Sampling

Previous phylogenetic studies on Pandanaceae included a limited sampling of Benstonea: nine spp. in Buerki et al. (2012) and three spp. in Gallaher et al. (2015). We greatly expanded the sampling of Benstonea by including 54 samples representing 36 species (60% of the species generic diversity). To assess the monophyly of Benstonea, representatives of all the other genera of Pandanaceae were included following results of Buerki et al. (2012) and Gallaher et al. (2015). Barbacenia elegans Pax (Velloziaceae) was used as the most external outgroup taxon to root the phylogenetic analyses (Buerki et al., 2012).Two other representatives of Velloziaceae were also included to serve as additional outgroup taxa. The full list of the species sampled, voucher information and DNA GenBank accession numbers is provided in Appendix 1.

DNA extraction, amplification and sequencing

Genomic DNA was extracted from both silica-gel dried and herbarium leaf material. Extractions of total DNA were performed using the same modified 2 × CTAB method as in our previous study on Pandanaceae (Buerki et al., 2012) and samples were stored at the Royal Botanic Gardens, Kew's DNA bank [ http://apps.kew.org/dnabank/homepage.html].

Phylogenetic relationships within Benstonea were reconstructed using six plastid DNA regions, of which one is coding (matK) and five are intergenic spacers (atpB-rbcL, trnQ-5'-rps16, trnL-trnF, trnV-ndhC, ndhF-rpl32). Information on primers and PCR protocols for matK, trnQ-5'-rps16 and trnL-trnF are described in Buerki et al. (2012). This latter information for atpB-rbcL, trnV-ndhC and ndhF-rpl32 is available respectively in Shaw et al. (2007), Schnitzler et al. (2011) and Forest et al. (2014). All PCR products were purified using DNA purification columns according to the manufacturers' protocols (QIAquick; Qiagen Ltd, Crawley, UK.). Dideoxy cycle sequencing was then performed using the chain termination method and ABI Prism Big Dye v. 3.1 reaction kit, following the manufacturer's protocols, but using 0.5µl of reaction mix (Applied Biosystems Inc., Warrington, UK).The products were prepared for sequencing using the ethanol-precipitation method and visualised on an ABI 3730 DNA Analyzer, also according to the manufacturer's protocols.

Alignment and phylogenetic analyses

The program Geneious v. 8.1.3 (Biomatters, Aukland, New Zealand) was used to assemble complementary strands, verify software base-calling and produce the alignments (using MUSCLE; Edgar, 2004). Single-gene and partitioned phylogenetic inferences were carried out employing both maximum likelihood (ML) and Bayesian Markov chain Monte Carlo (MCMC) analyses. In the case of the partitioned analyses, the dataset was divided into six partitions and each locus was allowed to have partition-specific model parameters. The phylogenetic analyses were done using the facilities offered by the CIPRES portal in San-Diego, USA [ http://www.phylo.org].

The ML analyses were performed using RAxML v. 8.1.11 (Stamatakis, 2006; Stamatakis et al., 2008) with a 1,000 rapid bootstrap analysis followed by the search of the best-scoring ML tree in one single run. The default model, GTRCAT, was used to perform the ML analyses. The Bayesian MCMC analyses were performed in MrBayes v. 3.2 (Ronquist et al., 2012) and the best-fit model for each DNA region was estimated using MrModeltest v. 2.3 (Nylander, 2004) and the Akaike Information criterion (see Table 1 for best-fit models). Three Metropolis-coupled Markov chains with an incremental heating temperature of 0.2 were run for 10 hours on the CIPRES portal (yielding 14,821,000 generations for the partitioned analysis) and sampled every 1,000^th generation. Each analysis was repeated twice starting with random trees. The MCMC sampling was considered sufficient when the effective sampling size (ESS) was higher than 200, as verified in Tracer v. 1.4 (Rambaut & Drummond, 2007). After a burn-in period of 25% per run, the remaining trees were used to construct a majority-rule consensus from MrBayes (half-compatible maximum credibility tree in BEAST) and its associated Bayesian posterior probabilities (BPP).

Divergence time estimations

Divergence time estimates were obtained using the Bayesian inference approach implemented in the package BEAST v. 1.8.0 (Drummond & Rambaut, 2007), applying the same partition delimitation and evolutionary models as those used for the MrBayes analysis. We used an uncorrelated relaxed molecular clock with a lognormal distribution of rates and a Yule speciation model. The analysis was run twice on the CIPRES portal for 10 million generations, sampling one tree every 1,000^th generation. Parameter convergence was confirmed following the same approach as in the MrBayes analysis (see above). Following a burn-in period of one million generations, a maximum clade credibility tree with median branch lengths and 95% highest posterior density (HPD) interval on nodes was reconstructed using TreeAnnotator v. 1.8.0 (Drummond & Rambaut, 2007). HPD was only inferred for nodes with BPPs ≥ 0.5.

The calibration of the phylogenetic tree to obtained absolute age estimates was performed using four secondary calibration points obtained by Gallaher et al. (2015; see Table 1 in reference and values of 95% HPD obtained from Cyclanthus and Pandanaceae fossils). Uniform priors were set as follows (in million years): a) Crown of Pandanaceae (upper: 97.5, lower: 41.9) and splits between b) Freycinetia and clade comprising Martellidendron, Pandanus and Benstonea (upper: 60.7, lower: 24.6), c) Martellidendron and clade comprising Pandanus and Benstonea (upper: 38.1, lower: 15.1) and d) Pandanus and Benstonea (upper: 33.0, lower: 12.5).

Biogeographical inference

Geographical areas were defined based on the current distribution of Benstonea species and paleogeological data (Hall, 2009). We recognized five areas: A. India and Indochina (the limit of this area is set at the Isthmus of Kra; Parnell, 2013), B. Sunda shelf (including Peninsular Malaysia, Sumatra and Java, but excluding the Philippines, which are here included in the Wallacea region), C. Borneo, D. Wallacea (including Sulawesi, the Philippines and the Moluccas Islands) and E. Sahul shelf (including New Guinea and northern Australia) and Pacific islands (here the Solomon Islands and Fiji).

Table 1.

Characteristics of the six DNA plastid regions used in the phylogenetic analyses of Pandanaceae.

The dispersal—extinction—cladogenesis (DEC) likelihood model implemented in Lagrange v. 2.0.1 (Ree et al., 2005; Ree & Smith, 2008) was used to investigate the biogeographical history of Benstonea (further details on this method are presented in Buerki et al., 2011). We followed the same approach as in Forest et al. (2014) and performed the Lagrange analysis on the BEAST maximum clade credibility tree excluding all taxa expect those belonging to Benstonea and pruning the dated tree at the species level (i.e. only one accession per species was kept).The maximum number of areas at nodes was constrained to two; however additional areas were included in the Lagrange analysis to account for the widespread species, B. affinis (Kurz) Callm. & Buerki, occurring in areas B, C and D (only species occurring in more than two areas). Ancestral area reconstructions for each node were plotted on the BEAST tree using pie charts and the biogeographical scenario was produced using a collection of R scripts following Buerki et al. (2013). This latter procedure (i.e. the type and frequency of transition events between ancestral and descendant nodes along the dated phylogenetic tree) was inferred according to the Q matrix implemented in the DEC model (Ree et al., 2005; Ree & Smith, 2008).

Phylogenetic analyses

The numbers of accessions per DNA region are provided in Table 1 together with various statistics.The ML and Bayesian MCMC single-partition analyses yielded congruent topologies, i.e. no incongruence with BPP > 0.95 or bootstrap support (BS) > 75% was found between the plastid regions, thus allowing performing a combined partitioned analysis. An identical situation was observed between the combined ML and Bayesian MCMC trees, where no supported incongruence was recognized. Only the halfcompatible maximum credibility tree from the combined Bayesian analysis is presented here (including BPP and BS values on nodes; Fig. 1).

Phylogenetic analyses support the monophyly of all the genera of Pandanaceae (Fig. 1). Within this framework, Sararanga (BPP: 1, BS: 100%) is inferred as the sister lineage to all other Pandanaceae with Freycinetia (BPP: 1, BS: 100%) and Martellidendron (BPP: 1, BS: 94%) as subsequent sister lineages to Benstonea + Pandanus (Fig. 1). Although moderately supported by the combined partitioned analyses (BPP: 0.95, BS: 74%), our phylogenetic inferences infer Pandanus (BPP: 1, BS: 100%) as sister clade to Benstonea (BPP: 1, BS: 85%) (Fig. 1).Three clades are retrieved within Benstonea: clade I (BPP: 1, BS: 85%), clade II (BPP: 0.98, BS: 88%) and clade III (BPP: 1, BS: 84%) (Fig. 1). Clade III is further divided into three moderately to strongly supported subclades to facilitate the discussion on the evolution and biogeography of the genus: clade IIIa (BPP: 1, BS: 99%), clade IIIb (BPP: 1, BS: 65%) and clade IIIc (BPP: 1, BS: 94%) (Fig. 1). Finally, the phylogenetic relationship between these subclades is only moderately supported (BPP: 0.97, BS: 79%).

Divergence time estimations and biogeographical inference

The BEAST maximum clade credibility tree of Pandanaceae is displayed on Fig. 2.The topology and node support are highly congruent with the MrBayes and RAxML analyses (Fig. 1). The temporal framework is in agreement with Gallaher et al. (2015).The biogeographical analysis suggested an origin of Benstonea on the Sunda shelf (areas B and C) sometime during the Miocene (Fig. 3, 4). Clade I remained on the Sunda shelf and two independent dispersals to area A (more specifically India, Burma and Sri Lanka) took place at the end of the Miocene (ca. 10 million years ago) (Fig. 3, 4). A vicariance event is inferred at the most recent common ancestor of clades II and III between Peninsular Malaysia and Borneo (Fig. 3). Clade II most likely originated on the Peninsular Malaysia (area B), with two subsequent dispersals to Borneo (area C) and another dispersal from Borneo back to Peninsular Malaysia (Fig. 3, 4). Clade III originated in Borneo and a dispersal event is inferred from this area to Wallacea (area D) at the most recent common ancestor of subclades IIIb and IIIc (Fig. 3, 4). A dispersal event into the Sahul shelf (New Guinea and Australia) from Wallacea (including the Philippines) is inferred at the origin of subclade IIIb. Within area E, the spacial origin of Fijian Benstonea remains unclear, but the most likely hypothesis suggests a dispersal during the Pleistocene from either Australia or New Guinea as shown by the close relationships between the Fijian endemic B. thurstonii (C.H. Wright) Callm. & Buerki and the Australia and New Guinean B. lauterbachii (K. Schum. & Warb.) Callm. & Buerki (Fig. 3). However this latter phylogenetic relationship is poorly supported (BPP<0.5) and has to be taken with caution. Subclade IIIc originated in Borneo and the two lineages underwent contrasting biogeographical histories (Fig. 3).The lower lineage (including the type species, B. affinis) dispersed three times towards Peninsular Malaysia and another time towards Wallacea (more precisely the Philippines in the case of B. affinis) from Borneo, whereas the upper lineage underwent sympatric speciation in Borneo (Fig. 3).

Systematics and evolution of Benstonea

Our phylogenetic analyses support the monophyly of Benstonea and suggest its close relationship with the largest genus of the family, Pandanus (Fig. 1).These results are in agreement with previous findings from Gallaher et al. (2015) based on very limited taxon sampling. Benstonea is subdivided into three clades exhibiting contrasting species diversities. Clades I and II have seven species each, whereas most of the species diversity occurs in clade III with 21 species (Fig. 1). Such uneven species diversity pattern is most likely not due to a bias in our taxon sampling since a taxonomic study on New Guinean species of Benstonea (all belonging to clade IIIb; Fig.1) indicated that this area is more speciose than previously expected (e.g. Callmander et al., 2014). None of the clades within Benstonea reflect the infrageneric classification proposed by Stone (1974, 1978, 1983) in Pandanus subg. Acrostigma. Species of the largest P. sect. Acrostigma (further subdivided into 15 subsections; see Stone, 1978) are scattered across the clades. Sampled species of P. sect. Pseudoacrostigma (two species, only B. platystigma was sampled) and P. sect. Epiphytica (one species, B. epiphytica sampled here) are nested within species of P. sect. Acrostigma in respectively clades I and II (Fig. 1). Pandanus sect. Fusiforma as defined by Stone (1968,1974) included six species. The same author later in his revision of the subgenus Acrostigma (Stone, 1978) only accepted four species, but stated difficulties in identifying clear morphological characters discriminating section Fusiforma from section Acrostigma subsect. Dismissistily (where Stone moved in 1978 other species previously placed in Fusiforma). Benstonea nana (Martelli) Callm. & Buerki, the accepted name of Pandanus magnifibrosus H. St.John (see Stone, 1978; Callmander et al., 2012), should be placed in sect. Fusiforma, a species sampled in our studies. Furthermore, another species accepted in this section by Stone (1978), Pandanus sobolfer B.C. Stone (currently accepted as Benstonea sobolfera (B.C. Stone) Callm. & Buerki) will be treated as a synonym of B. nana in the upcoming treatment of the family Pandanaceae for the “Flora of Peninsular Malaysia” (Beentje & Callmander, unpubl. data).The only species of section Fusiforma sampled here is also nested in section Acrostigma in clades II (Fig. 1).

Fig. 1.

Bayesian half-compatible consensus tree of Pandanaceae inferred from six plastid DNA regions. Bayesian posterior probabilities (BPP) and bootstrap support (BS) values are displayed at nodes. See main text for the discussion of the clades within Benstonea Callm. & Buerki.

Fig. 2.

BEAST maximum clade credibility tree of Pandanaceae. Bayesian posterior probabilities (BPP) and 95% highest posterior density (HPD) interval on nodes are displayed. See main text for more details on clades.

The polyphyly of the sections suggest that key morphological characters used by Stone for his infra-generic classification evolved multiple times independently. For instance, species with scaly surface pileus previously gathered in Pandanus subsect. Scabridi B.C. Stone belong to clade II (Benstonea kurzii (Merr.) Callm. & Buerki, B. atrocarpa) and clade III (B. gibbsiana (Martelli) Callm. & Buerki). Species of Pandanus subsection Ornati B.C. Stone, a group defined by its narrow and linear leaves with ventral pleats gradually attenuated with very small teeth and oblong or cylindrical syncarps (Stone, 1974: 525) are shared between clade I (B. ornata (Kurz) Callm. & Buerki) and clade IIIc (B. rustica (B.C. Stone) Callm. & Buerki).

On the other hand, it is interesting to note that several small understorey species distributed in lowland Bornean evergreen forests are restricted to clade IIIc, i.e. B. brevistylis (B.C. Stone) Callm. & Buerki, B. brunigii (B.C. Stone) Callm. & Buerki or B. rustica.The same is also true for another group with the same ecology in New Guinea, i.e. B. permicron (Kaneh.) Callm. & Buerki and B. rostellata (Merr. & L.M. Perry) comb. ined. in clade IIIb. This pattern suggests that species occurring in the same ecological conditions in Borneo and New Guinea radiated following similar processes and that most of the morphological characters used by Stone are homologous. Finally, the epiphyte habit evolved at least twice during the evolution of Benstonea in clades II and IIIc (Fig. 1). Phylogenetic evidence suggests that at least three facultative epiphytic species (clade II) are derived from acaulescent species (Fig. l).This shift of habit might have been key in enabling Benstonea species (especially B. thomissophylla (B.C. Stone) Callm. & Buerki) to colonize drier habitats on the Sunda shelf on limestone at low elevations (Callmander et al., 2012). Water supply on limestone is very limited due to high rates of drainage; however the epiphytic habit allows Benstonea species to trap water and debris into their leaves for long periods of time therefore maintaining a suitable growing environment (see Zona & Christenhusz, 2015 for a review).

Biogeographical history of Benstonea

Most islands in the Wallacea region were created from the Eocene-Oligocene boundary onwards (as a result of the collision between the Australian and Eurasian plates; Hall, 2009) with a peak of tectonic activities during the Miocene. A previous study on Sapindaceae showed that Wallacea acted as a hub connecting the Sunda and Sahul shelves and triggered the diversification of that group of plants (Buerki et al., 2013). A review published by Crayn et al. (2015) concluded that most plant lineages exhibiting the same distribution as Benstonea originated on the Sunda shelf during the Miocene and subsequently dispersed towards the Sahul shelf by using the islands in the Wallacea region as stepping-stones.

Our biogeographical analysis inferred an origin of Benstonea on the Sunda shelf during the Miocene followed by several (almost symmetrical) exchanges between Peninsular Malaysia and Borneo (Fig. 3, 4). The dispersals are not restricted to a specific clade and occurred throughout the phylogenetic tree of the genus (Fig. 3, 4). Only one lineage within subclade IIIc is restricted to Borneo and it apparently underwent significant diversification (seven species; Fig. 4). This endemic Bornean radiation occurred within the last five million years and might result from the isolation of Borneo from Peninsular Malaysia due to Quaternary sea level rise and/or orographic effects (see Cannon et al., 2009; Hall, 2009). Our analysis also inferred at least two northern dispersals in clade I from Peninsular Malaysia to Indochina and the Indian continent (Fig. 3, 4).

With the exception of B. affinis (belonging to clade IIIc, which colonized the Philippines from Borneo, most likely by taking advantage of low sea levels enabling species adapted to swamps to disperse across the Sunda shelf and parts of the Philippines; Cannon et al., 2009), all the species occurring in Wallacea, the Sahul and the Pacific islands are restricted to clade IIIb (Fig. 2, 3). Species of clade IIIb invaded the Wallacea region (including the Philippines) from Borneo during the Miocene (ca. 10 million years ago) and no back-dispersals were inferred (Fig. 3, 4). A study comparing the dispersal modes of Benstonea species would be required to test if the absence of back dispersals observed here is associated with adaptations to local dispersers. The fauna on each side of the Wallace line are significantly different (Huxley, 1868) and we could hypothesize that such difference might have played a role in shaping distribution patterns of Benstonea species.

Fig. 3.

Biogeographical scenario of Benstonea Callm. & Buerki inferred using the DEC model implemented in Lagrange and displayed on the BEAST maximum credibility clade tree. Abbreviations for the biogeographical areas: A. India and Indochina (the limit of this area is set at the Isthmus of Kra; Parnell, 2013), B. Sunda shelf (including Peninsular Malaysia, Sumatra and Java, but excluding the Philippines, which are here included in the Wallacea region), C. Borneo, D. Wallacea (including Sulawesi, the Philippines and the Mollucas Islands) and E. the Sahul shelf (including New Guinea and Northern Australia) and the Pacific islands (here the Solomon islands and Fiji). Please see figure 4 for more details on the geographical boundaries of the biogeographical areas. Other abbreviations: P: peripheral isolate; V: vicariance; D: dispersal event.

Fig. 4.

Biogeographical areas used for the Lagrange analysis of Benstonea Callm. & Buerki displayed on an elevation map retrieved from the WorldClim database [ http://www.worldclim.org/current]. Dispersal events inferred by the Lagrange analysis are also indicated. See legend of Fig. 3 for more details on biogeographical areas.