A total of 87 accessions were sampled, representing 43 species of Archidendron (68 accessions), five species of Archidendropsis (six accessions) and nine species (11 accessions) of the other genera in the Archidendron clade; two species of Old World Albizia Durazz. were included as outgroups (Table 1). In total 43% of the species of Archidendron were sampled including representatives of all eight series. Both subgenera of Archidendropsis were sampled covering 36% of species in the genus. Samples were collected in the field and from herbarium specimens sourced from AAU, BISH, BRI, CANB, CNS, KEP, KUN, L, NY, MEL and MELU (herbarium codes as per Thiers, updated continuously).

Total genomic DNA (gDNA) was extracted following the CTAB method of Doyle and Doyle (1987) with modifications as per Shepherd and McLay (2011). Isolated gDNA was quantified with a NanoDrop 2000 (ThermoScientific) spectrophotometer and cleaned with a 2.4 M sodium acetate wash. Recalcitrant herbarium material that failed using the CTAB method was extracted using the AccuPrep Stool genomic DNA extraction kit (Bioneer) using the manufacturer’s protocol with some modifications suggested by Schuster (pers. comm.). Only 30 mg of leaf material was used instead of the recommended 100–200 mg. A total of 600 µl of stool lysis buffer (SL) was added to the extraction tube instead of 400 µl, the incubation step was increased to one hour in total, centrifugation was done for 10 minutes at step five, and to maintain equal volumes, 600 µl of binding buffer was added. Two consecutive washes were performed using buffer 1 (W1). The final elution was done by adding 160 µl total elution buffer in two steps (first 60 µl, and then 100 µl) instead of a single elution with 200 µl.

Eight nuclear markers (low copy genes: AIGP, CYB6, Eif3E, SHMT, RBPCO, UDPG; nrDNA: ITS, ETS) and four chloroplast DNA intergenic spacer regions (trnK-matK, trnV-ndhC, psbD-trnT, trnL-rpl32) were assessed for variability between nine individuals spanning the series of Archidendron using Sanger sequencing.

PCR reagents, primers and cycling conditions are described in Suppl. material 1 (Johnson and Soltis 1994; Sun et al. 1994; Käss and Wink 1997; Baldwin and Markos 1998; Miller and Bayer 2001; Ariati et al. 2006; Choi et al. 2006; Shaw et al. 2007; Li et al. 2008). PCR products were visualised on a 1.5% agarose gel with Easy ladder I (Bioline) and cleaned with ExoSAP-IT (USB) as per the manufacturer’s protocol. The purified amplicons were sequenced on an AB3730xl sequencer (Thermo Scientific) at the Australian Genome Research Facility, Melbourne. Sequences were aligned in Geneious v.8.1.4 (Biomatters Ltd.) and assessed for variability between the samples. The most variable loci were then used in a targeted amplicon sequencing (TAS) approach (McLay et al. 2021), sequencing pooled amplicons on an Illumina MiSeq. For this, additional internal primers were designed for the five loci that had a total amplicon length greater than 500 bp, in order to produce shorter amplicons that could be fully sequenced using a 500-cycle sequencing kit. These primers were designed using Primer 3 v.2.3.4 (Rozen and Skaletsky 2000) implemented in Geneious v.8.1.4 (Biomatters Ltd.), selecting priming sites in conserved regions across the nine sequenced individuals.

Library preparation followed the two-step PCR process outlined in McLay et al. (2021). The first step used the region-specific primers to amplify each locus individually for each sample. Initial PCR reactions included 1 × MyTaq Buffer (Bioline), 1.2 µl of MgCl₂ 2.5 M (Bioline, 100 mg mL), 1.2 µl of dimethyl sulfoxide (DMSO, 99.5%; Sigma-Aldrich), 3 µl of each “tailed” primer (10 µM), 0.375 U of MyTaq (Bioline), 100 ng of gDNA, and ultra-pure water to make up for 16 µl volume. Variations in these reactions are noted in Suppl. material 1 for specific loci. Conditions for PCR were based on those of Choi et al. (2006), Shaw et al. (2007), and Ariati et al. (2006) with modifications as required to obtain successful amplifications (Suppl. material 1). To estimate amplicon concentration to decide the volume of PCR product for amplicon pooling, 2.2 µl of PCR product and 2.5 µl of molecular ladder (Easyladder I, Bioline) were run on 1.5% agarose. A total of 120 ng of each nuclear DNA (ncDNA) region PCR product and 20 ng of each chloroplast region PCR product were pooled in the same well of a 96-well plate. The ncDNA were pooled in a higher concentration to account for the possible presence of different alleles. Pooled samples were cleaned with 1.5 × Serapure beads (Rohland and Reich 2012).

The second step used qPCR to add unique Illumina indexing barcodes to each sample for the pooled amplicons. Indexing PCR reactions consisted of 5 µM of each of index primer (McLay et al. 2021), 3 µl of pooled amplicons, 1 × Kapa HiFi ReadyMix (Biosystems) and ultra-pure water to make up a total of 25 µl reaction. Conditions for PCR were 95 °C for 1 min, followed by 13 cycles of 98 °C for 50 sec, 67 °C for 50 sec, and 72 °C for 20 sec, and a final extension at 72 °C for 30 sec. Each sample was then cleaned with 1.4 × Serapure beads and concentrations were quantified using fluorescence in a EnSpire multimode plate reader. In total, 10 ng of each indexed and cleaned sample was pooled together. The final pooled library was cleaned with 1.5 × Serapure bead-to-sample ratio and the library was submitted to the Australian Genome Research Facility, Melbourne for sequencing on an Illumina MiSeq using a 500 cycle MiSeq v2 Nano Kit.

Sequences obtained by Sanger sequencing were aligned by individual locus in Geneious v.8.1.4 (Biomatters Ltd.) and a consensus sequence was generated and used as the reference for the reads obtained by TAS. The demultiplexed TAS Illumina MiSeq files were imported into Geneious v.8.1.4. Reads were trimmed to remove adapters and low-quality sequence. The map-to-reference option was selected to map reads for each sample to the different reference loci using High Sensitivity/Medium settings and a minimum mapping quality of 20. A consensus sequence for each locus was generated for each individual with Generate Consensus Sequence (Threshold = 65%, with Ns called if coverage was less than 10). The forward and reverse reads of the low-copy nuclear genes (LCNG) overlapped so it was possible to phase these loci into separate alleles, but this was not possible for the nuclear ribosomal DNA loci (ETS and ITS) as the reads were not overlapping due to unexpected length variation in both of these loci. Alignments of individual consensus sequences for each locus were generated using MUSCLE (Edgar 2004) in Geneious v.8.1.4 and adjusted manually. For each LCNG, samples with multiple alleles were assessed for topological concordance between the different copies using neighbour-joining trees (using the Geneious tree-builder, HKY model) and NeighbourNet networks (SplitsTree4, default settings, Huson and Bryant 2006), to ensure that a conflicting signal was not introduced from distantly related allelic variants (see Suppl. material 2: SHMT network and tree and Suppl. material 3: RBPCO network and tree). Allelic variants within samples were largely concordant with one-another permitting consensus sequences for those samples to be used for subsequent phylogenetic analyses.

Alignments of all nuclear loci (ncDNA; with consensus sequences for LCNG alignments) were analysed individually to explore gene tree topologies in IQ-TREE v.1.6.12 on the web server (http://iqtree.cibiv.univie.ac.at/, Trifinopoulos et al. 2016) with support estimated with 1,000 ultra-fast bootstrap replicates (UFBS) (Minh et al. 2013). After comparing topologies, four ncDNA loci (ETS, ITS, RBPCO, SHMT) were concatenated into a single matrix as no major incongruencies were observed. The combined ncDNA dataset was partitioned into six partitions corresponding to each locus with the ITS region further divided in ITS1, 5.8S and ITS2 for subsequent analyses. IQ-TREE was used to perform maximum likelihood (ML) analyses on the concatenated ncDNA alignment. The analysis was run with the alignment partitioned and allowing ModelFinder (Kalyaanamoorthy et al. 2017) to identify the optimal substitution models for each partition (Table 2). Node support was estimated using 1,000 UFBS. Bayesian Inference (BI) was performed, with the alignment partitioned by locus. The best model of substitution for each partition was estimated with IQ-TREE model selection using the options: selection criteria of Bayesian (BIC), candidate models JC, F81, K80, HKY, SYM, GTR, heterogeneity types I, G, I+G, and the genomic source of nuclear (Table 2). MrBayes v.3.2.7a (Ronquist et al. 2012) was run using the CIPRES Science Gateway (Miller et al. 2010). Two parallel runs each with eight Monte Carlo Markov Chains were run for five million generations, sampling a tree every 1,000 generations and a burn-in of 25%.

A consensus network of the combined ncDNA dataset was constructed in SplitsTree4 (Huson and Bryant 2006) using the last 101 sampled BI trees (edge weights = mean, threshold = 0.05). This method allows for the visualisation of conflict in a set of trees and provides an alternative method of interpretation to a single fixed topology of a consensus tree.

All chloroplast (cpDNA) loci were concatenated into a single matrix for phylogenetic analyses. IQ-TREE was used to perform ML analyses on the cpDNA matrix, with the alignment partitioned by locus, using ModelFinder to identify the optimal substitution model for each locus, and support was estimated using 1,000 UFBS replicates. The resulting topology was very poorly supported (though similar groups to the ncDNA phylogeny were discovered within the genus Archidendron). To further investigate cpDNA relationships within Archidendron, the outgroups were removed, and the IQ-TREE analysis was performed on the reduced dataset. The UFBS replicates were then used to create a consensus network in SplitsTree4 (edge-weights = mean, threshold = 0.20).

Pollen size and surface texture are key morphological features differentiating the subgenera of Archidendropsis but one of the three species of subg. Basaltica (A. xanthoxylon (C.T. White & W.D. Francis) I.C. Nielsen) was not examined by Nielsen et al. (1983b). To fill this gap and ensure consistency of results with published data, pollen from A. xanthoxylon (BRI AQ0199126, BRI AQ0874091, BRI AQ0199129 and BRI AQ0648303) and A. basaltica (F. Muell.) I.C. Nielsen (BRI AQ1003764, BRI AQ0199029, BRI AQ0625292 and BRI AQ0648454) of subg. Basaltica was examined. Pollen grains were obtained from flowers of herbarium specimens under a Zeiss dissecting microscope at the Queensland Herbarium (BRI) using clean forceps and a fine brush. Samples were mounted on aluminium stubs using double-sided carbon tabs and coated with gold using an Agar Scientific Automatic Sputter Coater. Pollen grains were observed and photographed using a Phenom G2 5keV (kiloelectron-volt) desktop scanning electron microscope (PhenomWorld). Pollen diameter for 10 grains of A. basaltica and eight grains of A. xanthoxylon was measured using ToupView (TOUPTEK PHOTONICS) software; overall fewer grains were available on specimens of A. xanthoxylon for microscopy.

Of the eight nuclear loci only four were included in the final phylogenetic analyses: SHMT, RBPCO, ITS and ETS. ETS and ITS amplified well, were variable, and are commonly used phylogenetic markers in Caesalpinioideae phylogenetic studies. Of the LCNGs, SHMT was the most informative, followed by RBPCO; allelic variation was found in some individuals for all LCNGs. Exploring allelic variation in the SHMT (36 samples with alleles) and RBPCO (24 samples with alleles) showed that for samples with more than one allele, the copies were closely related to each other (Suppl. material 2: SHMT network and tree and Suppl. material 3: RBPCO network and tree). Two LCNGs were excluded because few individuals of the target genera were successfully sequenced; only 12 sequences of Archidendron and two sequences of Archidendropsis were obtained for AlGP, and only 16 sequences of Archidendron and one Archidendropsis were obtained for Eif3E. The remaining two LCNG loci (CYB6 and UDPG) are not included in the analyses due to their short lengths, 240 bp and 202 bp respectively, and lack of variation.

Of the four chloroplast loci, trnK-matK was the most informative, followed by psbD-trnT and then trnV-ndhC. However, only one of the three blocks of trnV-ndhC was successfully sequenced. The internal primers designed allowed 100% coverage for the trnK-matK, 81% coverage for the psbD-trnT, and less than 30% coverage for the trnV-ndhC. It was not possible to obtain sequences for all samples for all blocks in which the three cpDNA regions were divided; as a result the cpDNA dataset was patchy. The trnL-rpl32 intergenic spacer did not amplify well, with 10 samples partially sequenced, and it was not included in final analyses.

The topologies of the combined ncDNA Bayesian and IQ-TREE analyses were congruent (nodes supported with UFBS ≥ 95; PP ≥ 0.90) and the Bayesian tree is presented (Fig. 2A,B). The Archidendron clade was recovered as monophyletic (PP 1.0) with six well supported clades (A–F) resolved within it. However, the relationships between clades A–F were not well resolved or supported with a polytomy in the backbone of the phylogeny. Clade A (PP 0.99) includes all three species of Archidendropsis subg. Basaltica, clade B (PP 1.0) includes the three samples of Pararchidendron pruinosum (Benth.) I.C. Nielsen, and clade C (PP 1.0) includes the two sampled representatives of Archidendropsis subg. Archidendropsis. Four monophyletic genera are grouped together in clade D (PP 1.0), with Acacia sister to Paraserianthes in clade D1 (PP 1.0) and Falcataria sister to Serianthes (PP 1.0) in clade D2 (Fig. 2A). Clade E (PP 1.0) comprises all but two sampled representatives of Archidendron ser. Clypeariae, and all other samples of Archidendron are placed in clade F (PP 1.0). Clades C, D and Wallaceodendron are related (PP 0.98) and together are sister to Clade E (PP 0.96; Fig. 2A).

Within Archidendron, only one of Nielsen’s eight series is resolved as monophyletic (ser. Ptenopae) within subclade F1 (Fig. 2A). Clade E, the Clypeariae clade had two main lineages and several smaller supported subclades within them. Clade F, the Archidendron s.s. clade is segregated into three well supported subclades: the lucyi subclade (F1, PP 1.0) that includes three fully supported lineages; the grandiflorum subclade (F2, PP 1.0) that is poorly resolved; and the vaillantii subclade (F3, PP 1.0) that comprises two well supported lineages (PP 0.99; Fig. 2A–C).

Of the 12 species of Archidendron that included more than one accession, seven are monophyletic (A. glabrum (K. Schum.) K. Schum. & Lauterb., A. kanisii R.S. Cowan, A. lucyi F. Muell., A. muellerianum, A. ramiflorum (F. Muell.) Kosterm., A. vaillantii (F. Muell.) F. Muell. and A. whitei), one is unresolved (A. lovelliae (F.M. Bailey) I.C. Nielsen), and four are not monophyletic (A. clypearia (Jack) I.C. Nielsen, A. grandiflorum (Sol. ex. Benth.) I.C. Nielsen, A. hendersonii (F. Muell.) I.C. Nielsen and A. hirsutum I.C. Nielsen). Three of the four samples of A. clypearia form a clade (within clade E, Fig. 2; PP 1.0) with A. borneense (Benth.) I.C. Nielsen nested among them. One sample of A. hendersonii (JA45) is related to A. grandiflorum within clade F2; all other samples of A. hendersonii (Z114, JA103, JA44) form a clade within F3 (PP 1.0; Fig. 2A). Another species falling in both subclades F2 and F3 is A. hirsutum, with one sample (JA46) related to A. forbesii Baker f. and A. lovelliae in subclade F2 (PP 0.99), and the other two (Z113 and JA86) forming a sister pair in subclade F3 (PP 1.0; Fig. 2A).

The consensus network of the final 101-sampled BI trees shows the degree of topological uncertainty between the genera in the Archidendron clade (Fig. 2C). While each respective genus is well-supported as monophyletic (except Archidendropsis and Archidendron as described above) the relationships between the genera are highly uncertain, reflecting the lack of support in the consensus phylogenies. However, the network reinforces the distinction between the two clades of Archidendropsis, and the distinction of the Clyperiae clade from the rest of Archidendron.

The phylogeny of the three cpDNA loci combined lacks support for nearly all nodes (Suppl. material 4: cpDNA tree). Of the supported nodes there are two that are incongruent with the ncDNA tree (Fig. 2): Paraserianthes is sister to Falcataria (UFBS 100), and A. harmsii Malm is supported in the grandiflorum subclade (UFBS 95) sister to A. grandiflorum JA100 (UFBS 97; Suppl. material 4: cpDNA tree). The consensus network of the UFBS replicates (with splits present in at least 20% of trees) reflects the patterns in the ncDNA phylogeny, with four distinct groupings within Archidendron (Fig. 3). Within these groupings, several individuals are placed in different clades to the ncDNA tree: A. hendersonii JA45 is placed in the vaillantii subclade rather than the grandiflorum subclade, and A. harmsii JA74 is in the grandiflorum subclade rather than the lucyi subclade (Fig. 3).

The pollen measurement results are consistent with Nielsen et al. (1983a, 1983b). The pollen of the two species examined (A. basaltica and A. xanthoxylon) are aggregated into symmetrical 16-celled polyads with a diameter of 55–62 μm for A. basaltica and 62–68 μm for A. xanthoxylon (Fig. 4). Fossules were present on the surface of all grains of both species, but they were fainter on the peripheral cells compared to the central ones and overall fainter on A. basaltica compared to A. xanthoxylon (Fig. 4).

Our study presents the most taxon-rich sampling of the Archidendron clade of any phylogenetic analyses to date. We confirm that the Archidendron clade sensu Koenen et al. (2020) of Indomalayan-Australasian genera (Acacia, Archidendron, Archidendropsis, Falcataria, Serianthes, Pararchidendron, Paraserianthes and Wallaceodendron) is robustly supported, yet the relationships between the constituent clades are poorly resolved and lack support. This result is not unexpected given we used only four ncDNA loci and that phylogenomic studies based on hundreds of loci also yield short branches with low support across the backbone of the Archidendron clade (Koenen et al. 2020; Demeulenaere et al. 2022; Ringelberg et al. 2022). It has been suggested that this lack of resolution may be the result of extremely rapid speciation and that the backbone of this clade could be best regarded as a polytomy within the Ingoid legumes (Koenen et al. 2020). The differences in published topologies of the Archidendron clade are illustrated in Demeulenaere et al. (2022) but it is clear that further work based on increased sampling of phylogenomic data is required to uncover the evolutionary history of the clade.

Despite the poorly resolved backbone of the Archidendron clade, many clades within it are robustly supported and corroborate published phylogenies, as well as shedding new light on the genera Archidendron and Archidendropsis (Fig. 2). Four genera of the Archidendron clade are confirmed to be monophyletic – Acacia (Miller and Bayer 2001; Luckow et al. 2003; Miller et al. 2003; Brown et al. 2008), Falcataria (Brown et al. 2011), Pararchidendron and Serianthes (Demeulenaere et al. 2022) – and the previously suggested non-monophyly of Archidendron and Archidendropsis (Brown et al. 2008, 2011; Iganci et al. 2016; Demeulenaere et al. 2022; Ringelberg et al. 2022) is confirmed and clarified by increased sampling within these genera.

The genus Archidendron is not monophyletic, and the eight series, while useful for identification purposes, do not coincide with evolutionary lineages (Fig. 2). The only series confirmed to be monophyletic was series Ptenopae from the island of New Guinea, the smallest series comprising just two species with two-winged leaf rachises and pinnae: A. ptenopum Verdc. and A. hispidum (Mohlenbr.) Verdc. (Nielsen et al. 1984b). The monophyly of series Calycinae and Pendulosae was not tested, as only one species of each was sampled, however, all other series (Archidendron, Bellae, Clypeariae, Morolobiae, and Stipulatae) are not monophyletic. Archidendron is instead resolved into two well supported lineages, one of which is primarily distributed in western Malesia and mainland Asia (the Clypeariae clade; clade E, Figs 1–3) and the other (the Archidendron s.s. clade; clade F, Figs 1–3) mostly restricted to eastern Malesia and Australia. These two lineages have been identified in previous phylogenetic studies but the sampling for each was extremely limited, with at most seven species of one lineage included (Brown et al. 2008, 2011; Iganci et al. 2016; Demeulenaere et al. 2022; Ringelberg et al. 2022). The further segregation of the Archidendron s.s. clade into three well supported lineages, the lucyi (F1), the grandiflorum (F2), and the vaillantii subclades (F3; Figs 2–3), is novel.

These three subclades of the Archidendron s.s. clade reflect geographic distributions to some extent, but no macromorphological characters have been identified to clearly delineate them. The grandiflorum and vaillantii subclades are predominantly Australian with some southern New Guinean species included, while the lucyi subclade is geographically more broadly distributed in the Lesser Sunda Islands, the Moluccas, through New Guinea to the Solomon Islands with only one species, A. lucyi, extending into northern Australia. Morphologically, the lucyi subclade includes all the sampled species lacking stipules that are not from ser. Clypeariae (i.e. A. calliandrum de Wit, A. harmsii, and A. glabrum), although stipules are reported for other species in this clade, three with stipular glands (A. lucyi, A. megaphyllum Merr. & L.M. Perry, Archidendron sp. nov. JA85), two with stipules only (A. ptenopum and A. hispidum) and A. parviflorum Pulle having both stipular glands and stipules (AAU Balgooy 6769; Nielsen et al. 1984b). All sampled species in the grandiflorum and vaillantii subclades have stipules, except A. arborescens (Kosterm.) I.C. Nielsen and A. forbesii, which have stipular glands (BM000946689; BRI AQ0380081; BRI AQ052589; Nielsen et al. 1984b) The placement of an undescribed species (Archidendron sp. nov. JA85) from the Aru Islands (Moluccas) in the lucyi subclade fits the geographic range. Ivan Nielsen noted this as a putative new species in October 1998 (AAU Balgooy 6769) but it does not align with any of the 20 imperfectly known species he outlined (Nielsen et al. 1984b), highlighting that further taxonomic work is required.

Three species in the Archidendron s.s. clade were not resolved as monophyletic (Fig. 2A), although it is unlikely these are issues with species delimitation. The paraphyly of A. grandiflorum (Fig. 2), a morphologically consistent species across a large geographic range (Brown pers. obs.), could be the result of potentially rapid and recent divergence or may be due to insufficient phylogenetically informative characters in this study. The latter could also apply to the polyphyletic species (A. hendersonii and A hirsutum), as A. hendersonii JA45, which is placed separately from the other conspecific samples is missing data for two of the four ncDNA loci (Table 1). However, this was not the case for A. hirsutum JA46. Re-examination of the vouchers of all accessions of A. hendersonii and A. hirsutum confirmed their identifications, suggesting that incomplete lineage sorting or paralogy problems associated with one or more nuclear loci could explain these non-monophyletic species; further data are required to investigate this.

The Clypeariae clade (clade E, Figs 2–3) includes all sampled species of ser. Clypeariae (19/51), except one accession of A. clypearia (JA95) from Papua New Guinea and A. pellitum (Gagnep.) I.C. Nielsen from Vietnam. Series Clypeariae was previously recognised in Pithecellobium as section Clypearia until Nielsen et al. (1984b) expanded Archidendron based on evidence from shared wood anatomy, inflorescence and pod morphology (Nielsen et al. 1984b). Characters of the pods are also useful to differentiate series Clypeariae from the rest of Archidendron. Nielsen et al. (1984b) described six pod types and most species of ser. Clypeariae have pod type 2 (long funicle, opens ventral suture first) or 6 (straight pods with overgrown seeds), while the other series primarily have pod type 1 (opens dorsal suture first, short funicles). Seeds of ser. Clypeariae are usually flattened and are not embedded in the pericarp, which is possibly linked to characteristics of the pod, such as dryness (de Wit 1942; Nielsen 1981, 1992; Nielsen et al. 1984b). Additionally, the combination of lack of stipules and solitary, stipitate ovaries delineates ser. Clypeariae (Nielsen et al. 1984b). Individually though, these characters are not diagnostic, as some species with sessile ovaries are placed in ser. Clypeariae (e.g. A. occultatum (Gagnep.) I.C. Nielsen and A. turgidum (Merr.) I.C. Nielsen), other species lacking stipules are placed in series Archidendron (e.g. A. harmsii and A. tjendana (Kosterm.) I.C. Nielsen), and two Philippine species of ser. Clypeariae (A. apoense (Elmer) I.C. Nielsen and A. merrillii (J.F. Macbr.) I.C. Nielsen) have more than one ovary but both are stipitate (Nielsen et al. 1984b). Given these morphological differences of ser. Clypeariae from the rest of Archidendron, together with the non-monophyly of the genus, there are grounds for segregating Clypeariae as a distinct genus; however, we are not proposing such a taxonomic change here for several reasons. First, there are many shared morphological characters between species of Archidendron s.l.; second, the shallow backbone of the ncDNA tree remains poorly supported with topological uncertainty between lineages; third, the placement of two species of ser. Clypeariae within the Archidendron s.s. clade (clade F; A. clypearia var. velutinum (Merr. & L.M. Perry) I.C. Nielsen and A. pellitum) raises further doubts; and fourth, phylogenetic sampling of species remains incomplete. All these issues suggest that denser taxon sampling and larger phylogenomic datasets are required before re-classifying Archidendron as two genera.

Archidendron clypearia is the most widespread species of Archidendron, found from India through to Papua New Guinea. The morphological variation within A. clypearia has been used to recognise four infraspecific taxa (Legume Phylogeny Working Group 2021): subsp. clypearia, subsp. subcoriaceum (Thwaites) M.G. Gangop & Chakrab., var. sessiliflorum (Merr.) I.C. Nielsen, and var. velutinum. The one accession of A. clypearia placed outside the Clypeariae clade (JA95) (Fig. 2A) has been identified as var. velutinum (Brown, pers. obs. of CANB525617; previously only identified to species level by the collector), the only infraspecific taxon found in eastern Malesia (Sulawesi, Moluccas and PNG). The three other samples of A. clypearia included in the phylogeny have not been assigned to infraspecific taxa but they are not likely var. velutinum, as they are from Malaysia and Vietnam and lack the woolly to velutinous hairs on the lower surface of the leaflets (Brown per. obs.). Taxonomic revision and denser phylogenetic sampling of A. clypearia from across its morphological and geographic range is required to verify this placement, delineate the taxa and investigate if var. velutinum should be raised to species level (Merrill and Perry 1942) or if there are intermediate forms as suggested by Kostermans (1966). The only other species of series Clypeariae that extends into eastern Malesia, A. palauense (Kaneh.) I.C. Nielsen, from the Moluccas through to the Solomon Islands (Nielsen et al. 1984b), was not sampled here. There are no obvious morphological characters that support placement of A. pellitum outside the Clypeariae clade, as it has the full combination of diagnostic characters of ser. Clypeariae: compressed pods with a long (3–5 mm) funicle, stipitate single ovary and no visible stipules (US 2515891; P01818442; Nielsen 1981). In addition, no evidence of paralogy in the nuclear loci of A. pellitum and A. clypearia var. velutinum (JA95) was noted in this study; all sequences suggest they fall in the A. lucyi subclade.

The last revision of the genus Archidendron (Nielsen et al. 1984b) significantly advanced our understanding of the genus but more detailed taxonomic study is still required, focusing especially on the large number of species known from incomplete material and widespread morphologically variable species, such as A. clypearia. To resolve the backbone of the Archidendron clade and inform decisions about generic delimitation to deal with the non-monophyly of Archidendron, we recommend further sampling of ser. Clypearia, particularly from the Wallacean region of Malesia (i.e. Moluccas, Sulawesi, Philippines), together with further genomic sampling.

While Archidendropsis is not monophyletic, its two subgenera (Archidendropsis and Basaltica) are (Fig. 2). The species within each subgenus have long been recognised as closely related (Bentham 1875; Nielsen 1981) but the two subgenera themselves have not always been associated with each other. For example, Bentham (1875) placed the species of each subgenus in different sections of Albizia based on inflorescence shape. Species of subgenus Archidendropsis that have flowers arranged in cylindrical spikes were placed by Bentham (1875) in Albizia section Lophantha Benth. (an illegitimate name later corrected to Albizia section Pachysperma (Benth.) Fosberg by Fosberg (1965)). Within this section they were separated from the other taxa, which are now recognised as Paraserianthes, into series Platyspermae Benth. because they have flattened, broadly orbiculate seeds (Bentham 1875). The two species of subgenus Basaltica known at that time (A. basaltica and A. thozetiana (F. Muell.) I.C. Nielsen) were placed by Bentham in his large section Eualbizzia distinguished by flowers in globular heads and flattened orbicular seeds (Bentham 1875). Within that section, these taxa were placed into series Obtusifolia, which corresponds to the Australian species with 1–2 jugate leaves, ovate, oblong or obtuse leaflets, short petioles, pedunculate heads in the axils, and small sessile flowers.

It was only recently that the species of the two subgenera were united within Archidendropsis by Nielsen (1983) based on characters of the fruit and seed: pods dehiscent along both sutures, and seeds that are winged, thin-walled and lack a pleurogram. However, Nielsen himself questioned whether the subgenera should be congeneric, noting that if they were not, “the evolution of the winged thin walled seeds without pleurogram should have happened twice” (Nielsen et al. 1983a: p. 337). The results presented here (Fig. 2) alongside two recent phylogenomic analyses (Demeulenaere et al. 2022; Ringelberg et al. 2022) show that the two subgenera of Archidendropsis do not form a monophyletic group, suggesting these seed characteristics are indeed the result of convergent evolution.

The presence of a pleurogram is common in mimosoid genera (Gunn 1984), and is considered to have evolved multiple times (Maumont 1993). Within the Archidendron clade, Archidendron and Archidendropsis are the only two genera whose seeds lack a pleurogram (Nielsen 1992). The absence of a pleurogram has been associated with short-lived ‘recalcitrant’ seeds (i.e. seeds which lack dormancy and can be viviparous; Nielsen 1992) and has been thought to be an adaptive response to humid environments (Corner 1951 in Nielsen 1992; Maumont 1993). Like the absence of a pleurogram, winged seeds are also rare in mimosoids occurring in only eight genera, including Archidendropsis (Gunn 1984). The possession of a winged seed has been suggested to be an adaptation for wind-dispersal but there have been no published observations of this in Archidendropsis (Gunn 1984; Nielsen 1992). The short viability of Archidendropsis seeds has been linked to the restricted geographic ranges of individual species (Nielsen 1983). However, humidity may be a more important determinant of these distributions, as the ranges of the two Australian species occurring in drier, non-rainforest habitats are more than 10 times larger than the rainforest species (e.g. A. basaltica ≥ 750,000 km² compared to A. xanthoxylon c. 8,750 km² (AVH 2021)). The habitats of A. basaltica and A. thozetiana are also more open than for A. xanthoxylon, but these two species generally have narrower wings on their seeds than the rainforest species A. xanthoxylon (Cowan 1998), suggesting that the wing is unlikely to have an impact on wind dispersal. Morphological features that have been used to unite the two subgenera in Archidendropsis are thus homoplasious and not useful for generic delimitation.

The non-monophyly and clear morphological distinctions between them means that the two subgenera can no longer be treated as congeneric and need to be placed in separate genera. As the type of Archidendropsis (A. fulgens (Labill.) I.C. Nielsen) is from subg. Archidendropsis, it is subg. Basaltica that requires a new name. No name exists at the generic level for these taxa, as they have previously been placed in Acacia, Albizia and Archidendropsis (Mueller 1859; Bentham 1875; Fosberg 1965; Nielsen 1983), names which are all typified by other taxa.

In addition to the aforementioned morphological differences between the two subgenera, species of subg. Basaltica are endemic to Australia, whereas those of subg. Archidendropsis are found in New Caledonia, New Britain, the Solomon Islands and on the island of New Guinea (Fig. 1B). Furthermore, there are several pollen characters separating the two subgenera (Nielsen et al. 1983a). Pollen of subg. Basaltica has isometric channels in the tectum and is aggregated into smaller polyads (55–68 μm), cf (80–120 μm) for subg. Archidendropsis where the tectum has non-isometric channels (Fig. 4; Nielsen et al. 1983a). The pollen surface of subg. Basaltica has fossules on the central cells, with either faint fossules or smooth peripheral cells, while in subg. Archidendropsis the surface of all pollen cells has small rounded areoles or deep fossules (Fig. 4; Nielsen et al. 1983a). Species of subg. Basaltica have sessile flowers arranged in globular pedunculate heads, rather than in spikes or racemes. Although one species of subg. Archidendropsis, A. fournieri (Vieill.) I.C. Nielsen, also has flowers arranged in globular pedunculate heads, it does not share the other diagnostic characters of subg. Basaltica, it is endemic to New Caledonia, its seeds are not winged, and the diameter of the pollen polyads is larger, fitting within the size range for subg. Archidendropsis (Nielsen 1983). Another character noted by Nielsen et al. (1983a) to differentiate the two subgenera, was the shape of the stipules, with those of subg. Basaltica being small and often developed into stipular spines (to 1.2 mm long; Brown pers. obs.; Fig. 5F) that are early caducous. However, the stipules of A. xanthoxylon were not recorded by Nielsen et al. (1983a) and are not like other Australian species being 1.2–3 mm long, ovate to triangular, dark gland-like and persistent (Brown, pers. obs., BRI AQ022813, BRI AQ0234095, BRI AQ0771148, BRI AQ199127, BRI AQ0199128; Fig. 5G). These stipules do differ, however, from those of the species of subg. Archidendropsis which, if present, are usually small (c. 1 mm), ovate or filiform and often caducous (Nielsen 1983).

Flowers arranged in globular heads, seeds lacking a pleurogram with a narrow peripheral membranous wing and flat, narrowly oblong, brown pods opening along both sutures distinguish this new genus from other Australian mimosoid legumes, and the keys in Flora of Australia (Cowan 1998) and available on KeyBase (Bean 2021; KeyBase 2021) still remain suitable.