Phylogeny of the Madagascar-centred tribe Danaideae (Rubiaceae) as a precursor to taxonomic revision: insights into its generic and species limits, affinities and distribution

Abstract Background and aims The tribe Danaideae (Rubiaceae) is almost exclusively endemic to the Western Indian Ocean Region (WIOR), and encompasses the genera Danais, Payera and Schismatoclada that occur in humid, sub-humid and mountain and mountain bio climate zones. Much of the species diversity is endemic to restricted, remote and/or mountainous areas of Madagascar and recent field work on the island indicates substantial unknown diversity of the Danaideae. Furthermore, the monophyly of the Malagasy genera Payera and Schismatoclada has been questioned in previous work, species delimitations and phylogenetic relationships within the genera are poorly understood, and the distribution and evolution of gross morphological features have not been assessed. Methods We conducted morphological investigations, and produced robust phylogenies of Danaideae based on nuclear and plastid sequence data from 193 terminals. Ample plant material has been newly collected in the WIOR for the purpose of the present study, including potentially new species unknown to science. We performed Bayesian non-clock and relaxed-clock analyses employing three alternative clock models of a dataset with a dense sample of taxa from the entire geographical ranges of Danaideae. Based on the results, we discuss species diversity and distribution, relationships, and morphology in Danaideae. Key results Our results demonstrate the monophyly of Danaideae, its three genera and 42 species. Nine species are resolved as non-monophyletic. Many geographically distinct but morphologically heterogeneous lineages were identified, and morphological features traditionally considered diagnostic of subgroups of the genera, used for example in species identification keys, are not clade-specific. Conclusions Our results demonstrate that Madagascar contains ample previously undocumented morphological and species diversity of Danaideae. Our novel approach to molecular phylogenetic analyses as a precursor to taxonomic revisions provides numerous benefits for the latter. There are tentative indications of parallel northward diversification in Payera and Schismatoclada on Madagascar, and of geographical phylogenetic clustering despite the anemochorous condition of Danaideae.


INTRODUCTION
The Western Indian Ocean Region (WIOR) is one of the world's biodiversity hotspots, encompassing Madagascar and the neighbouring Comoros, Mascarene and Seychelles archipelagos (Myers et al., 2000) (Fig. 1). Madagascar is by far the largest island (~593 000 km 2 ) in the region and its extant biota mostly encompasses the descendants of Cenozoic dispersers, often from mainland Africa (e.g. Yoder and Nowak, 2006), with few groups with Asian, Neotropical and Pacific origins (e.g. Rouhan et al., 2012;Razafimandimbison et al., 2017). Madagascar has a complex landscape dominated by mountainous areas running northsouth, stretching from the Tsaratanana Massif and its satellites in the north to the Anosy mountain chains in the southeast, and resulting in a central highland above 800 m (Fig. 1). The eastern slope of the central highland is narrow and falls steeply towards the Indian Ocean. By contrast, its western slope is occupied by a large plain declining progressively towards the Mozambique Channel (e.g. Tattersall and Sussman, 1975). The island has five distinct bioclimatic zones and diverse vegetation types, partly due to the impacts of the southeastern trade winds (Alizé) and the north-western monsoon from the Equator (Koechlin, 1972;Koechlin et al., 1974;Jury, 2003;Fig. 1). The eastern region (0-800 m in altitude) is humid, and harbours littoral forests (at sea level and on sandy soils) and lowland rainforests (including a narrow strip of rainforests along the east coast and the low-elevation rainforests from sea level up to 800 m). The central highland (800-2000 m in altitude) is sub-humid, and hosts montane (or highland) rainforests, sclerophyllous montane forests, woodland savannahs and ericoid thickets. These highland rainforests, woodland savannahs and sclerophyllous forests are intermixed, while the ericoid thickets are characteristics of the high (above 2000 m in altitude), isolated mountains of the Andringitra, Ankaratra, Marojejy and Tsaratanana Massifs and their respective satellites. Most of the western region and the north tip (except the Montagne d'Ambre Massif) of Madagascar are dry, harbouring deciduous, dry forests and woodland savannahs; part of northwestern Madagascar (including the Nosibe Island) is sub-humid, hosting semi-deciduous forests. The south and southwestern regions are sub-arid, hosting spiny thickets (Koechlin, 1972;Koechlin et al., 1974;Jury, 2003;Fig. 1).
Madagascar harbours about 11 400 endemic species, 1743 native genera and 253 families of vascular plants (as of November 2020, Tropicos database). The coffee family (Rubiaceae) is the second largest flowering plant family after the orchid family, with at least 1000 species in 95 genera and 27 tribes on the island. It thus encompasses about 11 % of the land plant diversity of Madagascar, with the level of specific endemism nearly 100 % (Razafimandimbison et al., 2022). Alberteae sensu Kainulainen et al. (2009) (subfamily Cinchonoideae sensu lato) and Danaideae sensu Bremer and Manen (2000) (subfamily Rubioideae) are the only two rubiaceous tribes with their species diversity centred in the WIOR. Danaideae (Fig. 2) contain a high proportion of microendemics (endemic species with very small areas of  occurrence; Ganzhorn et al., 2014) (Buchner and Puff, 1993;Puff and Buchner, 1994). On Madagascar, the eastern lowland and central highland rainforests of the humid and subhumid zones (Fig. 1) are the primary habitats of Danaideae, and its geographical range extends from the island's southeast tip to the northernmost tip (the Montagne d'Ambre Massif) (Fig. 1). Danaideae do not occur in the dry habitats, meaning that its members are entirely absent in the dry and sub-arid zones in western and southwestern Madagascar, respectively (e.g. Buchner and Puff, 1993;Puff and Buchner, 1994;Fig. 1); their centre of species richness is the highland rainforests above 800 m, with only few species found in the eastern littoral and lowland forests, the north-western semi-deciduous forests (from sea level to 800 m), and the ericoid thickets above 2000 m elevation.
Danaideae contain 69 currently accepted species (Supplementary Data Table S1) classified in three genera: the lianescent Danais Comm. ex Vent ( Fig. 2A, B) (38 species), and the two arborescent (shrubs or small trees) genera Payera Baill. (Fig. 2C, D) (10 species) and Schismatoclada Baker (Fig. 2E, F) (21 species). Danais is widely distributed in the WIOR (but absent in the Seychelles), with most species endemic to Madagascar (Puff and Buchner, 1994;Taylor and Rodgers, 2013), but two occur in the Comoros (one endemic and one species shared with Madagascar; Puff and Buchner, 1994), three are endemic species in the Mascarenes (Verdcourt, 1989) and one species is endemic to mainland Africa (Tanzania; Verdcourt, 1976). By contrast, Payera and Schismatoclada are Malagasy endemics. Danais ( Fig. 2A, B) was originally described by Ventenat (1799), but it was Persoon (1805) who validly published the first Danais species, Danais fragrans (Comm. ex Lam.) Pers. from the Mascarenes and Madagascar, and described D. sulcata Pers. from Mauritius. Specimens of D. fragrans from Madagascar and from the Mascarenes (Mauritius and Reunion Islands) have, however, been shown to be distantly related (Krüger et al., 2012), with the latter species forming a clade with the Mauritius D. sulcata. As a result, D. fragrans is now restricted to the Mascarenes, and the Malagasy D. lyallii Baker was resurrected to accommodate the Malagasy D. fragrans (Krüger et al., 2012). The first comprehensive morphological study on Danais and its allied genera Payera and Schismatoclada was conducted by Buchner and Puff (1993), who demonstrated their close affinities and proposed new generic circumscriptions. The authors rejected Danais as defined by Cavaco (1966), and transferred four arborescent Malagasy species (Danais bakeriana Homolle, D. decaryi Homolle, D. madagascariensis Cavaco and D. mandrarensis Homolle ex Cavaco) to Payera (Buchner and Puff, 1993), rendering Danais an exclusively lianescent group. While this taxonomic adjustment has yet to be assessed using molecular data, Danais as delimited by Puff and Buchner (1994) has been widely accepted by the Rubiaceae community. In their revision of the Malagasy and Comorian Danais, Puff and Buchner (1994) recognized a total of 26 species (plus two that were considered poorly known). Krüger et al.'s (2012) resurrection of D. lyallii and six new species from Madagascar described by Taylor and Rogers (2013) bring the total number of Malagasy species of Danais to 35. Two additional species, D. corymbosa Balf. F. and D. sulcata, were treated in the Mascarene Flora of Rubiaceae (Verdcourt, 1989), and the Tanzanian D. xanthorrhoea (K. Schum.) Bremek. was included in the East African Flora of Rubiaceae (Verdcourt, 1976), bringing a total of 38 currently accepted species of Danais (Supplementary Data Table S1).
The Malagasy genus Payera (Fig. 2C, D) was originally described by Baillon (1878) based on the holotype and the sole isotype (both at herbarium P) of Payera conspicua Baill. The type species P. conspicua may have been an extremely rare species that is now extinct; it is known only from the type specimens collected in 1841. The type locality on Madagascar is unknown and the species has not been recollected for 180 years despite recent efforts. As a result, the genus remained virtually unknown until Buchner and Puff (1993) conducted their detailed morphological study on the so-called 'genus complex Danais-Schismatoclada-Payera' (now Danaideae sensu Bremer and Manen, 2000), including the Malagasy monospecific genus Coursiana Homolle (Homolle, 1942). They report that the former director of the Paris herbarium, Professor J.-F. Leroy, based on his unpublished notes, was the first to have postulated close affinities of Payera with Danais, Coursiana and Schismatoclada. Leroy's hypothesis was endorsed by Buchner and Puff (1993), who used corolla pubescence, corolla aestivation type and fruit dehiscence as primary characters to recircumscribe the two arborescent Danaideae genera (Payera and Schismatoclada). As a consequence, Buchner and Puff (1993) formally transferred two dwarf or low-growing shrubby species of Schismatoclada (S. beondrokensis Humbert and S. coriacea Humbert, Humbert, 1955) endemic to the Marojejy Massif in northeastern Madagascar (Fig. 1) to Payera. Furthermore, Buchner and Puff (1993) formally merged the Malagasy monotypic genus Coursiana in Payera, and described two new species of Payera (P. glabrifolia J.-F. Leroy ex R. Buchner and Puff and P. marojejyensis R. Buchner and Puff). These taxonomic adjustments [together with Buchner's and Puff's (1993) transfer of the four arborescent Danais species to Payera] increased the total number of Payera species to 10 (Supplementary Data Table S1).
The monophyly of Danaideae is strongly supported by molecular data (e.g. Bremer and Manen, 2000;Krüger et al., 2012). Beside their isolated geographical range in the WIOR (with the exception of the Tanzanian Danais xanthorrhoea), the members of the tribe can be characterized by a combination of the following character states: presence of raphides in tissues, heterodistylous flowers, valvate corolla aestivation, dry capsular fruits and winged seeds (Buchner and Puff, 1993). Features such as growth habit, aestivation of corolla lobes, corolla pubescence and fruit type have been used for genus recognition in Danaideae (Buchner and Puff, 1993), but the plesiomorphic or apomorphic status of these features, i.e. their evolutionary history, have not been tested. For example, while Danais and Schismatoclada are more similar to each other than either is to Payera regarding the large bracts subtending the inflorescences (mostly absent vs. mostly present), aestivation of corolla lobes (valvate-reduplicate vs. valvate s.s.), and corolla pubescence (mostly absent vs. covered by appressed, silvery hairs) (e.g. Buchner and Puff, 1993;Schatz, 2001), Payera and Schismatoclada share an arborescent habit and capsular fruits that are conspicuously beaked (Fig. 2D, F), while Danais and Payera have capsular fruits that dehisce loculicidally (Buchner and Puff 1993: 62). Molecular phylogenetics is an important tool that allows systematists to assess the phylogenetic value of such characters.
The first molecular phylogenetic study that focused on Danaideae (Krüger et al., 2012) indicated that Payera and Schismatoclada as defined by Buchner and Puff (1993) are mutually paraphyletic, and thus untenable, but the authors refrained from making any taxonomic adjustment, because the type of Payera (P. conspicua) was not investigated and statistical support for results were partly weak. Furthermore, the sampling of Payera and Schismatoclada available to Krüger and colleagues at the time (2012) did not cover the entire geographical ranges of these genera. For example, none of the Schismatoclada species from southeastern Madagascar and the Payera species from central east of the island was included in the study, and many included specimens were not identified, or were misidentified, at the genus and/or species level. In association with ongoing taxonomic revisions, we have re-identified all the Danaideae specimens included in Krüger et al. (2012), and ample newly collected plant material is available from subsequent recent field work by us. It is thus timely and highly relevant to conduct a new phylogenetic study of Danaideae.
The main objective of the present study is to investigate species diversity and phylogenetic relationships in the tribe Danaideae based on morphology and molecular data. We included 193 samples, many of which are newly collected in remote areas of Madagascar representing 43 accepted species and 37 (potentially) new yet undescribed species of Danaideae identified during ongoing taxonomic revisions (Supplementary Data Tables S1 and S2). The resulting phylogenies are subsequently utilized to: (1) re-assess the monophyly of the three genera of Danaideae, (2) investigate species delimitations and relationships within the tribe, and (3) evaluate the distribution and evolution of gross morphological characters commonly used in floras and taxonomic work of Danaideae.

Taxon sampling
We investigated as many Danaideae species as possible for the present study, including those analysed in Krüger et al. (2012) and a substantial set of new plant material collected by us for the purpose of the present study during field work over the entire distribution area of the tribe. Most of the sampled species were represented by two-three individuals, while species known to be morphologically variable with wide geographical distributions (e.g. Danais cernua Baker, D. lyallii, Schismatoclada concinna, S. farahimpensis Homolle and S. psychotrioides) were represented by many individuals. We were not able to obtain sequenceable material for all species of Danaideae (i.e. species listed in grey in Supplementary Data Table S1). On the other hand, a total of 37 as yet undescribed and potentially new species of Danais, Payera and Schismatoclada were included in our study. Eight of the 31 included species of Danais, three of the 11 included species of Payera and 19 of the 31 included species of Schismatoclada are potentially new yet undescribed species. In addition to these 30 new and 43 currently accepted species, which were all investigated using morphology and molecular data, two potentially new species of Payera and five new species of Schismatoclada (for which no molecular data were produced) were studied morphologically. Six species from six tribes of the Spermacoceae alliance and seven species from seven tribes of the Psychotrieae alliance were included as outgroups. The species from the Psychotrieae alliance were used to root the trees. Of the total 193 analysed samples, 74 were from Danais (representing 23 of the 38 accepted species + eight newly proposed species, ~67 %), 21 samples were from Payera (representing eight of the ten accepted species + three of the five newly proposed species, ~73 %) and 85 samples were from Schismatoclada (representing 11 of the 21 accepted species + 19 of 24 newly proposed species, ~67 %). Voucher information, country of origin of these samples and sequence accession numbers are specified in Table S2.

Morphological studies
All sampled Danaideae species included in the present study have been carefully studied using stereomicroscopy and conventional techniques. The specimens were identified with the aid of species keys in Puff and Buchner (1994) and Taylor and Rodgers (2013) for Danais, in Buchner and Puff (1993) for Payera, and in Cavaco (1964) and Buchner and Puff (1993) for Schismatoclada. Scanned images of the type specimens of Danais, Payera and Schismatoclada species available through the databases of the Kew Royal Garden (WCSP, 2022), Missouri Botanical Garden (Tropicos, 2022) and Paris herbarium (MNHN, Chagnoux S, 2022) were accessed for comparison with the studied material, which comprises more than 800 specimens of these genera loaned from the G, BR, K, MO, P, S, UPS, TAN and TEF herbaria (Thiers, 2016). We placed particular emphasis on morphological variation within and among species regarding features used in these floral and classificational works, i.e. leaf size, type and phyllotaxy, stipule type and flower colour. These morphological investigations are in addition part of extended morphological studies conducted by the first author (S.G.R.) for ongoing taxonomic revisions of Payera and Schismatoclada.

Choice of molecular markers, molecular laboratory procedures and data assembly
We used three plastid (ndhF, matK including trnK, and trnT-F) and nuclear ribosomal ITS (nrITS) markers for our study. DNA extraction and amplification were achieved following the protocols outlined in Razafimandimbison et al. (2004) for nrITS, Bremer et al. (1999) for ndhF, Kainulainen and Bremer (2014) for matK, and Razafimandimbison and Bremer (2002) for trnT-F. The same primers used for PCRs were utilized for sequencing reactions, which were sent to Macrogen Europe (Amsterdam, the Netherlands) for sequencing. The new sequence data were assembled using the Staden package v.2.0.09b (Staden, 1996). For each marker, sequences were aligned using MUSCLE v.3.8.31 (default settings; Edgar, 2004), as implemented in AliView v.1.18.1 (Larsson, 2014). Manual adjustments were subsequently done for the nrITS and trnT-F datasets following the similarity criterion (Simmons, 2004) using AliView.

Phylogenetic non-clock analyses
Phylogenetic reconstructions were achieved using the Bayesian Markov chain Monte Carlo (MCMC) method (Yang and Rannala, 1997) as implemented in the software MrBayes v.3.2.7b (Ronquist et al., 2012). Data partitioning was selected using the software PartitionFinder v.2 (Lanfear et al., 2017), which indicated a best-fitting partitioning scheme employing two partitions for our data, one comprising trnT-F and a second comprising nrITS, matK and ndhF. The general time reversible substitution model with rate variation across sites modelled as a gamma distribution (GTR+Γ) was used for trnT-F. For the partition including the nrITS, matK and ndhF datasets, the general time reversible substitution model with rate variation across sites modelled as a gamma distribution and including a proportion of invariable sites (GTR+Γ+I) was used. Model selection was based on the corrected Akaike information criterion as calculated utilizing MrAIC v.1.4.6, (Nylander, 2004). Gaps were treated as missing data in all alignments. Inferred indels/ deletions were not coded as separate characters, and sites considered ambiguously aligned in the nrITS and trnT-F regions were removed from the analyses. Individual gene regions were initially analysed separately using Bayesian inference and substitution models as specified above.
All separate and combined Bayesian analyses were run on the CIPRES computing cluster (Miller et al., 2010). The combined data matrix is provided in nexus format (Supplementary Data File S1). Each analysis comprised two runs of four chains each that were run for 20 million generations, sampling trees and parameters every 2000th generation. For all analyses, stationarity and convergence of runs were checked using the program AWTY (Nylander et al., 2008). The effective sample size (ESS) of parameters was monitored using the program TRACER v.1.6 (Rambaut and Drummond, 2013). Trees sampled before the Bayesian posterior probability (BPP) of splits stabilized were excluded as a burn-in phase. All saved trees from the two independent runs were subsequently pooled for a consensus tree.

Phylogenetic relaxed-clock analyses
With the primary objective to infer phylogenetic relationships within a relaxed-clock framework (not to estimate node ages), we performed relaxed-clock analyses using three alternative relaxed-clock models, the Brownian motion model (TK02) described by Thorne and Kishino (2002), the white noise model (WN) described by LePage et al. (2007) and the independent lognormal model (ILN) described by Drummond et al. (2006). The fossil Morinda chinensis from the middle Eocene of south China (Shi et al., 2012) was, based on morphological investigations in Razafimandimbison et al. (2009a), used to constrain the split between Morindeae (here represented by Morinda citrifolia L.) and Gaertnereae (here represented by Gaertnera phyllosepala Baker). We specified a uniform prior age distribution for the Morindeae stem lineage with a minimum age of 38 Ma (middle-late Eocene; Shi et al., 2012) and a maximum age of 70 Ma. The root node (split between Psychotrieae and Spermacoceae alliances) was constrained to a minimum age of 0 Ma, a mean age of 61 Ma and a standard deviation of 4.6 Ma, which results in 95 % highest probability density (HPD) limits of 52-70 Ma, consistent with results in Wikström et al. (2015). A truncated normal prior age distribution was also specified for the Spermacoceae crown group with a minimum age of 0 Ma, a mean age of 48 Ma and a standard deviation of 5.1 Ma, resulting in 95 % HPD limits of 38-58 Ma, consistent with Wikström et al. (2015).
The relaxed-clock analyses were otherwise executed in the same way as the non-clock analyses (see above). Consensus trees were produced from the posterior distributions of each of these analyses, as well as from combining them into a single posterior distribution (in order to accommodate the variation in the BPP obtained in the three individual analyses using different clock models) (Supplementary Data Table S3).

Main topological results
Phylogenetic results and clade distributions based on the relaxed-clock analyses are summarized in Fig. 3 with BPP values indicated above nodes and those from the non-clock analyses added below nodes. The Danaideae and each of its three genera are strongly supported as monophyletic. Danais is sister to Payera + Schismatoclada but the statistical support for the latter clade is weak (BPP 0.72). Danais is resolved into four main clades, a Malagasy northwestern clade, a clade with broader distribution on Madagascar, a Mascarene clade, and a mostly Malagasy clade with a few species in the Comores and Tanzania (Fig. 3). Payera and Schismatoclada are resolved into four and five main clades, respectively, which are all geographically restricted to one or a few of the regions of Madagascar (Fig. 3). Detailed phylogenetic results within Payera, Schismatoclada and Danais are, respectively, depicted in Figs 4-6 (see below).

Morphological studies
We were able to identify all the Danaideae specimens used in our study to species (Supplementary Data Table S2). Our phylogenetic results show that gross morphological features that are frequently discussed in the taxonomic literature on Payera and Schismatoclada do not characterize monophyletic groups. In Payera, stipules can be fimbriate as in P. homolleana . The phylogeny shows a simplified 50% Bayesian majority-rule consensus tree retrieved from the combined tree distributions of three individual relaxed-clock analyses (using three different relaxed-clock models), and based on analyses of plastid matK, ndhF, trnT-F and nuclear ribosomal ITS from 193 samples. Values above and below nodes are Bayesian posterior probabilities from the relaxed-clock and non-clock analyses, respectively. Node numbers are indicated in bold type to the right of nodes. The mini-maps show geographical distributions of clades across the official Regions of Madagascar (B). Scale bar represents expected number of substitutions per site. of the Payera north-eastern clade, purple to dark purple as in P. marojejyensis in the Payera northeastern clade (and P. 'ambalabeensis' ined.; no molecular data), white corolla tubes with red-orange corolla lobes as in P. 'vangaindranoensis' in the Payera southeastern clade (Fig. 4), or white as in the remaining species of Payera.

Phylogenetic results based on analyses of molecular data
Analyses of separate gene regions. The 50 % Bayesian majority rule consensus trees generated from non-clock analyses of the nrITS, the non-coding plastid (trnT-F) and coding plastid (matK and ndhF) are presented as Supplementary Data (Fig.  S1). Visual inspection showed no supported topological conflict (supported clades are defined as having a BPP ≥ 0.95, Erixon et al., 2003). Accordingly, we merged the sequence data of these four markers into a large matrix, which contained a total of 193 samples and 7082 characters, of which 1115 were parsimonyinformative. GenBank accessions of the 677 newly produced  Fig. 3, drawn as a cladogram, to show relationships in the genus Payera. As in Fig. 3, support values above and below nodes are Bayesian posterior probabilities from the relaxed-clock and non-clock analyses, respectively. Node numbers are indicated in bold type to the right of nodes.  Fig. 5. Detailed phylogenetic results from the analysis presented in Fig. 3, drawn as a cladogram, to show relationships in the genus Schismatoclada. As in Fig.  3, support values above and below nodes are Bayesian posterior probabilities from the relaxed-clock and non-clock analyses, respectively. Node numbers are indicated in bold type to the right of nodes.
Analyses of the combined data set . Results from the combined non-clock and the three combined relaxed-clock analyses were congruent; none of the observed differences received strong support in either analysis. Figures 3-6 depict the Bayesian 50 % majority rule tree produced from the relaxed-clock analyses of the combined dataset, with the posterior tree distributions of the three relaxed-clock analyses merged into a single distribution. The resulting support values are shown above nodes (Figs  3-6). Support values from the non-clock analysis of the same dataset are plotted below nodes (Figs 3-6). Node numbers and BPP values calculated from the three individual distributions of the relaxed-clock analyses employing different clock models, as well as from the combined tree distribution, are given as Supplementary Data (Table S3).

Generic monophyly in Danaideae
The monophyly of the tribe Danaideae sensu Bremer and Manen (2000) is supported by all separate and combined analyses; our combined nuclear-plastid analyses (Figs 3-6) are the first to demonstrate with strong support (1/1) the monophyly of all the three Danaideae genera, Danais sensu Puff and Buchner (1994), and Payera and Schismatoclada both sensu Buchner and Puff (1993). Buchner and Puff (1993) formally transferred four arborescent Malagasy species of Danais with pubescent corollas (D. bakeriana, D. decaryi, D. madagascariensis and D. mandrarensis, all sensu Cavaco, 1966) to Payera [P. bakeriana, P. decaryi, P. madagascariensis and P. mandrarensis (Homolle ex Cavaco) R. Buchner & Puff], and this taxonomic decision is strongly supported by our results (Fig. 4). Danais mandrarensis was not investigated in the present study but we expect it to belong to Payera, as it has some of the salient features of the genus (e.g. valvate-reduplicate corolla aestivation and pubescent corollas). The mutual paraphyly of Payera and Schismatoclada suggested by Krüger et al. (2012) is mainly explained by misidentifications of many specimens of Payera and Schismatoclada in that study.
The inclusion of the Malagasy monotypic genus Coursiana, represented by C. homolleana, in Payera (i.e. Payera homolleana) proposed by Buchner and Puff (1993) receives strong support from the results of the present study (Fig. 4). On the other hand, Schismatoclada as delimited by Humbert (1955) and Cavaco (1964), both of which included the two low-growing shrubby species Schismatoclada beondrokensis (now Payera beondrokensis) and Schismatoclada coriacea (now Payera coriacea), is not supported by our analyses. These latter species are shown here to belong to Payera, consistent with Buchner and Puff (1993) and Krüger et al. (2012), and are resolved in the Payera northeastern clade (Fig. 4).

Phylogenetic relationships among the Danaideae genera
While all three genera of Danaideae are, for the first time, strongly supported as monophyletic in our study, their interrelationships remain unresolved. The sister-relationship between Payera and Schismatoclada received poor support (BPP 0.72) in the relaxed-clock analysis (Fig. 3), and the non-clock analysis instead provided weak support (BPP 0.61) for a sister-relationship between Danais and Schismatoclada. The difficulty in resolving the generic relationships in Danaideae with molecular data goes in concert with the conflicting generic relationships indicated by morphology (e.g. Buchner and Puff, 1993;Schatz, 2001). For example, the arborescent habit of Payera and Schismatoclada (as opposed to lianescent habit in Danais) seems to suggest their close affinity. By contrast, the capsular fruits of Danais and Payera have loculicidal dehiscence, whereas those of Schismatoclada bear septicidal dehiscence (Buchner and Puff, 1993). Furthermore, Danais and Schismatoclada appear more closely related to each other than either is to Payera based on the absence of the large bracts subtending the inflorescence (vs. mostly present in Payera), valvate-reduplicate corolla aestivation (vs. valvate in Payera) and glabrous corolla (vs. pubescent or puberulous in Payera).
Phylogenetic relationships in Payera. Our analyses resolved Payera in four well-supported monophyletic groups (Figs 3B  and 4). The first lineage to diverge, the Payera southeastern-tip clade, contains at least three morphologically distinct species of shrubs (P. decaryi, P. homolleana and P. madagascariensis) (Fig. 4) that grow sympatrically in the lowland rainforests of the Anosy Region (Fig. 3A: Area 17). Payera decaryi can easily be distinguished from the other two by its very large stipules and leaves, long-pedunculate, pendulous, racemose inflorescences, and flowers with white corolla tubes but orange corolla lobes. Both P. homolleana and P. madagascariensis have globose inflorescences but the former has relatively short peduncles, while the latter has pendulous, long peduncles up to 10 cm long. Payera decaryi and P. madagascariensis are endemic to this region, while P. homolleana sensu Buchner and Puff (1993) is known from the Anosy Region and the Alaotra-Mangoro Region (Fig. 3A: Area 8). While the samples of P. homolleana included in this study are from the Anosy Region, the type specimens of the species are from the Ambatondrazaka District in the Alaotra-Mangoro Region. Buchner and Puff (1993) treated the populations of P. homolleana from the Anosy and Alaotra-Mangoro Regions as the same species mainly because of their densely pubescent habits, but we do not necessarily expect them to form a monophyletic group. They can easily be distinguished based on their inflorescence types and the colour of their corollas (purple vs. white), and that they are geographically distinct. If future work based on molecular data confirms the populations of the Anosy and Alaotra-Mangoro Regions as distantly related, P. homolleana from the southeast tip would have to be described as a new species.
The next lineage to diverge, the Payera southeastern clade (Figs 3B and 4), encompasses at least three morphologically distinct species of shrubs (P. 'vondrozoensis' sister to P. 'vangaindranoensis' and P. 'bemangidyensis') (Fig. 4). The three species are newly proposed, have yet to be described and appear not to be sympatric. Payera bemangidyensis grows together with P. homolleana and P. madagascariensis in the Anosy Region ( Fig. 3A: Area 17), whereas P. 'vangaindranoensis' and P. 'vondrozoensis' are known only from separate districts within the Atsimo Atsinanana Region (Fig. 3A: Area 16) in southeastern Madagascar. We suspect that a fourth undescribed species (P. 'ambodivohitrensis' ined.), only known from the summit of the Ambodivohitra mountain (1200-1400 m) of the Vohibato District within the Matsiatra-Ambony Region (Fig.  3A: Area 14) belongs in the Payera southeastern clade. This new species appears to be closely related to P. 'bemangidyensis' and P. 'vangaindranoensis' based on its narrowly lanceolate, membranous leaves, but differs from the other two by having much smaller leaves, fimbriate stipules and subsessile head-like inflorescences bearing orange corollas. Payera 'bemangidyensis' has bifid stipules and rather lax inflorescences/infructescences with pendulous fruits, and its leaf size is intermediate between that of P. 'ambohivohitrensis' and P. 'vangaindranoensis'. Payera 'vangaindranoensis' is known from the Manombo Reserve (Farafangana District) and the Vohipaho forest (Vaingaindrano District) within the Atsimo-Atsinanana Region ( Fig. 3A: Area 16). This species differs from the other species in the Payera southeastern clade by its deeply fimbriate stipules and globose inflorescences bearing orange corollas. Payera 'vondrozoensis' is known from a single collection (De Block et al. 1979, BR) from the Vondrozo District in the Atsimo Atsinanana Region (Fig. 3A: Area 16). It is morphologically similar to P. homolleana but differs by its large stipules with laciniate margins and subsessile flowers and fruits (as opposed to stipules with laciniate margins and long-pedicellate flowers and fruits in P. homolleana).
The Payera centraleastern clade (Figs 3B and 4) contains at least two morphologically distinct species, P. bakeriana and P. glabrifolia, with adjacent geographical ranges. These two sister species can easily be distinguished morphologically; P. bakeriana has narrowly lanceolate leaves, small stipules and lax inflorescences, while P. glabrifolia has broadly lanceolate, glabrous leaves that are large with laciniate margins, and condensed inflorescences. Payera bakeriana has a wide distribution in the highland rainforests, ranging from the Andringitra Massif (Matsiatra-Ambony Region, Fig. 3A: Area 14), through the Ranomafana National Park (Ifanadiana District, Vatovavy Fitovinany Region, Fig. 3A: Area 13), to the Moramanga and Ambatondrazaka Districts (Mangoro-Alaotra Region, Fig. 3A: Area 8). Many remote areas of Madagascar have not yet been thoroughly searched by botanists, and we expect the species to also occur in the Marolambo and Anosibe An'ala Districts (Atsinanana Region, Fig. 3A: Area 9), which lie between Ifanadiana and the Moramanga Districts in the central east of Madagascar. Payera glabrifolia is known from the lowland rainforests in the Brickaville, Toamasina II and Vatomandry Districts (Atsinanana Region, Area 9), through the Ambatondrazaka District (Alaotra-Mangoro Region, Fig. 3A: Area 8) to the Maroantsetra District, and we expect it to occur also in the Soanerana-Ivongo and Mananara-Avaratra Districts (Analanjirofo Region, Fig. 3A: Area 4) although it has not yet been discovered from that area.
The Payera northeastern clade (Figs 3B and 4) encompasses at least three species of ericoid shrubs (P. coriacea, sister to P. beondrokensis, and P. marojejyensis) (Fig. 4). These species grow exclusively and sympatrically in the ericoid thickets near the summit of the Marojejy Massif in northeastern Madagascar (Figs 1 and 3A: Area 2), and are morphologically distinct. While they are all low-growing shrubs up to 50 cm tall, P. marojejyensis has dark purple flowers as opposed to yellow in the other two species. Payera coriacea has large, persistent, appressed stipules and glabrous leaves, whereas P. beondrokensis bears small stipules and pubescent leaves.
Phylogenetic relationships in Schismatoclada. Our analyses resolve Schismatoclada in five major lineages (Figs 3B and 5). The Schismatoclada southeastern-tip clade, sister to the remainder of the genus, comprises at least five species (S. 'bemangidyensis', S. citrifolia, S. 'papangoensis', S. mandrarensis and S. 'pseudopurpurea', Fig. 5), which are restricted to the Fort-Dauphin District within the Anosy Region ( Fig. 3A: Area 17). Based on their geographical proximity to these five species, we postulate that S. rupestris var. rupestris and S. rubra belong to this lineage as well (see below). Cavaco (1967) merged S. rubra with S. citrifolia and Buchner and Puff (1993) endorsed this taxonomic adjustment, but we disagree with this assertion based on both morphology and geography (see below), although molecular data from S. rubra are currently unavailable to us to confirm this. Schismatoclada citrifolia is confined to the sandy littoral forests, and it also frequently grows along or in the swampy areas north of Fort Dauphin town (e.g. Ampangalatsilo, Etazo, Mandena, Saint Luce). Schismatoclada citrifolia is similar to S. rupestris (i.e. S. rupestris var. rupestris) in its verticillate leaves, cymose inflorescences and mauve corollas. However, the latter is distinct by having smaller, coriaceous leaves with revolute margins, connate stipules with triangular appendages and foliaceous but shorter calyx lobes (as opposed to larger non-coriaceous leaves with non-revolute margins, triangular stipules and foliaceous, longer calyx lobes in the latter). Furthermore, S. rupestris is a mountain species (1300-1850 m in altitude) known from the Ivakoany and Kalambatitra Massifs of the Anosy mountain chains, and its geographical ranges are quite distant from that of S. citrifolia. We were unfortunately unable to sample representatives of S. rupestris var. rupestris; however, based on the evidence shown in this study, we expect it to belong to the Schismatoclada southeastern-tip clade (and thus not to group with S. rupestris var. brevicalyx resolved in the Schismatoclada northern clade, see further below). Schismatoclada 'bemangidyensis', strongly supported as sister to the remaining four species in the clade, is a new undescribed species known only from two collections in fruit from the lowland rainforest of Bemangidy in the northern part of the Tsitongabarika new protected area. It can be distinguished by its glossy, broad, glabrous obovate leaves and glossy young, beaked, elongated fruits ornamented by small triangle calyx lobes.
Schismatoclada 'papangoensis' is restricted to the ericoid thicket at the summit of the Papanga Mont, one of the satellites of the Beampingaratra Massif of the Anosy mountain chains. It is distinct from the other members of the southeastern-tip clade by its relatively small elliptical leaves. This species, yet to be described, is only known from a single specimen collected by Humbert in 1928, and has not been recollected since. Schismatoclada mandrarensis was collected from the same locality, but the species has a wider range and is known from more collections than S. 'papangoensis'. It is worth noting that Cavaco (1964) considered S. mandrarensis and S. aurantiaca to be conspecifics, and S. mandrarensis is currently not an accepted species name. Schismatoclada aurantiaca is, as defined by Homolle (1939), restricted to the Andringitra Massif (Matsiatra-Ambony Region, Fig. 3A: Area 14) and the Ivohibe Massif (Atsimo-Atsinanana Region, Fig. 3A: Area 16). It differs from S. mandrarensis by its membranaceous leaves and relatively large paniculate inflorescences (Homolle, 1939), as opposed to coriaceous leaves and smaller paniculate inflorescences in S. mandrarensis, and we have therefore (provisionally) treated the two as distinct species. We expect S. aurantiaca to belong in the Schismatoclada central clade given that its geographical range overlaps with that of S. aurea. However, no sample of S. aurantiaca from its type locality (Andringitra Massif) was analysed in this study, and whether these two species are conspecifics remains to be tested with molecular data. Finally, S. 'pseudopurpurea' is known only from two specimens from the summit of the Ambatofotsy Mont within the Andohahela Massif in the Anosy Region ( Fig. 3A: Area 17). The leaf shape and venation of this species are similar to those of S. purpurea, but S. purpurea is restricted to the Tsaratanana and Marojejy Massifs (Fig. 3A: Areas 1 and 2) in northern Madagascar and belongs in the Schismatoclada northeastern clade (Figs 3B and 5).
The Schismatoclada central clade (Figs 3B and 5) is the second clade to diverge, and it encompasses at least two species (S. 'andringitrensis' and S. aurea) (Fig. 5). The former species is known from two specimens from the highland rainforests of both the Andringitra and Ambondrombe Massifs (Ambalavao District) within the Matsiatra-Ambony Region (Fig. 3A: Area  14). By contrast, the latter species is restricted to high elevations of the Andringitra Massif and the Pic Ivohibe National Park (Vondrozo District, Atsimo Atsinanana Region, Fig. 3A: Area 16). Schismatoclada aurea has red stems, yellow-orange corollas (or whitish tubes and orange lobes) and globose fruits, as has S. homolleae. Based on morphology, we argue that S. homolleae and S. aurea are conspecifics. Schismatoclada 'andringitrensis' is markedly distinct among species of the central clade by its winged stems, small, spathulate leaves, and its long-pedunculate, elongate fruits with longitudinal ridges. Its flowers are unknown. Schismatoclada pubescens Cavaco is distinct by its pubescent stems, leaves and corollas. It was not sampled for the present study; it is confined to the lowland swampy forests in the Vondrozo District (Atsimo-Atsinanana Region, Fig. 3A: Area 16), and is known only from three specimens. However, based on the general phylogenetic patterns of Schismatoclada as revealed by our results, we postulate that S. pubescens belongs in the Schismatoclada central clade.
The Schismatoclada south-centraleastern clade (Figs 3B and 5), the next clade to diverge, contains at least three species (S. bracteata and the new undescribed species S. 'andohahelensis' and S. 'ialatsarensis';Fig. 5). Schismatoclada bracteata is known to occur in the Andringitra Massif (District Ambalavao, Matsiatra-Ambony Region, Fig. 3A: Area 14), the Ranomafana National Park (Ifanadiana District, Vatovinany-Fitovavy Region, Fig. 3A: Area 13), whereas S. 'andohahelensis' is known only from two recent collections from the north of the Tsitongabarika new protected area (Anosy Region, Fig. 3A: Area 17). Similarly, S. ialatsarensis is only known from two recent collections from the Ialatsara forest of the Ambohimahasoa District (Matsiatra-Ambony Region). The placement of S. 'andohahelensis' in this clade is strongly supported by molecular data but deviates from the geographical patterns of the major lineages of Schismatoclada, as shown in this study.
The Schismatoclada northern clade of the core Schismatoclada (Figs 3B and 5) is a species-rich clade formed by three geographically segregated subclades: clade 137 distributed in Areas 2 and 8; clade 115 distributed in Areas 1 and 2; and clade 117 distributed in Area 2 (Fig. 3A). Clades 137 and 115 are successive sisters to clade 117, which in turn is divided into clades 129 and 118, both distributed in Area 2 (Fig. 3A). The Schismatoclada clade 137 contains at least three species (S. 'didyensis' sister to S. 'anjanaharibeensis' ined. + S. rupestris var. brevicalyx) (Fig. 5). Based on morphology, we further expect S. lutea and the undescribed new species S. 'makira' ined. from the eastern Sofia Region (Fig.  3A: Area 3) and the Analanjirofo Region ( Fig. 3A: Area 4) to belong to this subclade. The clade (137) has a disjunct distribution; S. 'anjanaharibeensis' is from the Anjanaharibe-Sud and Marojejy Massifs in the Sava Region ( Fig. 3A: Area 2) in northeastern Madagascar, whereas S. 'didyensis', distinguished by its relatively small leaves, sessile head-like inflorescences and white slightly zygomorphic corolla tubes, is known only from the Didy forest of the Ambatondrazaka District within the Alaotra-Mangoro Region (Fig. 3A: Area 8). All other species of clade 137 have much larger leaves, and loose, sessile inflorescences. Schismatoclada rupestris var. brevicalyx was described by Cavaco (1967) (transferred at the varietal level from S. rubra var. brevicalyx Humbert), as an endemic to the Marojejy Massif in northeastern Madagascar (Sava Region, Fig. 3A: Area 2). The decision was seemingly justified based on morphological resemblance with S. rupestris (e.g. broadly elliptical leaves, short calyx lobes) even though it made S. rupestris the sole species of Schismatoclada with a disjunct distribution (since the type specimens of S. rupestris var. rupestris are all from the Beampiangatra Massif). However, while S. rupestris var. brevicalyx is resolved in the Schismatoclada northern clade (clade 137), based on geographical proximity we anticipate that S. rupestris var. rupestris belongs in the Schismatoclada southeastern-tip clade, and we consequently argue that S. rupestris var. brevicalyx should be recognized at the species level: S. 'brevicalyx' ined.
The Schismatoclada northern clade 115 contains at least two undescribed species: S. 'binara', from the Binara forest within the Daraina Commune of the Vohemar District (Sava Region, Fig. 3A: Area 2), and S. 'galokoensis', from the Galoko Massif in the Ambilobe District (Diana Region, Fig. 3A: Area 1). The geographical ranges of these species correspond to the northern and northwestern limits of Schismatoclada. It is important to note that the geographical range of S. 'binara' in clade 115 is quite far from that of the species endemic to the Marojejy Massif in the northeastern clade 117, which occur within the Andapa District (Sava Region, Fig. 3A: Area 2). Schismatoclada 'binara' and S. 'galokoensis', both in clade 115, are distinct based on their leaf size and shape. Schismatoclada 'binara' has relatively small, spathulate leaves and long, white corollas, while S. galokoensis has medium-sized, elliptical leaves and short white corollas. We expect S. 'manongarivoensis' ined.  'avaratra',S. spathulata,S. 'megastipula',S. 'ravelonarivoi' and S. 'violacea' (Fig. 5). The three latter species form a wellsupported monophyletic group that is distinct based on their head-like inflorescences. The Schismatoclada northeastern clade 118 includes five morphologically distinct species, S. marojejyensis, sister to S. longistipula, S. 'gracilis', S. purpurea and S. 'rotundifolia' (Fig. 5). Both S. marojejyensis and S. 'gracilis' have small leaves, but the former differs by being a low-growing shrub, its verticillate, coriaceous, subsessile leaves, persistent stipules, and very long, solitary and pendulous, light-yellow flowers, as opposed to opposite, membranaceous, petiolate leaves and cream flowers arranged in lax, pedunculate inflorescences. Schismatoclada longistipula is distinct based on its large leaves and fimbriate, persistent stipules, large paniculate inflorescences and small obovoid fruits. Schismatoclada purpurea and S. 'rotundifolia' have ovoid and coriaceous leaves.
The fifth clade, the Schismatoclada eastern clade of the core Schismatoclada (Figs 3B and 5), includes samples from six currently accepted and newly proposed species: S. farahimpensis, S. 'ranomafanensis ',S. concinna,S. 'ambalabeensis',S. 'keraudreniana' and S. psychotrioides (Fig. 5). All samples of S. farahimpensis form a monophyletic group (clade 108, Fig. 5) even though this species is widely distributed. It is the only widespread species of Schismatoclada with a geographical distribution ranging from the southeast tip (Beampiangatra Massifs, Anosy Region, Fig. 3A: Area 17) to northeastern (Marojejy and Anjanaharibe Sud Massifs, Sava Region, Fig. 3A: Area 2) and northwestern regions (Manongarivo Massif, Diana Region, Fig. 3A: Area 1). It is recognized by its membranaceous leaves and persistent triangular stipules covered by glands on its margins and yellow-orange corollas, and our results support two geographically distinct groups of S. farahimpensis, the northeastern group (Fig. 3A: Area 2) and the south-central eastern group (Fig. 3A: Areas 8 and 16). Schismatoclada psychotrioides is recorded from the Andramasina, Anjozorobe and Ankazobe Districts (Analamanga Region, Fig. 3A Fig. 3A: Area 9) and Ambatolampy District (northern Vakinankaratra Region, Fig. 3A: Area 11). This species is recognized by a combination of the following characters: elliptical leaves with reticulate venation, small triangular, paniculate inflorescences with each axis subtended by a pair of leaf-like bracts, flowers with enlarged cream corolla tubes that are tinged violet outside of the corolla lobes, lanceolate calyx lobes, fruits beaked and capsular, and seeds broadly winged at both ends but narrowly winged along both lateral sides. The monophyly of each of the remaining species of the Schismatoclada eastern clade (S. 'ranomafanensis', S. concinna, S. 'ambalabeensis', S. 'keraudreniana') is not supported in our results and their respective delimitations need further research.
Phylogenetic relationships in Danais. Danais sensu Puff and Buchner (1994) is resolved into four well-supported lineages (Figs 3B and 6): the northwestern Malagasy Danais 'nigra' group; the Malagasy Danais 'coronata' group; the Mascarene Danais 'fragrans' group; and the Afro-Malagasy-Comorian Danais 'microcarpa' group, which is centred on Madagascar but expands its geographical ranges to Tanzania (D. xanthorrhoea) and the Comores (D. comorensis). Also, the Danais 'coronata' group has a wide geographical range with many of its species widely distributed.
Danais nigra as defined in our study is resolved as sister to a very large clade formed by the remaining sampled Danais (core Danais, Figs 3B and 6). This is consistent with Krüger et al. (2012), although specimen Danais nigra with labID dc25 (voucher Kårehed et al. 254) included in both studies was not identified at the species level in Krüger et al. (2012) but was referred to as Danais sp. 1 in their work. According to Puff and Buchner (1994), Danais nigra has a disjunct distribution, the Sambirano area (including Manongarivo and Galoko-Kalobenono Massifs) within the Diana Region ( Fig. 3A: Area 1) in northwestern Madagascar and the Masoala National Park and the neighbouring forests, all within Analanjirofo Region ( Fig. 3A: Area 4) in the central-northeast of the island. We disagree with this assertion and argue that D. nigra is restricted and endemic to the lowlands between the Ambanja and Ambilobe Districts and around the Galoko-Kalabenono Massif within the Diana Region ( Fig. 3A: Area 1), because the type specimens (lectotype and isolectotypes) are from there. Danais nigra as delimited here is distinct from the so-called D. nigra plants from the central-northeastern region (Fig. 3A: Area 4) by having very lax inflorescences and flowers with very long pedicels and erect corolla lobes as opposed to much denser inflorescences and flowers with much shorter pedicels and recurved corolla lobes in the latter plants.
The Danais Mascarene clade was poorly supported (BPP 0.84) as sister to the Danais 'coronata' group in Krüger et al. (2012), while it is strongly supported as sister to the Danais 'microcarpa' group in the present study. The Danais 'microcarpa' group as defined by Puff and Buchner (1994) includes D. breviflora Baker, D. ligustrifolia, D. microcarpa, D. rhamnifolia and D. verticillata Baker, and is characterized by their terminal inflorescences, small flowers, fruits and seeds. This group is not monophyletic in our results; many species with large flowers, fruits and seeds are nested in this lineage and D. ligustrifolia is instead included in the Danais 'coronata' group (Fig. 6). Buchner and Puff (1993) also postulated D. brickavillensis, D. humblotii, D. longipedunculata and D. rubra to be closely related, but this is not supported by our results, as D. brickavillensis and D. longipedunculata fall in the Danais 'microcarpa' group whereas D. humblotii is included in the D. 'coronata' group (Fig. 6). Furthermore, Buchner and Puff (1993: 53) postulated close affinities between D. coronata, D. sulcata and D. volubilis Baker on the basis of their axillary inflorescences, much elongated calyx lobes and rather large fruits. We find no support for this group either, because D. sulcata and D. volubilis are nested in the Danais 'fragrans' and Danais 'microcarpa' groups, respectively (Fig. 6).

Species delimitation in Danaideae genera
Taxonomists identify morphologically diagnosable units using presumably autapomorphic characters. From a phylogenetic standpoint, these morphologically diagnosable units (also known as morphospecies) are viewed as species hypotheses, which can be tested using analyses of other sources of data (e.g. molecular data). The result may reveal members of these hypothesized morphospecies as either monophyletic or nonmonophyletic units (Rosell et al., 2010). When the result is the latter, non-monophyletic units, it is likely that the presumably autapomorphic morphological characters that were considered diagnostic of a species in reality comprised a combination of autapomorphic, synapomorphic and pleisiomorphic characters (e.g. Razafimandimbison, 2002), meaning that the included specimens represent an assemblage of non-related individuals that would be better understood as several different species. Our taxon sampling allows us to test the monophyly of many species of Danais, Payera and Schismatoclada, both currently accepted and new species proposed here.
We find that 42 species of Danaideae (including one currently described as a variety) are supported as monophyletic (see specification in the result section): seven species of Payera, 19 species and one variety of Schismatoclada, and 16 species of Danais (Figs 4-6). Of these 42, 15 species are yet to be formally described. By contrast, S. concinna and S. 'ranomafanensis' are polyphyletic, and possibly S. 'keraudreniana' too although the results are poorly supported. The morphologically variable S. concinna as defined by Baker (1885) may well be considered a 'morphospecies', members of which do not descend from a common ancestor. The non-monophyly of S. 'keraudreniana' is surprising given that this species is notably distinct from the other species in the same clade by its quadrangular, reddish stems, small spathulate leaves with inconspicuous secondary and tertiary veins, and reduced and sessile inflorescences with one to two flowers with white-greenish corolla tubes and white corolla lobes. Regarding Danais, D. cernua, D. hispida, D. lyallii and D. ligustrifolia seem to be non-monophyletic, and possibly D. pubescens and D. aurantiaca too although the results are poorly resolved (Fig. 6). The delimitations of all these taxa are probably in need of revision.

Geographically driven evolution of Danaideae
Our results show that there is a strong association between geography and evolutionary relationships in Danaideae. In all three genera, we recover many geographically segregated lineages. Species from the same geographical area tend to be more closely related to each other than they are to species from other more distant areas. The predominantly (sub)montane habitats of most Danaideae seem to be the driving factor, explaining the importance of geographical isolation in their evolution and the resulting patterns of endemism. Furthermore, the observed pattern also appears to fit the general pattern that rare, localized species (or narrowly endemic species) generally maintain the geographical signal of their diversification (e.g. Abellán and Ribera, 2017). Almost all species of Payera and Schismatoclada are narrow endemics (Buchner and Puff, 1993;this study). This geographical phylogenetic clustering indicates that the species of Danaideae seem to be limited in their dispersal ability, which is surprising given that all species of Danaideae produce capsular fruits and wind-dispersed seeds. In other words, their winged seeds may not be easily dispersed by wind compared to winged seeds of plants growing in open habitats. On the other hand, the restricted distributions of most Danaideae species might partly be related to limitations of their seed germination and/or seedling growth and survival in different environmental conditions.
Evidence of a southeast-tip origin and parallel northward diversification of Payera and Schismatoclada. Despite their disparity in species richness, Payera and Schismatoclada display an apparent congruence in geographical distribution and diversification patterns, with the southeastern-tip clade and south(central)eastern clades forming basal grades in both genera (Fig. 3). Interpreting the phylogenetic pattern, it is possible that the lowland rainforests and littoral forests of the southeast tip of Madagascar is the area of origin of both Payera and Schismatoclada, with a putative northward diversification. Few previous studies have been able to disclose potential geographical areas of origin and species diversification on Madagascar more precisely, in particular for plants, but this conclusion stands in contrast to the suggested northern origin of some Malagasy animal groups (e.g. Boumans et al., 2007;Gehring et al., 2013). It is also unexpected given that the southeast tip of Madagascar is the most isolated part of the island. Future biogeographical work may provide clarity.

CONCLUSIONS
The monophyly of the tribe Danaideae and its three genera (Danais, Payera and Schismatoclada) is supported. Numerous geographically separated but morphologically heterogeneous lineages and a high level of previously undocumented morphological and species diversity of Danaideae on Madagascar are revealed. This will result in two-fold and one-third increases of Schismatoclada and Payera species, respectively. Geographical proximity (i.e. geography) is in general better than morphologybased taxonomy in predicting evolutionary history (i.e. phylogeny) in Danaideae.
Systematists may often prefer to conduct molecular phylogenetic studies on groups of organisms that have been revised, taxonomically. Here, we have adopted a novel approach by performing rigorous molecular phylogenetic analyses of Danaideae based on a relatively dense taxon sampling, as a precursor to taxonomic revisions of its genera (Payera and Schismatoclada), classically achieved using morphological similarities and/or dissimilarities to delineate taxa. This approach enables integration of molecular and morphological evidence more efficiently during the process of taxonomic revisions. It allows us to test the monophyly of as yet undescribed taxa proposed based on morphology, using molecular data, and reveals their phylogenetic placements prior to the publication of the revisions. Such crucial information can thus be a part of the original decision of whether a species should be described as new. Furthermore, the molecular evidence for the monophyly of rare, new, undescribed species is a strong justification for formally describing such species even based on incomplete material. Finally, our approach allows systematists to formulate a more accurate species diagnosis, traditionally based solely on morphological similarities.

SUPPLEMENTARY DATA
Supplementary data are available online at https://academic. oup.com/aob and consist of the following. Figure S1: Bayesian majority-rule consensus trees based on non-clock analyses of single gene regions. Table S1: List of Danaideae species accepted in the present study, and provisionally proposed new species. Table S2: List of investigated plant material, voucher information, country origins and GenBank accession numbers. Table S3: Results from the individual relaxed-clock analyses, and from combining these distributions into one. File S1: The combined data matrix in nexus format.