We included 21 species of Acridocarpus (58% of the 36 accepted species), each represented by a single specimen, sampling all accepted subgenera and sections. A single species of Brachylophon was included as the sister group, and Byrsonima and Galphimia as outgroups of our analyses, according to Davis and Anderson (2010) and Almeida et al. (2024a). All sequences for the ITS, PHYC, matK, ndhF, rbcL, and trnL-F markers were retrieved from GenBank (Suppl. material 1), edited using Geneious v.4.8.4 (Kearse et al. 2012), and aligned using Muscle v.1.0 (Edgar 2004), with subsequent adjustments in the preliminary matrices made manually by eye.

Phylogenetic analyses were performed using Bayesian inference and Maximum Likelihood. The model selection used hierarchical likelihood ratio tests (HLRT) on jModeltest2 (Darriba et al. 2012). Both model-based methods were carried out with mixed models (GTR+G+I) and unlinked parameters, using MrBayes v.3.1.2 (Ronquist and Huelsenbeck 2003) and RAxML v.8 (Stamatakis 2014) and implemented on raxmlGUI v.2 (Edler et al. 2021). For the Bayesian inference, the Markov Chain Monte Carlo (MCMC) was carried out in two simultaneous independent runs with four chains each (one cold and three heated), saving one tree every 1,000 generations for a total of ten million generations. We excluded 20% of the retained trees as burn-in and checked for a stationary phase of likelihood and the Effective Sample Size (ESS) values higher than 200 for all parameters on Tracer v.1.7 (Rambaut et al. 2018). The posterior probabilities (PP) of clades were based on the majority-rule consensus, using the stored trees, and calculated with MrBayes v.3.1.2 (Ronquist and Huelsenbeck 2003).

For all species in the study, ten macromorphological characters were scored from Niedenzu (1928), Arènes (1945, 1954), and Almeida et al. (2020), and ten micromorphological characters were scored from Guesdon et al. (2019) and Santos et al. (2023). Character coding followed the recommendations of Sereno (2007) for morphological analyses. Primary homology hypotheses (De Pinna 1991) were proposed for 20 macro- and micromorphological characters (Suppl. material 2). All characters were optimised on the concatenated tree using the Maximum Likelihood function (mk1 model) of Mesquite v.3.61 (Maddison and Maddison 2019) and visualised on Winclada (Nixon 1999).

Dating estimates were based on a simplified ultrametric Bayesian combined tree generated with BEAST v.1.8.4 (Drummond et al. 2012). This analysis used a relaxed, uncorrelated lognormal clock and Yule process speciation prior to inferring trees. The calibration parameters were based on previous estimates derived from a comprehensive study of the whole Malpighiaceae (Davis et al. 2014) because no known fossil for Acridocarpus was available in the literature. We opted for calibrating at the crown node of Acridocarpus using a normal prior with mean initial values of 43.0 Mya and a standard deviation of 1.0 (Davis et al. 2014). Two separate and convergent runs were carried out, with 10,000,000 generations, sampling every 1,000 steps and 2,000 trees as burn-in. We checked for ESS values higher than 400 for all parameters on Tracer v.1.7 (Rambaut et al. 2018). Tree topology was assessed by selecting the maximum clade credibility tree of the runs with Tree Annotator and the graphic display generated in FigTree v.1.4.4 (Rambaut 2018).

Species distribution data were compiled from POWO (2025) and GBIF (2025). Ancestral areas of Acridocarpus and its relatives were estimated for continental (South America, Africa, India, Madagascar, and New Caledonia) and biome (dry forests, humid forests, and savannas) ranges using the R package (R Core Team 2025) BioGeoBEARS v.1.1.1 (Matzke 2018) and the models DEC, DIVALIKE, and BAYAREALIKE. Considering the possibility of underestimation of extinction due to the null ranges of species allowed by the models, analyses with the “*” option were also performed. The “j” option was used to consider founder speciation events. Different combinations of the three models and the two options generated a total of 12 models. The best-fitting models for analyses were determined by AIC weight ratios.

The combined dataset included 6,110 nucleotides. There were no major incongruences in the topologies produced by BI and ML analyses, so we chose to present the results on a single summarised tree (Fig. 1). Acridocarpus was resolved as monophyletic and sister to Brachylophon, with the Byrsonima + Galphimia clade as the outgroup of the analysis (Fig. 1). The BI analysis recovered an almost entirely resolved tree, while the ML analysis recovered a fully resolved tree with most clades well-supported (> PP 95% / BS 70) and a few poorly supported (< PP 95% / BS 70) (Fig. 1).

Within Acridocarpus, three well-supported clades were recovered: A. subg. Anophyllaris 1 (highlighted in pink on Fig. 1), A. subg. Acridocarpus, and A. subg. Anophyllaris 2 (highlighted in blue on Fig. 1). Only A. subg. Acridocarpus was recovered as monophyletic, with A. subg. Anophyllaris recovered into two distant clades representing sections Macrantha and Micrantha (Fig. 1). For both sections of Acridocarpus, all subsections were recovered as paraphyletic (black and white squares in Fig. 1).

Acridocarpus subg. Anophyllaris 1 (in pink on Fig. 1; = A. subg. Madagascariensis) was recovered with three synapomorphies (trees (0/0), leaves abaxial side uniformly and densely hairy with epidermis hidden (1/1), and cincinni bracteoles 2-glandular (18/2)) and a single homoplasy (leaves with midrib vascular system in an open arch without convoluted extremities (8/2)). Acridocarpus subg. Acridocarpus (in white on Fig. 1) was recovered with a single synapomorphy (lianas (0/2)) and homoplasy (leaf blade glands prominent in relation to the epidermis (17/2)). The clade comprising A. subg. Acridocarpus (in white on Fig. 1) + A. subg. Anophyllaris 2 (in blue on Fig. 1) was recovered with a single synapomorphy (sepals with 3 glands (19/3)). Finally, A. subg. Anophyllaris 2 (in blue on Fig. 1) was recovered with a single synapomorphy (leaves with midrib vascular system conformation in an open arch with convoluted extremities (8/3)) and homoplasy (leaf epidermis with cuboid cells (2/0)).

BioGeoBEARS selected the DIVALIKE + J model for both continental (LnL = -16.85) and biome (LnL = -23.56) analyses due to the lowest LnL scores (Figs 2, 3). The MRCA of the Brachylophon + Acridocarpus clade (i.e. tribe Acridocarpeae) arose at 43 Mya and was widespread in rainforests through Africa, the Indian subcontinent, and Madagascar (Figs 2, 3). At the same time, the MRCA of Brachylophon colonised the rainforests of the Indian subcontinent (Figs 2, 3). The MRCA of Acridocarpus arose at 41 Mya and was widespread in rainforests of continental Africa and Madagascar (Figs 2, 3). At the same time, the MRCA of A. subg. Anophyllaris 1 colonised the rainforests of Madagascar, diversifying at 32 Mya (Figs 2, 3). The MRCA of A. humbertii colonised the dry forests of Madagascar at the same age, while the MRCA of A. adenophorus diversified in the rainforests of Madagascar at 22 Mya (Figs 2, 3). The MRCA of A. excelsus + A. perrieri + A. austrocaledonicus clade diversified in the savannahs of Madagascar from 17–10 Mya (Figs 2, 3). The MRCA of A. austrocaledonicus colonised New Caledonia (Oceania) from a Malagasy ancestor at 10 Mya (Figs 2, 3). The MRCA of West African Acridocarpus (i.e. A. subg. Acridocarpus + A. subg. Anophyllaris 2) colonised continental African rainforests at 39 Mya (Figs 2, 3). Finally, the colonisation of continental African dry forests took place at three different times, starting at 39 Mya in the MRCA of A. plagiopterus in West Africa, 24 Mya in the MRCA of Acridocarpus in East Africa, and 11 Mya in the MRCA of the West African A. spectabilis (Figs 2, 3).

The infrageneric classification of Acridocarpus was proposed by Niedenzu (1928), who divided the genus into two subgenera (A. subg. Acridocarpus and A. subg. Anophyllaris), four sections (sect. Acridocarpus, Diacra, Macranthera, and Micranthera), two subsections (subsect. Adenochrea Nied. and Adenolepis Nied.), four series (ser. Hemioopterys Nied., Leptorrhachis Nied., Machaeropterys Nied., and Pachyrrhachis Nied.), and five subseries (subser. Actinadenia Nied., Anadenia Nied., Monadenia Nied., Pleiadenia Nied., and Zygadenia Nied.; Suppl. material 3). This author based his infrageneric classification on morphological characters such as habit type, shape of inflorescence axis, presence and number of glands in bracteoles and sepals, length of anthers, posture of the dorsal wing, and shape of the nuts in mericarps.

Acridocarpus. subg. Anophyllaris comprises trees to shrubs, with erect bracteoles, short anthers, mericarps with an erect dorsal wing, and mericarp nuts semi-lentiform in shape, while A. subg. Acridocarpus comprises lianas with patent bracteoles, long anthers, mericarps with a patent dorsal wing, and mericarp nuts semi-globose to semi-ovoid in shape (Niedenzu 1928; Suppl. material 3). Acridocarpus subg. Acridocarpus is divided into two sections based on the shape and confluence of the apex of anthers: A. subg. Acridocarpus sect. Acridocarpus (Monacra) is characterised by the acute apices of anthers and connivent anther sacs, while A. subg. Acridocarpus sect. Diacra Nied. is comprised of species with bicornate apices of anthers and divergent anther sacs (Niedenzu 1928; Suppl. material 3). Acridocarpus subg. Anophyllaris is divided into two sections based on the shape and presence of glands in the sepals and the length of anthers and filaments: A. subg. Anophyllaris sect. Macrantha Nied. with ovate-rotund, 1–8 glandular sepals and 3.5–7 mm long anthers; and A. subg. Anophyllaris sect. Micrantha Nied. with ovate, eglandular sepals and 1.5–2 mm long anthers (Niedenzu 1928). Niedenzu (1928) divided sect. Macrantha into two subsections (subsect. Adenochrea Nied. and Adenolepis Nied.), four series (ser. Hemioopterys Nied., Leptorrhachis Nied., Machaeropterys Nied., and Pachyrrhachis Nied.) and five subseries (subser. Actinadenia Nied., Anadenia Nied., Monadenia Nied., Pleiadenia Nied., and Zygadenia Nied.; Suppl. material 3). In contrast, Arènes (1945) divided sect. Micrantha Nied. into two subsections (subsect. Madagascariensis and Neocaledonici Arènes; Suppl. material 3).

All diagnoses presented for Niedenzu’s proposed subgenera were not comparable to each other, making it incredibly difficult to correctly assign any newly published species to his infrageneric rank circumscriptions. These incomparable diagnoses were also pointed out by Arènes (1945), who published two additional subsections for A. subg. Anophyllaris sect. Micrantha. Subsection Madagascariensis Arènes included all Acridocarpus species from Madagascar, while subsect. Neocaledonici Arènes only included the New Caledonia endemic A. austrocaledonicus. Arènes based his infrageneric classification on morphological characters such as habit type, presence, shape, and size of glands in bracteoles, petal consistency and shape, and length of filaments and anthers (Arènes 1945). Later, Morton (1968) found out that A. subg. Anophyllaris sect. Macranthera was an illegitimate name due to part of its species being previously published under Anomalopterys (DC.) G.Don. Thus, he published A. subg. Anophyllaris sect. Anomalopterys to accommodate this name and synonymised A. subg. Anophyllaris sect. Macranthera under it.

Our results showed that from all infrageneric ranks currently accepted in Acridocarpus, only A. subg. Acridocarpus, A. subg. Anophyllaris sect. Anomalopterys, and A. subg. Anophyllaris sect. Micrantha were resolved as monophyletic (Fig. 1). All remaining infrageneric ranks were found to lack monophyly. Most of the diagnostic morphological characters that were used for their circumscription were recovered as highly homoplastic (Fig. 1). Our analyses have recovered the presence of nectar-secreting sepal glands as a synapomorphy for continental African Acridocarpus (Fig. 1). Nonetheless, most Paleotropical lineages of Malpighiaceae also present nectar-secreting sepal glands, resulting from a pollination shift that occurred in the ancestor of these lineages when dispersing from the Neotropics to the Paleotropics (Davis et al. 2014). Sepal glands in species of Acridocarpus are small and restricted to the base near margins or are a single, large, and intersepaline-placed gland resulting from the fusion of glands from adjacent sepals (Guesdon et al. 2019). Thus, nectar-secreting glands are a homoplastic character in Malpighiaceae related to a pollination shift and cannot be regarded as an exclusive trait for Acridocarpus.

A recircumscription of the infrageneric classification of Acridocarpus is presented in the taxonomy section below.

Previous studies (Davis et al. 2002) just using dispersal-vicariance analysis pointed to the boreotropical route as the dispersal route taken by the MRCA of Acridocarpus. However, our results point to a Gondwana dispersal route, supported by 12 different statistical models (Matzke 2018) and a new fossil record (Ali et al. in press). Our results indicate that the MRCA of Acridocarpus colonised rainforests of Africa and Madagascar via the Gondwana route. This is not surprising since 35 species (out of 36 species) of the genus are currently restricted to countries in continental Africa and Madagascar with patches of rainforest (POWO 2025). Our analyses also corroborate the existence of a long-distance dispersal event at 43 Mya of the MRCA of Acridocarpus + Brachylophon from its Neotropical relatives (i.e. subfamily Byrsonimoideae), corroborating Davis et al. (2014). This dispersal event occurred from the Americas to the Indian subcontinent + Madagascar + Africa through the Gondwana route instead of the Boreotropical route (Davis et al. 2004, 2014). The Boreotropical route is greatly supported by fossils (Givnish and Renner 2004) and the calibrated phylogenies of several groups of plants, such as Clusiaceae (Meseguer et al. 2018), Annonaceae (Thomas et al. 2015), Symplocaceae (Fritsch et al. 2015), Leguminosae (Estrella et al. 2017), Lycopodiaceae (Bauret et al. 2018), and Malpighiaceae (Davis et al. 2014; Almeida and van den Berg 2022), among others.

The MRCA of Acridocarpus, represented by the Acridocarpus crown node, started to diversify in rainforests of the African continent around 43 Mya, with at least four different colonisation events in dry forests and a single colonisation event in savannas. These results corroborate the presence of a pan-African rainforest from West to East during the Eocene (Couvreur et al. 2021), which was likely the migration route for the MRCA of continental African Acridocarpus to disperse from Madagascar to West Africa and back again to continental East Africa (Fig. 3). This is a well-documented pattern in African rainforests, which substantiates the breakup of this continuous pan-African rainforest in the cooler and drier Miocene (Couvreur et al. 2021). Our results exhibit that most dry forest lineages in Acridocarpus started to diversify in the Oligocene, matching the Oligocene origins of most neotropical dry forest lineages of Malpighiaceae, both in Central and South America (Willis et al. 2014; Almeida et al. 2018). The same Eocene/Oligocene pattern of African dry forests diversification is also observed in Apocynaceae (Bitencourt et al. 2021), Zygophyllaceae (Wu et al. 2018), and Rhamnaceae (Rickenback et al. 2024). Additionally, the only savanna lineage of Acridocarpus diversified throughout the entire Miocene in Madagascar, corroborating the Miocene origins of American, African, and Malagasy savanna species of Malpighiaceae (Almeida et al. 2018, 2024b; Almeida and van den Berg 2020). The remarkable congruence of New and Old World savanna origins has already been stated by Pennington and Hughes (2014), but additional dated phylogenies are still needed to properly corroborate the timeline of the origins of savannas worldwide.

Numerous studies on the age of Malagasy savannas can be found in the literature, drawing on various types of evidence. Based on dated molecular phylogenies of C3 grasses, some authors showed that savannas originated in the Oligocene to Miocene (Hackel et al. 2018). The C3 grasses continued to radiate into the Pliocene and Quaternary, while C4 grasses originated later in the Miocene to Pliocene (8–3 Mya) and radiated expansively into the Pleistocene (Hackel et al. 2018; Hagl et al. 2021). The Malagasy grass lineages are linked to dispersal events primarily from mainland Africa, a pattern also observed in Acridocarpus. Moreover, the radiation of C4 grasses in Madagascar coincided with the origin and global spread of open, grassy, fire-prone landscapes during the Miocene (Cerling et al. 1997; Edwards et al. 2010; Strömberg 2011). Silander Jr et al. (2023) evidenced 41 woody plant lineages endemic to the grassy ecosystems of Madagascar. From this initial list, we were able to find studies with dated molecular phylogenies for at least 14 native lineages (Arecaceae, Asparagaceae, Asteraceae, Cyperaceae, Fabaceae, Malpighiaceae, Melastomataceae, Oleaceae, and Poaceae) showing a distinct pattern of Miocene radiation in Malagasy savannas (Table 1).