A new generic system for the pantropical Caesalpinia group (Leguminosae)

Abstract The Caesalpinia group is a large pantropical clade of ca. 205 species in subfamily Caesalpinioideae (Leguminosae) in which generic delimitation has been in a state of considerable flux. Here we present new phylogenetic analyses based on five plastid and one nuclear ribosomal marker, with dense taxon sampling including 172 (84%) of the species and representatives of all previously described genera in the Caesalpinia group. These analyses show that the current classification of the Caesalpinia group into 21 genera needs to be revised. Several genera (Poincianella, Erythrostemon, Cenostigma and Caesalpinia sensu Lewis, 2005) are non-monophyletic and several previously unclassified Asian species segregate into clades that merit recognition at generic rank. In addition, the near-completeness of our taxon sampling identifies three species that do not belong in any of the main clades and these are recognised as new monospecific genera. A new generic classification of the Caesalpinia group is presented including a key for the identification of genera, full generic descriptions, illustrations (drawings and photo plates of all genera), and (for most genera) the nomenclatural transfer of species to their correct genus. We recognise 26 genera, with reinstatement of two previously described genera (Biancaea Tod., Denisophytum R. Vig.), re-delimitation and expansion of several others (Moullava, Cenostigma, Libidibia and Erythrostemon), contraction of Caesalpinia s.s. and description of four new ones (Gelrebia, Paubrasilia, Hererolandia and Hultholia), and make 75 new nomenclatural combinations in this new generic system.

Th e Caesalpinia group epitomises this generic fl ux, with persistent doubts about the delimitation of genera over the last 35 years Fig. 1). Th is has been due to the diffi culties of identifying diagnostic morphological synapomorphies and obtaining adequate sampling of taxa and genes in phylogenetic studies for this large pantropically distributed clade. Th e group is placed in the newly re-circumscribed subfamily Caesalpinioideae (LPWG submitted; equivalent to the Mimosoideae-Cassieae-Caesalpinieae, MCC clade sensu Doyle (2012); see also LPWG 2013), forming one of the informal groups in tribe Caesalpinieae. Th e Caesalpinia group was defi ned by Polhill and Vidal (1981) to include the genera with species that have a large variety of glandular trichomes, prickles and spines as a defense mechanism, and possessing zygomorphic fl owers with a somewhat modifi ed lower sepal and stamens crowded around the pistil. It is currently classifi ed into 21 genera (Lewis, 2005), but recent studies, and notably Gagnon et al. (2013Gagnon et al. ( , 2015, have demonstrated the non-monophly of some  Polhill and Vidal (1981), Lewis (2005), Gagnon et al. (2013), and this study; names in bold represent new genera described here; prefi x * indicates that the description of the genus is emended; prefi x # indicates that the genus is being re-instated; prefi x ? indicates that the status of the genus is uncertain. of these and the need for a new generic classifi cation (Fig. 1). Th e group comprises ca. 205 species of small trees, woody shrubs and herbaceous subshrubs, with extremely diverse pollination and seed dispersal syndromes (the diversity of plant forms, fl owers and fruits is extensively illustrated for all genera in the taxonomic acount), occurring predominantly in seasonally dry tropical forests and shrublands, but extending in a subset of clades into tropical and warm temperate savannas, tropical wet forests and tropical coastal habitats.
Th e genus Caesalpinia L. itself has been particularly problematic having been variously circumscribed by diff erent authors. In its broadest sense Caesalpinia comprises ca. 150 species but these have had a tumultuous taxonomic and nomenclatural history, having been placed in up to 30 diff erent genera since the description of the genus in 1753. Th ese changing generic concepts illustrate the diffi culties in establishing a stable classification of the group. Th e proliferation of generic names associated with Caesalpinia s. l. is due in part to the often complex, confusing and highly homoplastic nature of many morphological characters across the group, as well as the occurrence of many narrowly restricted endemics in a group with a pantropical distribution spanning fi ve continents.
Previous molecular and morphologically-based phylogenetic analyses (Lewis and Schrire 1995, Simpson and Miao 1997, Nores et al. 2012, including data from fl oral ontogeny (Kantz andTucker 1994, Kantz 1996), phytochemistry (Kite and Lewis 1994), wood anatomy (Gasson et al. 2009), and leaf anatomy and secretory structures (Rudall et al. 1994, Lersten and Curtis 1994, Herendeen et al. 2003, attempted to more clearly delimit monophyletic genera within the Caesalpinia group. However, none of these studies achieved the comprehensive taxon sampling needed to fully understand and synthesize morphological diversity across the group as a whole. Other studies have focused on particular genera or clades, such as Hoff mannseggia Cav. (Simpson et al. 2004(Simpson et al. , 2005, Pomaria Cav. , Mezoneuron Desf. (Clark and Gagnon, 2015), and Arquita E. Gagnon, G. P. Lewis & C. E. Hughes (Gagnon et al. 2015). Th e most recent phylogenetic study ), based on a single plastid marker (rps16) and sampling 120 of ca. 205 species (i.e. 58% taxon sampling), suggested that at least 23 genera would need to be recognised due to the non-monophyly of several genera, but lacked suffi cient resolution and support as well as critical taxa (notably Lophocarpinia Burkart, Stahlia Bello, Stenodrepanum Harms, Caesalpinia pearsonii L. Bolus and C. glandulosa Bertero ex DC.), to confi dently propose a comprehensive new generic classifi cation. Here we present a new phylogenetic analysis that samples the full morphological diversity and nearly the entire geographical range of the Caesalpinia group. Th is analysis is based on fi ve plastid loci and the nuclear ribosomal ITS region, providing improved resolution and support over Gagnon et al. (2013). We use this densely sampled phylogenetic analysis to propose a new generic classifi cation of the Caesalpinia group, in which we recognise 26 genera (with one additional clade tentatively suggested as a 27 th genus to be recognised pending additional taxon sampling), provide new or emended generic descriptions, a key to genera and, for genera where no further ambiguity as to species placements exists, the new nomenclatural combinations for species as required.

Taxon sampling
DNA was extracted from herbarium specimens and fi eld-collected silica-dried leaves from wild and, in a few cases, cultivated plants. When possible, multiple individuals per species from diff erent localities were sampled. In addition, previously published sequences (Bruneau et al. 2001, Haston et al. 2005, Marazzi et al. 2006, Marazzi and Sanderson 2010, Manzanilla and Bruneau 2012, Nores et al. 2012, Babineau et al. 2013 were downloaded from GenBank (Appendix 1). All 21 genera belonging to the informal Caesalpinia group (sensu Lewis 2005), including all their type species (except for Mezoneuron Desf.), were sampled.
A total of 429 accessions representing 172 of the ca. 205 species (83.9%) of the Caesalpinia group, and including 131 species previously ascribed to the genus Caesalpinia s. l., were sequenced (Appendix 1). Th is sampling represents the full geographical range and morphological diversity of the group, with the important exception of seven species from mainland China for which no material was available for study. Several key species, whose phylogenetic and taxonomic affi nities were previously unclear, including Caesalpinia digyna Rottler, C. tortuosa Roxb., C. pellucida Vogel, C. glandulosa, and C. pearsonii

Molecular methods
Th ree protocols were used to extract DNA: (1) a modifi ed CTAB protocol (Joly and Bruneau 2006); (2) QIAGEN DNeasy Plant Mini Kit (Mississauga, ON, Canada); or (3) a 4% MATAB protocol (Ky et al. 2000). Six genetic markers were amplifi ed, including the 5.8S subunit and fl anking internal transcribed spacers, ITS1 and ITS2, of nuclear ribosomal DNA, and fi ve plastid loci: rps16, the trnD-trnT intergenic spacer, ycf6-psbM, the matK gene and fl anking 3'-trnK intron, and the trnL-trnF intron-spacer region. Th e fi rst four markers were amplifi ed using both standard and nested-PCR protocols, described in Gagnon et al. (2015). Th e matK-3'-trnK region was amplifi ed using the primers trnK685F (Hu et al. 2000), trnK4La (Wojciechowski et al. 2004), trnK2R* and KC6 (Bruneau et al. 2008), following the protocols described in Bruneau et al. (2008). Because of initially poor amplifi cations, we designed a new primer, matK-C6-Caesalpinia (5'-GAA TGC TCG GAT AAT TGG TTT-3'), which improved the amplifi cation of the 5'section of this locus. Th e trnL-trnF intron-spacer region was amplifi ed using the primers trnL-C, -D, -E and -F (Taberlet et al. 1991), using the same protocols as for the rps16 locus , with annealing temperatures varying between 50 and 53 °C. While we attempted to amplify the fi rst four loci for all available material, for the matK-3'trnK and trnL-trnF regions we sequenced a targeted subset of taxa to complement existing data. For problematic samples, including those presenting sequencing problems due to mononucleotide repeats, we used a protocol with Phusion Hot Start II High-Fidelity DNA polymerase (Th ermo Scientifi c, United States), as described by Gagnon et al. (2013), which yields more accurate and longer quality mononucleotide sequence reads (Fazekas et al. 2010).
PCR amplifi cations were sequenced by Genome Quebec (Montreal, Canada), with Big Dye Terminator 3.1 chemistry on an ABI 3730xl DNA Analyzer (Applied Biosystems, Carlsbad, CA, USA). Geneious Biomatters,Auckland,New Zealand) was used to assemble chromatograms and inspect and edit contigs. All sequences were submitted to BLAST (Altschul et al. 1990) to verify for non-specifi c amplifi cation, and eliminated if they did not match Leguminosae sequences in Gen-Bank. GenBank numbers with corresponding locality details and herbarium vouchers are listed in Appendix 1.
Phylogenetic analyses were carried out on each of the six loci individually and on two concatenated matrices, one with the fi ve plastid loci and a second matrix with all six loci (plastid + ITS). Matrices were concatenated using SequenceMatrix (Vaidya et al. 2011). We used a Maximum Likelihood (ML) approach using RaxML 8.0.0 (Stamatakis 2014) on the CIPRES gateway v.3.3 (Miller et al. 2010). Th e analyses were conducted using the GTRGAMMA model for the DNA sequences and the BIN-CAT model for the indel partitions. Bootstrap support was assessed through 1000 non-parametric bootstrap replicates.
Because topological confl icts amongst the six individual gene trees were minimal, and where diff erences were found these were always only weakly supported (< 60% BS), all subsequent analyses were done on the six-locus concatenated matrix. Initial analyses of this six-locus matrix keeping all accessions of species as separate terminals resulted in a matrix with signifi cant missing data because not all accessions were sequenced for all loci (see Tables 1 and 2). To reduce missing data, multiple accessions of the same species were concatenated if they occurred in the same clade in the pre- Table 1. Character statistics for the six loci analysed, with the number of accessions for each locus, aligned length (including ambiguous alignment regions), number of indels scored, numbers and % of parsimony informative characters (for both DNA and indel characters), and critical missing genera and taxa. liminary RaxML analyses to maximize the number of loci represented for a species. When more than one sequence per species was available for a given locus, the longest sequence was selected, because we never found any sequence variation in the overlapping sections. Th is resulted in concatenation of accessions for 16 species (see Appendix 1): Caesalpinia cacalaco Bonpl., C. caladenia Standl., C. caudata (A. Gray) Fisher, C. colimensis F. J. Herm., C. epifanioi J. L. Contr., C. exilifolia Griseb., C. madagascariensis (R. Vig.) Senesse, C. melanadenia (Rose) Standl., C. mimosoides Lam., C. pringlei (Britton & Rose) Standl., C. sappan L., C. sessilifolia S. Watson, Libidibia sclerocarpa (Standl.) Britton & Rose, Haematoxylum brasiletto H. Karst., H. dinteri Harms and Tara spinosa (Molina) Britton & Rose. In addition to concatenating sequences obtained from diff erent accessions of a species, preliminary analyses showed lack of resolution for a few accessions for which only one or two loci were sequenced. To explore the impacts of diff erent levels of missing data, a series of matrices that progressively excluded accessions with fi ve, four, three, two and one missing loci were generated, resulting in six diff erent concatenated matrices ( Table 2). Because the matrix containing sequences with no missing data lacked representatives from a number of genera and critical clades or species, a seventh matrix was generated (with 39 taxa) that added an accession from each of these critical taxa to maximise taxonomic representation while minimizing missing data. For these seven concatenanted matrices, phylogenetic analyses were carried out using ML, maximum parsimony (MP) and Bayesian methods. For the ML analyses, we used RaxML (Stamatakis 2014) as described above. For MP analyses, PAUP* (Swoff ord 2003) was used with a two-step approach (Davis et al. 2004) as described in Gagnon et al. (2013), but saving a maximum of 50,000 trees with 5,000 bootstrap Table 2. Statistics for the seven combined matrices, with the number of accessions, number of ingroup and outgroup species, % missing data, and missing genera/ critical taxa. Th e results of the parsimony analyses are indicated, with the number of trees retained, the length of the shortest trees (length), consistency index (CI), and retention index (RI).  (Nylander 2004) to select the GTR + I + G model for all six loci and the F81-like model for the indel partition. Analyses were run on a high performance computer cluster (Calcul Québec, Université de Montréal, Canada) with two parallel runs of eight Markov Chain Monte Carlo (MCMC) chains, four swaps per swapping cycle, and trees sampled every 1000 generations. Th e stop criterion was set to an average standard deviation of split frequencies that dropped to below the critical value of 0.01. Tracer v.1.6 (Rambaut et al. 2014) was used to ensure eff ective sample sizes were above 200 and that chains mixed appropriately, with 510,000 and 27 million generations, depending on the size of the matrix. Th e "burn-in" fraction for all analyses was set to 10%.

Results
Of the six loci, ITS had the highest proportion of parsimony-informative characters (61.7%), followed by rps16,. Th e concatenated six-locus matrix (aligned length = 8803 bp) included 429 accessions, which was reduced to 408 when accessions were combined for 16 species (see above). Table 2 summarises the number of accessions and species per locus, the percentage of missing data, the number of trees, tree length, CI and RI obtained in the MP analyses for the series of seven concatenated matrices with successively lower numbers of taxa with missing loci. With the exception of the least informative (trnL-trnF) gene tree, which is poorly resolved (data not shown), the Caesalpinia group is monophyletic in all analyses, generally with high bootstrap and PP support (see Suppl. material 1). Th e 23 major clades identifi ed from the rps16 phylogeny by Gagnon et al. (2013;Fig. 1) are also generally recovered in each of the individual ML gene trees (Suppl. material 1), as well in the analyses of the matrices combining all six loci, with two notable exceptions. First, in the MP and ML analyses, Lophocarpinia is nested within Haematoxylum, but in the Bayesian analyses Lophocarpinia is sister to Haematoxylum. Second, the genus Pterolobium is also sometimes recovered as non-monophyletic, with Caesalpinia crista nested within it in some of the MP, ML and Bayesian analyses, while in other analyses it is recovered as monophyletic, but with poor to moderate support in the ML and Bayesian analyses of all six loci, with a minimum of 2 to 3 loci per accession (Suppl. material 1).
In addition to these 23 clades ( Fig. 1; see Gagnon et al. 2013), four other clades or monospecifi c lineages were consistently recovered in the MP, ML, and Bayesian analyses of the matrices with all six loci (Suppl. material 1): the three monospecifi c C. echinata, C. mimosoides and C. pearsonii lineages, and the C. crista clade, corresponding to Caesalpinia sect. Nugaria, represented by C. crista and C. vernalis in the rps16 gene tree of Gagnon et al. (2013), although it is important to note that C. vernalis was excluded from later analyses of the concatenated matrices due to missing data and does not appear in Fig. 2 or Fig. 3. In total, this resulted in 27 possible genera in the Cae-Tetrapterocarpon geayi 1049 (6) Pterogyne nitens 13XII971 (6) Gymnocladus chinensis IIV021 (6) Cassia javanica 6 (6)

Guilandina bonduc 1342 (5)
Given this congruence among the ML, MP and Bayesian analyses, only the Bayesian topology is presented ( Branch support values are indicated as follows: branches in bold indicate that maximum support has been attained in the MP, ML and Bayesian phylogenetic analyses; otherwise, posterior probabilities are indicated above in bold, with bootstrap support from ML analyses (italicised) and parsimony analyses separated by a slash below the branches; for each terminal, the species name is followed by the collector number of the corresponding voucher (see Appendix 1 for full voucher details); the suffi x ** indicates that several sequences from diff erent accessions of the same species were concatenated for analysis (see Appendix 1 for details); for major clades and genera, the names used by Gagnon et al. (2013) are indicated, as well as the corresponding new genera.    Although all 27 lineages and all 26 genera are robustly supported, the precise relationships amongst a few genera remain unresolved or are not supported. For example, the position of C. echinata lacks support in both the MP and ML analyses (bootstrap support below 50%), while in the Bayesian analyses it is sometimes resolved as sister to Caesalpinia s. s. (PP between 64 and 97), emphasising that this species is phylogenetically isolated and justifying its recognition as a new genus (see below). Similarly, the relationships between C. mimosoides, the C. trothae clade, and Guilandina are sometimes resolved, but generally with low support, again pointing to the phylogenetic distinctiveness of C. mimosoides. Within the core Poincianella-Erythrostemon clade, the relationships of C. placida and C. glandulosa are unstable, being placed either as sister to a Central American lineage or to a South American lineage. Finally, the position of Stenodrepanum as sister to Hoff mannseggia is consistent across all analyses, but always with low branch support (Fig. 3C).

Discussion
In his generic classifi cation of Caesalpinia s. l., Lewis (2005) suggested that molecular phylogenies with increased taxon sampling were needed to rigourously test the monophyly of the genera he was reinstating and to resolve the relationships of a group of 12 to15 Asian species that could not be placed in any of the proposed segregates. Whilst several recent studies based on single DNA sequence loci or morphology have partially addressed this problem , Nores et al. 2012, the results presented here, based on combined analyses of six DNA sequence loci totaling 8.8 kb of DNA sequence data, and sampling 84% of species, provide the most comprehensively sampled and robust phylogeny of the group to date. As seen in many other species-level phylogenetic studies of legume taxa (e.g. Moura et al. 2016, Rando et al. 2016, the most informative DNA sequence locus is ITS, which has at least twice as many informative characters as the plastid loci included in this study. Near-complete sampling of species across the Caesalpinia group, provides a much more stringent and comprehensive assessment of the monophyly of the subclades, as well as of the homology and interpretation of morphological character evolution within the group. Furthermore, as found in both empirical and simulation studies of other taxa (Wiens 2003, Phillipe et al. 2004, Pyron et al. 2011, Johnson et al. 2012, Hinchliff and Roalson 2013, the concatenated supermatrix approach used here is shown to be robust to missing data. Of the 21 genera proposed by Lewis (2005; Fig. 1), it is clear that some of these groups, such as the Poincianella-Erythrostemon group (Lewis, 1998), Caesalpinia sensu Lewis (2005) and Cenostigma are non-monophyletic. Our analyses also reveal additional clades of Asian species that do not correspond to any of the genera in the Lewis (2005) classifi cation system. In addition, three species (C. echinata, C. mimosoides and C. pearsonii) are placed outside the clades corresponding to the genera proposed by Lewis (2005) or Gagnon et al. (2013) and comprise phylogenetically isolated monospecifi c lineages. Based on this new and much more comprehensively sampled phylogeny, thorough review of the literature and detailed survey of the morphological diversity of the group, we propose a new classifi cation recognizing 26 genera corresponding to robustly supported clades found across analyses regardless of the amount of missing data. We also discuss the possibility of recognizing a 27 th genus, but more molecular and fi eld sampling, especially of freshly collected fi eld specimens, are needed before naming this clade at generic rank.

Phylogenetic relationships and generic delimitation
In their description of the Caesalpinia group, Polhill and Vidal (1981) remarked that this was one of the most distinctive of the nine informal generic groups in tribe Caesalpinieae, based on several morphological characters, and notably the presence of a lower cucullate sepal on the calyx. Although they included the genera Conzattia, Lemuropisum and Parkinsonia in the Caesalpinia group, these were subsequently shown to belong to the Peltophorum group (Haston et al. 2005). Th e Caesalpinia group, as circumscribed by Lewis (2005), is here shown to form a robustly supported clade (Figs 2 and 3). All of the 13 genera outside Caesalpinia s. l. form robustly supported monophyletic groups, except Moullava and Cenostigma, which are both recircumscribed and expanded to include extra species that were previously placed in Caesalpinia s.l. Of the original eight genera re-instated by Lewis (2005), fi ve (Tara, Coulteria, Guilandina, Mezoneuron, and Libidibia) also form robust clades in our analyses. Th ese fi ve genera are clearly defi ned by diagnostic morphological synapomorphies, as discussed in Gagnon et al. (2013).
Libidibia shares many similarities with the monotypic Stahlia from the Caribbean, the two together forming a robustly supported clade (Figs 2 and 3C), prompting re-evaluation of their status as distinct genera. Stahlia has been distinguished by its somewhat fl eshy red fruits (Fig. 32A) and singly pinnate leaves. However, the pods of Stahlia are similar to those of some species of Libidibia (especially L. sclerocarpa and some South American species) in terms of shape and lack of dehiscence ( Fig. 32A-C and F). All other closely related genera have dehiscent pods. Stahlia has also been differentiated from Libidibia by the presence of pinnate rather than bipinnate leaves as in Libidibia, but the dark punctate gland dots on the undersurface of the leafl ets, which are distinctively aligned parallel to the midvein, are also observed in certain species of Libidibia, including L. coriaria and L. ferrea , Nores et al. 2012. Elsewhere in the Caesalpinia group, leaf pinnation and the occurrence of pinnate vs. bipinnate leaves can be extremely labile within genera (e.g. Haematoxylum and Cenostigma), within species (e.g. Stuhlmannia moavi), and even within individuals (e.g. Haematoxylum sousanum Cruz Durán & J. Jiménez Ramirez (Durán and Ramirez 2008)). Given these morphological similarities and the apparent lability of leaf division, we conclude that there is no justifi cation for retaining Stahlia and Libidbia as separate genera.
As found previously by Gagnon et al. (2013Gagnon et al. ( , 2015, the other three genera recognised by Lewis (2005), Poincianella, Erythrostemon and Caesalpinia s. s., are not supported as monophyletic (Fig. 3A, D). Although Lewis (1998) considered that Poincianella and Erythrostemon together formed a clade, Gagnon et al. (2013Gagnon et al. ( , 2015 plus the more densely sampled phylogeny presented here (Fig. 3), show that their species fall into unrelated clades, providing the basis for recognition of three genera. First, a subset of Poincianella species corresponding to the Poincianella B group of Lewis and Schrire (1995) group with Cenostigma (Fig. 3C), as found in the morphological cladistic analysis of Lewis and Schrire (1995). Th ese Poincianella B species diff er from the remaining Poincianella and Erythrostemon species in wood anatomy (Gasson et al. 2009) and in their alternate to subopposite leafl ets (De Queiroz 2009). While Cenostigma was originally considered as a distinct genus, in part based on its pinnate leaves, two species of the Poincianella B clade (Caesalpinia marginata and Caesalpinia pinnata) also have pinnate leaves. More importantly, several species of Poincianella B have internal secretory cavities in the leafl et lamina and infl orescences (Lersten and Curtis 1994;Rudall et al. 1994), as well as a stellate indumentum on the stems, leaves and/or infl orescences, both of which are considered as diagnostic characters of Cenostigma. Th ese leaf traits are completely lacking in the core Poincianella-Erythrostemon group. In addition, Poincianella B and Cenostigma share robust pods with conspicuously thickened margins , which are absent in the other species of the Poincianella-Erythrostemon group and provide a diagnostic synapomorphy for an expanded Cenostigma including the Poincianella B species. It thus appears that in this group morphological homoplasy (pinnation of leaves, alternate to subopposite leafl ets, the presence/absence of stipitate glands, stellate indumentum) has obscured relationships resulting in non-monophyletic genera. Here we expand Cenostigma to include the subset of Poincianella-Erythrostemon group species formerly assigned to Poincianella B by Lewis and Schrire (1995;Fig. 3C).
Th e remaining species of the former Poincianella and Erythrostemon are placed either in an Andean clade of fi ve species, which is sister to Pomaria, or are part of another lineage containing the type species of both Poincianella and Erythrostemon (Fig. 3D). Th e Andean clade has recently been recognised as the new genus Arquita, based on a combination of morphological, ecological and geographical characters (Gagnon et al. 2015, Fig. 39I-O). In the other lineage, two robustly supported subclades are resolved, one including the type species of Erythrostemon (E. gilliesii), and the other the type of Poincianella (P. mexicana; Fig. 3D). While these two subclades could potentially be retained as distinct genera, the unresolved relationships of Caesalpinia glandulosa and Caesalpinia placida at the base of this Poincianella-Erythrostemon lineage in the current phylogeny ( Fig. 3D) would entail recognizing two additional monospecifi c genera to account for these species. We prefer to treat this large Poincianella-Erythrostemon clade as a single genus which comprises a morphologically and ecologically coherent group of shrubs and small treelets in Neotropical seasonally dry tropical forests with a bicentric amphitropical distribution (Lewis 1998. Although there are currently more species under the name Poincianella Britton & Rose (1930), the older name Erythrostemon Klotzsch (1844) takes precedence. As such, Erythrostemon is here re-circumscribed to include Poincianella but excludes the subsets of Poincianella species now transferred to either Cenostigma or Arquita.
Caesalpinia s.s., as delimited by Lewis (2005), is also non-monophyletic and comprises three independent lineages. Th e most distinctive of these-the C. trothae cladeclearly is not closely related to the remaining Caesalpinia s. s. species (Fig. 3B). Th is clade consists of African species found in dry forests and thickets from the Horn of Africa, across Tanzania, Botswana, Mozambique, and South Africa to Namibia. Species in this clade share a number of diagnostic morphological synapomorphies: they are all spiny, multi-stemmed shrubs with racemes of reddish-pink to whitish-pink fl ow-ers (Fig. 11J); have distinct pyriform pods, with large, rounded, oblique bases and an acute apex; bracts with an aristate tip; and leafl ets with translucent dots on the lower surface. However, species delimitation needs to be re-examined. For example, Brenan (1963Brenan ( , 1967 remarked that the rostrate appendage on the calyx, which distinguishes C. rostrata, is also found on some specimens of C. rubra, bringing into question the distinction of these two species. Despite uncertainty about the number of species, this clade is phylogenetically, morphologically and geographically distinct, clearly meriting recognition as a new genus, here named Gelrebia after the Somali vernacular name for C. trothae, which means camel trap and evidently alludes to the highly thorny and impenetrable habit of these plants. Th e other two clades containing members of the former Caesalpinia s. s. lack obvious diagnostic morphological synapomorphies. Both clades include species of shrubs or small treelets that are eglandular and generally spiny (except for one species in each clade), and have explosively dehiscent pods with twisting valves. Th e type species of Caesalpinia s. s., C. brasiliensis, is placed within a clade that includes a set of Caribbean species, most probably pollinated by bats (Koch et al. 2004), the Central American / Mexican C. pulcherrima, pollinated by butterfl ies (Fig. 11F), the northern Andean C. cassioides with red, laterally-compressed, tubular corollas, likely pollinated by birds (Fig. 11C), and C. nipensis, endemic to the Sierra de Nipe in Cuba, which has a fl ower morphology and a yellow corolla suggestive of bee pollination (Fig. 11B). As recircumscribed here, a reduced Caesalpinia s. s. is now restricted to the Neotropics with no species now ascribed to this genus in Africa or Asia. Th e other group, the C. erianthera clade (Fig. 3A), contains only yellow-fl owered species, but these occur across a strikingly disjunct geographic range in Madasgascar (C. madagascariensis, Fig.  11I), Ethiopia, Somalia and the Arabian Peninsula (C. erianthera), South America (C. stuckertii), Mexico (C. sessilifolia), and the Caribbean (C. buchii, C. paucifl ora (Fig. 11G,H) and C. rosei). Th e C. erianthera clade is morphologically distinct from its sister clade, the combined Tara + Coulteria clade. Th is latter clade includes species that are characterised by fl owers having a distinctive lower sepal with a cucullate-pectinate margin (although the pectinate margin is absent in C. vesicaria, and in C. cacalaco the margin is only obscurely pectinate), and pods which are thick and indehiscent (Tara), or thin, chartaceous and indehiscent to tardily and passively dehiscent (Coulteria). Species from the C. erianthera clade lack the cucullate-pectinate lower sepal margin and have pods that are explosively dehiscent, with twisting valves. Given the distant phylogenetic placement of the C. erianthera clade from both Gelrebia and the recircumscribed Caesalpinia s. s., and its morphological distinctiveness from its sister group, it is clear that the C. erianthera clade should also be recognised as a distinct genus. Within this clade, C. madagascariensis, endemic to Madagascar, was formerly placed in the monospecifi c genus Denisophytum, here reinstated with an emended circumscription that includes all species of the C. erianthera clade.
Th e majority of the rest of the currently unclassifi ed Old World species fall into two main clades, the C. decapetala clade and a clade that groups the monospecifi c genus Moullava, C. welwitschiana and two species of Caesalpinia section Cinclidocarpus, which Gagnon et al. (2013) suggested to be closely related to Moullava. Th e species in these two Old World clades consist of lianas and scrambling shrubs, but are distinguished from the other liana taxa in the Caesalpinia group (which are concentrated in clade C, see Figs 2 and 3B) by their distinctive pods. In the C. decapetala clade, the pods are oblong and somewhat laterally compressed, dehiscent along the dorsal suture, and slightly enlarged and truncate towards the apex. In the second clade, all four species have similar rounded, sub-torulose indehiscent pods, with thickened margins, and an exocarp and endocarp that are strongly adnate when dried. It is apparent that both clades merit recognition at the generic level. Based on the preliminary results of Gagnon et al. (2013),  reinstated the genus Biancaea Todaro (1860) for the C. decapetala clade and provided new combinations for three species within the genus. Here we transfer an additional species of Caesalpinia to Biancaea and emend the description of the genus, which was not included in the treatment of . We also emend the description of Moullava to include three additional species in that genus ( Fig. 3B) (see Taxonomic treatment for details).

Monospecific genera
With near-complete taxon sampling and robust support across the phylogeny, it is now clear that the three species, C. mimosoides, C. pearsonii and C. echinata, do not nest within any of the well resolved clades of the Caesalpinia group even though all six loci were sequenced for these species (except for ITS in C. mimosoides). Th e taxonomic placements of these taxa have been problematic in the past, and each species is morphologically unique within the Caesalpinia group, especially with respect to pod morphology. To incorporate these unusual taxa in our generic classifi cation, we propose three new monospecifi c genera, Hultholia,Hererolandia and Paubrasilia,respectively. Caesalpinia mimosoides (Figs 17,18) is a liana found in India, Bangladesh, Th ailand, Vietnam, Laos, Myanmar and South-West China. It is morphologically distinct from all other liana species in the Caesalpinia group, because the stem, calyx and fruits are covered in glandular dots, and the pods are falcate, chartaceous and infl ated. Th e robust, needle-like trichomes in C. mimosoides, which are present on the stem, infl orescence rachis and pedicels, are also distinctive, and quite diff erent from the more robust and strongly recurved prickles found on stems (and sometimes sparsely at the base of the infl orescences) of other Asian species of the Caesalpinia group. We propose the new generic name Hultholia, to honour the Cambodian taxonomist Dr. Salvamony Hul Th ol (see Taxonomic treatment).
The second unplaced taxon, C. pearsonii, differs from the rest of Caesalpinia s. l. primarily by its unusual flattened, circular or semi-circular one-seeded pods, covered in patent red trichomes up to 6 mm long (Fig. 5D). The precise rela-tionships of this rarely collected species, endemic to Namibia, remain uncertain and weakly supported. Our analyses provide only weak support for a sister group relationship to the Lophocarpinia + Haematoxylum clade (Fig. 2), and in most analyses C. pearsonii remains unresolved (Fig. 3A). Caesalpinia pearsonii differs from Lophocarpinia and Haematoxylum in having pinnate leaves arranged in fascicles on short brachyblasts, as opposed to the alternate pinnate or bipinnate leaves typical of these latter two genera. In addition, the secondary leaflet venation in C. pearsonii is not visible, whereas in Haematoxylum the secondary veins are ascending, and form a sharp angle with the primary vein. Furthermore, armature among these genera differs, with curved and deflexed prickles on the stems and inflorescence rachis in C. pearsonii, straight spinescent shoots in Haematoxylum, and straight, conical spines scattered along the branches in Lophocarpinia, which also has distinctively modified lateral, short, spine-tipped branchlets (Fig. 5H). Given the apparently isolated phylogenetic position of this taxon and its morphological distinctiveness, we recognise this species as a new genus, Hererolandia, a name referring to the type locality of H. pearsonii, which Bolus originally described as coming from "Hereroland" in Namibia, and also chosen to honour the Herero people of that country.
Th e third unplaced taxon, C. echinata, also has several unusual morphological features. Th e pods of C. echinata combine characteristics of Guilandina and Caesalpinia s. s. Th e patent, sub-woody bristles on the pod valves ( Fig. 9B) are reminiscent of Guilandina pods ( Fig. 20D and E), but the fruit is laterally compressed with lunate-falcate valves that twist after dehiscence and the seeds are fl attened, as in many species of Caesalpinia s. s. In contrast to Caesalpinia s. s. and Guilandina, C. echinata has reddish heartwood ( Fig. 9F) which is a source of red dye (also found in C. sappan in the C. decapetala clade and in Haematoxylum). Caesalpinia echinata forms a medium-sized to large tree ( Fig. 9E) with unusual upcurved prickles arising from woody protuberances on the trunk and branches (Fig. 9C). In our analyses, multiple accessions of C. echinata form a clade in the ITS and ycf6-psbM gene trees and in the combined analysis ( Fig. 3A), but in the other plastid gene trees there is no resolution amongst these accessions, suggesting lack of time for coalescence sensu Pennington and Lavin (2016) (Suppl. material 1). Caesalpinia echinata populations along the Atlantic coast of Brazil have been shown to be strongly diff erentiated genetically (Cardoso et al. 1998, Lira et al. 2003) and morphologically variable (Lewis 1998, De Lima et al. 2002. Denser sampling and detailed phylogeographical analyses are needed to assess whether these morphotypes represent a continuum or a set of discrete entities worthy of taxonomic recognition. Regardless, we consider that C. echinata should be recognised as a distinct genus based on the available morphological and phylogenetic evidence. We propose the genus name Paubrasilia, based on the common name pau-brasil and in reference to the fact that Paubrasilia is the national tree of Brazil with a long and important association with the country.

Unresolved generic relationships
Th ree areas of the phylogeny remain unclear and warrant greater sampling before making further adjustments to the generic classifi cation. We hypothesise, based on morphology and preliminary phylogenetic results, that nine species from mainland Asia will form a well-supported clade with C. crista (previously referred to as the C. nuga clade; Gagnon et al. 2013), which is sister to Pterolobium and which also remains sparsely sampled (Fig. 3B). However, only two of these nine species, C. crista and C. vernalis (the latter not included in the combined analysis due to missing data, but placed in this clade in the rps16 gene tree in Gagnon et al. (2013)), have been sampled so far. If this putative C. crista clade is indeed supported as monophyletic with greater taxon sampling, the oldest available generic name for the clade would be Ticanto Adans. It is notable that two of the species from mainland China (C. caesia and C. sinense) sometimes have a small wing on the fruit suggesting a fruit intermediate between the typical samara of Pterolobium and the wingless pods of species of the Ticanto clade. Th is morphological variation highlights the need for thorough sampling and detailed study to arrive at a better understanding of generic delimitation of this group (for more details see Clark 2016).
Th e other questionable taxa are the monospecifi c genera Lophocarpinia and Stenodrepanum, both of which could potentially be sunk into other genera. However, because only trnL-trnF and matK-3'trnK, the two least informative markers in our study, were sequenced for these two genera, their phylogenetic placements remain weakly or moderately supported. As found by Nores et al. (2012), Lophocarpinia is moderately supported as sister to Haematoxylum (Figs 2 and 3A, clade A). Burkart (1944Burkart ( , 1952 proposed that Lophocarpinia could be synonymised under Haematoxylum due to the strikingly similar vegetative morphology of the two genera, and despite the very distinctive lomentaceous and coarsely serrate-margined winged fruits of Lophocarpinia (Figs 5I and 6). Similarly, Stenodrepanum and Hoff mannseggia are weakly supported as sister taxa, and are distinguished morphologically only by their fruits which are cylindrical and torulose in Stenodrepanum and fl attened in Hoff mannseggia (Fig. 34 F,H and K). Although these two generic pairs are diff erentiated on fruit characters alone, we refrain from proposing any taxonomic changes until additional sequence data can be obtained.

Morphological variation in the Caesalpinia group
Th e Caesalpinia group has long been considered a morphologically heterogeneous group, in which morphological homoplasy and convergence have plagued previous attempts to provide a satisfactory generic system (see Lewis and Schrire 1995, Lewis 1998. As circumscribed here, the Caesalpinia group includes 27 robustly supported major lineages (26 of which are formally recognised here as genera). Although there are no unique diagnostic morphological synapomorphies for the clade as a whole, the Caesalpinia group can be recognised by a combination of features, including the presence of glandular trichomes, prickles and spines, bilaterally symmetrical fl owers with a somewhat modifi ed lower sepal, and free stamens crowded around the pistil; fl owers vary greatly and can be strongly modifi ed depending on pollination system, and fruits across the clade are extremely diverse refl ecting a striking variation in seed dispersal strategies. Our new molecular phylogenies (Figs 2,3) suggest that a number of leaf, armature and fruit characteristics can be used to distinguish genera and delimit the major clades, being exclusive, with minor exceptions, to particular clades. For example, bipinnate leaves with a terminal pinna occur almost exclusively in species of clade II, whereas almost all the species having bipinnate leaves without a terminal pinna are members of clade I. Similarly, clade II contains only species that lack thorns, spines or prickles, and almost all species that lack idioblasts in their leafl ets (the latter are also absent in C. mimosoides in clade I (Lersten and Curtis 1996) and in Haematoxylum), and almost all species in clade II are characterised by the presence of multi-cellular glandular structures on the stems, leaves and infl orescences (although Haematoxylum dinteri, Caesalpinia mimosoides, and members of Coulteria in clade I also have glandular structures on the margin of the pectinate lower cucullate sepal). In contrast, clade I contains all the species that are armed with spines and prickles along the branches (although Coulteria, C. madagascariensis and C. nipensis lack thorns, spines or prickles), and which have idioblasts in the lamina of their leafl ets. Th e nearly mutually exclusive distribution of external glands vs. spines+idioblasts gives some support to the idea that these structures constitute alternative plant defense strategies against herbivory Curtis 1994, 1996), even though the role and function of idioblasts and secretory glands in the Caesalpinia group have never been studied in detail.
At the generic level, fruits are highly variable and taxonomically more useful than fl owers. Several of the genera we recognise here can be diff erentiated based on fruit characteristics. For example, the pods of Balsamocarpon, Cenostigma, Guilandina, Haematoxylum, Hererolandia, Hultholia, Libidibia, Lophocarpinia, Moullava, Mezoneuron, Paubrasilia, Pterolobium and Zuccagnia are all distinctive and provide useful diagnostic synapomorphies for these genera (Figs 5,9,14,18,20,24,30,34). In contrast, only a few fl oral synapomorphies are diagnostic at the generic level: Guilandina species have sepals that are valvate in bud; in the Balsamocarpon, Zuccagnia, and Hoff mannseggia clade, sepals are persistent until fruiting (Fig. 34), except in Stenodrepanum (Fig.  34); and in Pomaria species, the androecium and gynoecium are cupped in the lower cucullate sepal ( Fig. 39A-C, F). In general, however, fl oral morphology within clades is highly variable refl ecting diff erences in pollination syndromes, including examples of melittophily, chiropterophily, psychophily, phalaenophily and ornithophily, sometimes occurring among closely related congeneric species (e.g. Caesalpinia s. s., as emended here, and Erythrostemon-see above and Figs 11 and 42). Th ese repeated fl oral morphologies across disparate members of the Caesalpinia group suggest convergent evolution of similar pollination modes in multiple clades across the group.

Taxonomy
Here we present a comprehensive phylogenetically-based and signifi cantly revised generic classifi cation of the Caesalpinia group recognizing 26 genera, including reinstatement of two previously described genera, re-circumscription of eight genera and description of four new genera. A 27 th genus (Ticanto) is provisionally indicated, but not formally reinstated. A key to the identifi cation of genera, full generic descriptions, and illustrations of all genera are presented. In addition, we provide new combinations where necessary and where we are confi dent about species affi nities and taxonomy (Biancaea, Cenostigma, Erythrostemon, Hererolandia, Hultholia, Libidibia, Moullava, Paubrasilia) and/or lists of accepted species names (in bold) associated with each genus, as well as references to recently published species-level taxonomic accounts. For the genera Guilandina, Coulteria and Ticanto, only a preliminary list of species names (not bold) is indicated, with no nomenclatural combinations provided. Th ese genera remain poorly understood taxonomically and work is currently ongoing in Coulteria to clarify and delimit species (Sotuyo et al., submitted).

Key to the genera of the Caesalpinia group
Genus 27 Ticanto is provisionally indicated, pending further studies to establish the status of the genus
Description. A multi-stemmed shrub to 2 m, but usually less than 1 m tall, armed with curved, defl exed, 7 mm long prickles scattered along the branches; bark white or brown; stems terete and slightly sinuous, with a fi ne silvery indumentum on the young twigs, older stems glabrescent. Stipules not seen. Leaves pinnate, 7-17 mm long, subsessile, borne in fascicles on short woody brachyblasts that are usually subtended by a pair of tiny (sometimes obscure) prickles; leafl ets opposite, (4-) 5-7 (-9) pairs per pinna, eglandular, covered in a fi ne silvery pubescence, 5-6.5 × 2.5-3 mm, elliptic to oblong-elliptic, apex obtuse, with an acuminate tip, main vein prominent, secondary venation not visible. Infl orescence a short raceme of bisexual fl owers, about 5 cm long, usually borne on brachyblasts, covered in a fi ne silvery pubescence, with prickles along the infl orescence rachis; bracts about 2-3 × 1.5 mm, ovate, apex acute, caducous. Flowers zygomorphic; calyx with a short hypanthium, and 5 free sepals, c. 3-5 mm long, fi nely white pubescent, with the lower sepal cucullate and covering the other 4 sepals in bud, all sepals caducous, but hypanthium persistent as a ring around the stipe of the fruit; petals 5, yellow, free, c. 6-9 mm long, obovate; stamens 10, free, up to 10 mm long, eglandular, pubescent on the lower half; ovary pubescent, stigma a fringed and slightly indented chamber. Fruit a thinly woody, laterally compressed, almost circular to strongly sickle-shaped pod, c. 2-2.3 × 1-1.5 cm, dehiscing along the sutures, fi nely pubescent and covered in robust trichomes up to 6 mm long, usually 1-seeded. Seeds laterally compressed, about 6-8 mm long.
Geographic distribution. A monospecifi c genus endemic to Namibia, on the Great Escarpment.
Habitat. Semi-desert and desert areas, on stony, sandy soils. Etymology. Semiarid Hereroland, a region of eastern Namibia, is the type locality of H. pearsonii. Th e Herero people who inhabit this region are nomadic cattle herders  and it is they and their region that are honoured in the name proposed for this monospecifi c genus, endemic to this restricted area of Namibia.
Geographic distribution. A monospecifi c genus restricted to Argentina and Paraguay (possibly also occurring in Mato Grosso do Sul, Brazil, pers. comm. H. C. de Lima).
Habitat. Chaco woodland and seasonally dry tropical to subtropical forest. Etymology. From lopho-(Greek: combed or crested) and carpos (Greek: fruit), the fruit has 4 crested wings, the ending -inia signifi es a close relationship with Caesalpinia.
Geographic distribution. Haematoxylum comprises fi ve species: two in Central America (Salvador to Costa Rica), Mexico, South America (Colombia and Venezuela) and the Caribbean (perhaps introduced), two endemic to Mexico and one in Southern Africa (Namibia).
Habitat. Deserts, seasonally dry tropical semi-deciduous scrub and thorn scrub, sandy river beds and dry rocky hillsides. One species (H. campechianum) is known to grow in frequently inundated marshy areas by rivers.
Etymology. From haemato-(Greek: bloody) and xylon (Greek: wood), alluding to the blood-red heartwood of H. campechianum L. which produces a brilliant red dye.
Geographic distribution. A monospecifi c genus endemic to Eastern Brazil, in the states of Pernambuco, Bahia, Espirito Santo and Rio de Janeiro. Widely cultivated in Brazil as an ornamental street or park tree, and sometimes in plantations.
Habitat. Dry coastal cactus scrub often on rocky outcrops, inland in Mata Atlântica, and in tall restinga on well-drained sandy soil.
Etymology. "Pau-brasil" is the national tree of Brazil, and has long been associated with the country. Its red sap was once used for dying cotton and cloth and its wood is much prized for the manufacture of high quality violin bows. Originally described as Caesalpinia echinata by Lamarck in 1785, it is appropriate that this phylogenetically isolated taxon should be placed in its own monospecifi c genus and a Latinization of its well-known and much used common name recognises the importance of the species to Brazil. For a detailed account of this iconic species refer to Pau-brasil by E. Bueno [et al.], São Paulo, Axis Mundi (2002).
Type. Caesalpinia brasiliensis L. Emended description. Shrubs or small trees, usually 1-6 m tall, armed with curved defl exed prickles (except C. nipensis which is unarmed), these either in pairs at the base of leaves, or scattered along the shoots (or both), or sometimes on woody protuberances at the base of trunks and stems; young shoots terete, glabrous and eglandular. Stipules not seen. Leaves alternate, bipinnate, c. 4-30 cm long, ending with a pair of pinnae, unarmed, or sometimes with a pair of prickles at the insertion of the pinnae on the leaf rachis, sometimes also at the insertions of the leafl ets on the pinna rachis; pinnae opposite, in (1-) 2-6 pairs per leaf; leafl ets alternate to opposite, in 3-13 pairs per pinna, short-petiolulate, blades suborbicular, obovate or elliptic, apex mucronate, rounded or emarginate, base cuneiform, rounded or oblique; main vein centric, secondary veins reticulate. Infl orescence a terminal or axillary raceme or panicle of pedicellate, bisexual fl owers, c. 5-37 cm long, unarmed; bracts lanceolate or ovate, apex acute to acuminate, caducous. Flowers zygomorphic, c. 13-25 mm long; calyx comprising a hypanthium with 5 sepals, that are each c. 7-17 mm long, glabrous to occasionally fi nely puberulous, always eglandular, the lower sepal strongly cucullate and covering the other 4 sepals in bud, all sepals caducous, but hypanthium persistent as a free ring around the pedicel as the fruit matures; petals 5, variable in colour (yellow, white, red, orange, or green; certain horticultural varieties are also pink), the corolla also variable in shape (related to diff erent pollination systems: bees, butterfl ies, birds and bats); stamens 10, free, c. 10-65 mm long, the fi laments pubescent, eglandular; ovary glabrous and eglandular. Fruit a wingless, unarmed, coriaceous, glabrous, eglandular, oblong-elliptic, or linear pod, with a marcescent style forming an acute  apex, c. 34-120 × 7-26 mm, explosively dehiscent, with twisting valves, 3-7-seeded. Seeds laterally compressed, obovate, up to 10 mm in diameter.
Geographic distribution. Caesalpinia, as re-circumscribed here, is reduced to around nine species (a detailed taxonomic revision is needed to properly delimit species), and is now restricted to the Neotropics (apart from the pantropically cultivated C. pulcherrima). All the Old World species previously included in Caesalpinia s.s. sensu Lewis (2005) are here transferred to other genera. One species (C. cassioides) occurs in the northern Andes from Peru to Colombia, one (C. pulcherrima) is likely native in Guatemala and the state of Sonora in Mexico), two occur in the Caribbean (one, C. nipensis, is endemic to Cuba, the other widely distributed and possibly divisible into six separate species, all of which are listed below). Caesalpinia pulcherrima is a widely cultivated ornamental throughout the tropics. It includes red, orange, pink, and pure yellow-fl owered forms and cultivated specimens are usually unarmed and lack bristles (unlike wild specimens which are armed and bristly).
Habitat. Seasonally dry tropical forests, coastal thicket, bushland and thorn scrub, dry plains and riparian woodland, on soils derived from limestone or sandstone.
Etymology. Named by Linnaeus for Andrea Cesalpino (1519-1603), Italian naturalist, botanical collector, systematist and philosopher, physician to Pope Clement VIII, professor of medicine and botany in Pisa and Rome.
References. Britton and Rose (1930); Macbride (1943: 191, 194-195); Ulibarri (1996); Barreto Valdés (2013). Diagnosis. Denisophytum is closely related to Tara (Fig. 3), but diff ers in having fl owers with a lower cucullate sepal with an entire margin (vs. a lower cucullate sepal with a pectinate margin), and dehiscent, coriaceous, laterally compressed pods (except for D. madagascariense which has infl ated fruits) (vs. indehiscent, somewhat fl eshy, coriaceous pods that are slightly turgid). Morphologically, species of Denisophytum are most likely to be confused with those of Caesalpinia s.s., but no reliable diagnostic characters have been found to diff erentiate these two genera. Th e corolla of Denisophytum species is consistently yellow and the fl owers are bee pollinated, whereas Caesalpinia s.s. species display a wide range of fl ower colour (yellow, orange, red, green and white) and pollination syndromes (chiropterophily, ornitophily, psychophily and mellitophily).
Geographic distribution. Denisophytum comprises nine taxa in eight species, found across North America, South America and Africa, including Madagascar, a classical highly disjunct trans-continental distribution typical of lineages occupying the succulent biome sensu Schrire et al. (2005). Th ree species are distributed in Mexico, Florida, and the Caribbean, one species is endemic to Paraguay and Argentina, one is endemic to northern Madagascar, and the other three occur in northern Kenya, Somalia and Arabia. An evaluation of species limits is needed in this group.
Habitat. Low deciduous seasonally dry tropical woodland or scrubland, also in open pineland or coastal plains and foothills. Species in Madagascar and Africa grow in limestone soils.
Etymology. Th ere is no indication of the etymology of Denisophytum in the posthumous publication of the generic name. Nevertheless, it is quite likely that the author, René Viguier, had intended to honour his friend and collaborator, Marcel Denis, a botanist with expertise in the genus Euphorbia in Madagascar. Sadly, M. Denis passed away prematurely at the age of 33 in 1929 (Allorge and Allorge 1930).
Geographic distribution. A genus of three species, one in South America (T. spinosa thought to be native to Peru and Ecuador), one in Mexico (T. cacalaco) and one in Mexico, Guatemala, Nicaragua and extending into the Caribbean (T.vesicaria). Tara spinosa is also widely cultivated across the tropics and subtropics (including in the Canary Islands) as a source of tannins and occasionally as an ornamental.
Habitat. Seasonally dry tropical forest to semi-arid thorn scrub. Etymology. Derived from the vernacular name 'tara' in Peru, Bolivia and Chile. Notes. Based on Gagnon et al. (2013), Molinari-Novoa and Sánchez Ocharan (2016) transfered C. cacalaco and C. vesicaria to the genus Tara, but did not emend the description of the genus, which we provide above.
Type. No type designated in the original publication, nor since. Type designated here: Coulteria mollis Kunth.
Geographic distribution. A genus of approximately seven species in Mexico and Central America, one species extending to Cuba, Jamaica and Curaçao, one to Venezuela (including Isla Margarita) and Colombia.
Habitat. Seasonally dry tropical forest, deciduous woodland and dry thorn scrub, some species occurring on limestone.
Etymology. Named by Kunth for the Irish botanist Th omas Coulter (1793-1846) who collected in central Mexico (1825Mexico ( -1834 and was curator of the herbarium at Trinity College, Dublin, Ireland.
Notes. A revision of the genus has been submitted by S. Sotuyo, J. L. Contreras, E. Gagnon, and G. P. Lewis. Th e list of species names presented here simply includes all names associated with the genus Coulteria and will be reduced in the forthcoming taxonomic account.
References. Britton and Rose (1930: 320-322); Ulibarri (1996); Zamora Villalobos (2010) Diagnosis. Gelrebia is morphologically similar to Caesalpinia s. s. but the two genera diff er somewhat in habit, with Gelrebia species being erect to scrambling shrubs (vs. erect shrubs or small trees), in having dark pinkish mauve to light pinkish-white fl owers (vs. fl owers that are variable in colour, from yellow, white, red and orange to green), and coriaceous, broadly oblong-ovoid to obliquely pyriform pods, with a large, oblique, rounded base (vs. coriaceous, oblong-elliptic to linear pods, with an oblique cuneate base).
Geographic distribution. A genus of nine taxa in eight species, restricted to Africa, in Namibia, Botswana, South Africa, Northern Kenya, Ethiopia, and Somalia. One species also found in the Democratic Republic of the Congo (Zaire, Katanga).
Habitat. Deciduous bushland, dry woodlands, on rocky ridges, often along dry river beds, or on sandy valley fl oors. One species also found in degraded savanna, close to termite mounds.
Geographic distribution. Th e single species is distributed across Asia, in China (Yunnan), Bangladesh, India, Laos, Myanmar (Burma), Th ailand and Vietnam.
Habitat. In secondary thickets and clearings, often on roadsides, up to 1500 m elevation. More information on the ecology of this genus is needed.
Etymology. Th e name Hultholia honours the Cambodian botanist Dr. Sovanmoly Hul Th ol (born 1946), whose doctoral thesis, "Contribution à la révision de quelques genres de Caesalpiniaceae, representés en Asie" (1976), is an important revision of the Asian species and genera of the Caesalpinia group, and particularly the genus Pterolobium. Dr. Hul Th ol retired from the Museum National d'Histoire Naturelle, Paris in 2014, but continues as an honorary researcher. She is a specialist on the fl ora of Cambodia and South East Asia, directed the publication of multiple volumes of the Flora of Cambodia, Laos and Vietnam from 1995, and is one of the co-founders of the National Herbarium of Cambodia, Royal University of Phnom Penh.
Notes. Although Hultholia mimosoides is not known to be cultivated, the young, pungent, fl owering shoots are sold as a vegetable in markets in Vientiane (Laos) (Vidal and Hul Th ol 1976).
References. Vidal and Hul Th ol (1976); Chen et al. (2010a: 42-43   Description. Lianas, woody climbers, scrambling or trailing shrubs, often forming dense tangled clumps, densely armed with recurved prickles on branches and shoots, as well as in pairs at leaf bases (except Caesalpinia murifructa and closely related species in the Caribbean which are unarmed). Stipules foliaceous to subulate, sub-persistent or caducous. Leaves bipinnate, ending with a pair of pinnae, prickles present in pairs at the insertion of pinnae and scattered on the leaf rachis, and at the insertion of leafl ets on the pinnae rachises; leafl ets oblong, apex obtuse and mucronulate to acuminate, base rounded. Infl orescences supra-axillary or terminal racemes, 30-60 cm long; bracts narrow, lanceolate, aristulate, 1 mm long, to conspicuous and exceeding fl oral buds, caducous. Flowers unisexual, segregated on separate male and female racemes, the female fl owers cryptically bisexual with 10 fully formed stamens, but these produce no pollen; male fl owers with a highly reduced, non-functional pistil, zygomorphic to sub-actinomorphic; calyx with a hypanthium and 5 almost equal sepals, these valvate in bud, the lower sepal slightly cucullate, the hypanthium and sepals caducous, leaving no persistent calyx ring, eglandular, without spines (except Madagascan Caesalpinia delphinensis in which the calyx is armed with slender prickles); petals 5, free, yellow, barely exceeding the sepals; stamens 10, free, pubescent near the fi lament base; ovary usually covered in bristly trichomes, except in a few species, including Caesalpinia solomonensis and Caesalpinia murifructa. Fruits oblongelliptic, infl ated pods, usually armed with 5-10 mm long spinescent bristles, apex terminating in a beak, base acute, 1-4-seeded. Seeds obovoid to globular, c. 2 cm in diameter, smooth, grey, pale to dark brown, or orange, with parallel fracture lines concentric with the small apical hilum.
Geographic distribution. Th is pantropical genus lacks a recent global taxonomic account and there are doubts about the number of species, with previous estimates ranging from seven to as many as 19. Species occur from as far north as Japan, south to South Africa, with three species in the Caribbean, one in China, India, Myanmar (Burma), Indo China, Hong Kong and Taiwan, one endemic to Madagascar, one in Australia, and two widespread across the Old and New World tropics.
Habitat. Coastal thickets on sand, in secondary forest, and lowland rain forest, occasionally on limestone.
Etymology. Named by Linnaeus for Melchior Wieland (1515-1589), Prussian naturalist, traveller and scholar from Königsberg, who settled in Italy and italianised his name to 'Guilandini', or Guilandinus in Latin; he was sent to the Levant, Asia and Africa (1559-1560), was captured by pirates and fi nally ransomed by Gabriele Falloppio.
Notes. Pending a complete taxonomic revision, the list of 19 names presented below provides a guide to potential species content in Guilandina, but includes no synonymy and no information on types, nor any new nomenclatural combinations for the fi ve species of Caesalpinia that as yet have no published name in Guilandina. References. Britton and Rose (1930: 336-341); Wilczek (1951); Brenan (1967); Gillis and Proctor (1974); Hattink (1974); Vidal and Hul Th ol (1976); Du Puy and Rabevohitra (2002: 46-48
Geographic distribution. A genus of four species, three in south Asia: India, Nepal, Myanmar (Burma), Th ailand, Laos, Cambodia, Sri Lanka, southern China (Yunnan and Hainan), and the Malay Peninsula and Archipelago, and one in Africa: Cameroun, Gabon, the Democratic Republic of Congo, Angola, Zambia (Kabompo Dist.), Uganda and Tanzania (Kigoma Dist.).
Habitat. Th e Asian species are found in seasonally dry tropical semi-evergreen forest margins, secondary thickets, and on mountain slopes, up to 1200 m elevation. Th e African species occurs mostly in riverine habitats in lowland rainforests.

Diagnosis.
Biancaea is closely related to Mezoneuron, diff ering principally in its fruit, a coriaceous, laterally compressed, wingless, dehiscent pod (except B. decapetala, which has somewhat infl ated, boat-shaped pods, often with a narrow wing or ridge along the upper suture). In contrast, Mezoneuron has chartaceous, coriaceous or ligneous pods, which are also laterally compressed, but indehiscent, and with a wing along the upper suture. In addition, the ovary of Biancaea species always has a velvety indumentum (vs. glabrous to pubescent in Mezoneuron).

Geographic distribution.
A genus of six species widespread across southern Asia, from India, to Myanmar (Burma), Th ailand, Cambodia, Vietnam, south China, Japan, the Philippines, and the Malay Peninsula and Archipelago, one species endemic to Sabah (near Sandakan). Biancaea decapetala, native to Asia, has been widely introduced across the tropics as a hedge plant or ornamental and is considered to be invasive in South Africa and Hawaii.
Habitat. Primary forest and forest margins, grasslands, scrub vegetation, riverine habitats, secondary thickets and clearings. From the coast to mountain slopes.
Notes. Based on the study of Gagnon et al. (2013), Molinari-Novoa et al. (2016) provided some, but not all, of the required nomenclatural transfers to the genus Biancaea. Furthermore, they did not emend the description of the genus, as provided here.
Geographic distribution. A genus of 10 species; one in southern tropical Africa, East Africa and Arabia, nine in SE Asia (one endemic to India, two in China, four in Indo-China [one endemic to Th ailand, two extending to Malesia], three restricted to the Malay Peninsula and Archipelago [one endemic to the Philippines]).
Habitat. Seasonally dry tropical upland evergreen forest, riverine and humid forest, woodland and wooded grassland.
Etymology. From ptero-(Greek: wing) and lobion (Greek: pod, fruit), in reference to the fruit which is a samara.
Notes. Vidal and Hul Th ol (1974) published a revision of Pterolobium, with a key to species. We provide below a list of species currently accepted in the genus, taking into account the treatment of P. sinense as a synonym of P. macropterum (Chen et al. 2010b).

Pterolobium borneense
Geographic distribution. A genus of 24 extant species, mainly in Asia, extending to Australia, Polynesia, Madagascar and Africa; two species on mainland Africa (one widespread in West Africa, the other in both West, East and Southeast Africa); one endemic to Madagascar; fi ve endemic to New Caledonia; one endemic in Hawaii; one in Vietnam; four endemic to Australia (Queensland and New South Wales); one endemic in the Philippines; one in Australia and Papua New Guinea; nine species more widespread across Asia.
Habitat. Tropical and subtropical riverine forest, lowland rain forest, swamp forest, seasonally dry forest, thicket, vine forest and wooded grassland, especially along forest and river margins.
Etymology. From meso-(Greek: middle) or meizon (Greek: greater) and neuron (Greek: nerve), the upper suture of the fruit is bordered by a usually broad longitudinal wing so that the suture appears as a prominent sub-central nerve or vein.
Geographic distribution. A monospecifi c genus in E Africa (Kenya and Tanzania) and N Madagascar.
Habitat. Seasonally dry tropical forest, woodland on limestone and in riverine forest.

Diagnosis. Cenostigma is morphologically most similar to the genus Erythrostemon.
It diff ers from the latter by its leaves with alternate to subopposite (occasionally opposite) leafl ets (vs. leafl ets consistently opposite in Erythrostemon). A number of other characters can help to distinguish between the two genera, but these are not constant across species of Cenostigma. For example, a stellate indumentum on the leafl ets, infl orescences, and/or sepals is found on some, but not all Cenostigma species, but is always lacking in Erythrostemon. Black subepidermal glands (visible with a × 20 lens) can be found scattered in the undersurface of leafl ets and/or on sepals in Cenostigma (vs. these always lacking in Erythrostemon). Cenostigma pods are generally woody with thickened margins or an adaxial, proximal woody ridge or crest (vs. less robust pods lacking any woody ridge or crest in Erythrostemon).
Geographic distribution. We recognise 20 taxa in 14 species, all of them neotropical; only two of these taxa do not require new names, while the rest are species of Caesalpinia here transferred to Cenostigma. Th e majority of species are found in central and NE Brazil, including parts of the Amazon. Two species extend around the circum-Amazonian arc of dry forests and adjacent cerrado, including in Paraguay, Argentina and Bolivia, and one taxon is also found in the seasonally dry inter-Andean valleys of Peru. Species  are also found throughout Central America, from Panama northwards and in Mexico, extending to the Caribbean, with endemics in Cuba and Hispaniola.
Etymology. From ceno-(Greek: empty) and stigma, presumably alluding to the chambered stigma (a character of many species of the Caesalpinia Group, and not restricted to Cenostigma).

Diagnosis.
Libidibia is related to Hoff mannseggia, Stenodrepanum, Balsamocarpon and Zuccagnia but diff ers in being a genus of medium to tall trees, 6-20 m in height (versus woody based perennial herbs to shrubs, 10 cm to 5 m tall), most species have a distinctive, smooth patchwork bark in shades of white, grey and green ("snake skin bark") a characteristic not found in the other four genera. Libidibia (except L. monosperma) has bipinnate leaves (Balsamocarpon and Zuccagnia are pinnate) and coriaceous or woody, glabrous, eglandular, indehiscent fruits which dry black (red in L. monosperma) versus thick, turgid, glandular, resinous, indehiscent fruits (Balsamocarpon), or laterally compressed, gall-like, ?indehiscent fruits covered in trichomes (Zuccagnia). Stenodrepanum and Hoff mannseggia are bipinnate but the fruits of most species of Hoff mannseggia are dehiscent with twisting pod valves and persistent sepals (in Libidibia sepals are caducous in fruit); the fruits of Stenodrepanum are narrow, cylindrical and torulose.
Geographic distribution. A genus of ten taxa in seven species in the Neotropics. One species in Mexico, one widespread in Brazil, one in Colombia, Venezuela and the Antilles, one in Colombia, Ecuador and Peru, one in Paraguay, Bolivia, Argentina and SW Brazil, one (Libidibia monosperma, previously in the monospecifi c genus Stahlia) endemic to Puerto Rico and the Dominican Republic, and L. coriaria widespread throughout Mexico, Central America, the Caribbean and NW South America. Other species perhaps waiting to be discovered and described, both in the fi eld and in herbaria; the genus needs revising.
Habitat. Seasonally dry tropical forest and thorn scrub (including Brazilian caatinga) and savanna woodland. Libidibia monosperma occurs along the margins of mangrove swamps and in marshy deltas, in drier edaphic conditions. Etymology. Th e name Libidibia is derived from the vernacular name 'libi-dibi' or 'divi-divi' used for some species.
Geographic distribution. A monospecifi c genus endemic to northern Chile, from the Coquibo and La Serena valleys.
Geographic distribution. A monospecifi c genus restricted to Chile, NW and central-W Argentina.
Habitat. Dry temperate upland and montane bushland and thickets on sandy plains. Etymology. Named by Cavanilles for the Italian physician, traveller and plant collector, Attilio Zuccagni (1754-1807).

Zuccagnia punctata Cav.
Geographic distribution. A monospecifi c genus endemic to central and western Argentina.
Habitat. Subtropical wooded grassland and scrub, especially close to salt pans. Etymology. From steno-(Greek: narrow) and drepano-(Greek: sickle), in allusion to the narrow sickle-shaped fruit.
Description. Perennial woody herbs, most species forming a basal rosette, or subshrubs to 3 m, unarmed, often arising from bud-bearing and tuberous roots, shoots pubescent and with gland-tipped trichomes. Stipules not seen. Leaves alternate, bipinnate, ending in a pair of pinnae plus a single terminal pinna (except for H. aphylla); pinnae opposite, in 1-13 pairs; leafl ets small and numerous, in 2-15 (-18) pairs per pinna, glabrous to pubescent, and glandular. Infl orescences terminal or axillary racemes; bracts often caducous. Flowers bisexual, zygomorphic; calyx comprising a hypanthium and 5 sepals, these weakly imbricate, persistent as pods mature  (except in H. microphylla and H. peninsularis, where they are not always persistent); petals 5, free, yellow to orange, the median petal often with red markings; stamens 10, free, fi laments pubescent; ovary glabrous to pubescent, eglandular to glandular, stigma apical, concave. Fruit a laterally compressed, straight or sometimes falcate pod, the sutures almost parallel, papery to leathery, glabrous to pubescent, eglandular or with glandular trichomes, indehiscent or dehiscent, with twisting valves. Seeds compressed, ovoid.
Geographic distribution. Hoff mannseggia comprises 25 taxa in 23 species and occupies a classical amphitropical distribution in the New World with 10 species restricted to North America (southern USA and Mexico), 12 in South America (Peru, Bolivia to south-central Argentina and Chile, mainly Andean), and one species (H. glauca (Ortega) Eifert) widespread throughout the range of the genus.
Habitat. Subtropical desert and semi-desert grassland, often in open areas and on disturbed sites, on sandy, rocky or calcareous soils.

Hoff mannseggia aphylla (Phil.) G.P. Lewis & Sotuyo 23.2 Hoff mannseggia arequipensis
Geographic distribution. Th e genus Arquita comprises six taxa in fi ve species restricted to the Andes in South America, in disjunct inter-Andean valleys, in Ecuador, Peru, Bolivia and Argentina. Habitat. Seasonally dry, montane, rupestral habitats in inter-Andean valleys. Etymology. Th e name Arquita derives from the vernacular name of A. trichocarpa in Argentina (Ulibarri 1996).

Type. Pomaria glandulosa Cav.
Description. Small shrubs, subshrubs or perennial herbs, with a moderate to dense indumentum of simple curled hairs, sometimes also scattered plumose trichomes, intermixed with sessile, oblate glands (drying black) on stems. Stipules laciniate, pubescent, glandular, persistent. Leaves alternate, bipinnate, pinnae in 1-8 (-11) pairs plus a terminal pinna; leafl ets small, in 2-16 (-27) pairs per pinna, always with multiple sessile glands on their lower surface (these orange in the fi eld, drying black). Infl orescence a terminal or axillary raceme; bracts caducous. Flowers bisexual, zygomorphic; calyx comprising a hypanthium and 5 lanceolate sepals, the lower sepal cucullate, covering the other 4 in bud, and closely embracing the androecium and gynoecium at anthesis, sepals not persistent in fruit; petals 5, free, yellow, white, red or pink; stamens 10, fi laments pubescent; ovary sparsely to densely hairy and glandular, stigma lateral. Fruit a linear or sickle-shaped, laterally-compressed pod, apex acute, with a sparse to dense covering of plumose/dendritic or stellate trichomes (these sometimes obscure and restricted to the fruit margin) intermixed with sessile oblate glands (drying black), elastically dehiscent, with twisting valves. Seeds laterally compressed.
Geographic distribution. A genus of 17 taxa in 16 species: nine in North America (south-eastern USA, central and northern Mexico), four in South America (south-east-  ern Brazil, Paraguay, and Argentina), and three in southern Africa (Namibia, Botswana and South Africa).
Habitat. Mainly in subtropical dry grassland and in degraded sites, many on limestone. Etymology. Named by Cavanilles for Dominic Pomar, botanist from Valencia, and doctor to Philip III (1598-1621), King of Spain.
Notes. Revisions of the species of Pomaria are available for North America (Simpson, 1998), South America and Africa (Simpson and Lewis 2003), and southern Africa (under the name Hoff mannseggia, Brummit and Ross 1974). A list of accepted species is given below, but excludes types and synonymy which are available in the aforementioned revisions.

Diagnosis.
Erythrostemon is closely related to Pomaria, but diff ers in habit, consisting of large shrubs and small to medium sized trees, or occasionally suff rutices (vs. shrubs, suff rutices, or perennial herbs in Pomaria). It also diff ers by its ovatelanceolate to orbicular sepals (vs. linear, laciniate sepals in Pomaria), leafl ets that are either eglandular or with conspicuous black sessile glands along the margin, these sometimes sunken in the sinuses of the crenulated margin (vs. leafl ets with multiple glandular dots on the lower leafl et surfaces, that are orange in the fi eld, drying black), the androecium and gynoecium free from the calyx (vs. the androecium and gynoecium cupped in the lower cucullate sepal), defl exed petals (vs. the two lower petals forming a horizontal platform above the lower cucullate sepal), and oblong-elliptic pods, the valves chartaceous to slightly woody, glabrous to pubescent, eglandular or with stipitate glands (vs. linear to sickle-shaped pods, the valves glabrous or with plumose trichomes and stipitate glands).
Geographic distribution. Th e genus comprises 34 taxa in 31 species. Its circumscription is emended here to include many species previously placed in Central American and Mexican Poincianella. 22 species are found across the southern USA, Mexico and Central America, one occurs in the Caribbean (Cuba and Hispaniola), eight occur in South America, with one endemic in the caatinga vegetation of Brazil, and the other seven in Argentina, Bolivia, Chile, and Paraguay.
Habitat. Low-elevation seasonally dry tropical forests across Mexico, Central America, the Caribbean and in caatinga vegetation in Brazil; also in patches of dry forest, deserts, yungas-puna transition zones, and chaco-transition forests in Argentina, Bolivia, Chile and Paraguay.
Etymology. From erythro-(Greek: red) and stemon (Greek: stamen), the type species E. gilliesii (Wall. ex Hook.) Klotzsch has long red exserted stamens, but this is unusual in the genus as circumscribed here.
Notes. Species descriptions (under Caesalpinia binomials) are available in Lewis (1998). A key is also available in that revision, but it includes species now considered to belong in Cenostigma, Arquita, and Hoff mannseggia.
References. Britton and Rose (1930); Burkart (1936: 82-84, 97-108); Ulibarri (1996); Lewis (1998);De Queiroz (2009: 120-121).   Notes. More work is needed to determine whether the species listed below form a clade and merit reinstatement as a distinct genus, or alternatively if the name Ticanto should be synonymised under another genus in the Caesalpinia group. Th e list of species presented below includes names that most probably belong in Ticanto, but revisionary and phylogenetic work are needed to accurately delimit species, and determine types and synonyms.

Authors' contributions
EG, AB, CEH and GPL were involved in study conception and design; EG, AB, CEH, GPL and LPdQ collected and provided herbarium and fi eld samples for analysis; EG generated and assembled all the data, which she was also responsible for analysing and interpreting; EG drafted the manuscript, and critical revisions were provided by AB, CEH, GPL and LPdQ; EG also wrote the key, generic descriptions and provided the list of species belonging to each genus. Th ese were all critically revised by GPL, who completed this work by verifying the nomenclature and identifying types for all species names and synonyms. GPL was also the main instigator behind the new generic names (Paubrasilia, Hultholia, Hererolandia and Gelrebia).
who helped us with the typifi cation of Paubrasilia echinata. We also thank Heather Lindon, for verifying that all our nomenclatural combinations respected the latin and greek rules of grammar. We thank the following herbaria for loans of material and permission to sample leafl ets for DNA: Accessions included in this study. Species of the Caesalpinia group are classifi ed sensu Lewis (2005), and the number of species sampled over the total number of species recognized in the genus is given in parentheses. Type species for genera in the Caesalpinia group are preceded by an asterisk (*). Collector names, numbers (and herbarium acronym) of voucher specimens are listed for all material that was taken from herbarium specimens and for the voucher specimens of seed collections and silica-dried leaf samples, if known. Collection locality indicates the country where the specimen originated, and indicates which accessions were from cultivated specimens; N/A indicates that locality data was not available. Accession numbers are provided for published sequences downloaded directly from Genbank; with the exception of 22 sequences for the species Caesalpinia crista, C. decapetala, C. sappan, Cenostigma gardnerianum, Coulteria platyloba, Guilandina bonduc, Libidibia coriaria, P. exostemma, P. bracteosa, P. pyramidalis and Pterolobium stellatum, the majority of the sequences come from the following published studies: Bruneau et al. (2001), , Haston et al. (2005), Simpson et al. (2005), Marazzi et al. (2006), , Sanderson (2010), Babineau et al. (2013), Gagnon et al. (2013), and Gagnon et al. (2015). Furthermore, certain accessions were combined together in phylogenetic analyses: these accessions are in bold. If there were redundant sequences between these combined accessions, the longest sequence for each available marker were selected (Genbank numbers in bold).
Genus (no. of species sampled/total no. species) Species

Voucher specimen (Herbarium)
Collection locality : Accessions with the same prefi x were combined together.