ABSTRACT
The yam genus Dioscorea comprises circa 650 species of tropical vines with starch rich tubers, usefull as an energy source and often containing secondary metabolites. The Neotropical Region holds the highest diversity of species and morphology of yams. We generated a time-calibrated tree for Dioscorea using, for the first time, a dense sampling of Neotropical species (64 sp., 20% of all Neotropical sp. and 22 sections) to trace the biogeography of these plants in this region. Four origins of Dioscorea in the neotropics were estimated since the Eocene. The two most diverse lineages originated between the Eocene and Oligocene, respectively in the Southern Andes and eastern South America. Both lineages occupied the South American ‘Dry Diagonal’ after the Miocene, but New World II clade remained associated with forest habitats. Several exchanges between Dry Diagonal and adjacent forested biomes occurred, corroborating the interchange between these vegetation types. Dispersals to Central America occurred before the closure of the Panama Isthmus. We highlight two important events of long distance dispersal, the colonization of Central American before the closure of Isthmus of Panama and the dispersal of D. antaly lineage to Madagascar. In addition, our phylogenetic tree evidenced the unnatural nature of the classical infrageneric classification of Dioscorea. The taxonomic implications of our results are also discussed.
INTRODUCTION
Dioscorea L. comprises approximately 95% of the known species of Dioscoreaceae (Govaerts, Wilkin, & Saunders, 2007), as a reflection of this larger number of species the genus possess a tremendous morphological diversity, as several critical traits to distinguish from the other genus of the family (eg.: hermaphroditism in the other genera and dioecism in Dioscorea), and a considerable richness of chemical and genetic characters. Most species of Dioscorea are known as yam (and variations in different languages: inhame, ñame, igname, niam, enyame, nyami, etc.). They are mostly dioecious vines, usually with small flowers and starch-rich tubers as the underground organ. Due to the great nutritional value of its underground system, they are an important source of energy for human populations worldwide (Wilkin et al., 2005). Dioscorea species have their history linked to humans for thousands of years by a slow and gradual process of domestication (Ayensu & Coursey, 1972). In addition, Dioscorea is economically important due to secondary metabolites present in the tubers; diosgenin, for example, has a molecular structure similar to steroidal hormones and has been used to synthesize steroids for the production of birth control pills, among other uses (Coursey, 1967). Yams have been studied by its secondary metabolites, however studies related on prospection of useful substances in tubers of Neotropical species are restricted to only a few species (eg.: D. composita, D. floribunda and D. mexicana) even with the great diversity of neotropical species (Mignouna, Abang, & Geeta, 2009; Coursey, 1967).
The family Dioscoreaceae currently includes 653 accepted species, distributed in four genera: Dioscorea (633 species), Stenomeris Planch. (2 species), Tacca J.R. Forst. & G. Forst. (17 species), and Trichopus Gaertn. (1 species) (Govaerts, Wilkin, & Saunders, 2007). A total of 1600 names are attributed to Dioscorea, among species, varieties and subspecies, mostly considered as synonyms (The Plant List, 2013). Dioscorea species occur mostly in tropical areas with some representatives in subtropical and temperate regions of the planet, but are especially diverse in the Neotropics, where about 50% of the species occur (e-Monocot team, 2017).
According to Viruel et al. (2016) in its study with 135 taxa and four plastid DNA markers, Dioscorea originated in the Laurasian Palearctic region between the late Cretaceous (57.7 – 85.9 mya) and the Mid Eocene (47.6 – 49.1 mya), with subsequent radiations to the Southern regions by long-distance dispersal or migration by land bridges in the Oligocene-Miocene (33.9 to 5.332 Mya) (Viruel et al., 2016). They occur in several Neotropical environments, from dry restinga at sea level to the Andean paramos, including edges and interior of humid forests, natural grassland ecosystems, rupicolous areas, and even semi-desertic environments (Dorr & Stergios, 2003; Couto et al., 2014). As a consequence of the great variety of environmental conditions in which they occur, Dioscorea species exhibit a wide range of ecological responses, evidenced by the large morphological variability found in the family, both in vegetative and reproductive organs. They range from large climbing vines (40m high) to dwarf species, monoecious or dioecious plants, and they can present impressively colored leaves and flowers, among other distinctive characters (Fig. 1).
Knuth (1924) proposed 58 sections and four subgenera for Dioscorea: D. subgenus Helmia (Kunth) Uline (seeds winged basally), D. subgenus Dioscorea (Pax) Uline (seeds winged all around), D. subgenus Stenophora (Uline) Knuth (seeds winged apically) and D. subgenus Testudinaria (Salisb.) Uline (seeds winged apically). This classification has undergone subsequent changes and additions (Burkill, 1960; Barroso et al., 1974), although these were not based on phylogenetic studies. With great morphological variation and scarce knowledge on anatomy, ecology, chemistry and palynology of Dioscorea (Ayensu, 1972; Caddick et al., 1998; Schols et al., 2001, 2003; Wilkin et al., 2009), external morphology was the base for the delimitation of taxonomic groups. Recent phylogenetic studies based on molecular data suggest eight main lineages distributed in the tropical and subtropical regions of the globe: ‘Enantiophyllum’, ‘Compound Leaved’, ‘Malagasy’, ‘Birmanica’, ‘Africa’, ‘European’, ‘New World’ and ‘Stenophora’ (Wilkin et al., 2005; Viruel et al., 2016). Viruel et al. (2016) obtained some clades not found in previous analyses: Dioscorea section Rajania (L.) Raz, a group of species endemic to the Caribbean islands (Raz, 2016), the clade Epipetrum, proposed by Philippi (1864) as genus for dwarf-sized species endemic to Chile (Viruel et al., 2010), and the clade Nanarepenta, for non-winged seeds species exclusive to Mexico, also previously proposed as a separate genus (Téllez-Valdés & Dávila-Aranda, 1998). Those clades fund by (Viruel et al., 2016) were sampled by previus studies (Caddick et al., 2002a,b; Wilkin et al., 2005)(), but (Viruel et al., 2016)presented a expanded the number of species and more a robustly supported results.
Full understanding of relationships among species of Dioscorea and a natural infrageneric classification will only be possible with strongly supported phylogenetic studies. However, the phylogenetic results to date do not allow a new formal and complete classification, especially for the Neotropical region, which is still poorly sampled in most recent published works (five neotropical taxa in Caddick et al., 2002a); three in Caddick et al., 2002b; 43 in Viruel et al., 2016). It should also be noted that the genera mentioned above — which are now known to be part of Dioscorea — still require a change of status regarding their positioning in an infrageneric taxonomic category within Dioscorea, with exception of Rajania that has been reduced to a section by Raz (2016).
Previous phylogenetic studies points to three events of radiation of Dioscorea into the Neotropics (Wilkin et al., 2005; Viruel et al., 2016), however their sampling did not completely reflect the diversity of Neotropical region. Species of Dioscorea arrived at least thrice in the Neotropics since the Oligocene and greatly diversified there, occupying various habitats in this region (Viruel et al., 2016). Since previous phylogenies did not include a dense sampling of Neotropical species, much of the taxonomic, morphological and geographical variation of Neotropical Dioscorea has not been covered, hampering inferences regarding its origin and diversification in this biogeographical region.
The present study increases by 45% the sampling of Neotropical species compared to previous phylogenetic studies in Dioscorea, including wider taxonomic and morphological diversity, and summed with sequences available in GenBank (mostly from Wilkin et al. (2005) and Viruel et al. (2016)), it represents the most densely sampled phylogeny of Dioscorea to date. This neotropical sampling (64 species in total) represents around 20% of all neotropical species. In terms of possible neotropical lineages, Knuth (1924) presented 56 sections for D. sect. Helmia and D. sect. Dioscorea (17 for Helmia and 39 for Dioscorea), 38 of these feature neotropical distribution (67%, 13 for Helmia and 23 for Dioscorea). In our study, we cover 22 of those sections (almost 40% of all sections, 7 for Helmia and 15 for Dioscorea) and four more species Incertae sedis, on which 14 of these lineages (10 sections and four Incertae sedis) had not been sampled in previous studies.
By reconstructing the phylogeny of Dioscorea, with special addition of Neotropical samples, we aimed to infer phylogenetic relationships among Neotropical taxa, as well as their arrival and divergence times in this region. As the richest lineages of the genus, we seek to test the infrageneric classification proposed Knuth (1924) (reestablishing any of the sections if supported by phylogeny) and to obtain a clear view of the lineages existing now at the Neotropical region, especially in cases where the Knuth, (1924) classification does not present clear delimitations between sections. Also, the understanding of how they occur in past, up to the present environments can provide tools for a better comprehension of the evolution of this group so rich morphologically. We also offer a more robust scenario of phylogenetic relationships in the genus, facilitating future proposals of an updated classification of Dioscorea.
MATERIALS AND METHODS
Taxon sampling, DNA sequencing and alignment
Species sampling aimed to cover a wide geographic and morphological range within Dioscorea. We produced new sequences for 34 species of Dioscorea: 12 spp. representatives of D. subg. Helmia, 18 spp. of D. subg. Dioscorea, and four spp. incertae sedis, plus Tacca artocarpifolia Seem as an outgroup representative of Dioscoreaceae (Table S1). The taxon sampling aimed to cover a great part of the lineages proposed as sections by classic authors (sensus the revisionmade by Knuth, 1924), covering 22 of the 38 sections proposed to the neotropical region, attending the sections with great number of species with more than one taxa (eg.: D. section Dematostemon Griseb., five species sampled in diferent morphological strata).
Most new sequences were generated from specimens collected in the field by R. S. Couto, prioritizing taxonomically well-delimited species from well-known populations. Vouchers were deposited in the Herbarium of National Museum (R), with duplicates at the Herbarium of the Botanical Garden of Rio de Janeiro, Brazil (RB).
Additionally, we included 143 sequences from GenBank: 129 from Dioscoreaceae (119 species of Dioscorea, two species Trichopus, seven of Tacca and one Stenomeris), and five from other families (three species of Burmanniaceae, one Stemonaceae and one Thismiaceae), totaliing 177 species sampled. All sampled species with geographic origin, herbarium vouchers and GenBank accession numbers are listed in Table S2.
Total genomic DNA was extracted from leaf samples, fresh or silica-dried following the 2x CTAB protocol (Doyle & Doyle, 1987), without the addition of RNase A and scaled to 2 ml tubes. The extracted DNA was measured in 1% agarose gel with a DNA mass ladder and deposited at the DNA collection of Laboratory of Systematics and Molecular Ecology of Plants, Federal University of Paraná (UFPR), Brazil, associated to the reference vouchers deposited in K, R, RB, RFA and UPCB (Herbarium acronyms follow Index Herbariorum (Thiers, continuously updated)). The plastid genome regions matK and rbcL were amplified and sequenced using the following primer pairs, respectively: 3F_KIM-f (cgtacagtacttttgtgtttacgag) and 1R_KIM-r (acccagtccatctggaaatcttggttc) (Ki-Joong Kim, pers. com.), rbcLa_f (atgtcaccacaaacagagactaaagc) (Levin et al., 2003) and rbcLa_r (gtaaaatcaagtccaccaccrcg) (Kress & Erickson, 2007). Additionally, the primers ITS92 and ITS75 or ITS18F and ITS26R, which is widely used in angiosperms (Bolson et al., 2015), was used to amplify the nuclear ITS region, however without sucesss in sequencing. This negative result and the scarcity of ITS sequences for Dioscorea in GenBank lead us to use only plastidial genome markers. PCR amplifications were performed using initial 94°C pre-melt for 1 min followed by 40 cycles of (i) 94°C denaturation for 30s, (ii) 53°C annealing for 40s, and (iii) 72°C extension for 40s, followed by 72°C a final extension for 5 min. Following PCR, the samples were purified with 20% PEG and sequenced with Big Dye Terminator version 3.1 (Applied Biosystems, California, USA) by the company Macrogen Inc. (South Korea). Forward and reverse sequences were assembled using the Staden package v.2.0.0b11 (Staden, Judge, & Bonfield, 2003). Sequences were aligned with Clustal W using default parameters (Thompson et al., 1997) implemented in the software MEGA6 (Tamura et al., 2013).
Geographical and morphological data
We examined over 4,000 specimens deposited in 79 herbaria, in addition to field observations, with special effort on the Neotropics, for the selection of taxons sequenced, to obtain morphological comparisons between the analyzed species (detailed in the discussion of the clades) and especially for coding the geographic distributions used in biogeography. Specimens from the following herbaria were examined: B, BAA, BAFC, BR, C, CAY, CEPEC, CESJ, COAH, COL, CR, CTES, CUVC, CVRD, ESA, F, FAA, FCAB, FURB, GUA, HAL, HAS, HB, HCF, HMUC, HRCB, HST, HUCP, HUEFS, HUPG, HVASF, HXBH, IAC, ICN, INPA, IPA, IRBR, JE, JVR, K, L, LPS, M, MBM, MEXU, MG, MNHN, MO, MVFA, MVFQ, MVM, NY, OPUR, P, PACA, PEL, PH, R, RB, RBR, RFA, RFFP, S, SI, SMDB, SP, SSUC, U, UFP, UFPR, ULS, UNR, UPCB, US, UV, WU, XAL, Z, and ZT. Herbarium acronyms follow Index Herbariorum (Thiers, continuously updated).
Phylogenetic analysis
Maximum Likelihood, Parsimony and Bayesian inference were used to estimate tree topologies on the concatenated matrix of 177 taxa and 1658 nucleotides from rbcL and matK genes. Maximum-likelihood tree searches were performed using raxmlGUI v.1.0 (Silvestro & Michalak, 2012) under the model GTR+I+G and statistical support for nodes were assessed with 1,000 bootstrap replicates, consistent with that used by Viruel et al. (2016). Parsimony analyses were conducted in PAUP 4.0b10a (Swofford, 2002) using heuristic tree searches with tree bisection-reconnection (TBR), 2,000 random-taxon-addition replicates holding 20 trees per replicate. Branch support was estimated with 2,000 bootstrap pseudo-replicates (Felsenstein, 1985). Bayesian phylogenetic inference with the Metropolis-coupled Markov Chain Monte Carlo (MCMC) was used to estimate tree topology and posterior probability distribution as implemented in MrBayes v.3.3.4 (Ronquist et al., 2012), using two parallel runs, each with four chains and 10 million generations, with parameters sampled every 1,000 generations. Each gene was considered as one partition and the best-fitting models under the Akaike Information Criterion in the software Mega 6 (Tamura et al., 2013) were GTR+G (matK) and K2+G (rbcL). Convergence of runs was assessed in Tracer 1.6 (Rambaut, Suchard, & Drummond, 2014). A 25% burnin was applied to eliminate trees prior to convergence of chains and a 50% majority rule consensus tree was constructed from the remaining trees.
Molecular dating
The Dioscorea tree molecular clock was estimated using fossil and secondary calibrations and the full matrix of 177 taxa including 162 representatives of Dioscorea and 15 outgroups. The fossil record of Dioscoreaceae was recently reviewed by Raz (2017), who analyzed twenty fossils attributed to this family from different time periods and geographical origins, mostly leaves. Only three can be attributed to sections within Dioscorea and therefore are more suitable for molecular dating: Dioscoroides lyelli (Eocene), Dioscorea wilkinii (Oligocene) and Dioscorea sp. from Kenya (Sect. Asterotricha) (Miocene). Other fossils are attributed only to genus level or would require further information to confirm their position, and eight of them are not Dioscoreaceae (Raz, 2017). Three Dioscoreaceae fossils used to calibrate the tree were chosen by stratigraphic reliability and confidence of taxonomic assignment (Raz, 2017). We follow the fossil calibration described by Viruel et al. (2016): A. The fossilized leaf of Dioscoroides lyelli from the Eocene at the Paris basin, France (Potonié, 1921), dated from the Ypresian age, was assigned to the Stenophora stem due to similarities with the extant Stenophora species; B. The fossil seed attributed to Tacca buzekii from the Upper Eocene from Putschirn, Czech Republic (Gregor, 1983), was assigned to the crown node of Tacca; C. The fossilized leaflet attributed to Dioscorea wilkinii, from the Middle Oligoceneof Ethiopia (Pan, Jacobs, & Currano, 2014), was assigned to the crown node of section D. Section Lasiophytum A lognormal distribution prior was applied to all fossil calibrations nodes, with values also following Viruel et al. (2016), as follows: Dioscoroides lyelli (mean= 48.2, sd= 0.008); Tacca (mean: 35.85, sd= 0.028); Dioscorea wilkinni (mean: 27.23, sd= 0.002). Calibration points are depicted in Fig. S3.
Divergence times were inferred under a relaxed uncorrelated lognormal clock model in BEAST 1.8.3 (Drummond et al., 2012) implemented in the CIPRES server (Miller, Pfeiffer, & Schwartz, 2010), using a Yule tree model of speciation, and HKY+I+G substitution model with empirical base frequencies. The MCMC chains ran for 50 million generations, sampled every 10,000 generations. Convergence was assessed using effective sample size (ESS) values ≥ 200 in Tracer 1.6 (Rambaut et al., 2014). Two separated runs were performed and their results were combined in LogCombiner, totalizing 100 million generations. The maximum clade credibility tree was generated in TreeAnnotator (BEAST package), and visualized and edited in FigTree v 1.4 (Rambaut, 2009).
Biogeographic analysis
In order to estimate ancestral distribution areas of Neotropical Dioscorea we defined six major geographic areas: A= Central America, B= Northern Andes, C= Southern Andes, D= Amazonia, E= Dry Diagonal and F= Atlantic Forest. Definitions were based on the current distribution of Neotropical species of Dioscorea, considering areas with more than 10 endemic species, seeking to exclude areas that present only occasional endemisms. The delimited areas were also based on the Neotropical regional classification proposed by Morrone (2014) and a study on Rubiaceae by Antonelli et. al. (2009). Although another six species that also occur in the Neotropics were included in our analysis, we focused our biogeographical analysis on the most diverse clades, New World I and New World II (hereafter NWI and NWII).
We estimated ancestral range probabilities on the multimodel approach performed by the R package BioGeoBEARS (Matzke, 2013; R Core Team, 2016) using the three available models: DEC, BAYAREALIKE and DIVALIKE. The dispersal-extinction-cladogenesis model (DEC) considered cladogenetic processes as the evolution of range at speciation events and allows the estimation of free parameters d (dispersal) or range extension and e (extinction) or range loss by maximum likelihood (Ree & Smith, 2008). Dispersal-vicariance-analysis (DIVA) is a parsimony-based method that allows dispersal and extinction in anagenetic processes and vicariance in cladogenetic processes (Ronquist, 1997). The model is called DIVALIKE in BioGeoBEARS because it is a maximum likelihood implementation of DIVA. The BayArea method is a Bayesian approach specifically designed to analyze a large number of areas efficiently (in reasonable computer time) (Landis et al., 2013). BAYAREALIKE implemented in BioGeoBEARS is a maximum likelihood interpretation of BayArea. The founder speciation event parameter j was also added to all analyses, creating the models DEC+j, DIVALIKE+j, BAYAREALIKE+j The parameter j adds the possibility of a new cladogenesis event, where an individual ‘jumps’ to an area completely outside the ancestral range, founding a new genetically isolated lineage (Matzke, 2013, 2014). For selection of best-fit model, we relied on the best likelihood value as well as the Akaike Information Criterion (ΔAICc).
RESULTS
Phylogenetic analysis
Dioscorea species form a strongly supported clade (bootstrap support value (BS) ML and MP 100%, Posterior Probability (PP) 1) (Fig. 2, Fig. S1 and S2). Among the internal clades, Dioscorea section Stenophora has strong support (BS-ML and BS-MP 100%, PP 1) and represents a sister lineage to the remaining Dioscorea. Traditional classification systems and phylogenetic analysis results are compared in Table S4.
Most Neotropical species grouped into two clades, here indicated as ‘New World I’ and ‘New World II’. New World I groups the species corresponding to D. subg. Dioscorea, restricted to the Neotropics. Within this clade it is possible to recognize another three small clades: Epipetrum, with high support (BS ML and MP 100%, PP 1), Microdioscorea, and Nanarepenta, with low support. New World II has strong support (BS-ML 99%, BS-MP 95%, PP 1) and groups species corresponding to D. subg. Helmia restricted to the Neotropics. The Monadelpha clade can be recognized in the New World II clade, as a strongly supported clade (BS-ML 98%, BS-MP 90%, PP 1) with sister species. Neotropical species that were external to NWI and NWII, D. dodecaneura Vell., D. stegelmanniana R.Knuth, and species representatives of D. section Rajania (L.) Raz form the clade ‘New World III’, appearing associated with clades of African and Asian species (Africa and Compound Leaved). The species of D. section Rajania, however, form a well supported clade (BS-ML 90%, BS-MP 75%, PP 1) within New World III. Only one species with Neotropical occurrence appears in this analysis outside the clades mentioned above, D. mollis, which is sister to D. antaly, and related to Asian species.
The clade Shannicorea appears for the first time and brings together the seven species from Southeast Asia, with moderate support (BS ML and MP 81%, PP 1). The remaining Old World species are organized in a large clade, where the inner clades ‘Mediterranean’, ‘Africa’, ‘Compound Leaved’, ‘Birmanica’, ‘Malagasy’ and Enantiophyllum can be highlighted, following Viruel et al. (2016) nomenclature. The results obtained in this analysis are congruent with Viruel et al. (2016), just with some differences in the support values of the clades, those values can be retrieved in Figure 1 and others in supplementary material (Fig. S1, S2, S3).
Bayesian and Maximum Likelihood analysis results did not differ considerably, except for small differences in the most recent clades in ‘Birmanica’, D. section Shannicorea and ‘Malagasy’. Maximum Parsimony analysis also resulted in high support for most of the clades that were well-supported in other analyses (Fig S2).
Divergence times and biogeography
According to our divergence time analysis in BEAST, the most recent common ancestor of Dioscoreaceae (including Tacca) originated in the Cretaceous around 99 Mya (95% Highest posterior density interval (HPD): 72.7–132.4 Mya) (Fig. 3). All divergences in the genus level occurred in the Paleocene. The most diverse genus in the family, Dioscorea, is estimated to have originated in the Cretaceous-Paleocene boundary, around 66 Mya (54.8 –84.6 Mya) (stem age). The two main Neotropical clades originated between the Eocene and Oligocene: the crown age for New World I is 37.2 Mya (28–44.3 Mya) and around 28 Mya (20.8–37.8 Mya) for New World II. The third Neotropical clade comprises only four species and originated in the Oligocene, around 30 Mya (20.3 –37.5 Mya). The clade that includes Neotropical Dioscorea mollis plus the Malagasy D. antaly originated in the Miocene at 18.5 Mya (6–31.5 Mya).
Biogeographic analyses performed separately on clades New World I and New World II yielded the same model as the best scored model (i.e. with lowest likelihood): BAYAREALIKE+j (See Table S3 for all parameter results). The ancestral distribution of the MRCA for New World I clade in Dioscorea is predicted to be the “Southern Andes” biogeographic region (Fig. 3), a region that nowadays comprises Northern Argentina and Chile and corresponds to the southern part of the mountain range, not present at the time of origin of NWI clade. Some clades remain endemic to this area (for example, the D. fastigiata – D. humilis clade). Dioscorea expanded its distribution eastward during the Oligocene, diversifying in today’s Atlantic Forest. Expansions to Central America probably occurred in the Miocene at least twice. Further occupations of Dry Diagonal plus Amazonia and/or Atlantic Forest occurred in the Middle Miocene (around 10 Mya).
In contrast, New World II is estimated to have originated in the eastern part of South America, i.e. in a region comprising the Dry Diagonal and Atlantic Forest (Fig. 3). Most of the extant species in this clade still inhabit these regions. In the Miocene, two independent occupations of Northern Andes occurred, corresponding to the D. coriacea and D. larecajensis stems. Approximately at the same time, around 15 Mya, the D. convolvulacea – D. galeottiana lineage dispersed into Central America, with D. convolvulacea still being distribuited in Northern South America.
Two species of very restricted distribution in completely distinct regions appear in a small recent clade in the phylogenetic analysis of Dioscorea, D. mollis and D. antaly, which are endemic species of Southeastern South America (Atlantic Forest) and Madagascar (Tropophylles Forest = Deciduous Dry Forest), respectively. Dioscorea antaly is the only endemic species of Madagascar to appear outside the Malagasy clade, and D. mollis is the only endemic species of the Neotropics to emerge outside the New World clades (NWI, NWII and NWIII).
DISCUSSION
Phylogenetic analysis and classification of Dioscorea and position of Neotropical clades
Eleven major clades were obtained for the monophyletic genus Dioscorea, which was consistent with the clades observed in other recent studies. The nine clades and sections of Dioscorea already established in other studies (Wilkin et al., 2005; Viruel et al., 2016) were also recovered in this analysis, with the emergence of internal clades such as D. section Shannicorea (even though weakly supported) and the position of Neotropical species external to the clades New World I and II (Fig. 2).
The taxa of unique occurrence in the Neotropics form basically two clades, as also stated by Wilkin et al. (2005) and Viruel et al. (2016). The New World I clade (NWI) groups (although, with low support) the species with circular or semicircular winged seeds that occur in the tropical region of the Americas, once placed by Knuth (1924) in D. subgenus Dioscorea along with other Dioscorea of circular or semicircular winged seeds with wider, pantropical distribution. The seed morphology is part of the Dioscorea taxonomy since the the first classifications proposed, Kunth (1850) proposed the Helmia genus using (beside others) this character, as other follow in theirs infrageneric classification. Although Knuth (1924) knew that the use of this character to split the Dioscorea genus almost at half was not completely adequate, nor reflected a systematic relationship, it was used, at least for practical reasons, for many years as the base of the infrageneric classification of Dioscorea. Wilkin et al. (2005) had already shown that this character and subgenus classification were not a reflection of the evolutionary relations of the group, however, they obtained several clades where one of the states of this character were fixed in all specis sampled in their analisys. The autors used this condition to help to explain and separate the especies in the two internal clades of Madagascar and as a plesiomorfic condition of all species of their clade “Compund-leafed”, yet they didn’t discussed this character in the scope of the “New World” clade (probably by the lack of knowledge of the fruit morphology of neotropical species at the time).
Allthough, this clade does not include all the species placed by Knuth (1924) in this subgenus of pantropical distribution, it contemplates species from different regions and habitats of the Neotropics, as well as great morphological diversity (Fig. 1). This could mean that this character, had only a few changes between the the two states, at least in the Neotropics, where almost half of all species of the genus are distributed, and could point to a probable sinapomorphy to the NW I and II clades.
The Chilean species in the NWI clade are basically arranged in three smaller clades (Epipetrum, Microdioscorea and Nanarepenta, with exception of D. brachybotrya), with poor internal resolution. In spite of the evidence for the relationship between the taxa endemic to Chile, the Epipetrum clade groups dwarf species that present non-winged seeds, characteristic to the dry regions of Chile, as seen in Viruel et al. (2016). Besides these species, we have six taxa placed by Knuth (1924) in the D. section Microdioscorea, a section composed mostly of Chilean species (only two taxa for Peru and Brazil) with stems of reduced length and six stamens. Dioscorea section Microdioscorea appears as monophyletic with low support in this analysis and also in Viruel et al. (2016).
Another clade within the NWI group is the group endemic to Mexico, previously designated by Matuda (1961) as the genus Nanarepenta, which also has non-winged seeds. In Viruel et al. (2016), D. minima appeared along with the other three species of Nanarepenta, but in the present analysis this species is distantly related. This lack of resolution could be a reflection of the increased number of taxon and the lower amount of markers used in this study. Due to the great diversity found in the Neotropics is desirable that a greater number of species, as well as the DNA regions, could be analysed to solve the internal relations of the largest Neotropical clade.
Whithin the species of NWI clade is possible to highlight D. perdicum Taub., one of the dwarf species endemic from Brazil that does not fit any specific infrageneric classification devotaded to this issue. Dioscorea perdicum is mistakenly placed by Knuth (1924) into the section Cycladenium Uline (D. subg. Helmia), as the autor didn’t know the species fruits or seeds, that place D. perdicum in D. subg. Dioscorea by its all-round winged seeds. Burkill (1960) pointed out the mistake made in the allocation in the Cycladenium section suggesting to be placed in D. sect. Pedicellatae R. Knuth, but most importante he proposed this species as a possible link between the Neotropical and Paleotropical species, with relation to D. sect. Stenophora, specifically by the presence of what he called a rhizome. As a matter of fact, the underground organ of D. perdicum is a tuber of rhizomatous aspect (Fig. 2), as have all anatomical features of a tuber (Tenorio, pers. comm.) but the aspect is similar to a rhizome, as doesn’t have cataphylls but it branches and produces roots and shoots from more than one point. This unique morphology leaded Burkill (1960) to assume the close relation to the Paleotropical species of Dioscorea, aspect not confirmed in our analisys, as D. perdicum appears without any special relation to the Stenophora clade.
The other species that are endemic to the Neotropics appear here as the strongly supported clade NWII (BS-ML 99%, BS-MP 95%, PP 1), which contains part of the formerly designated D. subg. Helmia (sensu Knuth, 1924), unlike the NWI clade, this group has been retrieved in all phylogenetic studys that have sampled the neotropical species (Wilkin et al., 2005; Viruel et al., 2016) and also strog support, demonstrating the likely single origin of the elongated seed wings in the Neotropical region. Even though the sections sampled here have been increased only by one, the number of species within key sections (D. sect. Dematostemon Griseb. and Monadelpha Uline) has been increased in order to contemplate more comprehensively the diversity of this group. Dioscorea sect. Dematostemon is one of the biggest sections of the Neotropical region, comprising 45 taxons of great morphological diversity and high distribution range, going from dwarf species (eg.: D. anomala (Kunth) Griseb. and D. maianthemoides Uline ex R. Knuth) endemic to the brazilian “Cerrado” to typical Atlantic Forest species (D. campanulata Uline ex R. Knuth and D. cinnamomifolia Hook.). Knuth (1924) also placed in this section species (D. moyobambensis R. Knuth, D. galiiflora R. Knuth and D. triangularis (Griseb.) R. Knuth) closely related to others in D. sect. Centrostemon Griseb., showing that the boundaries of this group are not well defined. With our increased sampled analisys, this section is shown to be polyphyletic.
In addition, we have a small clade with strong support, formed by D. section Monadelpha species, which present unusual characteristics, such large staminate flowers, three fertile stamens forming a fleshy column, free and entire stylus in pistillate flowers. Besides those characteristics, this section possess an almost unique feature related to sexuality in the genus, the monoecy, all species from this section present staminated inflorescences in the lower part of the plant and pistillated on the higher part. This character is only shared with a few species from D. section Cycladenium (probably misplaced) and D. margarethia G.M. Barroso, E.F. Guim. & Sucre (poorly know compoud-leafed species with six stamens on a column instead of three), and besides some unstable dioecy in D. sect. Stenophora and D. convovulacea Schltdl. & Cham. (Wilkin et al., 2005). The monoecious species from D. section Cycladenium aro not well know and the boundaries of the section are not clear, as is a section with great diversity (i.e. monoecious and dioecious species). In our analisys, the only species from D. section Cycladenium, D. coriacea Humb. & Bonpl. ex Wild., (dioecious), doesn’t appear to be closely related to this clade. Addition of more taxa from this group could clarify relationships the two sections and reinforce the monophyly of D. section Monadelpha.
The Monadelpha clade also appear as monophyletic in other studies with less dense taxon sampling for the Neotropic (Wilkin et al., 2005; Viruel et al., 2016), however, with no mention of the section, due in part to the smaller sample of the Monadelpha clade and due to a misidentification of one of the species used in Viruel et al. (2016) (D. monandra Hauman was iderntified as D. calderillensis, from D. section Cycladenium). Therefore, we understand that monoecy probably originated only once in the in the family, with origin in the Neotropical region. The position of D. margarethia should be tested to reinforce this organization, but the species is only know from its type specimens and a second gathering (that could not be sequenced), and even the fact that it has six stamens doesn’t seem to be a problem as this character is volatile in the NW clades.
Even with the increased number of taxa sampled to the NWI clade from 23 to 64 species and 15 Knuth’s sections for D. subg. Dioscorea and seven for D. subg. Helmia, plus four incertae sedis species the resolution in the Neotropical clades is not yet close to be solved. It has been increased the number of internal clades monophyletic (Monadelpha, Nanarepenta, Epipetrum, Microdioscorea) and species of most sections of Knuth (1924) are shown not to be phylogenetically related (e.g.: Dematostemon, Apodostemon Uline, Cryptantha Uline).
A small group of Neotropical species emerged outside the main Neotropical clades (NWI and NWII), being composed of three species from D. section Rajania, plus two South-American species, D. dodecaneura and D. stegelmanniana. Wilkin et al. (2005) already presented Dioscorea cordata (L.) Raz as a distinct lineage, separate from the NWI and NWII clades, which can also be observed in Viruel et al. (2016), also associated to the neotropical edible species D. trifida L. f.. Those species have very different morphological characteristics when compared to other Neotropical species, some of these characteristics being shared with those found in species from Asia and Africa. The common ground to this species is the presence of annualy renewed tubers, those are the only neotropical species sampled so far that have this character (absent in the NW clades and rather rare in the neotropical species). The matter of annual tubers has been addressed by Wilkin et al. (2005), showing that is a paleotropical characteristic (only present in their B clade), being these clades the ones related here to NWIII.
Dioscorea dodecaneura and D. stegelmanniana are morphologically very similar to each other, but they present marked differences in one key aspect of Knuth’s classification, the fruit (transversely oblong and oblong, respectively) and seeds shape (circular and oblong, respectively). This indicates that besides the NWI and NWII clades the seed wing shape is not stable character, having closely related species on the NWIII clade with both of the states of this characteristic. Additionally, these two species present a particular pattern of organization of the vascular bundles of the aerial stem (Tenorio et al., 2017), similar to that described by Ayensu (1972) as the typical Old World pattern. It is noteworthy that D. trifida and D. stagelmanniana have been sorted by Knuth (1924) to D. sect. Macrogynodium Uline, reflecting in his view the close relationship of this species.
Dioscorea section Rajania is composed of 18 species, besides one non-described species, with occurrence restricted to the West Indies (Raz, 2016; Raz & Pérez-Camacho, 2016)(Raz, 2016). The species belonging to D. section Rajania are distinguished by the samaroid fruits, although this is not an exclusive feature of this section, as pointed out by Raz (2016) in the most recent taxonomic treatment on this group. The morphological characteristics exhibited by this section exemplify the diversity found in the Neotropical region, even though it is a clade with lower morphological diversity, it presents great differences for the rest of the neotropical species and with a more recent arrival when compared to the other two Neotropical clades (NWI and NWII), with the maintenance of characteristics typical of species of the Asian and African region.
Dioscorea mollis shares several characteristics with paleotropical species, such as phyllotaxis ranging from alternate to subopposite or even opposite, a characteristic found in less than 2% of the species of the American continent, where most species presents alternate leaves. The species also present an underground system composed of several fibrous nodules from which numerous aerial stems appear (similar to a rhizomatous system), stems of woody aspect and prickles in the basal stem, which are also unusual characteristics for Neotropical species. These characteristics are shared with one closely related species of D. section Chondrocarpa Uline, D. chondrocarpa Griseb., not sampled here by the unsuccessful amplification of matK gene. Dioscorea chondrocarpa was also sampled by Viruel et al. (2016) but did not reached the final publication by lack of genes amplified successfully, however its position in their inicial analisys topology is congruent to the one fund here to D. mollis (Raz, pers comm.). Anatomically these species also presents similarities to the Old World species, as verified by Tenorio et al. (2017). All these evidences strongly indicate a fourth lineage of Dioscorea in the Neotropical region, more related to paleotropical species.
The Paleotropical clades obtained in our analysis were similar as those recovered in previous studies (Wilkin et al., 2005; Maurin et al., 2016; Viruel et al., 2016), consisting in the Africa clade with the inner clade Testudinaria (composed of species occurring in the mountainous regions of eastern and southern Africa), the Malagasy clade (with all endemic species from the island of Madagascar, except for D. antaly), and the clade Enantiophyllum (composed by several species proposed to the D. section Enantiophyllum Uline). Previous and the present phylogenetic results contradict the main infrageneric classifications of Dioscorea (Uline, 1897; Knuth, 1924; Burkill, 1960), which grouped the species in various sections. Some aspects of those clade of these clades are interesting to emphasize, such as the Malagasy clade is internally organized into two small clades, one presenting circular winged seeds, and the second grouping the remainder species with elongated winged seeds, and that Enantiophyllum (with its enormous diversity and imprecise delimitation) has some polytomies in the present and previous phylogenetic analyses, indicating the need of more data to elucidate the internal relationships
Dioscorea section Shannicorea was proposed to group six species of occurrence restricted to Asia, mostly China. The taxa share the left twining stem, the staminate inflorescences composed of small scorpioid cymes, stamens inserted at the base of the tube segments and the elongated seed wings. Knuth (1924) treats the same species, with the addition of two taxa, as D. section Shannicorea, but subordinated to D. subg. Stenophora. In contrast to the initial position proposed by Uline (1897) for the Stenophora section, Knuth (1924) elevated Stenophora to subgenus and further organized it internally into two sections, Eustenophora R.Knuth and Shannicorea Prain & Burkill.
In our analyses, the six species listed by Prain & Burkill (1914) in D. section Shannicorea (D. hemsleyi, D. martini, D. nitens, D. subcalva, D. velutipes and D. yunnanensis) are grouped in a single clade, presenting similar internal relationships in all analyses, but positioned differently in the Dioscorea tree. It is the first time that this clade is supported in a widely sampled Dioscorea tree. Hsu et al. (2013) also recovered this clade, but their sampling included only species from East and Southeast Asia. Viruel et al. (2016) also analysed some of these species (D. nitens and D. subcalva), retrieving them inside the Birmanica clade, as they are related. In our analysis, D. section Shannicorea is distantly related to D. section Stenophora, demonstrating that Knuth’s (1924) proposal to treat D. section Shannicorea as part of the subgenus Stenophora has no phylogenetic support and Prain & Burkill’s (1914) proposal could be more accurate. This indicates that the section as proposed by Prain & Burkill (1914) could be monophyletic, but more DNA markers in a phylogeny that includes species from both clade that is closely related (Shannicorea and Birmanica) are needed to solve uncertainties in this part of the topology.
The placement of D. sansibarensis at the base of the Malagasy clade raises some questions regarding the evolution of this group in an insular environment. According to Viruel et al. (2016), Madagascar was colonized by Dioscorea species from Asia, and not from Africa as many angiosperms. On the other hand, the sister species D. sansibarensis occurs in several areas of Africa, besides Madagascar, presenting a high vegetative dispersal ability: they massively produce small aerial tubers in the leaf axils, which possibly facilitated their invasive behavior in several countries (Raz, 2002; Choo, 2009; Hsu & Wang, 2012). The presence of D. sansibarensis in Madagascar could be product of a recent natural dispersal event or human introduction, since this species has been used for food and for the production of venom (Wilkin et al., 2005).
We have new evidence suggesting that within both the biggest New World clades, some of the sections proposed by Uline (1897) may be supported and can fit in a phylogenetic sound revised classification of Dioscorea to come. The position and monophyletism of D. section Microdioscorea and D. section Monadelpha within a widely-sampled phylogenetic analysis of Dioscorea shows that the proposed infrageneric classification by classical works (Kunth 1924; Uline 1897) mostly do not represent natural lineages, but some of them still may be used in modern systematics of the genus. Increasing the Neotropical species sampling evidenced the role of this group of species as key to provide a complete and accurate infrageneric classification of Dioscorea, as it is the most diverse and taxonomically complex region, and at the same time, the most underrepresented in phylogenetic studies until the present study. Nevertheless, the Neotropical species still lack a wider sampling to reach a better resolution of these clades, 16 sections of the Knuth (1924) classification still don’t have been used in any phylogeny up to the present (22 of 38 were coverd here), and this should be goal to persue to better undertand the infrageneric classification of Neotropical Dioscorea.
Dioscorea lineages originated four times independently in the Neotropics since the Eocene
The phylogenetic analysis presented here, focused on the Neotropical clades of Dioscorea, provides a new perspective on the biogeographical history of this genus in South and Central America. The biogeographic analysis has shown that four independent lineages of Dioscorea diversified into the Neotropical Region, two of them becoming highly diverse and wide spread. The Neotropical species of Dioscorea present at least four different origins. The New World I and II clades are more diversified and widely distributed (Fig. 2), while another two species are grouped in the predominantly Caribbean group Rajania, and D. mollis is sister to the Malagasy D. antaly. With the exception of D. mollis, which has an independent origin, all the other hypotheses on origins of Neotropical clades of Dioscorea had been described before (Viruel et al., 2016). For the first time, however, we presented a more detailed view on the biogeographic history of the group after the colonization of the Neotropics.
The pantropical genus Dioscorea putatively originated in Laurasia during the Late Cretaceous – Early Eocene, later dispersing into South America, Africa and Madagascar (Viruel et al., 2016). The colonization of America could have been facilitated by the existence of land bridges during the Palaeocene-Eocene thermal maximum (Zachos, Dickens, & Zeebe, 2008), such as the North Atlantic Land Bridge (NALB) and Beringean Land Bridge (BLB), presumably during early Oligocene (Viruel et al., 2016), further reaching South America by occasional island chains such as the proto-Greater Antilles (Antonelli et al., 2009). Nevertheless, ancestral area reconstruction suggests a South American origin for the Neotropical Dioscorea with further dispersals towards Central America (Fig. 3), a result that was also found in the global analysis by Viruel et al., (2016). Exchanges between Laurasia and South America are reported for other plant groups, such as Malpighiaceae, which is similarly pantropical, greatly diverse in South America, and possibly migrated through Laurasia after having originated in South America (Davis et al., 2002).
At least three different origins are estimated for the Neotropical clades. The NWI clade originated in the Eocene-Oligocene boundary, in what is now the Southern Andes Region, and dispersed to eastern South America and Central America. At this time most of the Andes were not formed yet, which could allow eastwards expansions of range; however, the South American continent was partially occupied by marine incursions from the Caribbean and from the Pacific seas, which could be barriers to expansions (Antonelli et al., 2009). Expansions towards the East in South America occurred only after the Oligocene, around 30 Mya. Since the end of Cretaceous, the South American humid forests dominated the terrestrial habitats and no evidence of dry vegetation exists for this period until the decrease in temperatures that took place after the Miocene Medium Climatic Optimum (Davis et al., 2005; Hoorn et al., 2010).
Occupation of dry vegetation biomes occurred in both clades, but only sparsely in NW II, which remained almost exclusive to rainforests. The MRCA of the clade that occurs in Dry Diagonal is ambiguous and could have been Amazonian or from the Dry Diagonal. It is known that the origin of the South American Dry Diagonal must have taken place only after 10 Mya, with climate gradual cooling and drying in the Miocene, and before the establishment of rainforests (Simon et al., 2009). Some species of Dioscorea occur in both open vegetation formations (Dry Diagonal) and forested biomes, i.e. Amazonia (D. acanthogene, D. pohlii and D. amaranthoides) and Atlantic Forest (D. sinuata, D. leptostachya and D. trisecta). The most generalist species can occur in the Dry Diagonal and the two forested biomes (D. multiflora, D. marginata and D. laxiflora). Species with occurrence in forested habitats usually will appear in arboreous vegetation patches in the Cerrado, like the “Cerradão” and Gallery forests or Caatinga forest, and not in open vegetation. Connections between Amazonia and Cerrado occurred many times in history, not only because of the geographic proximity of the regions occupied by the two vegetation types, but also because during episodes of climatic fluctuations, forests are known to have expanded or retracted (Costa, 2003). Quaternary cooling and drying episodes during the glacial times favored the expansions of savanna-type vegetation (Cerrado), decreasing the extent of tropical forests (Werneck et al., 2012). Only NWII presents species with distribution restricted to the open field formations (D. maianthemoides) and rocky savannas (D. campos-portoi) within the Cerrado biome.
Dispersals towards Central America occurred in Middle Miocene in both NW I and II, much earlier the estimated closure of the Ishtmus of Panama (3 Mya), which linked the still “isolated” biota of South America to North America (for a review of the long process of formation of Isthmus of Panama see O’Dea et al., 2016). Flora exchange before the Isthmus formation is best explained by long distance dispersals (Antonelli & Sanmartín, 2011; Freitas et al., 2016) and this seems to be the case of Dioscorea. Transoceanic dispersals explained much of plant biogeographic patterns and is hypothetically more explanatory than plate tectonics (Renner, 2004; Christenhusz & Chase, 2013). Besides the plant’s dispersal capability, which in the case of Dioscorea is facilitated by seed and fruit morphology and production of aerial tubers, wind and sea currents also facilitate the dispersal across oceans, generating a “dispersal pattern” (Renner, 2004). Many plant and animal groups dispersed from South to North America (or the opposite direction) starting in the Eocene, such as Malpighiaceae (Davis et al., 2002), Hedyosmum (Antonelli & Sanmartín, 2011) and Rubiaceae (Antonelli et al., 2009).
The sister group relationship between the Neotropical D. mollis and Malagasy D. antaly is quite unusual, however there are a few examples of sister clades occurring in Neotropical region and Africa or Madagascar in ferns, such as Leucotrichum (Polypodiaceae) (for more examples see Rouhan et al., 2012). Fern spores are efficient dispersal agents and greatly facilitated trans-atlantic dispersals and colonization (Rouhan et al., 2012). In Solanaceae, the genus Tsoala also dispersed from South America to Madagascar, probably by long distance dispersal facilitated by sea currents (Olmstead, 2013). In this family, most fruits are fleshy, animal dispersed, but many are dry and could have been dispersed by wind and sea currents, and dispersal occurred in both cases, but more frequently in the fleshy fruited lineages (Olmstead, 2013; Dupin et al., 2017). Dioscorea antaly could have dispersed from South America to Madagascar (or in the opposite direction) facilitated by its anemocoric dispersion syndrome, and specially by the shape of the seed, that is winged towards the base of the capsule, being more effective in high speed winds (Maurin et al., 2016). In gramitid ferns and Tsoala the long distance dispersal event occurred from the Neotropics towards Madagascar. Stem anatomy results indicated the proximity between D. mollis and other Neotropical species (Tenorio et al., 2017), possibly indicating that the direction of dispersal could have been from the Neotropical Region to Madagascar. Further phylogenetic analysis including more Neotropical species morphologically similar to D. mollis could further test this relationship hypothesis and clarify a scenario of dispersal between Neotropical Region and Madagascar.
ACKNOWLEDGEMENTS
We would like to thank the Coordenação de Aperfeiçoamento de Pessoal de Nível Superior (Capes) for the research grants of RSC and ACM. ECS thanks Conselho Nacional de Desenvolvimento Científico e Tecnológico (CNPq) for Bolsa de Produtividade em Pesquisa CNPq-Nível 2 (processo 311001/2014-9). We would like to acknowledge Sylvain Razafimandimbison, for the contribuitions concerning his knowledge of Madagascar and its biogeography, Anamaria Dal Molin, for the revision of the English, Gabriel Rezende, for the preparation and revision of the figures, and the anonymous reviewers for they contribution on greatly improving the discussion of this paper.