipetaly have evolved at least three times in Tradescantia, and seem to be at least partially dependent to each other, as previously hypothesized by Rohweder (1956) and Hunt (1975). Still, tubular flowers are not necessarily correlated to the aforementioned characters. In the present topology it is possible to observe species that possess clearly tubular flowers, but no sign of sympetaly, clawed petals, or epipetaly (e.g. C. fragrans (Lindl.) Woodson, C. gentlei Matuda, and C. repens). There are also cases of species that possess tubular flowers, clawed petals, epipetaly, but free petals (e.g. T. soconuscana). Finally, there are species that possess tubular flowers and sessile and free petals, but present filaments basally conate to the petals (e.g. T. mirandae, T. orchidophylla, T. rozynskii, and T. sillamontana). These results support my decision to code each of these four characters as independent, since the same result was recovered when these characters were coded as a single character. Alternatively, the statistical support for some branches increased greatly when these characters were coded independently.

Androecium morphology has historically been the most prominent character in the taxonomy and classification of Commelinaceae (Clarke 1881; Bentham and Hooker 1883; Brückner 1930; Woodson Jr. 1942; Pichon 1946; Brenan 1961; Faden and Hunt 1991; Faden 1998). Indeed, a great deal of variation is observed in the family as a whole, with some patterns characteristic or synapomorphic to several taxonomic ranks (Brückner 1930; Handlos 1975; Hunt 1983; Faden 1991, 1998; Faden and Hunt 1991; Pellegrini et al. 2016; Pellegrini and Faden 2017). The androecium morphology of Commelinaceae is known to vary regarding the: (1) symmetry of the androecium as a whole; (2) number of stamens; (3) fertility of the stamens; (4) similarity between the inner and outer whorl of stamens; (5) similarity within each whorl of stamens; (6) connation with the corolla; (7) filament connation; (8) position, curvature and torsion of the filaments; (9) pubescence of the filaments; (10) insertion of the anthers; (11) morphology of the connectives; (12) morphology of the anther sacs; (13) relative position of the anther sacs; (15) dehiscence of the anther sacs; and (16) fertility of the pollen grains (Faden 1998; Evans et al. 2000, 2003; pers. observ.). In Tradescantiinae, this plasticity is easily observable in most genera (Figs 5, 6). Nonetheless, different androecium features have been shown to be highly homoplastic by different phylogenetic studies (Evans et al. 2000, 2003; Wade et al. 2006; Burns et al. 2011; Zuiderveen et al. 2011; Hertweck and Pires 2014). This is not surprising, considering that all Commelinaceae lack nectaries of all types, and pollen is the only floral resource available for pollinators (Faden 1992, 2000). Thus, it should be expected for androecium morphology to be strongly connected to key shifts in pollination syndromes in different lineages of the family (Faden 1992, 2000; Pellegrini and Faden 2017).

On the other hand, when coupled with different macro and micromorphological characters, androecium morphology can be successfully used to circumscribe monophyletic groups in the family (e.g. Pellegrini et al. 2016; Pellegrini and Faden 2017). For instance, subfamily Cartonematoideae was characterized based on the absence of several morphological features present in subfamily Commelinoideae, such as raphide-canals and glandular microhairs (Faden and Hunt 1991). Nonetheless, Cartonema R.Br. and Triceratella Brenan can be uniquely characterized by the their cormose habit, thyrsi composed of several one-flowered cincinni (i.e. a pseudoraceme), enantiostylous flowers, persistent sepals completely enclosing and/or surpassing the capsules, petals generally yellow and with tannin cells, six fertile stamens with inconspicuous connectives, poricidal and connivent anthers, parallel anther sacs, bowl-shaped seeds with prominent embryotega, and the loss of the micropylar collar (Barker et al. 2001; Panigo et al. 2011; pers. observ.). These androecium characters are not exclusive to Cartonematoideae, but combined with the aforementioned morphological characters, uniquely circumscribe the subfamily. Correspondingly, as recovered in the present phylogeny, androecium characters combined with different morphological characters, successfully circumscribe several lineages (Figs 3–6). For instance, Tradescantias.s. can be easily differentiated from Gibasis and Elasis based on inflorescence morphology and anatomical features, but especially due to its basifixed anthers with expanded connectives. Tripogandras.l. can be differentiated from Callisias.l. by its inconspicuous cincinni bracts and flowers with a 60° display torsion, coupled with the presence of dimorphic stamens, and dorsifixed anthers.

The present study recovered the same five clades pattern as Burns et al. (2011), and Hertweck and Pires (2014), despite the slightly different relationship between the three clades of Core Tradescantia (Figs 3–5). The present results also reveal that more than 60% of the non-monospecific sections proposed by Hunt (1975, 1980, 1986b) for Tradescantia are paraphyletic. The congruence between the phylogenetic hypotheses recovered based on the two-previous molecular datasets, with the hypothesis provided by the herein presented multiapproach dataset is indeed intriguing, but not at all surprising. All three topologies evidence an ecological and biogeographical pattern for each of the five main lineages of Tradescantia, reflecting on the probable relevance of these features in the evolutionary history of the genus. This data also corroborates the phylogenetic hypothesis proposed for the genus by Martínez and Martínez (1993), based exclusively on phytochemical characters. This scenario suggests a South American origin for Tradescantia in the Atlantic Forest domain, with a subsequent occupation of the Andes, Central America and the West Indies by the Campelia clade, and a latter occupation of North America, in Seasonally Tropical Dry Forests (STDF) and savanna formations by Core Tradescantia, with some probable dispersions to STDF in Central and South America.

The Austrotradescantia clade is restricted to South America, more precisely to Southeastern and Southern Brazil (i.e. Brazilian Atlantic Forest, especially in moist areas), Argentina, Paraguay, Uruguay, and Bolivia. Tradescantia sect. Austrotradescantia is invariably recovered as monophyletic, regardless of the number of species sampled for the group. The Austrotradescantia clade is also consistently recovered as sister to the remaining species of Tradescantias.s. The species of the Austrotradescantia clade possess a pronounced floral conservatism, with flat and small flowers, petals elliptic to ovate to broadly ovate, commonly white but sometimes in shades of pink and lilac, equal stamens with basally densely bearded moniliform hairs, glabrous gynoecium, and punctate stigma. During field and cultivation studies, flowers from this group were observed to rarely be visited by any insects at all. The few observed insects consist of generalist pollen-collectors such as hoverflies (Diptera, Syrphidae) and less commonly sweatbees (Hymenoptera, Halictidae) (pers. observ.), and point to a non-specialized floral syndrome, which might also lead to the formation of some putative hybrids observed during the development of the taxonomic revision of the group (Pellegrini 2015) and evidenced by Martínez (1984) with his cytological work on the group.

The Campelia clade is mostly restricted to understory environments of SDTF and rainforests, ranging from Mexico to Argentina. The species in this clade possess a wide range of variation regarding vegetative morphology, but share similar reproductive specializations (e.g. the presence of coflorescences, spathaceous cincinni bracts, distinct floral display position, bigger flowers, sepals zygomorphic and partially connate, petals variously colored and shaped, showy androecium with very enlarged connectives, white pollen in vivo, and semilateral embryotega), that might indicate a key shift in the reproductive strategy in this lineage. These reproductive features might help in the attraction of pollinators and could also be related in avoiding hybridization (pers. observ.). Aside from that, T. zanonia is the only species in the genus confirmed to be zoochoric dispersed, with its berry-like capsules being dispersed by birds. Despite the unique collection of reproductive peculiarities in the Campelia clade, no reproductive study has ever focused on it. As in most Commelinaceae, almost nothing is known regarding the group’s reproductive biology (Pellegrini and Faden 2017).

Core Tradescantia has Central and North America (especially Mexico and southern USA) as its diversity center, with almost all of its species being restricted to deserts, savanna formations, or STDF. In Core Tradescantia the increase in floral size is obvious in almost all species, being the clade with the most widely cultivated species due to their showy flowers. Despite the flowers ranging from medium to very large in this group, little floral specialization is observed, with androecium and gynoecium morphology being rather constant. Almost all floral specializations in Core Tradescantia are also synapomorphic to the group. Alternatively, most of the reproductive diversity recorded in this group is related either to synflorescence structure or petal conation. The Mandonia clade is especially interesting, ranging from South to North America, but with a peculiarly disjunct distribution restricted to the dry environments across the American continent. Its species seem to be greatly adapted to seasonality and longer dry periods, since the tuberous roots are extremely well-developed in all species (Fig. 9A). The inflorescence architecture in the Mandonia clade is unique in the genus, where a great number of axillary coflorescences is produced, and the main florescence is consistently sessile, with reduced cincinni bracts. This collection of features, gives the impression that the fertile individuals are large, many-branched thyrse with alternate double-cincinni. Furthermore, the flowers in the Mandonia clade possess characteristically long filaments and styles, that become spirally-coiled at post-anthesis (Fig. 9H–K). Nonetheless, since it is hard to infer how these features may affect or may have been selected by a shift in the group’s pollination syndrome, this group should also be the focus of reproductive biology studies (pers. observ.). The Setcreasea and Tradescantia clades are restricted to North America, with few species naturally reaching Central America. Both clades possess rather similar floral morphologies, differing mostly in the shape of their perianths (i.e. tubular in the Setcreasea clade vs. flat in the Tradescantia clade). Once again, the species from these two groups seem to present a generalist floral syndrome, with its flowers being visited by a wide range of insects, such as: hoverflies, sweatbees, honeybees (Hymenoptera, Apidae), bumblebees (Hymenoptera, Apidae), and occasional small unidentified beetles (pers. observ.).