Revisiting the taxonomy of Dioclea and related genera (Leguminosae, Papilionoideae), with new generic circumscriptions

Abstract The Dioclea clade comprises four genera and aproximately 60 species of the tribe Diocleae: Cleobulia (4 species), Cymbosema (1), Dioclea (ca. 50), Luzonia (1) and Macropsychanthus (3–4). Dioclea has been demonstrated to be a non-monophyletic genus, but low sampling in previous phylogenetic studies hampered the adoption of new taxonomic arrangements. We carried out densely sampled phylogenetic analyses of the Dioclea clade using molecular markers that had performed well in previous studies: the ITS and ETS nuclear ribosomal regions and the plastid trnK/matK. Our results support the maintenance of the genera Cleobulia and Cymbosema with their current circumscriptions, but confirmed the polyphyly of Dioclea, with its species falling into three different positions: (1) the puzzling species, Dioclea paniculata, was highly supported as a member of the Galactia clade; (2) Dioclea subg. Dioclea appeared as sister to a clade composed of Cleobulia and Cymbosema; and (3) the species of Dioclea subgenera Pachylobium and Platylobium composed a paraphyletic grade nesting the genera Luzonia and Macropsychanthus. We thus propose that the circumscription of Dioclea should be restricted to Dioclea subg. Dioclea, with 13 species and that the limits of Macropsychanthus should be widened to include the genus Luzonia, as well as the Dioclea subgenera Pachylobium and Platylobium, with 46 species. Taxonomic summaries, new combinations and synonyms are presented for all genera of the Dioclea clade. Cleobulia and Cymbosema were retained in their original circumscriptions. We presented an illustrated taxonomic conspectus of all genera of the Dioclea clade including 44 new combinations, one new name, ten new synonyms, two re-established holotypes, 38 lectotypes, two epitypes and one neotype.


Introduction
The genus Dioclea Kunth is one of the most important groups of tropical rainforest lianas. It includes some of the largest plants in primary forests, which are capable of spreading over wide areas on the canopies of the highest trees, often at heights above 30 m. With approximately 50 species in its current circumscription, the genus is distributed throughout the humid tropics of the Americas, Africa, Asia and the Pacific Islands. Dioclea is included in Diocleae, a tribe of Papilionoid legumes with 14 genera and approximately 200 species (Queiroz et al. 2015). Together with four other small genera, it composes the Dioclea clade, a monophyletic lineage that includes the geographically restricted genera Cleobulia Mart. ex Benth. (four species from the Neotropics), Cymbosema Benth. (one Amazonian and Mesoamerican species), Luzonia Elmer (one species from the Philippines) and Macropsychanthus Harms (2-3 species from New Guinea and neighbouring islands) (Queiroz et al. 2015).
In addition to a woody, coarse lianescent habit, the genera of the Dioclea clade also share trifoliolate leaves with stipellate leaflets, a pseudoracemose inflorescence with woody multiflorous nodes, rather large and robust firm flowers, a pseudomonadelphous androecium (i.e. with the 10 stamens joined in a tube, but with the vexillary stamen free at the base, forming fenestration via two holes at the base of the staminal tube) and a fleshy and robust intrastaminal nectary disc. Their large flowers are mostly pollinated by large carpenter bees, but some species are adapted for bird pollination (Arroyo 1981;Franco 1995;Peçanha 2014). Most species have large fruits and large seeds with long and linear (or short and oblong) hilum (Lackey 1981;Maxwell and Taylor 2003;Queiroz et al. 2003) and disperse their seeds through autochory, but some species have buoyant sea-drifted seeds (Muir 1933;Armstrong 2001).
The Dioclea clade is one of three highly-supported major lineages of the tribe Diocleae, as revealed by a multilocus molecular phylogeny using the nuclear ITS/5.8S and ETS regions and the plastid matK gene and the trnT-Y region (Queiroz et al. 2015). Previous studies, based on either morphological (Maxwell and Taylor 2003;Queiroz et al. 2003) or molecular (nrITS) data with sparser sampling (Varela et al. 2004), suggested its existence, but with low support. None of the previous studies supported the monophyly of the genus Dioclea and, instead, it was recovered as a biphyletic group roughly corresponding to long-recognised infrageneric taxa: the species of Dioclea sect. Dioclea grouping with the New World genera Cleobulia and Cymbosema (Maxwell & Taylor 2003;Queiroz et al. 2003Queiroz et al. , 2015Varela et al. 2004;Sede et al. 2009) and the species belonging to sections Pachylobium Benth., Platylobium Benth. and Macrocarpon Amshoff nesting the representatives of the Old World genus Macropsycanthus (Maxwell and Taylor 2003;Queiroz et al. 2015). More recently, we included a sequence of the plastid matK gene of Luzonia purpurea Elmer in a broader phylogenetic analysis of the Leguminosae and it appeared as a sister to Macropsychanthus, nested within the second lineage of Dioclea, but with low support (LPWG 2017).
The morphological recognition of the two major lineages that include the species of Dioclea can be traced back to Bentham (1837), who divided the genus into the sections Dioclea (as Eudioclea) and Pachylobium. He later added a third section, Platylobium (Bentham 1859). Those three sections were diagnosed by a combination of just a few morphological traits: sect. Dioclea with stipules not prolonged beyond their base, keel petals straight and erostrate, all anthers fertile and uniform, fruits elastically dehiscent and seeds with a linear hilum; sect. Platylobium sharing with sect. Dioclea non-prolonged stipules, but with the keel strongly incurved, anthers alternately fertile and sterile, fruits flat compressed and obovate with 2-3 seeds near the apex and seeds with a short and oblong hilum; and sect. Pachylobium sharing with sect. Platylobium flowers with an incurved and rostrate keel and the anthers alternately fertile and sterile, but with stipules prolonged beyond their base, fruits indehiscent or partially dehiscent and seeds with a linear hilum encircling more than half of the seed's circumference.
The circumscriptions of Bentham's sections became less clear with the discovery of some Amazonian species that combined the diagnostic features of different sections, as was the case with Dioclea macrocarpa Huber and D. erecta Hoehne, which have androecia typical of sect. Dioclea and seeds typical of sect. Platylobium. Amshoff (1939) then created sect. Macrocarpon to include the species of Dioclea with stipules not prolonged beyond their base, androecium with uniform anthers, fruits mostly oblong with 4-5 seeds evenly distributed along their length and seeds with a short, oblong hilum.  elevated those three sections created by Bentham to subgenera and included Amshoff's sect. Macrocarpon into subg. Platylobium.
Despite the existence of phylogenetic studies focusing on the tribe Diocleae, there has been no re-appraisal of the taxonomy of the Dioclea clade incorporating those findings. We can speculate that the situation probably reflects the rather sparse sampling of taxa across the morphological and geographical ranges of the included genera. Here, we thus provide a re-assessment of the taxonomy of the Dioclea clade in light of robust and densely-sampled phylogenetic analyses. These analyses sought to: (1) test the previous findings of paraphyly of the genus Dioclea and its relationships with the remaining genera of the Dioclea clade; (2) re-examine the monophyly of the infrageneric groups of Dioclea; and, (3) provide a new generic classification that reflects the phylogenetic structure of the Dioclea clade.

Materials and methods
Taxon sampling was designed to test the monophyly of the Dioclea clade of the tribe Diocleae as identified by Queiroz et al. (2015), to test the monophyly of its genera and to explore relationships between the genera. The sampling included 62 accessions corresponding to: one species of the monospecific Cymbosema, four species of Cleobulia (100% of all species in the genus), one species of the monospecific Luzonia, one species and two varieties of Macropsychanthus (50% of the species and 33% of all taxa) and 36 described species (+ six inedit) of Dioclea (60%  Table 1. The DNA regions used in this study are the same as those used by Queiroz et al. (2015): the plastid trnK/matK (the matK gene and partial flanking trnK introns) and ribosomal nuclear ETS (partial 3' end of the External Transcribed Spacer) and ITS (5.8S and flanking Internal Transcribed Spacers 1 and 2) ( Table 2).
Total genomic DNA was extracted from silica gel-dried leaves using the 2× CTAB protocol of Doyle and Doyle (1987). For herbarium samples, DNA was extracted using the DNeasy Plant Mini Kit (QIAGEN GmbH, Hilden, Germany). PCR reactions were performed using the TopTaq Master Mix Kit (QIAGEN GmbH, Hilden, Germany) according to the manufacturer's protocols, with a final volume of 10 µl. For herbarium samples, the PCR reactions also included 2 µl of TBT-PAR [trealose, bovine serum albumin (BSA), polysorbate-20 (Tween-20)] (Samarakoon et al. 2013) and, for ITS, they also included 0.2 µl of 99.5% DMSO (dimethyl sulphoxide) to avoid secondary conformations. Primers and PCR conditions are summarised in Table 2.
The PCR products were cleaned using 11% PEG (Paithankar and Prasad 1991) and then sequenced in both directions using the Big Dye Terminator v3.1 Cycle Sequencing Kit (Applied Biosystems, Austin, Texas, USA) according to the following protocol: a hot start followed by 3 min of initial denaturation at 96 °C, 30 cycles of 96 °C denaturation for 20 s, 50 °C annealing for 15 s and a 60 °C extension for 4 min. Sequencing products were cleaned using 80% isopropanol and 70% ethanol and analysed on a 3130×l Genetic Analyser (Applied Biosystems/HITACHI, Tokyo, Japan) at the Laboratório de Sistemática Molecular de Plantas of the Universidade Estadual de Feira de Santana (LAMOL/UEFS).
The original electropherograms were assembled into final sequences using the Geneious platform (Drummond et al. 2012). The sequences were automatically aligned in MUSCLE with default settings (Edgar 2004) and then checked using Geneious for manual adjustments. We carried out maximum parsimony (MP), maximum likelihood (ML) and Bayesian analyses for both individual and combined (nrITS, nrETS and trnK/ matK) DNA datasets. Conflicts amongst datasets were evaluated by the incongruence length difference test (ILD; Farris et al. 1995), performed in PAUP v.4.0b10 (Swofford 2002) between nuclear regions and between the nuclear and plastid regions, using a heuristic search with 1000 replicates, random taxa-addition and tree bisection and reconnection (TBR) branch-swapping, saving 15 trees per replicate.
The search for the most parsimonious trees was carried out in PAUP v. 4.0b10 (Swofford 2002). Heuristic searches were made with 1000 random taxon-addition and tree bisection-reconnection (TBR) branch swapping, saving 15 trees per replicate. The trees saved in this first round were used as starting trees for a subsequent round of TBR swapping. All character state transformations were weighted equally and unordered (Fitch 1971). Non-parametric bootstrap resampling was used to estimate clade support (Felsenstein 1985), which was assessed through 2000 replicates (Hedges 1992;Müller 2005), simple taxon-addition and TBR algorithm, saving 15 trees per replicate. Only bootstrap percentages > 85% were considered as strong support (Kress et al. 2002). Bayesian analyses were performed using MrBayes v.3.2.7a (Ronquist et al. 2012) in CIPRES Science Gateway v.3.3 (Miller et al. 2010). Nucleotide substitution models were selected using the Akaike Information Criterion (AIC) in MrModeltest v.2.3 (Nylander 2004) for each DNA region (Table 3). Two runs using the Metropoliscoupled MCMC (Markov Chain Monte Carlo) algorithm, each with four randominitiated chains (one 'cold' and three 'heated'), involved 10 million generations and those were sampled every 1000 generations. The convergence of the runs was assessed by checking if the standard deviation of split frequencies reached a value below 0.01. The first 2500 trees of each run were excluded as burn-ins and the effective sample size (ESS) of all parameters was checked to verify if the values were > 200. The remaining trees were summarised into a majority-rule consensus tree including the posterior probabilities (PP) as branch support estimates. Only PP values ≥ 95 were considered as strong support (Erixon et al. 2003). Deguelia nitidula was chosen as the outgroup in the Bayesian analyses.
Maximum likelihood analyses were carried out using RAxML v.8.2.12 (Stamatakis 2014) in CIPRES Science Gateway v.3.3 (Miller et al. 2010) under a GTRGAMMA model, with the '-f a' option (search for the best-scoring ML tree and a rapid bootstrap analysis) and 1000 bootstrap replicates. The MP strict consensus trees, ML trees and Bayesian 50% majority-rule consensus trees were visualised and partially edited in FigTree v.1. 4.4 (Rambaut 2018).

Results
We generated 51 new sequences for the Dioclea clade (19 of the nuclear ETS, 20 of the nuclear ITS and 12 of the plastid trnK/matK). The most variable dataset was ETS, followed by ITS and trnK/matK, respectively (Table 3). In terms of informativeness as measured by the retention index (RI) of each dataset, the ETS and the ITS performed similarly and slightly worse than trnK/matK, suggesting that part of the variation in the nuclear datasets are homoplasious (Table 3).
The individual phylogenetic analyses demonstrated similar results in recovering the same major clades and presenting no strongly-supported incongruences (Suppl. material: Figs S1-S3). The ETS trees were better resolved than those from ITS and trnK/matK (Suppl. material: Figs S1-S3). However, resolution within the main clades of the tree (see below) varied amongst datasets and thus a better overall topology was obtained in the combined analyses. Since the ILD test indicated no incongruence between nuclear datasets (p = 0.3) or between nuclear and plastid datasets (p = 0.5), we performed combined analyses, which provided a better overall topology and higher support values for the nodes. Thus, we present and discuss the results from the combined analyses (Fig. 1).
The Dioclea clade, comprising the genera Cleobulia, Cymbosema, Dioclea, Luzonia and Macropsychanthus, was recovered as monophyletic with high support with the exclusion of Dioclea paniculata (Fig. 1). Two major clades were recovered: clade A, including the genera Cleobulia and Cymbosema, together with Dioclea subg. Dioclea; and clade B, including the genera Luzonia and Macropsychanthus, together with Dioclea subgs. Pachylobium and Platylobium. Dioclea paniculata (subg. Platylobium) grouped with the genera of the Galactia clade. The genus Dioclea, therefore, appears polyphyletic, while the rest of genera in Dioclea clade were resolved as monophyletic with high support.
Within clade A, Cleobulia and Cymbosema comprise a highly-supported clade, sister to Dioclea subg. Dioclea. Clade B presents two major clades: C and D. Clade C brings together species of Dioclea subg. Platylobium; and clade D includes species of Dioclea subg. Pachylobium together with D. huberi (subg. Platylobium) and nests the representatives of the genera Luzonia and Macropsychanthus within it.
The phylogenetic structure of Clade D shows some geographical and ecological trends in its two major clades, E and F. Clade E includes species mostly from eastern South America, including a subclade of species found in Atlantic rainforests (clade G), which is a sister to a clade of species found in seasonally dry forests (clade H). Clade F is mostly composed of species found in rainforests of the Amazon region, but includes the pantropical sea-drifted D. reflexa and D. wilsonii, as well as the Australasian genera Luzonia and Macropsychanthus.

Criteria for genera circumscriptions
As the genus Dioclea has been demonstrated here (and elsewhere) as non-monophyletic (Varela et al. 2004;Maxwell and Taylor 2003;Queiroz et al. 2003Queiroz et al. , 2015LPWG 2017), it should be reclassified to preserve the principle of monophyly. In deciding which monophyletic groups should be named, other principles besides monophyly should be taken into consideration to maximise support for monophyly, for phylogenetic information and for ease of identification (diagnosability; Backlund and Bremer 1998).
One possible taxonomic solution for resolving the non-monophyly of Dioclea would be to merge all of the genera of the Dioclea clade into a widely-circumscribed Dioclea, thus subsuming the genera Cleobulia, Cymbosema, Luzonia and Macropsychanthus within Dioclea. Although having high phylogenetic support, such a broadly-circumscribed genus would lack diagnosability with respect to other genera of the tribe Diocleae because it would result in a highly-heterogeneous genus, presenting variations in almost all of the characters used to diagnose the genera in the tribe Diocleae. At the other extreme, another taxonomic solution would be to split Dioclea into several smaller genera to preserve Luzonia and Macropsychanthus in their current circumscriptions (Queiroz et al. 2015;LPWG 2017). That option presents several drawbacks, however, as some of the smaller clades within clade B lack support and such narrowly-circumscribed genera would be highly redundant, as they would be defined by the same set of morphological traits and would therefore lack diagnosability.
We opted for the intermediate solution of splitting Dioclea into two genera corresponding to the two major clades, A2 and B. Clade A2 then corresponds to Dioclea subg. Dioclea and includes D. sericea Kunth, the type species of Dioclea and would, therefore, retain the name of the genus. Clade B then corresponds to the subgenera Pachylobium and Platylobium, plus the genera Luzonia and Macropsychanthus. The genus name Macropsychanthus has priority for this clade. Both of the proposed genera are monophyletic, have high phylogenetic support ( Fig. 1) and are diagnosed by clear macromorphological characters -thus presenting low redundancy (as will be discussed below).

The genus Dioclea with a narrower circumscription
The circumscription of Dioclea is restricted here to the subg. Dioclea (sensu Maxwell 2011) or sect. Dioclea (sensu Bentham 1837). This group had been recovered as monophyletic in most phylogenetic studies, based on either morphological (Queiroz et al. 2003) or DNA data (Varela et al. 2004;Queiroz et al. 2015). It has also been supported as sister to a clade composed of the genera Cleobulia and Cymbosema (Queiroz et al. 2015) or to the genus Cymbosema (Varela et al. 2004; Cleobulia was not sampled in that study).

Inflorescence
Axillary and with an arched axis.
Axillary and erect. Axillary and erect. Erect, mostly axillary but frequently cauliflorous.

Flower position
Resupinate (i.e. the standard petal backwards and the set wing-keel petals upwards).

Calyx
Cylindrical, 4-lobed, the lobes shorter than the tube and of the same length; upper lobe entire and truncate (wider than longer).
Campanulate, 4-lobed, the lobes having almost the same length and mathching the length of the tube; upper lobe triangulate.
Campanulate, 4-lobed, the lobes having almost the same length and mathching the length of the tube; upper lobe triangulate.
Campanulate, rarely cylindrical, upper edge humped or convex, 4-5-lobed or deeply bilabiate, the lower lobe much longer than the remaining.

Standard petal
Pink or purple, pubescent towards the apex, ecallose and spreading or reflexed ca. 90°.
Bright red, pubescent towards the apex, ecallose and spreading.

Wing petals
Dwarf, much shorter than the other petals and sagittate.
As long as the keel. As long as the keel. About twice as long as the keel. Mostly dimorphic, 5 fertile alternating with 5 sterile or 6 fertile and 4 sterile or anthers monomorphic and all 10 fertile.

Intrastaminal disc
10-lobed. Entire with a smooth rim. Entire with a smooth rim.

Fruit
Oblong-linear, elastically dehiscent; thin ribs at the margins Dioclea, as re-circumscribed here (hereafter Dioclea s.s.), Cleobulia and Cymbosema compose a clade of morphologically-similar genera, sharing fruits mostly oblong-linear, smaller than those of clade B (ranging from 9 to 13 cm long and 1.5 to 2 cm wide in clade A vs. 10 to 34 cm long and 3.5 to 6.5 cm wide in clade B), with flat and elastically-dehiscing valves. The seeds of those genera are also quite similar, being relatively small (ranging from 7 to 10 mm long, 4 to 7 mm wide and 2 to 4 mm thick in clade A vs. 20 to 35 mm long, 22 to 30 mm wide and 4 to 15 mm thick in clade B), with narrowly elliptic or oblong outlines, lenticular (i.e. slightly laterally compressed -elliptic in cross section), a linear hilum encircling almost half of the seed's circumference and a hard, bony testa (mostly marbled). All species of those genera also share an androecium with ten fertile stamens (Table 4).
Cymbosema was placed within Dioclea by Zamora (2000). It was found to be supported, however, as sister to Cleobulia and merging it into Dioclea would require that Cleobulia should likewise be placed into Dioclea s.s. Cymbosema can be differentiated from Dioclea s.s. by having diadelphous androecium, with the vexillary stamen free (vs. joined into a pseudomonadelphous androecium in Dioclea s.s.), petals bright red (vs. purple, white or reddish-purple), standard petal spreading (vs. reflexed > 90°), keel petals with margins entire (vs. upper margin serrate to fimbriate) and fruits short and oblong, ca. 2.5× longer than wide, with a long, downcurved persistent style and about 4 seeds (vs. fruits linear, ≥ 5× longer than wide, with 6-10 seeds). Maxwell (1970) reported the standard petal as spreading in D. fimbriata Huber and D. macrantha Huber, but the examination of more specimens than were available before evidenced that the flowers in anthesis of those species show a reflexed standard.
Cleobulia is quite distinct from Dioclea s.s. and Cymbosema in terms of flower and fruit traits. The flowers of Cleobulia are functionally resupinate due to the downcurved inflorescence rachis and show dwarf wings of less than half of the keel length that barely exceed the calyx (vs. wings and keel petals ± the same size in Dioclea s.s. and Cymbosema), a strongly upcurved keel bent ca. 90° (vs. keel straight), short calyx lobes with the upper ones broad and emarginate (vs. all calyx lobes triangulate and acute) and the base of the androecium pubescent (vs. androecium glabrous). The fruits of Cleobulia lack the distinct ribs (or wings) close to the upper suture that are characteristic of Dioclea s.s. fruits (Maxwell 1977).
With the exclusion of the species of the subgenera Pachylobium and Platylobium, Dioclea s.s. can be diagnosed by having the standard petal ecallose and pubescent towards the apex on the outer surface, wing and keel petals approximately the same length, keel petals straight with rounded apices and serrate to fimbriate upper margins, fruits oblong-linear with flat and elastically dehiscent woody valves, seeds 6-10, lenticular, with a linear hilum encircling almost half of the seed's circumference.
In their original circumscriptions, both Luzonia and Macropsychanthus have distinctive calyx morphologies. Luzonia (sensu Elmer 1907) has a very distinctive calyx, with the lobes joined into two deeply separate, entire and obtuse lips. Macropsychanthus (sensu Harms 1900) has a cylindrical calyx with five subequal and obtuse teeth. Dioclea subgenera Pachylobium and Platylobium typically have a 4-lobed campanulate calyx, with the upper lobe shorter and broader than the others, with the lower lobe longer, upcurved and long acuminate.
The highly-supported clade C corresponds to Dioclea subg. Platylobium, as defined by Maxwell (2011), including both sections Platylobium and Macrocarpon (but with the exclusion of D. huberi, which appeared nested in clade D). A clade, composed of taxa of subg. Platylobium, was recovered only in analyses using molecular data (Queiroz et al. 2015); in analyses using morphological data, the taxa belonging to that subgenus comprised a paraphyletic grade nesting the representatives of Dioclea subg. Pachylobium (Queiroz et al. 2003), as well as the genera Luzonia and Macropsychanthus (Maxwell and Taylor 2003). The enigmatic species Dioclea paniculata Killip ex R.H. Maxwell, tentatively placed in subg. Platylobium by Maxwell (1978), appeared more closely related to the Galactia clade (and its phylogenetic and taxonomic position will be addressed in another article).
Thus, in the new circumscription presented here, Macropsychanthus is polymorphic in both androecium and calyx traits, but can be diagnosed by woody and robust pseudoracemes with the peduncle up to 1.5 cm thick, inflorescence nodosities stalked and secundiflorous, calyx with a humped or convex tube on the upper side, standard petal glabrous and bicallose towards the blade base, keel petals strongly upcurved, intrastaminal disc 10-lobed, ovary sessile and large fruits and seeds.

Taxonomic treatment
Key to the genera of the Dioclea clade

Type.
[lectotype, designated by Britton and Wilson (1924)]. Dioclea sericea Kunth. Description. Woody vines along forest edges, trailing or shrubby in open habitats. Stipules basifixed, not prolonged beyond their bases. Leaves pinnately trifoliolate, stipellate, leaf rachis short, mostly < 5 mm long. Inflorescence an erect pseudoraceme, nodes multiflorous, woody, sessile, secundiflorous; bracteoles chartaceous or membranous. Flowers with calyx chartaceous, campanulate, the four lobes having almost the same length, upper lobe entire, triangulate, obtuse or acute, the other three lobes triangulate, acute, the lower lobe as long as the upper lobe; petals membranous, mostly purple, rarely withish-purple or reddish-purple, standard petal reflexed, ecallose, but slightly thickened near the base, provided with two basal and reflexed auricles, pubescent towards the apex on the outer surface, wing petals as long as the keel, oblong to obovate, provided with a basal spur on the upper margin, keel petals straight, elliptic to obovate, upper margin dentate, serrate or fimbriate; androecium pseudomonadelphous, the 10 stamens joined into a tube but the filament of the vexillary stamen free at the base, anthers monomorphic, all 10 stamens fertile; intrastaminal nectary disc entire, collar-shape; pistil sigmoid, ovary mostly 7-15-ovulate, stipitate, style not swollen. Fruits linear, mostly 5× longer than wide, up to 2.5 cm wide, elastically dehiscent, the thin woody valves explosively twisting to release the seeds, upper margin straight and provided with a longitudinal rib or wing to each side of the suture. Seeds small, up to 14 mm long and 8 mm wide, lenticular (slightly biconvex); testa hard (bony), smooth, mostly mottled; hilum linear, encircling almost half of the seed's circumference (Fig. 2G-K).
Discussion. Dioclea was described by Kunth (1823 [1824]) with two new species based on specimens collected by Humboldt and Bonpland: D. apurensis, from a depauperate fruiting specimen and D. sericea, with four flowering specimens and illustrated in plate 576. Dioclea sericea was selected as the type for the genus by Britton and Wilson (1924).
A few months after Kunth's publication, Sprengel (1825) used the name Dioclea Spreng. for a genus of Boraginaceae. Later, Sprengel (1827) created the genus Hymenospron to which he transferred both of Kunth's species, together with a species currently ascribed to Galactia [G. rubra (Jacq.) Urb.]. Dioclea Spreng. is a later homonym in relation to Dioclea Kunth and thus illegitimate. Hymenospron Spreng. is a superfluous name with respect to Dioclea Kunth. The genus Crepidotropis was created by Walpers (1840) with just one species (C. brasiliensis) that is conspecific with Dioclea virgata (Rich.) Amshoff.
The genus Dioclea was named after Diocles of Carystus, a Greek philosopher from the 3rd century BC., probably because he associated the word 'beans' with the genus Dolichos L., which, in its original circumscription, included species now ascribed to Dioclea (Candolle, 1825: 379-380).
Dioclea is diagnosed by the combination of flowers with a pseudomonadelphous androecium, standard petal reflexed and pubescent towards the apex, fruits with an oblong-linear, flat compressed body and explosive dehiscence and seeds elliptic-oblong, lenticular, with a long and linear hilum encircling about half of their circumference.
As circumscribed here, Dioclea includes 13 species from the tropical Americas, ranging from coastal central Mexico to northern Argentina and Paraguay. Dioclea virgata was introduced into the Old World and became a garden escape plant in Malaysia, Borneo and Ethiopia (Maxwell 1969;Adema 1998   Note. The specimen in P provides no information concerning its collector, but that information is recorded on the duplicate at G and agrees with the information of the protologue (Richard 1792).
Discussion. Our results support the recognition of Cymbosema as a monospecific genus, as originally proposed by Bentham (1840Bentham ( , 1859 and maintained by Maxwell (1970). Zamora (2000) synonymised Cymbosema in Dioclea, a proposal that is not supported by our results, which recovered Cymbosema as sister to Cleobulia rather than to Dioclea. Cymbosema is diagnosed as having flowers with a diadelphous androecium with the vexillary stamen free, petals bright red, the standard petal spreading (only rarely reflexed), keel petals with smooth margins and fruits oblong and falcate.

Cymbosema roseum
Discussion. Since first being described, Cleobulia was distinguished from Dioclea by having dwarf wings with a semi-sagitate blade (Bentham 1837;see Fig. 2B). Cleobulia could likewise be diagnosed by having an inflorescence with a long and arching peduncle, leaving its flowers resupinate (i.e. with the standard petal in a lower position and the keel above), a pseudomonadelphous androecium, the base of the staminal tube pubescent, with uniform anthers, a 10-lobed intrastaminal disc, and a sessile and straight ovary.
Three species are found from eastern Brazil to the eastern Brazilian Amazon and one species from western-central Mexico, all mostly in semi-deciduous forests.

Note.
A link between Dolichos coccineus Vell. and Cleobulia multiflora Mart. ex Benth. was established by Maxwell (1977), who speculated that they could be synonymous.
The description provided by Vellozo (1829: 321) is exceedingly brief, but presents some traits characteristic of this species, such as flowers small and perianth purpureum. The illustration provides more elements to confirm its identity as C. multiflora as it shows resupinate flowers with the standard spreading, the wing petals sagittate and much shorter than the others and the pistil with a straight ovary and style upcurved ca. 90°. There are issues regarding the publication dates of several sections of the Florae Fluminensis but the main text in volume 1 (pages 1 to 329) is considered as having been distributed between 7 September to 28 November 1829 and the illustration volumes on 29 October 1831 (Carauta 1969(Carauta , 1972Stafleu and Cowan 1985;Lima 1995), thus predating and having priority over Cleobulia multiflora published by Bentham in 1837.
To avoid misinterpretation of the name proposed by Vellozo (1831), we are designating an epitype with leaf, flowers and fruits. Note. The holotype is the only remanant of the material used by Bentham (1859) for describing C. diocleoides. The material now consists of a branch with leaves and a dissected flower bud within an envelope. A detached calyx from a mature flower is the only element that allows us to check that this plant presents flowers much larger than the other species of Cleobulia as described by Bentham (1859) and Maxwell (1977). We selected an epitype from a more complete material with flowers and immature fruits. Note. When describing the new species C. leiantha, Bentham (1859) cited the specimen collected by Spruce near Santarém. We selected as the lectotype the specimen with a handwritten label and with the collection number 1003 found in other duplicates.

Type. Macropsychanthus lauterbachii Harms.
Description. Stout, high-climbing lianas with twining stems, less frequently shrubs or woody vines in open habitats. Stipules medifixed and prolonged below their insertion (peltate) or basifixed and not prolonged below their insertion. Leaves pinnately trifoliolate, stipellate or estipellate. Inflorescence a stout, woody, erect pseudoraceme, nodes multiflorous, woody, stalked and secundiflorous; bracteoles fleshy. Flowers massive; calyx with the tube fleshy coriaceous, upper edge convex or humped, 4-lobed, with the upper lobe either entire and triangulate to obtuse or emarginate and then with the resulting tips rounded or 5-lobed with the two upper lobes rounded and the other three lobes triangulate, the lower lobe much longer than the remaining lobes or deeply bilabiate with two oblong lips; petals firm, the standard petal reflexed, somewhat fleshy, bicallose, provided with two basal and folded auricles, wing petals ca. twice as long as the keel, obliquely oblong, obliquely ovate, obovate, elliptic to almost quadrate, basal spur at the upper margin present or lacking, keel upcurved, the keel petals triangular or semi-lunar, extending distally into a slender, obtuse or truncate beak; androecium pseudomonadelphous, the 10 stamens joined in a tube, but the filament of the vexillary stamen free at the base, anthers mostly dimorphic, 5 fertile alternating with 5 sterile or 6 fertile and 4 sterile or anthers uniform and all 10 fertile; intrastaminal nectary disc 10-dentate or 10-lobed; ovary sessile, style usually swollen distally. Fruit indehiscent, passively dehiscent or elastically dehiscent with twisting woody valves, turgid, slightly compressed or flat compressed, valves coriaceous, fleshy or woody, upper margin smooth or provided with ribs or wings. Seeds 3-5 to 9, massive, either orbiculate and slightly compressed with a hard testa or soft overgrown and without a definite shape, with flat contact planes or elliptic and flat compressed; hilum linear, encircling 1/2 to 4/5 of the seed's circumference or short and oblong. Fig. 3.
Discussion. Macropsychanthus Harms is the earliest validly-published genus name for this group. Two older names, Lepidamphora Zolling. and Taurophtalmum Duchaiss., were not validly published. Lepidamphora volubilis Zolling. was published as a synonym of Dioclea javanica Benth. with the citation of two specimens ("Herb. n. 763 et 867 Z.";Miquel 1855: 217). Lepidamphora volubilis was probably just a name on herbarium sheets and is invalid because it was published as a synonym (ICN Article 36.1; Turland et al. 2018) and because it was published as a species, but the genus to which it was assigned was not validly published at the same time or was not validly published previously (Art. 35.1; Turland et al. 2018).
The Panamanian Taurophtalmum pulchrum Duchaiss. was another invalidly-published name that could be related with Macropsychanthus as defined here. It was originally published as a synonym of Canavalia miniata (Kunth) DC. by Griesebach (1866: 76). However, Urban (1899: 473) placed T. pulchrum as a synonym of Dioclea reflexa Hook. f. (= Macropsychanthus comosus), based on the calyx description provided earlier by Grisebach (1866). The only specimen of Canavalia or Dioclea collected by Duchassaing that we were able to track is the type of Dioclea panamensis Duchaiss. ex Walp.
(Duchassaing s.n. [GOET 004985]), which is a synonym of Dioclea guianensis Benth. and thus does not belong to Macropsychanthus as circumscribed here. There is a plate from Duchaissang housed at GOET (and annotated as Canavalia miniata by Griesebach) that probably represents the only remnant of the original material of Taurophtalmum pulchrum. It is a watercolour painting of a fruit and a seed with a pencil sketch of a flower and a detailed description by Duchaissang (Fig. 4). The fruit represented probably belongs to Macropsychanthus megacarpus and not to M. comosus as supposed by Urban (1899). The name Taurophtalmum literally means "bulls eye" and was probably derived from the Spanish name "ojo de buey" for several species of Macropsychanthus (also common in Portuguese as "olho-de-boi"), but not for species of Dioclea. In the absence of a specimen and taking the painting in GOET as evidence, we are considering Taurophtalmum as related to Macropsychanthus, although it is an invalid name.
Two major clades were recovered corresponding to the circumscription of Macropsychanthus proposed here. One (clade D) brings together species formerly ascribed to the genera Luzonia and Macropsychanthus, as well as to Dioclea subg. Pachylobium and Dioclea huberi (subg. Platylobium sect. Macrocarpon; Maxwell 2011). Clade C comprises all of the other species formerly ascribed to Dioclea subg. Platylobium. Clade D includes species with mostly medifixed stipules, fruits indehiscent or passively dehiscent and turgid seeds with a long, linear hilum; clade C includes species with basifixed stipules, fruits flat-compressed and elastically dehiscent and seeds with a short and oblong hilum. Our finding that the puzzling Dioclea huberi (formerly classified in subg. Platylobium sect. Macrocarpon) is part of clade D blurs the distinction between those major clades, because it shares basifixed stipules and flat-compressed fruits and seeds with D. subg. Platylobium, but seeds with a long linear hilum with D. subg. Pachylobium. Likewise, Dioclea macrocarpa, recovered in clade C, shows the basifixed stipules and the short and oblong hilum of D. subg. Platylobium together with the turgid fruits and seeds of D. subg. Pachylobium. Thus, clades B and C are diagnosed by only a few morphological traits (see below) and we chose to recognise them as subgenera of a largely polymorphic genus instead of treating them as two separate genera.
Macropsychanthus is a pantropical genus with 46 species. It is most diverse in the New World (36 species), with eleven species from the Philippines and Indonesia to New Guinea and two Pantropical sea-drifted species extending to continental Africa and Madagascar. Description. Stipules medifixed, prolonged below their insertion. Leaves stipellate, stipels mostly setaceous. Fruit indehiscent or passively dehiscent, turgid, slightly compressed (elastically dehiscent with twisting woody valves only in M. huberi). Seeds with a long and linear hilum encircling 1/2 to 4/5 of the seed's circumference (Fig. 3A-F). The distribution of this section is the same as that of the genus. Species of subg. Macropsychanthus are typical rainforest elements, where they occur as high-climbing lianas over the tallest trees. Few species are found in the savannahs of central Brazil or in the seasonally-dry woodlands of South America.    Adema (1998). As the name Dioclea ferruginea was already occupied by D. ferruginea Ducke, Adema (1998) Merrill (1910) did not refer to the herbarium where the type is housed and we were unable to track it. The PNH herbarium curator confirmed that the holotype was housed at PNH (as PNH 4697) but that it was destroyed during World War II (L. Evangelista, Philippine National Herbarium, National Museum, pers. comm.). Adema (1998) speculated that it could be more closely related to (or conspecific with) M. ferrugineus as it was described as having ten fertile stamens.   LPQ work on legume systematics was supported by CNPq (processes 303585/2016-1 and 440487/2015-3) and FAPESB (PTX0004/2016 and APP0096/2016). The use of DNA from the Brazilian species is authorised by SISGEN n° AEB0728.