Phylogenetic analyses of Begonia sect. Coelocentrum and allied limestone species of China shed light on the evolution of Sino-Vietnamese karst flora

The picturesque limestone karsts across the Sino-Vietnamese border are renowned biodiversity hotspot, distinguished for extremely high endemism of calciphilous plants restricted to caves and cave-like microhabitats that have functioned as biological refugia on the otherwise harsh habitats. To understand evolutionary mechanisms underlying the splendid limestone flora, dated phylogeny is reconstructed for Asian Begonia, a species-rich genus on limestone substrates represented by no less than 60 species in southern China, using DNA sequences of nrITS and chloroplast rpL16 intron. The sampling includes 94 Begonia species encompassing most major Asian clades with a special emphasized on Chinese species. Except for two tuberous deciduous species and a species with upright stems, a majority of Sino-Vietnamese limestone Begonia (SVLB), including sect. Coelocentrum (19 species sampled) and five species of sect. Diploclinium, Leprosae, and Petermannia, are rhizomatous and grouped in a strongly supported and yet internally poorly resolved clade (Clade SVLB), suggesting a single evolutionary origin of the adaptation to limestone substrates by rhizomatous species, subsequent species radiation, and a strong tendency to retain their ancestral niche. Divergence-time estimates indicate a late Miocene diversification of Clade SVLB, coinciding with the onset of the East Asian monsoon and the period of extensive karstification in the area. Based on our phylogenetic study, Begonia sect. Coelocentrum is recircumscribed and expanded to include other members of the Clade SVLB (sect. Diploclinium: B. cavaleriei, B. pulvinifera, and B. wangii; sect. Leprosae: B. cylindrica and B. leprosa; sect. Petermannia: B. sinofloribunda). Because species of Clade SVLB have strong niche conservatism to retain in their ancestral habitats in cave-like microhabitats and Begonia are generally poor dispersers prone to diversify allopatrically, we propose that extensive and continuous karstification of the Sino-Vietnamese limestone region facilitated by the onset of East Asian monsoon since the late Miocene has been the major driving force for species accumulation via geographic isolation in Clade SVLB. Morphologically species of Clade SVLB differ mainly in vegetative traits without apparent adaptive value, suggesting that limestone Begonia radiation is better characterized as non-adaptive, an underappreciated speciation mode crucial for rapid species accumulations in organisms of low vagility and strong niche conservatism.


Background
Karsts are distinctive landforms and hydrological systems formed by the dissolution of highly soluble and porous bedrock (e.g., carbonate rocks such as limestone) with geological ages ranging from the Cambrian to the Quaternary (Ford and Williams 2007). Around the world, carbonate rocks cover ca. 11% of world's land surface; however, with the expansion of subterranean hydrological systems, the karst realm makes up more than 14% of the global land areas (Williams 2008). With more than 800,000 km 2 , Southeast Asia and southern China contain the most extensive limestone karsts on earth (Gillieson 2005). In particular, the vast karst terrain stretching across the Sino-Vietnamese bordering region (southern China [Guangxi, western Guangdong, southern Guizhou, and southeastern Yunnan] and northern Vietnam; Xu et al. 2012a), renowned for spectacular landscapes of fengcongs (cone karst; Waltham 2008), fenglins (tower karst; Waltham 2008), and caves, is the largest area on earth with pure carbonate bedrock (Xu 1995) and has been considered as the model for karst studies (Sweeting 1978).
Behind these picturesque landscapes, limestone karsts of SE Asia and southern China also abound in rich and marvelous flora with strikingly morphological variation and exceedingly high endemism (Xu 1995;Clements et al. 2006;Zhu 2007). In China, 61 of the ca. 250 Chinese endemic plant genera occur solely in Guangxi (Qin and Liu 2010), a majority of which are associated with the limestone ecosystem. Of the ca. 470 species (58 genera) of Gesneriaceae recorded from China, at least 200 species (41 genera) are found in Guangxi, with most species confined to limestone karsts. Among them, ca. 100 species and 10 genera are endemic to Guangxi (Qin and Liu 2010). Of the 18 Begonia L. species from Sabah, Borneo associated with limestone substrates, 12 (67%) are only known from a single locality (Kiew 2001). In Kuching (Sarawak, Boreno), all 15 limestone Begonia species are endemic to the local hills only (Kiew 2004).
Despite their remarkable biodiversity, however, a great proportion of karsts remain to be explored and very little is known regarding the evolutionary processes that have generated the marvelous floras on these limestone landscapes (Clements et al. 2006). Since Begonia is one of the best represented plant groups on tropical limestone karsts (Clements et al. 2006), it provides an important model for understanding the role that limestone karsts have played in generating high species diversity and endemism.
With recent estimates ranging from ca. 1500 (e.g., Goodall-Copestake et al. 2010; Thomas et al. 2011a) to more than 1600 species (Hoover et al. 2004), the pantropically distributed Begonia sits firmly as one of the ten most species-rich flowering plant genera (Frodin 2004). However, despite their almost ubiquitous presence in the tropical and subtropical forest ecosystems across Africa, America, and Asia, with few exceptions (e.g., B. grandis Dryand., B. longifolia Blume, and B. palmata D. Don; Gu et al. 2007), most Begonia species have very narrow distribution ranges (Tebbitt et al. 2006;Hughes and Hollingsworth 2008) and single-site endemic species are common, especially in the limestone karsts (e.g., Kiew 2001;Peng et al. 2008b;Dewitte et al. 2011).
Begonia taxonomy is notoriously challenging because of the genus' enormous size and often poorly preserved morphological characters in herbarium specimens (Hoover et al. 2004;Hughes and Girmansyah 2011), generating great obstacles for further studies. To deal with the unwieldy number of species in the genus, sectional classifications have long been adopted (Doorenbos et al. 1998;Ku 1999;Shui et al. 2002;Kiew 2005;Gu 2007). Traditionally these infrageneric classifications have relied primarily on morphological and anatomical characters of fruits and ovaries, with each of the ca. 70 recognized sections confined to one continent only (Doorenbos et al. 1998), with one known exception (B. afromigrata J.J. de Wilde, sect. Tetraphila A. DC.; de Wilde et al. 2011).
Morphologically well circumscribed by its parietal placentation and rhizomatous perennation (Shui et al. 2002;Gu et al. 2007), Begonia sect. Coelocentrum is one of the most characteristic limestone plants confined to cave-like microhabitats (i.e., caves, crevices, and fissures) of the Sino-Vietnamese karst region (Peng et al. 2008a;Qin and Liu 2010), with most species known from a single or a few localities and differing from one another by leaf shape, pubescence, texture, and variegation ). In the past decade, explorations and taxonomic studies in the region have more than tripled the number of species in this section from 15 (Ku 1999) to over 50 Liu et al. 2007;Ku et al. 2008;Peng et al. 2007Peng et al. , 2008aPeng et al. , b, 2012Peng et al. , 2013Averyanov and Nguyen 2012;), representing a remarkable example of species radiations across the Sino-Vietnamese limestone karsts comparable to those on oceanic islands and tropical high mountains (Rundell and Price 2009).
Given the sectional classification schemes, it is tempting to speculate that Begonia should have adapted to the limestone substrates at least seven times in China. However, Chinese Begonia species, especially those found on limestone substrate, were very poorly sampled in recent phylogenetic analyses (Tebbitt et al. 2006;Rajbhandary et al. 2011;Thomas et al. 2011aThomas et al. , 2012. For example, despite its high diversity, only two species of sect. Coelocentrum were sampled in the studies by Thomas et al. (2011aThomas et al. ( , 2012 and its relationships to other limestone Begonia in the region remain to be explored. Additionally, although sect. Diploclinium and Platycentrum, both major representatives of Chinese Begonia, have been proven non-monophyly (Thomas et al. 2011a), it remains unclear as to what extent the infrageneric taxonomic framework of Chinese Begonia has to be revised, and what mechanisms could have been responsible for the high Begonia diversity on the Sino-Vietnamese limestone karsts.
This article studies phylogenetic relationships of limestone Begonia of Sino-Vietnamese karsts and concurrently evaluate the monophyly of major Begonia sections present in China. To gain further insights into the history of Begonia diversification, molecular divergence time estimates are also performed. This study focuses on continental Asian Begonia species, complementing recent studies by Thomas et al. (2011aThomas et al. ( , 2012) that emphasized the evolutionary patterns of SE Asian Begonia.

Taxon sampling
Our sampling strategy aimed to increase the number of species of limestone Begonia of continental China not represented in previous molecular phylogenetic analyses, while maximizing representatives of all major clades in Asia (Thomas et al. 2011a). In total, 94 Asian species Symbegonia (SYM), and B. boisiana Gagnep., which is assigned to sect. Petermannia by Kiew (2007) but placed as unassignable (UA) to existing section in Doorenbos et al. (1998) and Hughes and Pullan (2007). Two South African species were chosen as outgroups based on previous phylogenetic analyses (Goodall-Copestake et al. 2009Thomas et al. 2011a). Of the 96 species sampled, 49 species (51%) have never been investigated using molecular data. The majority of the sampled species were field-collected plants maintained in the experimental greenhouse of the Academic Sinica, Taipei, Taiwan. Voucher information and GenBank accession numbers are detailed in Additional file 1.

Molecular data
DNA of Begonia species is difficult to extract, especially for those collected from limestone substrates. Additionally, silica-gel dried leaves of Begonia tend to yield poor quality DNA. To circumvent these issues, total genomic DNA was extracted from fresh leaves of living collections using a modified CTAB protocol optimized for Begonia (Kopperud and Einset 1995).
Two DNA sequence regions were used: the nuclear ribosomal DNA (nrDNA) internal transcribed spacer (ITS) region and the chloroplast DNA rpL16 intron. Previous analyses based on ITS (e.g., Tebbitt et al. 2006) suggested its usefulness in resolving interspecific relationships while chloroplast sequences were more suitable for reconstructing the backbone structure of the Asian Begonia phylogeny (Thomas et al. 2011a). Although the rpL16 intron has thus far never been used in reconstructing the Begonia phylogeny, this region amplified and sequenced easily and shows adequate phylogenetic resolution in Asian Begonia. For each polymerase chain reaction (PCR), amplification was performed in a total volume of 25 μl, including 12.5 μl of Taq DNA Polymerase Master Mix Red (Ampliqon, Copenhagen, Denmark), 1 μl of each forward and reverse primer (10 μM), 2 μl of template DNA, and 8.5 μl of ddH 2 O. For ITS, the primers 5P and 26S1Rev primers (Clement et al. 2004) were used for both PCR amplification and sequencing. For rpL16 intron, three primers (rpL16-F: GCT ATG CTT AGT GTG TGA CTC G; rpL16-R: CGT CCY GCT TCT ATT TGT CTA G; Beg_rpL16: GTT TCA CAT TAT CTG GAT CG) were designed to optimize PCR amplification (rpL16-F and rpL16-R) and sequencing (Beg_rpL16 and rpL16-R) in Begonia. PCR reactions were carried out by a denaturation-step in 94°C for 5 min, 30 thermo-cycles of 94°C for 30 s, 60°C for 30 s, and 72°C for 90 s (60 s for ITS), and a final extension in 72°C for 5 min. PCR products were purified using QIAquick PCR purification Kit (Qiagen, Valencia, California, U.S.A.) and then sequenced using an ABI PRISM dye terminator cycle sequencer, model 3700 (Applied Biosystems, Foster City, California, U.S.A.).

Phylogenetic analyses
DNA sequences were aligned using the program MUSCLE implemented in the software MEGA5 (Tamura et al. 2011) with subsequent manual adjustments. The aligned matrix was analyzed using maximum parsimony (MP) and maximum likelihood (ML) optimality criteria and Bayesian Inference (BI). MP analyses were performed using PAUP* v.4.0b10 (Swofford 2002). All characters were unordered and equally weighted and gaps were treated as missing. Heuristic searches were conducted with 1000 replicates of random addition, 10 trees held at each step during stepwise addition, tree-bisection-reconnection (TBR) branch-swapping, and deepest descent option in effect. Brach supports (parsimony bootstrap: PB) were assessed by 1000 bootstrap replicates with 10 random taxon additions and heuristic search options. To assess the congruence among the two data sets, the incongruence length difference test (ILD test for partition-homogeneity test in PAUP*) was performed. One thousand homogeneity replicates of heuristic searches were conducted with 100 random sequence additions and MulTrees not in effect. Ten ML analyses with 5000 bootstrap (likelihood bootstrap: LB) resamplings were conducted using RAxML-HPC2 v7.2.7-3 (Stamatakis et al. 2008) via the CIPRES Portals (http://www.phylo.org/index.php/portal/). The matrix was partitioned (i.e., ITS vs. rpL16 intron) with a gamma model of rate heterogeneity and proportion of invariable sites estimated by the program. The program MrBayes v3.1.2 (Ronquist and Huelsenbeck 2003) was employed for BI analyses. The aligned matrix was partitioned into ITS and rpL16 data sets. Prior to the analysis, DNA substitution models for each partition were selected by MrModeltest (Nylander 2004) based on the Akaike information Criterion (AIC). BI analyses were conducted using mixed models specified to each data partition. Four runs of Metropolis-coupled Markov chain Monte Carlo (MCMCMC) analyses were performed, with a random starting tree and four chains for each run (one cold and three heated). The MCMCMC length was ten million generations and the chain was sampled every 1000th generation from the cold chain. To assure that the cold chain swapped successfully with the heated chains and the Metropolis coupling functioned well, the resulting log files were by inspecting the trace plots of parameters using the program Tracer v1.5 (Drummond and Rambaut 2007). The function 'Compare and Cumulative' of the program AWTY (Nylander et al. 2008) was employed to check topological convergence and convergence of posterior probabilities (PP) of splits from different runs. Bayesian clade posterior probabilities and average branch lengths were calculated based on the sampled trees after the first 25% of the sampled trees were discarded as burn-in.

Molecular dating
The software package BEAST v1.6.1 (Drummond and Rambaut 2007) was employed to perform divergence-time estimates. The aligned matrix was formatted for BEAST using BEAUti with the data partitioned to allow specification of DNA substitution models for each partition (see BI analysis above). Because suitable fossils in Begoniaceae are not available, the phylogenetic tree was calibrated using two normally distributed priors based on divergence time estimated by Thomas et al. (2012) for Asian Begonia: the mean of estimated divergence ages of the crown node of Asian Begonia (Clade 10 in Figure three of Thomas et al. 2012; the node marked with a rhombus sign in Figure 1 (Rubite et al. 2013). For these two clades, the calibration priors were modeled with the mean of 16.1 mya and the standard deviation of 3.265 (95% probability interval: 9.7-22.5 mya) for the Asian Begonia Clade and the mean of 13.0 mya and the standard deviation of 2.7 (95% probability interval: 7.7-18.3 mya) for Clade 49 to account for HPDs of age estimates of the original analysis. The tree prior was set to the Yule process and a single uncorrelated-rates relaxed molecular clock model assuming a lognormal distribution of rates (UCLD) was applied to all partitions. Two independent runs of BEAST analyses were conducted, each with 50 million generations and sampling every 1,000th generation.

Results
Information of sequence variability in each partition is summarized in Table 1. The alignment file is available upon request to the corresponding author (CIP). The combined data matrix included 1974 aligned positions (ITS: 864 bp; rpL16: 1110 bp), of which 29% of the aligned sequences (31.6% in ITS and 26.9% in rpL16) were excluded because of uncertain alignment. MP analyses yielded 149503 equally parsimonious trees of 3480 steps (CI = 0.4086; RI = 0.5542; RC = 0.2264). For BI analyses, AIC selected the model GTR+G+I for ITS and GTR+G for rpL16.
The result of ILD test (P = 0.001) indicates significant incongruence between ITS and the rpL16 data sets. However, because ILD test has been doubted for generating misleading conclusions under certain circumstances (Yoder et al. 2001), topologies of MP analysis of ITS, rpL16, and the concatenated data sets were compared (Wiens 1998). While relationships generated from ITS and the combined data sets are highly congruent, topology of rpL16 is largely incongruent with the formers (data not shown). Nevertheless, supports for the conflict relationships between ITS (and the concatenated) and rpL16 topologies are all less than 50% and the combined data set has much higher support values for most clades, suggesting a 'soft incongruence' (Wiens 1998). Therefore, the two data sets were combined for all subsequent analyses.
For MP and ML analyses, 70-79%, 80-90%, and 90-100% bootstrap supports (LB and PB) are considered as moderate, well, and strongly supported relationships, respectively. For BI analyses, the threshold for a well-supported clade is 95% PP. The best scored ML tree (ML optimization likelihood = −20967.197060), with bootstrap support values (PB and LB) and the posterior clade probabilities (PP), is depicted in Figure 1. Clades designated in Thomas et al. (2011a) and sectional classification are also marked to the right of the ML tree in Figure 1 to allow easier cross-comparisons between studies.
Topology of the best-scoring ML tree is largely compatible with those uncovered in recent phylogenetic analyses of similar scale (Rajbhandary et al. 2011;Thomas et al. 2011a), though relationships among deeper nodes are very poorly supported (Figure 1). In the best-scoring ML tree, all sampled Asian Begonia species form a moderately supported clade (LB: 74) with strong support in BI analysis (PP: 0.97). Within the Asian Clade, the Indian species B. dipetala Graham (sect. Haagea) is sister to a clade composed of all the remaining Asian species, though lacking support (Figure 1). Within this latter clade, species are grouped with low supports into two clades (Clades C and D; Figure 1) corresponding to Clade C1 + C2 and Clade D in Thomas et al. (2011a).
(See figure on previous page.) Figure 1 Best-scoring maximum likelihood phylogram. Clade support values (LB: likelihood bootstrap/PB: parsimony bootstrap/PP: posterior probability) larger than 50% are indicated at each node. Dashed branches indicate LB, PB, and PP all smaller than 50%/0.5 while thick branches denote those present in the strict consensus tree of MP analysis and PP ≥0.95. Species in bold denote limestone species distributed in the Sino-Vietnamese limestone karsts . Arrows point to clades and sections discussed in the text. Taxon names are followed by sectional placement and distribution (in parentheses). The rhombus (◇) and star (☆) signs denote calibration points for molecular dating. Sectional classification and clade names in Thomas et al. (2011a)   Within Clade C, eight sampled species of tuberous and deciduous habit form a poorly-supported successive grade (Figure 1) sister to a moderately to strongly supported Clade PLA-SPH (LB: 75, PB: 78, PP: 1). This grade, corresponding well (except for sect. Parvibegonia) to the DIP I grade in Thomas et al. (2011a) Liang & X.X. Chen), and two species of sect. Diploclinium (B. cavaleriei H. Lév. and B. pulvinifera C. I Peng & Yan Liu). Within the strongly supported Clade SVLB, phylogenetic resolutions are low and weakly supported (Figure 1).
The maximum clade credibility chronogram and the summary of the divergence time estimates and clade supports (PP) generated by the BEAST software are shown in Figure 2 and Table 2 The crown age of Clade 38, corresponding to Clade D of Thomas et al. (2011a) and Clade D49 of Thomas et al. (2012) that includes sect. Coelocentrum, Petermannia, and Baryandra (= DIP II), was estimated to be 12.64 mya, with the HPD date ranging from 8.54 to 16.67 mya. For Clade SVLB the mean divergent crown age was estimated to be 8.06 mya, with the HPD date ranging from 5.09 to 11.21 mya ( Figure 2; Table 2).

Discussion
Although relationships among deep nodes of Asian Begonia remain poorly supported in our current study, the basic topology and major lineages identified (Figure 1) are largely congruent with previous studies (Tebbitt et al. 2006;Rajbhandary et al. 2011;Thomas et al. 2011aThomas et al. , 2012 with some exceptions. For instance, our ML tree (Figure 1) indicates a sister relationship between the Indian species B. dipetala (sect. Haagea) and the rest of Asian Begonia, while B. dipetala was placed sister to Clades B + C1 + C2 with weak support in Thomas et al. (2011a). With the exclusion of B. sinofloribunda, sect. Petermannia (Clade PET) of Clade D is monophyletic with strong support ( Figure 1); however, in Thomas et al. (2011a) this section is polyphyletic, separated into two strongly clades (Clade PET I and PET II) with Clade PET I basal to the subclade encompassing Clades PET II, BRA (sect. Bracteibegonia A. DC.), DIP II (= Clade BAR in Figure 1), REI, and B. kingiana (sect. Ridleyella). It is worth to mention, however, that Clade BAR is the basal clade of the Platycentrum-Sphanenthera clade in Tebbitt et al. (2006). Additionally while Clade SVLB is placed sister to sect. Petermannia in present study (Figure 1), sect. Coelocentrum is the basal clade sister to the rest of Clade D in Thomas et al. (2011a). However, except for the monophyly of Clade PET (PET I + PET II) in present study, supports for relationships of these deep nodes are low, indicating that they are mostly soft incongruence (Wiens 1998). Additionally conflicts between our and previous studies could also be attributable to the limited phylogenetic signals of the molecular markers employed in the current analyses. Alternatively, past hybridization events could have also led to the incongruence between the chloroplast-based relationships in Thomas et al. (2011a) and those unveiled by our study using the combination of ITS and chloroplast markers (Goodall-Copestake et al. 2009;Dewitte et al. 2011;Thomas et al. 2011a), as reports on interspecific hybridizations have become increasingly common in Asian Begonia (e.g., Peng et al. 2010). Because a full discussion of phylogenetic relationships of Asian Begonia is beyond the scope of present study, our discussion is limited to well-supported clades pertaining to the evolution of limestone Begonia of China.

Taxonomic implications of the phylogenetic analyses
Based on a very different set of taxonomic sampling, our analyses corroborate results of previous analyses (Tebbitt et al. 2006;Thomas et al. 2011a) in confirming the non-monophyly of most major Asian sections, including sect. Diploclinium, Leprosae, Petermannia, Platycentrum, Reichenheimia, and Sphenanthera (Figure 1). Additionally, our expanded sampling of Chinese Begonia species further demonstrates the paraphyly of sect. Coelocentrum, dominating in the strongly supported Clade SVLB otherwise also composed of two species of sect. Diploclinium (B. cavaleriei and B. pulvinifera), two species of sect. Leprosae (i.e., B. cylindrica and B. leprosa), and one species of sect. Petermannia (B. sinofloribunda). Despite being one of the biggest (47 species) and morphologically most diverse sections in China, only three rhizomatous species of sect. Diploclinium (B. cavalerieri, B. pulvinifera, and B. wangii) are known from the Sino-Vietnamese limestone karsts Peng et al. 2006), none sampled in previous phylogenetic analyses. Our study thus further strengthens the polyphyly of sect. Diploclinium and the homoplasious nature of its diagnostic characters, i.e., bilamellate, 3-locular axillary placentation, in Asian Begonia (Thomas et al. 2011a).
Sect. Leprosae, raised for three species (B. cylindrica, B. leprosa and B. longicarpa) bearing cylindric and berry-like fruits in southern China (Shui et al. 2002), has been questioned to be artificial when B. bataiensis Kiew, a tuberous species bearing the characteristic fruit type from limestone karsts of southwestern corner of Vietnam described (Tam et al. 2005). Subsequently molecular study sampling two species (B. leprosa and B. longicarpa) showed that sect. Leprosae is polyphyletic with its type species B. leprosa (limestone species) placed sister to sect. Coelocentrum and B. longicarpa (not limestone species) nested within clade dominated by sect. Platycentrum and Sphenanthera, suggesting that the cylindric and berry-like fruits evolved independently (Tebbitt et al. 2006). With all four recognized species sampled, our analyses attested further the polyphyly of sect. Leprosae, with two limestone species (B. cylindrica and B. leprosa) in the Sino-Vietnamese karsts nested within Clade SVLB, B. longicarpa in Clade PLA-SPH, and B. bataienssi sister to B. kingiana (Figure 1).
Known only from hills of limestone karsts of the southern Guangxi adjacent to the Vietnamese border ), B. sinofloribunda, a rare species uniquely characterized by branched aerial, subwoody stems and pendulous inflorescences with small flowers and ovaries showing axillary placentas, has been variously classified in sect. Platycentrum (Gu 2007;Ku 1999), Diploclinium (Shui et al. 2002), and Petermannia (Shui and Chen 2004). Our analysis nevertheless indicates strongly that B. sinofloribunda is derived from within Clade SVLB (Figure 1), greatly increase the morphological diversity of Clade SVLB.
Based on our study, although species of Clade SVLB vary greatly in their ovary anatomy and fruit type that are crucial for sectional classification in Begonia, they are all rhizomatous distributed exclusively in cave-like microhabitats of the limestone karsts of the Sino-Vietnamese bordering areas. Additionally all available data, including an unpublished chromosome count of B. sinofloribunda (Peng, unpublished data), indicate that members of Clade SVLB are uniformly characterized by the chromosomal number 2n = 30 Peng et al. 2010;Thomas et al. 2011a) or its triploid number 2n = 45 . Given the obvious cohesiveness in their perennation organs as well as ecology, biogeography and chromosome cytology, we thus propose to expand the definition of sect. Coelocentrum (see Taxonomic Treatment) to include B. sinofloribunda, three species of sect. Diploclinium (B. cavaleriei, B. pulvinifera, and B. wangii; see Taxonomic Treatment) and two species of sect. Leprosae (B. cylindrica and B. leprosa).

Phylogenetics and molecular dating of Clade SVLB
Except for species of Clade SVLB, the only Begonia species found on the Sino-Vietnamese limestone karsts are the tuberous B. chingii Irmsch. and B. grandis  of the DIP I grade (Figure 1) and B. boisiana that possesses upright stems (Tebbitt 2005). The phylogenetic relationships therefore indicate that, in the Sino-Vietnamese limestone karsts, Begonia had evolved to adapt to the harsh limestone habitats four times, twice evolved with the tuberous and deciduous habit (or once and secondary invading the none-limestone habitat by B. labordei, Figure 1), (See figure on previous page.) Figure 2 Maximum clade credibility chronogram estimated by BEAST. The rhombus (◇) and star (☆) signs denote calibration points for molecular dating. Dashed clades indicate PP < 0.75 and thicken clades indicate PP ≥ 0.95. Node heights indicate mean ages, with 95% highest posterior density (HPD) date ranges shown by the node bars. Clades with PP ≥ 0.75 are numbered and their clade support and divergent ages mean (95% HPD) are shown in Table 2. once by the species with upright stem (B. boisiana), and once by the rhizomatous ancestor of Clade SVLB (Figure 1). Interestingly, although tubers and a deciduous habit appear to be suitable for surviving in the desiccationprone limestone habitats such as most karst Begonia in Madagascar (Grubb 2003;Keraudren-Aymonin 1983), our data strongly suggest that it is the rhizomatous Clade SVLB that succeeded and proliferated in the Sino-Vietnamese limestone karst terrains.
Within the strongly supported Clade SVLB, phylogenetic resolution is poor (Figure 1), possibly indicating a rapid species radiation (Hughes and Eastwood 2006;Kozak et al. 2006). Based on the divergence times calculated by BEAST, the crown age of Clade SVLB (Clade 35 in Figure 2; Table 2) was estimated to be ca. 8.06 (HPD 5.09-11.21) mya during the middle to late Miocene, with the stem age extending back to ca. 11 mya (Figure 2). Although these age estimates are likely  problematic because of the employment of secondary calibration, even within the lowest end (ca. 5 mya), the crown age of the Sino-Vietnamese limestone Begonia is much older than those estimated in recent phylogenetic literature such as species radiations triggered by the Pleistocene uplift of the South American Andes (e.g., Hughes and Eastwood 2006;Hoorn et al. 2010). Interestingly, the estimated crown age of Clade SVLB at 8.06 ) mya (Table 2; Figure 2) appears to coincide well with the most extensive stage of karstifications in Guangxi during the Miocene (Liu 1997(Liu , 2003 and the onset and intensification of the East Asian monsoon at ca. 7.2 (An 2000) to 11 mya (Zheng et al. 2004). In Asia, the climate has been affected strongly by the altitude of the Himalayas and the Tibetan Plateau. By the late Miocene, the coeval uplift of Himalaya-Tibetan region had attained significant heights that greatly altered Asian atmospheric circulation, resulting in the onset of the warm and humid East Asian summer monsoon . Because high temperature and precipitation brought by the summer monsoon are favorable conditions for the development of karsts (Zhang 1989), the late Miocene onset of the East Asian monsoon would have greatly facilitated the development of limestone karsts across the Sino-Vietnamese limestone regions (Liu 1997(Liu , 2003, resulting in the remarkable landscapes and providing ample habitats for the spreading and diversification of limestone karst plants such as Begonia. Recently the uplift of Himalaya-Tibetan region and intensification of East Asian monsoon have also been suggested to be responsible for triggering the rapid diversification of both seasonally and wet adapted Begonia in continental Asia (Rajbhandary et al. 2011) and the fern genus Lepisorus in China and Japan (Wang et al. 2012).

Diversity pattern of Clade SVLB
Species of the Clade SVLB, comprising ca. 60 species (see Taxonomic Treatment) mainly distributed in Guangxi, inhabit the 'twilight zone' (Poulson and White 1969) of limestone caves and cave-like microhabitats such as shaded fissures and crevices abundantly occurring throughout the Sino-Vietnamese limestone karst region (Sweeting 1978;Schindler 1982;Xu 1995;Zhu 2007). Such cave-like microhabitats, characterized by constant temperature, high humidity, and indirect and low light (Liang et al. 2010), function as shelters on the limestone bedrock characterized by highly alkaline conditions, thin soil layers, and severe desiccation due to high porosity (Clements et al. 2006). In addition to Begonia, these cave-like microhabitats are also inhabited by several species-rich, calciphilous herbaceous plant groups such as Aspidistra Ker Gawl. (e.g., Liu et al. 2011), a number of genera of gesnerids (e.g., Weber et al. 2011;Xu et al. 2012aXu et al. , 2012b, Elatostema L. (e.g., Wei et al. 2011), Impatiens (e.g., Yu et al. 2009), and the fern genus Polystichum (e.g., Zhang and He 2011), etc. Many of these limestone plants are often highly ornamental and have long been commercially exploited (Vermeulen and Whitten 1999). Biogeographically most of these calciphilous cave plant species have a very limited distribution ranges and site-endemics are very common (Liu et al. 2007;Peng et al. 2008aPeng et al. , 2012Xu et al. 2012a;Zhang and He 2011). In individual caves or crevices, species diversity is often low; however, across the landscapes, plant composition varies greatly from one cave to another, resulting in low α diversity and yet extremely high β diversity across the landscape (Hughes and Hollingsworth 2008).

Phylogenetic niche conservatism of Clade SVLB
The monophyly of the rhizomatous Clade SVLB suggests a single evolutionary origin of the adaptation to the cave-like limestone microhabitats, and shows that ecology is good indicator of taxonomic relationships than traditional morphological characters used to defined sections. Inspecting the chronogram ( Figure 2) reveals a 3 MY-time lag between the estimated stem and crown age of Clade SVLB, which might signify a long period needed for their ancestor to gradually adapt to the calcium-rich and cave-like microhabitats on the limestone karsts before the commencement of species radiation. Its monophyly and prolonged existence since the late Miocene also suggest the strong tendency in the evolution of this clade to retain the ancestral ecological niche, presenting an apparent case of strong 'phylogenetic niche conservatism' (Kozak et al. 2006;Wiens 2004).

Cave-like microhabitat as paleorefugia
Limestone caves and crevices/fissures are formed through the development of underground hydrological systems with maximum discharge and dissolutional aggressiveness (Palmer 1991). The development and 'collapse' of caves are essential processes shaping the karst topography (Ford and Williams 2007;Palmer 1991). In more recent histories, anthropogenic factors have also greatly accelerated the destruction of caves and cave-like microhabitats in the karsts (Clements et al. 2006). Together with the observations of the unique ecology of limestone Begonia, limestone caves and crevices currently inhabited by rich plant communities can be viewed as "biological refugia, habitats that support populations not able to live elsewhere in a landscape (Nekola 1999)". More precisely, these cave-like microhabitats in the Sino-Vietnamese limestone karsts qualify as "paleorefugia, now-fragmented relicts of a formerly widespread matrix community (Nekola 1999)". On the contrary, "neorefugia represent habitats that are younger than the surrounding biological matrix (Nekola 1999)".

Clade SVLB as an example of a non-adaptive radiation on limestone karsts
The onset of Asian summer monsoon in the late Miocene ) provided ideal conditions for the karst development (Zhang 1989) across the limestone terrains of the Sino-Vietnamese bordering region (Liu 1997), resulting in numerous crevices, fissures, and caves and the splendid fenglins and fengcongs across the landscapes. Although limestone karsts are an unsuitable growth substrate for most plants, over time, some plant groups (e.g., Aspidistra, Begonia, Impatiens, various gesnerid genera, Elatostema, etc.) adapted to and flourishing in the cave-like microhabitats. However, cave-like microhabitats are fragile and collapsed constantly due to the processes of karstification (Palmer 1991) associated with the persistence of East Asian monsoon Liu 1997), resulting in fragmentation and isolation of these microhabitats. Moreover, because of the strong niche conservatism (Wiens 2004) to the cave-like microhabitats, populations of the calciphilous plants had been forced to track the paleorefugia (Wiens 2004;Kozak et al. 2006;Comes et al. 2008). Moreover, because Begonia species are poor dispersers (Hughes and Hollingsworth 2008), gene flow among populations would be retarded, greatly augmenting effects of random genetic drift (Comes et al. 2008;Hughes and Hollingsworth 2008). As geological weathering processes have occurred continuously throughout the Sino-Vietnamese limestone karsts, population diversification likely would have resulted in steady accumulation of species (Kozak et al. 2006;Comes et al. 2008;Hughes and Hollingsworth 2008). However, because the major driving force of this evolutionary process is geographic isolation, the diversity pattern is better characterized as 'non-adaptive radiation', an underappreciated speciation mode crucial for rapid species accumulations in organisms of low vagility and strong niche conservatism (Gittenberger 1991;Kozak et al. 2006;Comes et al. 2008;Rundell and Price 2009). Morphologically, species of Clade SVLB all have very similar flowers, distinguishing from one another mainly by vegetative characters such as leaf morphology in shape, pubescence, texture, and variegation ). Recent studies, however, show that different variegations in Begonia leaves have no apparent adaptive differences (Sheue et al. 2012;Zhang et al. 2009), suggesting that the species accumulation in Begonia was mainly a non-adaptive process (Gittenberger 1991;Kozak et al. 2006;Comes et al. 2008;Rundell and Price 2009), though further studies are definitely needed to test the non-adaptive nature of leaf morphology in Clade SVLB.

Conclusions
The proposed scenario of species diversification in Clade SVLB driven by geological and climatic conditions in the Sino-Vietnamese limestone karst should have also affected other calciphilous cave plants in the region, resulting in concordant phylogenetic patterns across lineages. It is also worth noting that the biology of those calciphilous limestone karst plants appears to be highly similar to some amphibians (e.g., salamanders and frogs) whose populations mainly differentiate via isolation-by-distance (IBD) and allopatric speciation has been the primary mode responsible for species diversification because of their low vagility and strong niche conservatism in fragmented landscapes (Kozak et al. 2006;Fouquet et al. 2007). As such, a positive correlation between genetic and geographic distances is expected among populations of a species and IBD might still be observed among closely related and recently diversified species (Fouquet et al. 2007), resulting in geographically constrained monophyly in species (Fouquet et al. 2007;Hughes and Hollingsworth 2008;Dewitte et al. 2011). The predicted phylogenetic and population genetic patterns thus form testable hypotheses for further population genetic studies and better sampled Begonia phylogenetic analyses as well as other limestone cave plants in the Sino-Vietnamese limestone karst region and other major karst ecosystems worldwide.  Barkley and Baranov 1972).-Begonia sect. Leprosae (T.C. Ku) Y.M. Shui, Bot. Bull. Acad. Sin. 43: 321. 2002, syn. nov. Herbs monoecious, terrestrial, perennial, rhizomatous without erect stems and tubers, acaulescent, rhizome herbaceous or slightly woody. Leaves alternate, variable, mostly broadly ovate, oblique at base, rarely palmate, palmately compound, or occasionally peltate, glabrous to tomentose, often variegated. Inflorescence axillary, cymose (dichasial or dichasial at base and monochasial at apex), rarely thyrsoid, staminate flowers basal and carpellate flowers distal, protandrous. Perianth segments free, usually white or pink, rarely greenish or yellowish. Staminate flower with 4, or rarely 2 tepals. Carpellate flowers usually with 3 tepals, rarely 2 or 5 (-6); ovary 3-winged, or rarely unwinged, wings equal to unequal in fruit, mostly 1-locular, parietal placentation, rarely 3-locular with axile placentation. Fruit a 3-winged capsule, nodding, or rarely berry-like, cylindric.
Distribution and Habitat. Currently 61 species, two varieties, and one natural hybrid distributed in the caves, fissures and forest floors of limestone karsts in southern China (Guangxi, southern Guizhou, and southeastern