A New Classification of Ficus Subsection Urostigma (Moraceae) Based on Four Nuclear DNA Markers (ITS, ETS, G3pdh, and ncpGS), Morphology and Leaf Anatomy

Ficus subsection Urostigma as currently circumscribed contains 27 species, distributed in Africa, Asia, Australia and the Pacific, and is of key importance to understand the origin and evolution of Ficus and the fig-wasp mutualism. The species of subsection Urostigma are very variable in morphological characters and exhibit a wide range of often partly overlapping distributions, which makes identification often difficult. The systematic classification within and between this subsection and others is problematic, e.g., it is still unclear where to classify F. amplissima and F. rumphii. To clarify the circumscription of subsection Urostigma, a phylogenetic reconstruction based on four nuclear DNA markers (ITS, ETS, G3pdh, and ncpGS) combined with morphology and leaf anatomy is conducted. The phylogenetic tree based on the combined datasets shows that F. madagascariensis, a Madagascan species, is sister to the remainder of subsect. Urostigma. Ficus amplissima and F. rumphii, formerly constituting sect. Leucogyne, appear to be imbedded in subsect. Conosycea. The result of the phylogenetic analysis necessitates nomenclatural adjustments. A new classification of Ficus subsection Urostigma is presented along with the morphological and leaf anatomical apomorphies typical for the clades. Two new species are described ─ one in subsect. Urostigma, the other in Conosycea. One variety is raised to species level.


Introduction
Despite substantial effort, the origin and evolution of Ficus L. and the fig-wasp mutualism remain unclear due to lack of resolution of the backbone phylogeny of Ficus [1,2,3]. One of the key clades of uncertain placement is Ficus subsection Urostigma [4,5]. Ficus subg. Urostigma Leaf anatomy [16] appeared to show more consistent characters and less variation within species than the morphological characters previously studied [see identification key in 5] and, especially when combined with morphology, leaf anatomical characters provided a highly accurate tool for species recognition, enabling recognition of some of the morphologically highly variable species (e.g., F. virens). Leaf anatomical evidence also suggested that F. amplissima more closely resembles F. arnottiana (Miq.) (subsection Urostigma) Miq. than F. rumphii (former sect. Leucogyne). A result that contradicted the classification presented in [4,5].
Therefore, the main aims of this study are (1) to create a comprehensive phylogenetic hypothesis of subsect.Urostigma by analysing several molecular markers (ITS, ETS, G3pdh, and ncpGS) for almost all known species of subsect. Urostigma and related groups, and (2) to propose a new classification of subsect. Urostigma based on the resulting phylogenetic hypothesis.
Dried leaf samples from 37 herbarium collections and 26 leaf samples dried on silica gel were used for DNA extraction(for voucher information see S1 Appendix). The silica gel samples together with vouchers were collected in non-protected areas for the access of which no permits were needed (no specific permissions were required for these locations/activities and the field studies did not involve endangered or protected species); see Table 1 for localities. The species involved are non-CITES protected. DNA sequence data were sampled for four nuclear DNA markers (ITS, ETS, G3pdh, ncpGS). In total, 234 sequences were used in the analysis, including 199 new sequences and 35 sequences downloaded from GenBank. All new sequences are available from GenBank (S1 Appendix).

DNA extraction, amplification, and sequencing
About 20-50 mg of dried leaf tissue from each sample was used for extraction using the Qiagen DNeasy Plant Kit and following the manufacturer'sprotocol. We sequenced the nuclear encoded ITS, ETS, G3pdh and ncpGS regions following protocols in previous studies [1,2,3,17,18]. The primer sequences for all markers are shown in Table 2. The Polymerase chain reactions (PCR) were performed with 1μL of DNA product, 10 μL of Red-Sigma buffer (Qiagen Inc.), 2μL of each 10 μM primers(forwardand reverse), 0.4 μL of BSA (Promega, Madison, Wisconsin, USA) and 6.6 μL of H 2 O, in a total volume of 20 μL. The PCR programmes followed are summarised in Table 3.PCR fragments were checked for length and yield by gel electrophoresis on 2% agarose gels and cleaned using the Qiagen PCR clean-up kit before sequencing on an ABI 377 Genetic Analyzer according to the manufacturer's protocols (Applied Biosystems). Both strands were sequenced for each region for the majority of taxa.

DNA sequence alignments
Sequences were initially edited and improved by eye using CodonCode Aligner (CodonCode Corporation, Dedhem, USA) and MacClade 4.08 OSX [19], and both forward and reverse sequences were assembled. All assembled sequences were blasted via GenBank database to check for possible contamination with non-Ficus DNA. The alignment of whole sequences was done online with Phylogeny.fr, option MUSCLE [20], and SeaView 3.2 [21]. Gaps were treated as missing data and indels were excluded from the alignments, because they were not informative or only supported clades that already received high support. Missing markers were also coded as missing data.

Morphological and leaf anatomical data
The morphological data matrix was constructed using the most recent taxonomic revision of Ficus subsection Urostigma [5]. The specimens used in the revision were also the primary source for compiling the data matrix. In addition, specimens, stored in L, representing the species from other infrageneric taxa were also used to score data. In total, 43 qualitative morphological characters were coded for analysis (see S2 Appendix for characters, and S3 Appendix for the data matrix). The leaf anatomical data are based on recent work by Chantarasuwan et al. [16], to which the character states of non-subsect. Urostigma species were added, either studied (F. cf.rumphii) or extracted from Berg and Corner [7]. In total 23 qualitative characters were coded for analysis (see S2 Appendix for characters, and S3 Appendix for the data matrix). All characters were treated as unordered and of equal weight, missing data were coded as unknown. Characters 8,9,11,12,13,18,21,33,34,37,45, and 63 are in fact continuously Table 2. Sequences of primers used in this study. ITS  ITS_5F: 5´-GGA AGT AAA AGT CGT AAC AAG G-3´, ITS_4R: 5´-TCC TCC GCT TAT TGA TAT GC-3´,  ITS_17SE: 5´-ACG AAT TCA TGG TCC GGT GAA GTG TTC G-3´, ITS_26SE: 5´-TAG AAT TCC CCG  GTT CGC TCG CCG TTA C-3´[   38], [38], [17], [17] ETS G3pdh GPDX7F: 5´-GAT AGA TTT GGA ATT GTT GAG G-3´, GPDX9R: 5´-AAG CAA TTC CAG CCT TGG-3´ [18], [18] ncpGS GS_3F: 5´-GTT GTG ATT WAC CAT GCT-3´, GS_4R: 5´-AGA TTC AAA ATC GCC TTC-3´ [1], [1] Notes: The primer combinations ITS_5F plus ITS 4R and ITS17SE plus ITS26SE were used interchangingly with about equal success corresponding to standard protocols at C and CNRS. The combination of the Ficus specific internal primer ETS_Fig1_F plus 18S_ETS was only used for amplification of 13 accessions across the subsection, which could not be amplified with the standard primers.

Phylogenetic analysis
In total five analyses were made. The analyses of the four combined molecular DNA markers were performed with Maximum Parsimony (MP) and Bayesian Inference (BI) methods. The morphology and leaf anatomy dataset was analysed under Maximum Parsimony (MP). Both datasets, molecular and morphology/leaf anatomy, were subsequently combined (total evidence approach) and analysed under MP and BI. The MP analyses were run using PAUP Ã v4.0b10 [22] and heuristic searches with 3000 replicates,ten random taxon additions, tree-bisection-reconnection branch swapping (TBR), Mul-Trees option active, and no more than 10 trees saved per replicate. Branch support was performed in PAUP with bootstrap analyses [23] with 1000 replicates and all other settings similar to the phylogeny analysis. Bootstrap percentages(BS) are defined as high (85-100%), moderate (75-84%), low (50-74%) or no support (<50%).
Model selection for the Bayesian analysis was conducted using the model selection tool available through the online HIV sequence database site (http://www.hiv.lanl.gov/content/ sequence/findmodel/findmodel.html) checking all 28 models and constructing the initial tree with Weighbor (default) [22]. The chosen models were HKY+G for ITS, GTR+G for ETS, HrN +G for G3pdh, and HKY+G for ncpGS (JC for the morphological and anatomical data after the manual of MrBayes [24]). The datasets were analysed online using MrBayes v.3.1.2 [24] with 100,000,000 generations via the Cipres science gateway(http://www.phylo.org). The default values of 4 chains (3 heated, 1 cold, temperatures default) and two parallel runs were used, whereby every 1,000 th cladogram was sampled. A 10% burn-in was executed after Tracer 1.6 [25] was used for each tree file to check whether or not the effective sampling sizes (ESS) of all parameters exceeded 200, indicating that they are a good representation of the posterior distributions. The Potential Scale Reduction Factors (PSRF) in the MrBayes SUMP output were 1 or close to 1, which also indicates correct convergence. Bayesian inference produces posterior probabilities that are relatively higher than the corresponding bootstrap frequencies [26], thus we only used posterior probabilities (PP) above 0.9 as (high) support. TreeAnnotator v.1.8.0 (part of BEAST v.1.8.0 package [27,28]) was used to create a Maximum Clade Credibility (MCC) tree from every run. These did not differ in topology, only somewhat in support. The MCC tree of the first run was selected.
Mesquite v.2.7.5 [29] was used to show the changes in morphological and anatomical characters on the MCC tree from the Bayesian analysis of the combined datasets (see discussion for the preferred MCC tree, the molecular one or the one based on combined data one).

Nomenclature
The electronic version of this article in Portable Document Format (PDF) in a work with an ISSN or ISBN will represent a published work according to the International Code of Nomenclature for algae, fungi, and plants, and hence the new names contained in the electronic publication of a PLOS ONE article are effectively published under that Code from the electronic edition alone, so there is no longer any need to provide printed copies.
In addition, new names contained in this work have been submitted to IPNI, from where they will be made available to the Global Names Index. The IPNI LSIDs can be resolved and the associated information viewed through any standard web browser by appending the LSID contained in this publication to the prefix http://ipni.org/. The online version of this work is archived and available from the following digital repositories: PubMed Central, LOCKSS.

Analysis of DNA datasets
Seventy six taxa were included in the combined dataset with varying amplification success for the four DNA regions targeted as also found previously [1]: 74 taxa provided ITS data, 68 taxa ETS sequences, 53 taxa G3pdh sequences, and only38 taxa provided ncpGS sequences. MP analyses of the separate markers did not show major incongruences in topology, therefore all were united. The combined aligned data matrix was 2674 bp long with 472 potentially informative characters. The MP analysis resulted in 1300 most parsimonious trees (MPTs) with a length = 1636, consistency index (CI) = 0.68, and retention index (RI) = 0.78 (all characters included, informative and uninformative). The strict consensus tree (not shown) of 1300 most parsimonious trees (MPTs) contains two major clades.
The same two clades are present in the MCC tree (Fig 1) of the Bayesian analysis. Clade A comprises all members of subsect. Urostigma with a support of BS = 88 and PP = 1. Ficus madagascariensis is sister to the rest of this clade (high support, BS = 92 and PP = 1). Within clade A most internal nodes show low support, except for the higher support for most nodes that unite the various specimens of a species. Several of these species are not monophyletic, F. arnottiana, F. caulocarpa, F. prasinicarpa, F. virens are polyphyletic (different ancestral nodes included), and F. geniculata is paraphyletic (shared ancestral node, not all descending lineages included). Clade B (BS = 100, PP = 1) contains the members of sect. Americana, sect. Galoghycia, sect. Malvanthera, subsect. Conosycea, and F. rumphii of sect. Leucogyne.

Analysis of morphological and leaf anatomical data
A total of 43 morphological and 23 leaf anatomical characters were used. The MP analysis resulted in 1368 most parsimonious trees with a length = 280, CI = 0.25, and RI = 0.77 (including uninformative characters). The resulting strict consensus tree is a single extended polytomy (not shown).

Analysis of DNA markers combined with morphology and leaf anatomy
A total of 2740 characters, 2674 molecular (ITS, ETS, G3pdh, and ncpGS) and 66 morphological and leaf anatomical characters were used; of these 538 characters were parsimony informative. The MP analysis resulted in81 most parsimonious trees with a tree length = 1964 (including uninformative characters), CI = 0.5988, and RI = 0.7560 (strict consensus not shown).
Tracer [26] showed that all variables in the results of the BI analysis had an effective sampling size far above 200 (326-1851). The MCC tree is shown in Fig 2. The cladogram (Fig 2) shows the same two distinctive subclades as found in the analysis of the four combined DNA markers (Fig 1). Clade A (BS = 97, PP = 1) is composed of all species of subsect. Urostigma with F. madagascariensis as the first divergent lineage. Similar as with the molecular data analysis (Fig 1), relationships within the remainder of clade A are not well supported in the combined analysis. The species that are represented by several samples usually form monophyletic groups (with high support) except for F. caulocarpa, F. geniculata, F. prasinicarpa and F. virens. Ficus prasinicarpa is paraphyletic because of the inclusion of F. pseudoconcinna; the clade itself has low support (BS = 53, PP = 0.4), but F. prasinicarpa 2 and F. pseudoconcinna have high support (BS high = 87, PP high = 1). Ficus geniculata 3 groups with F. caulocarpa 2 and 3 and F. subpisocarpa Gagnep. subsp. pubipoda, but with very low support (BS<50, PP = 0.6).

Character mapping
The morphological and leaf anatomical character state changesare summarised in Fig 3. Subsect. Urostigma(clade A in Fig 3) is supported by the following apomorphies: intermittent growth(character 3, state 2; shared in parallel with F. rumphii of subsect. Conosycea, clade B),

Phylogenetic circumscription of Ficus subsect. Urostigma
Our results based on comprehensive sampling of subsection Urostigma are consistent with recent previous studies at the genus level supporting a narrow concept of subsect. Urostigma s.s. excluding former sect. Leucogyne [1,2,3,15]. Unfortunately the extraction of DNA from F. amplissima, the other species of sect. Leucogyne, was unsuccessful in our study, but a partial ITS sequence of F. amplissima(Rønsted, unpublished; specimen Matthew 20582 (K)) forms a clade together with F. rumphii embedded in the Conosycea clade. This is supported by evidence from the pollinators, because F. amplissima and F. rumphii are pollinated by the same wasp genus (Eupristina), a genus only known to be associated with species of subsect. Conosycea [7,30], which is indicative of co-evolution [1]. Based on these two independent pieces of evidence were classify F. amplissima in subsection Conosycea, which means that the complete sect. Leucosyce should now be synonymised with subsect. Conosycea. Corner [31] originally considered F. prolixa, a Polynesian species, to be related to the American hemi-epiphytic figs of sect. Americana, because of the scattered position of the staminate flowers in the fig. However, F. prolixa has three basal bracts and not two as in sect. Americana. Our phylogenetic results clearly show that there is no close relation between F. prolixa(clade A) and sect. Americana (clade B).
Relationships within subsection Urostigma s.s. are still not well supported based on four nuclear genes, morphology and leaf anatomy, and further work (e.g., with massive parallel sequencing) is needed before subdivision of the subsection. In the cladogram from the combined, total evidence approach (Fig 2), the species with multiple samples are more often grouped together than in the molecular phylogeny (Fig 1), (specimens F. arnottiana, F. prasinicarpa(paraphyletic) grouped together in Fig 2, both polyphyletic in Fig 1). Moreover, Fig 2 provides a much better historical biogeographic scenario than Fig 1  (not elaborated here); for instance the Madagascan and African taxa group are more grouped together and basal in Fig 2 than in Fig 1 (F. madagascariensis, F. cordata, F. densifolia, F. lecardii, F. salicifolia, F. verruculosa). In general, the support, especially in the terminal branches, is much higher in the total evidence approach (Fig 2) than in the molecular analysis (Fig 1). Based on these three reasons we prefer the results of the total evidence approach (Fig 2) above the results of the molecular data only (Fig 1). This conclusion supports the idea of Wiens [32] that morphology and leaf anatomy add valuable data to the phylogeny reconstruction when combined with molecular data.

Comparing the phylogeny with traditional classifications
To some degree, our phylogenetic results support the geographical implications of the classification made by Miquel [9], with the taxa arranged per continent (e.g., a group of African species separate from Asian species). However, there are a few exceptions. In our results (Fig 1) one African species, F. ingens, is placed among Asian species, and Sino-Himalayan F. hookeriana and F. orthoneura are among African species. Thus, a purely continental classification is not attainable. Corner [12,33] divided sect. Urostigma (similar to subsect. Urostigma here) of Asia and Australia into four series, Religiosae Miq., Superbae Corner, Caulobotryae (Miq.) Corner, and Orthoneurae Corner. However, species in the various series of Corner do not form monophyletic groups, but are mixed in our phylogenetic tree and the relationships among clades are not well supported. Moreover, Corner never included the African species, precluding direct comparison with his subdivision. Berg [4] re-classified sect. Urostigma and included African species, only recognising two subsections, Urostigma and Conosycea, and no series. Berg's classification compares well with ours and previous work [1,2,3,15] results of two clades, which cannot easily be subdivided into recognisable subgroups (low support for most branches and no distinct character combinations in Fig 3). Berg included F.amplissima and F. rumphii (formerly in Leucosyce) in subsect. Urostigma, which is not consistent with our results, which point at inclusion in subsect. Conosycea (see below).

Homoplasy in characters used or suitable for recognising subsect. Urostigma
The character mapping showed three unique apomorphies for the subsect. Urostigma clade (Fig 3), one morphological character (40.1: staminate flowers near ostiole), and two leaf anatomical characters (44.1: epidermis simple; 47.1: enlarged lithocysts only abaxially). Two morphological characters (3.1: intermittent growth present; 7.1: leaves deciduous) show parallel apomorphies in Conosycea, though the combination is unique. All characters were previously used for the recognition of subsect. Urostigmaby [4,5,7]. These resultsimply that the morphological data used here are not sufficient to separate both subsections, whereas the combination with leaf anatomy allows a distinct subsectional recognition.
Intermittent growth (char. 3, Fig 4A) was always the main character used to recognise subsect. Urostigma, but also occurs in parallel in F. amplissima and F. rumphii(subsect. Conosycea).Thus this character is homoplasious in our phylogeny and can only be used in combination with other characters to recognise subsect. Urostigma.
Deciduousness (char. 7, Fig 4B) is also homoplasious, and shows reversals in subsect. Urostigma: F. verruculosa is evergreen and F. religiosa are becomes evergreen when growing in wet areas. Moreover, several species of subsect. Conosycea are also deciduous. This character can respond to climatic conditions, either through phenotypic plasticity or through adaptive response over evolutionary time.
The character staminate flowers around the ostiole (char. 40, Fig 4C), the only typical morphological character, shows parallel reversals in F. arnottiana, F. hookeriana, F. orthoneura, F. prolixa, and F. virens 4 and 5. The character was used to recognise the subsection by different authors [4,5,7]. However, it may be that the character dispersed staminate flowers has evolved repeatedly within subsect. Urostigma in response to shifts from active to passive pollination.
Of the leaf anatomical characters, Corner [12] and Berg and Corner [7] used the enlarged lithocysts only on the abaxial surface (chr. 47, Fig 4D) as typical for subsect. Urostigma. However, the leaf anatomical work of Chantarasuwan et al. [16] revealed that F. arnottiana and F. virens 4 and 5 show enlarged lithocysts on both sides, which is similar to subsect. Conosycea. Thus, this character also is not unique for subsection Urostigma.
The articulation of the leaf(char. 4) only occurs in Asian and Australian species, for which it is a unique apomorphy within the Urostigma clade, but again there are reversals to absence in F. hookeriana and F. orthoneura (perhaps related to their non-deciduousness).
Because of the reclassification of the species of former sect. Leucogyne the recognition of subsect. Urostigma and subsect. Conosycea changes compared to [4] and [7].
Typical for subsect. Urostigma are: deciduous plants, intermittent growth, articulated leaves usually present, petioles relatively long (more than 1/4 th of lamina long), leaves with enlarged lithocysts generally abaxially, staminate flowers usually near the ostiole.
Typical for subsect. Conosyceaare: evergreen or deciduous plants, growth continuous, nonarticulated leaves, petioles relatively thick and short (less than 1/4 th of lamina long), enlarged lithocysts present at both sides of the leaf lamina, figs more frequently sessile than pedunculate, staminate flowers dispersed.

Non-monophyletic species within subsect. Urostigma
The sampled specimens of several species appear to be para-or polyphyletic in the results of our analysis: Ficus caulocarpa. Three specimens of F. caulocarpa var. caulocarpa were included in this study of which F. caulocarpa1 was separate in a clade with F. tsjakela with high PP support (Fig  2: PP = 1, BS = 79). The three specimens share many morphological characters, but F. caulocarpa 1 deviates in a few characters from F. caulocarpa2 and F. caulocarpa3 such as the stipule forming an ovoid terminal bud, the figs present on short spurs on the branches only, and the figs solitary or in pairs. Based on these differences F. caulocarpa 1 is described here as a separate species, F. pseudocaulocarpa (see below). However, in our phylogenetic analysis, the full genetic variation within F. caulocarpa is still not covered, because only samples with a narrow leaf form could be included.
Ficus geniculate. Four specimens of F. geniculata were analysed, three belong to F. geniculata var. geniculata and one to F. geniculata var. insignis. The three samples of var. geniculata are in different clades (paraphyletic), but var. insignis groups separately with F. virens 1, but with low support (Fig 2: PP = 0.8, BS = 52). Both varieties can be recognised at the species level, but because the support for the clades was low we refrain to make this decision until more molecular information becomes available.
Ficus geniculata var. geniculate. The two samples of F. geniculata var. geniculata (1 & 2) form a clade but with low support (Fig 2: PP = 0.8, BS = 68), while the other one (F. geniculata 3) forms a clade with F. caulocarpa and F. subpisocarpa subsp. pubipoda, also with low support (Fig 2: PP = 0.6, BS < 50). Because of the low support at the internal nodes, we refrain from changing the species concepts until more molecular information will be present.
Ficus prasinicarpa. The sample of F. prasinicarpa 1 forms a well-supported clade with F. pseudoconcinna (Fig 2: PP = 1, BS = 87). The two are sister to F. prasinicarpa 2, but with low support. Morphologically, the two specimens of F. prasinicarpa show a difference in the leaf apex (caudate versus acute to acuminate), but because of the low support for the clade we do not make any decision about possible cryptic species.
Ficus virens. Chantarasuwan et al. [5] recognised four varieties within the F. virens complex, var. virens, var. glabella, var. matthewii, and var. dispersa. Unfortunately, we only succeeded to amplify DNA sequences from two varieties (var. virens and var. glabella). Both varieties are separated in the resulting cladogram (Fig 2), and the five samples of var. virensare even polyphyletic. The clade of F. virens var. glabella has maximum support and its morphological circumscription is clear. Therefore, we will reinstate this taxon at the species level. We will maintain F. virens with three varieties, var. virens, var. dispersa, and var. matthewii. Ficus virens var. virens was represented by five samples in our analyses, which became divided into three groups (Figs 1 and 2), see above. Ficus virens 1 shows some morphological differences with F. virens 2-5, but the support is low (Fig 2: [35] accepted as synonym of F. virens. Therefore, we will reinstate F. wightiana.

Taxonomic Treatment
In this part we will officially make the changes in taxonomy on the basis of our phylogeny. Much of the nomenclature and descriptions can be found in Chantarasuwan et al. [5]. Ficus virens Aiton var. virens Corner [5]. Ficus virens Aiton var. dispersa Chantaras. [5].
Distribution and Habitat: Philippines. In lowland rain forest at altitude 60-80 m.
Distribution and Habitat: Thailand, on limestone in dwarf community, at elevation of c. 30 m.Figsin September-November.