The genus Albizia Durazz. has a complicated taxonomic history but has generally been treated as a pantropical genus with 120–140 species, of which 36 are endemic to Africa, with c. 30 species in Madagascar, of which c. 24 are endemic, c. 35 species in Asia, one in Australia, and 22 in tropical America (Lewis and Rico Arce 2005; Rico Arce et al. 2008). All species are woody, forming trees of variable stature and inhabit a wide range of lowland tropical biomes (Figs 1 and 2), including rain forests, seasonally dry tropical forests, and savannas, with one species, Albizia julibrissin Durazz., the type species of the genus, in subtropical and warm temperate forests in Asia. However, Albizia remains poorly defined; its delimitation remains one of the most challenging taxonomic problems in the legume family, and it is currently considered the main “dustbin” genus in tribe Ingeae (Koenen et al. 2020). In the past, the most problematic genus of tribe Ingeae was Pithecellobium Mart., but its taxonomy has been gradually clarified (Barneby and Grimes 1996). Resolution of the taxonomic status of Albizia has lagged behind that of Pithecellobium and only really started at the end of the twentieth century. For example, several new neotropical genera have been segregated from Albizia: Balizia Barneby & J.W. Grimes; Hesperalbizia Barneby & J.W. Grimes, and Hydrochorea Barneby & J.W. Grimes. Barneby and Grimes (1996) also re-established the genus Pseudosamanea Harms, which previously had been treated as a synonym within Albizia (Table 1). However, at the time they were established, the monophyly of these new and re-established genera had not been tested using phylogenetic analyses of molecular data.

Barneby and Grimes (1996) placed the remaining New World species of Albizia in their section Arthrosamanea (Britton & Rose) Barneby & J.W. Grimes. They characterized this section as forming a group that is homogeneous in most respects, but diverse in the late developmental stages of the fruit, which vary in: 1) fruit opening type: dehiscent, indehiscent, or breaking, 2) lateral shape: flat to conspicuously raised above the seed chambers, 3) texture and consistency of the valves: papery, chartaceous or woody, 4) longitudinal shape: straight to weakly falcate (Barneby and Grimes 1996) (Figs 1, 2 and 4). Within section Arthrosamanea, four series were proposed by Barneby and Grimes (1996): series Paniculatae with papery, plano-compressed, inertly dehiscent pods with continuous valves (13 species); series Arthrosamanea comprising 3 species with lomentiform, plano-compressed pods, where the ripe valves crack transversely between seeds but the wiry sutures persist at maturity; series Multiflorae (2 species) characterized by lomentiform fruits only reluctantly separating into articles, the thick-textured valves and sutural keels breaking transversely under pressure; and the monospecific series Inundatae which bears crypto-lomentiform pods, dehiscent through the sutures and with the valves differentiating into a continuous exocarp and a segmented endocarp separating into 1-seeded segments (Barneby and Grimes 1996).

Series Paniculatae is widespread across Mexico, Central and South America, occurring mainly in seasonally dry forests, grasslands, and less often in humid forests (in South America). All species of series Paniculatae have papyraceous, dehiscent fruits with one exception, A. berteroana (DC.) Fawc. & Rendle (the earlier combination A. berteroana (DC.) M. Gómez was invalidly published due to incorrect citation of the basionym, see Barneby and Grimes 1996), whose fruits are indehiscent and fall to the ground entire. In contrast, the other three series are distributed from Panama to South America and are most diverse in the Amazon basin (Barneby and Grimes 1996), and usually have more or less woody fruits, which are articulated and indehiscent, some dividing into monospermous segments through the grooves of the valves, considered to be an adaptation for hydrochory, i.e., seed dispersal in riparian and seasonally inundated forests (e.g., A. inundata (Mart.) Barneby & J.W. Grimes, A. pistaciifolia (Willd.) Barneby & J.W. Grimes, and A. subdimidiata (Splitg.) Barneby & J.W. Grimes).

The segregate genera established by Barneby and Grimes (1996) have not been universally accepted. For example, in the most recent taxonomic treatment of Albizia for Mexico and Central America (Rico Arce et al. 2008), the genera Balizia, Hesperalbizia, and Pseudosamanea were not recognized (Table 1). However, subsequent phylogenetic analyses have confirmed that these genera were rightfully segregated as distinct evolutionary lineages, Hesperalbizia being more closely related to Lysiloma Benth. (Duno de Stefano et al. 2021), and the closely related Balizia and Hydrochorea (the former reduced to synonymy of the latter, Soares et al. 2022) being placed as the sister-group of Jupunba Britton & Rose (Iganci et al. 2015; Soares et al. 2021, 2022). Phylogenomic analysis of the mimosoid clade, based on DNA sequences of 964 targeted nuclear genes confirmed these findings and, furthermore, showed that the species from the Americas (i.e., sect. Arthrosamanea) form a separate lineage from the African, Madagascan and Asian species (Koenen et al. 2020). Koenen et al. (2020) also showed that Pseudosamanea, although difficult to place in any clade, is not closely related to either Albizia s.s. or section Arthrosamanea and is perhaps most closely related to Samanea Merr. and Chloroleucon Britton & Rose ex Record.

While the data of Koenen et al. (2020) provided a robust phylogenomic backbone for the mimosoid and ingoid clades, and clearly demonstrated the non-monophyly of Albizia by sampling 25 species of that genus, only three of the 19 species from Central and South American sect. Arthrosamanea were included in that study, leaving doubts about the monophyly of that section and whether it should be segregated under the circumscription of Barneby and Grimes (1996), or whether there are further potential segregates, given the possibility that some of these species are more closely related to other neotropical genera. That possibility was suggested by the occurrence of lomentiform fruits in some species of section Arthrosamanea as reflected in the classification into separate series by Barneby and Grimes (1996). Similar lomentiform fruits also occur in Hydrochorea (Barneby and Grimes 1996; Soares et al. 2022) and Albizia s.s. (Albizia dolichadena (Kosterm.) I.C. Nielsen, Albizia moniliformis (DC.) F. Muell., Albizia rosulata (Kosterm.) I.C. Nielsen and Albizia umbellata (Vahl) E.J.M. Koenen), as well as a few other ingoid lineages (Barneby and Grimes 1996: 204; Koenen 2022a). For this reason, several species of neotropical Albizia have homotypic synonyms in Arthrosamanea Britton & Rose, Samanea or Cathormion Hassk., genera which previously had been recognized and defined based mainly on characters of fruit texture and dehiscence (Barneby and Grimes 1996). Barneby and Grimes (1996: 204) considered these fruit types to have arisen multiple times in parallel in different genera, and this was confirmed by subsequent phylogenetic (Iganci et al. 2015; Soares et al. 2021) and phylogenomic studies (Koenen et al. 2020), although the neotropical lomentiform Albizia species were not included in these studies, or remained unresolved.

Here we investigate whether Albizia sect. Arthrosamanea is monophyletic and thereby provide a more rigorous basis for recognizing its evolutionary distinctiveness from Albizia s.s. as a segregate genus. We infer a new phylogeny with emphasis on the neotropical species and make use of further insights offered by the phylogenomic analysis of Ringelberg et al. (2022). In addition, we use a tree topology inferred from data from the latter study to evaluate whether lomentiform fruits in Albizia sect. Arthrosamanea are independently derived from other lineages in which this fruit type occurs. Based on our phylogenetic results and the recent findings of Koenen et al. (2020) and Ringelberg et al. (2022), we update the taxonomy of neotropical Albizia by resurrecting the genus Pseudalbizzia Britton & Rose.

We used the nuclear ribosomal External and Internal Transcribed Spacer (ETS and ITS) regions that previously been used to study sister-group relationships within tribe Ingeae (Brown et al. 2008; Iganci et al. 2015; Souza et al. 2016). Our combined dataset included 123 accessions, of which 50 are from Genbank and 73 are newly sequenced here, including 25 species of Albizia s.l. sequenced for the first time. The outgroup, Vachellia farnesiana (L.) Wight & Arn., was designated to root the tree (Table 2). The plastid trnK region was initially explored but preliminary analyses suggested it is not sufficiently phylogenetically informative and these data were excluded from this study.

Fresh leaf material collected in the field plus herbarium material from the Jardín Botánico Regional Roger Orellana (CICY) were used for DNA extraction. Herbarium specimens used in these analyses came from AAU, CICY, FCME, MA, MEXU, and MO (acronyms as in Thiers 2016). Additional sequences were downloaded from GenBank (Table 2).

DNA from leaf fragments was obtained using the DNeasy Plant Mini Kit (QIAGEN Inc., Valencia, California) following the manufacturer’s specifications. To assess concentration and relative quality of DNA, 3 µl of final volume plus 2 µl loading buffer were run for 30 minutes at 6 V cm^-1 on a 1% agarose gel prepared with 0.5× TBE. The resulting gel was developed by immersion for 20–30 minutes in a 0.1 µg ml^-1 ethidium bromide solution and later observed in a DigiDoc-It Imaging System (version 6.7.1; UVP, Inc., Cambridge, UK) transilluminator. DNA purity and concentration were quantified with a NanoDrop 2000c. Afterwards, DNA samples were standardized to 10 ng µl^-1.

PCR amplifications were performed in an Applied Biosytems Veriti 96 Well Thermal Cycler. Volumes of reagents and conditions for the amplifications were as follows: ITS: 30 µL of mix containing 3 µl 10× Buffer, 2.5 µl MgCl_2,, 0.6 µl (~10 ng) primer, 4 µl Q solution, 1 µl 1.25 mM L^-1 dNTP, 0.2 µl (1 U) TAQ polymerase, 2 µl (~10 ng) DNA, then completed to volume (approx. 16.1 µl) with ultra-pure water. PCRs were conducted under the following protocol: 94 °C × 3 min + 30 cycles (94 °C × 1 min + 60.5 °C × 1 min + 72 °C × 2 min) + 72 °C × 7 min. Primers were S3 (AACCTGCGGAAGGATCATTG) (Käss and Wink 1997), and 26S (TAGAATTCCCCGGTTCGCTCGCCGTTAC) (Sun et al. 1994). ETS: 30 µl of mix containing 3 µl 10× Buffer, 2.5 µl MgCl₂, 0.6 µl (~10 ng) primer, 4 µl Q solution, 1 µl 1.25 mM l^-1 dNTP, 0.2 µl (1 U) TAQ polymerase, 2 µl (~10 ng) DNA, then completed to volume (approx. 16.1 µl) with ultra-pure water. PCR amplifications were conducted under the following protocol: 94 °C × 3 min + 30 cycles (94 °C × 1 min + 60.5 °C × 1 min + 72 °C × 2 min) + 72 °C × 7 min. Primers used were 18S-IGS (5’-GAGACAAGCATATGACTACTGGCAGGATCAACCAG-3’) and 26S-IGS (5’-GGATTGTTCACCCACCAATAGGGAACGTGAGCTG-3’) (Baldwin and Markos 1998).

The quality of the PCR products was evaluated by agarose electrophoresis (3 μl of final volume plus 2 μl of bromophenol blue, gel prepared with 0.5× TBE and 1% agarose, run at 120 volts and 25 amperes 30 min). PCR products were sequenced at Macrogen (http://www.macrogen.com/eng/) using the same amplification primers. The sequencing products were assembled and edited using the Sequencher v. 5.2.3. An initial automated alignment was conducted with MAFFT (Katoh et al. 2002) using the E-INS-i algorithm option, a 100PAM/k = 2 scoring matrix, a gap opening penalty of 1.3, and an offset value of 0.123. The alignments were visually inspected and manually edited for further improvement. The Akaike Information Criterion (AIC), implemented in jModeltest (Posada 2008) was used to select the best model of nucleotide substitution for each alignment. The selected models were TVM+I+G for ETS and GTR+I+G for ITS. Phylogenetic analyses were performed with MrBayes v.3.2.5 (Ronquist et al. 2012) separately for each dataset, and subsequently concatenated with each partition treated as independent and associated with its own evolutionary model. Analyses were performed using default parameters for 5 million generations. Two independent threads were run. Convergence was assessed with both MrBayes and Tracer (Rambaut et al. 2018). Posterior Probabilities (PP) ≤ 0.95 were considered weakly supported whereas PP of 0.95–1.0 were deemed strongly supported (Alfaro et al. 2003).

To examine the evolution of fruit types within New World Albizia we utilize a phylogeny derived from a new analysis based on data of Ringelberg et al. (2022) for the Jupunba clade, as described in Soares et al. (2022), because it is based on a large set of 560 nuclear exons and flanking non-coding regions, and therefore shows enhanced resolution within this clade compared to the ITS + ETS phylogeny.

Alignments of our combined datasets recovered by MAFFT required few manual adjustments. The ETS sequences had 381 bp and, once aligned, 52% of the data were informative. In the case of ITS, the sequences were slightly longer, 551 bp but only 34% were informative.

None of the molecular-based analyses (ETS, ITS, and ETS+ITS) using Bayesian inference recovered the genus Albizia as monophyletic. The combined ETS + ITS phylogeny (Fig. 3) is used as the basis for discussing the results in detail. The outgroup Vachellia farnesiana, plus Acaciella angustissima (Mill.) Britton & Rose, Senegalia berlandieri (Benth.) Britton & Rose and two species of Mariosousa Seigler & Ebinger form a paraphyletic grade subtending a fully supported clade (PP = 1) that includes all members of the tribe Ingeae as well as Acacia s.s.

The most relevant clade from the perspective of this study is highly supported (PP = 0.96) and includes all members of the genus Albizia and a few other genera of tribe Ingeae (clade A). The genus Albizia, as currently circumscribed, is non-monophyletic with species placed in two separate, strongly supported clades (Fig. 3). As in Koenen et al. (2020), species of Albizia sect. Arthrosamanea are placed in clade B (PP = 0.96), which equates to the Jupunba clade of Koenen et al. (2020). All species that were sampled from this section are included in this clade (Fig. 3 clade D), which received full support (PP = 1), and as in Koenen et al. (2020) this section is sister to clade C (PP = 1) comprising the genera Jupunba, Balizia and Hydrochorea.

Within sect. Arthrosamanea, three clades are well supported, one comprising Albizia polycephala (Benth.) Killip ex Record and Albizia edwallii (Hoehne) Barneby & J.W. Grimes of ser. Paniculatae, a second clade comprising species of ser. Paniculatae endemic to Mexico, Central America, and the Caribbean, and a third clade that includes Albizia niopoides (Benth.) Burkart (also ser. Paniculatae) and the species from the other three series. The phylogeny of Ringelberg et al. (2022) presented here in Fig. 4 based on a new analysis of the Jupunba clade accessions from Soares et al. (2022), has greater resolution within sect. Arthrosamanea and shows that ser. Paniculatae forms a well-resolved paraphyletic grade in which the other three series are nested. The other two non-monospecific series also appear to be non-monophyletic, with the monospecific ser. Inundatae nested inside ser. Arthrosamanea and these together in turn nested in ser. Multiflorae, although support for the paraphyly of ser. Multiflorae is only 0.72 pp.

The Old World species of Albizia form a monophyletic group (PP = 1) placed in clade E (PP = 0.96) (Fig. 3), with the genera Enterolobium Mart., Samanea, and Pseudosamanea. Within clade E Pseudosamanea (PP = 0.96) is sister to clade F which includes Enterolobium, Samanea and Old World Albizia. All these clades have high support (PP = 1). These analyses also support the transfer of Cathormion umbellatum Kosterm., which is placed in the Old World Albizia clade (Fig. 3), to Albizia, as proposed by Koenen et al. (2020).

Our analyses also confirm that the monotypic genus Hesperalbizia: H. occidentalis (Brandegee) Barneby & J.W. Grimes is sister to Lysiloma, in the Cojoba clade (sensu Koenen et al. 2020), unrelated to either New World Albizia (clade D) or Old World Albizia (clade F), as previously shown by Duno de Stefano et al. (2021). Furthermore, Pseudosamanea guachapele (Kunth) Harms (previously Albizia guachapele (Kunth) Dugand in Rico Arce et al. (2008)) is also unrelated to New World Albizia but is instead a member of clade E, sister to clade F which includes Enterolobium, Samanea and Old World Albizia.

In both Albizia sect. Arthrosamanea and the closely related Balizia and Hydrochorea, these phylogenies suggest that lomentiform fruits were independently derived from indehiscent fruits that are septate between the seeds, as species with the latter fruit type form paraphyletic grades to the lomentiform species in both cases (Fig. 4). In turn, these indehiscent septate fruits are nested within paraphyletic assemblages of species with fruits that dehisce along one or both sutures in both groups. Interestingly, in both cases, a single species with crypto-lomentiform fruits is found, but it is not clear whether these were derived from the same ancestral fruit type or not. In Hydrochorea this crypto-lomentiform species appears as an intermediate between indehiscent and lomentiform species, while in Albizia sect. Arthrosamanea the crypto-lomentiform-fruited A. inundata appears to be derived from a lomentiform-fruited ancestor. Another difference between these two groups is that the follicular dehiscence of Balizia pedicellaris (DC.) Barneby & J.W. Grimes fruits appears to be secondarily derived from indehiscent fruits, but we note that similar dehiscence is also found in a few species of Jupunba.

This study addresses the non-monophyly of the genus Albizia and our results provide important insights into the evolutionary history of the neotropical species placed in sect. Arthrosamanea, with implications for their taxonomic classification. We show that sect. Arthrosamanea, with expanded taxon sampling relative to Koenen et al. (2020), and as also shown by the study of Ringelberg et al. (2022), is monophyletic with only two exceptions: Albizia leonardii Barneby & J.W. Grimes which is placed among taxa of the ‘senegalioid grade’ (Ringelberg et al. 2022; Terra et al. 2022) and Albizia carbonaria Britton that is more closely related to Pseudosamanea (Koenen 2022b; Ringelberg et al. 2022).

The geographically-based splitting of a large genus in tribe Ingeae, such as Albizia, which occupies a pantropical distribution, is not unprecedented, nor unexpected, especially given the lack of pantropical monographic synthesis or geographically widely sampled phylogenies for the mimosoid clade. For example, the genus Pithecellobium, once the largest genus of tribe Ingeae, has been progressively divided during the last 50 years into multiple genera (see Brown et al. 2008 for a general history of the tribe). Another example is the genus Calliandra Benth. for which the New World species of Calliandra. ser. Laetevirentes were segregated into Zapoteca H.M. Hern. (Hernández 1986), and almost all Old World species allocated progressively to other segregated genera: Viguieranthus Villiers (Villiers 2002), Thailentadopsis Kosterm. (Lewis and Schrire 2003), Sanjappa E.R. Souza & M.V. Krishnaraj (Souza et al. 2016) and Afrocalliandra E.R. Souza & L.P. Queiroz (Souza et al. 2013). All these taxonomic rearrangements were supported by morpho-anatomical and molecular phylogenetic analyses. Molecular data have also demonstrated that Abarema is polyphyletic (Iganci et al. 2015), prompting reinstatement of the genera Jupunba and Punjuba Britton & Rose (Soares et al. 2021). Finally, neither Zygia P. Browne nor Marmaroxylon Killip are monophyletic, although a new generic classification for those genera has yet to be proposed (Ferm et al. 2019). The non-monophyly of Albizia documented here and elsewhere (Koenen et al. 2020; Ringelberg et al. 2022) is thus not a surprise and reflects the state of flux surrounding generic delimitation in mimosoids, especially within the ingoid clade.

Some of the taxonomic proposals of Barneby and Grimes (1996) relative to the American segregates of Albizia s.l. are confirmed by our results. The genera Balizia and Hydrochorea form part of the Jupunba clade (sensu Koenen et al. 2020) (Figs 3 and 4) (Iganci et al. 2015), although neither Hydrochorea nor Balizia are monophyletic in our phylogeny (Figs 3 and 4, see Soares et al. 2022). Two neotropical species, included in Albizia by Rico Arce et al. (2008) are also placed outside New World Albizia (Fig. 3): Hesperalbizia occidentalis is closely related to Lysiloma, in agreement with previous results (Iganci et al. 2015; Duno de Stefano et al. 2021); similarly, Pseudosamanea guachapele, is also placed outside Albizia in our phylogeny, emerging, as expected, together with the other species Pseudosamanea cubana (Britton & Rose) Barneby & J.W. Grimes, although relationships within this clade are unresolved (Fig. 3).

Here we show that the dehiscent, papery, plano-compressed fruit type is ancestral within Albizia sect. Arthrosamanea (Fig. 4) and is associated with species growing predominantly in seasonally dry tropical forest and woodland, with successive shifts to septate indehiscent fruits and then lomentiform fruits with hydrochorous seed dispersal associated with species growing in seasonally inundated varzea forest, riverine habitats and low-lying margins of palm and white-sand savannas (Fig. 4). Interestingly, in the sister group of Albizia sect. Arthrosamanea, the mainly neotropical clade composed of Jupunba, Punjuba, Balizia, and Hydrochorea, a similar parallel evolutionary transition in fruit types is apparent. In Jupunba and Punjuba, fruits are always dehiscent, while a transition to septate indehiscent fruits occurred in Balizia, an exception being Balizia pedicellaris which has follicular dehiscence and a newly described species with crypto-lomentiform fruits (Fig. 4, and Soares et al. 2022). Nested within the paraphyletic Balizia is a clade comprising the genus Hydrochorea plus two African species of Cathormion, all species of which have indehiscent lomentiform fruits adapted for hydrochory and are found in riparian or other periodically flooded habitats in the Amazon basin, West Africa and the Congo basin (Fig. 4, and Soares et al. 2022).

Barneby and Grimes (1996) pointed out that a radiation of species with similarly heterogeneous fruit types to that seen in section Arthrosamanea occurs in Madagascan Albizia s.s. and that the association between lomentiform fruits, hydrochorous seed dispersal, and seasonally flooded habitats is also apparent in Old World Albizia s.s. For example, Albizia dolichadena, A. moniliformis, A. rosulata, and A. umbellata from Australasia also have lomentiform fruits and are distributed near streams or in riparian and swamp forests (Rico Arce et al. 2008). Furthermore, as indicated above and pointed out by Barneby and Grimes (1996), similar transitions to lomentiform fruits have occurred in parallel in several other lineages across the ingoid clade, including Cathormion altissimum (Hook.f.) Hutch. & Dandy (sometimes referred to as Albizia altissima Hook.f.; Koenen 2022a) and Senegalia rostrata (Humb. & Bonpl. ex Willd.) Seigler & Ebinger (syn. Dugandia rostrata (Humb. & Bonpl. ex Willd.) Britton & Killip, syn. Manganaroa articulata Speg.; Barneby and Grimes 1996: 204) in all cases apparently also closely associated with riparian and/or periodically inundated habitats. These repeated parallel derivations of similar, but not strictly homologous fruit types attest to the high evolvability of the mimosoid fruit more generally. In the light of phylogenetic evidence, it is now clear that these evolutionarily highly labile morphological adaptations of the fruit related to seed dispersal syndrome do not provide reliable characters for generic delimitation, supporting inclusion of the species that were placed in ser. Arthrosamanea, ser. Inundatae and ser. Multiflorae within Albizia sect. Arthrosamanea by Barneby and Grimes (1996), i.e., the clade of New World Albizia that is recovered in our analysis.

GAP, RR, GCFC, IVM, and RDD designed the study. GAP, LLCI, ELC, RDD contributed labwork. RDD, GCFC, IRM contributed data by supervising students in the lab. EJMK, RDD, XC, SM and CEH contributed taxonomic knowledge, JR contributed species distribution data and the maps. GAP, RDD, ITC, JRP, and RR undertook the phylogenetic analyses. EJMK, RDD, RR, GCFC, CEH and JR contributed to writing the manuscript.