Taxonomic Status, Phylogenetic Affinities and Genetic Diversity of a Presumed Extinct Genus, Paraisometrum W.T. Wang (Gesneriaceae) from the Karst Regions of Southwest China

Background The karst regions in South China have an abundance of endemic plants that face high extinction risks. The Chinese Gesneriaceae endemic Paraisometrum mileense ( = Oreocharis mileensis), was presumed extinct for 100 years. After its re-discovery, the species has become one of five key plants selected by the Chinese forestry government to establish a new conservation category for plants with extremely small populations. For conservation purposes, we studied the phylogenetic and population genetic status of P. mileense at the three only known localities in Guangxi, Guizhou and Yunnan. Methodology/Principal Findings We collected 64 samples (52 species) of Oreocharis and 8 samples from three provinces of P. mileense and generated molecular phylogenies, and inferred that P. mileense represents a relatively isolated and derived taxonomic unit within Oreocharis. Phylogeographic results of 104 samples of 12 populations of P. mileense indicated that the populations in Yunnan have derived from those in Guangxi and Guizhou. Based on AFLP data, the populations were found to harbor low levels of genetic diversity (He = 0.118), with no apparent gradient across the species’ range, a restricted gene flow and significant isolation-by-distance with limited genetic differentiation among the populations across the three provinces (F ST = 0.207, P<0.001). The 10 populations in Yunnan were found to represent two distinct lineages residing at different altitudes and distances from villages. Conclusion/Significance The low levels of genetic diversity found in P. mileense are perhaps a consequence of severe bottlenecks in the recent past. The distribution of the genetic diversity suggests that all populations are significant for conservation. Current in situ and ex situ measures are discussed. Further conservation actions are apparently needed to fully safeguard this conservation flagship species. Our work provides a model of an integrated study for the numerous endemic species in the karst regions with extremely small populations.


Introduction
Paraisometrum mileense W.T.Wang was until relatively recently regarded as extinct in the wild because it had not been recollected for more than 100 years since its type specimen collection in 1906 [1,2]. In 2006, the plant was rediscovered in Shilin county, next to Mile county, Yunnan, where the type specimen was collected [1,3]. It was described as critically endangered (CR), possessing only 101-1000 individuals in a single population [4]. Interestingly, already in 2009, a second locality had been discovered in Longlin county in Guangxi, the province neighbouring Yunnan to the East [5], and soon afterwards, a third location, in Xingyi, in Guizhou province to the North of Guangxi, had been found [6] (Fig. 1), tripling the number of occurrence points. The total number of plants was estimated at .30,000 individuals [7], but up to 2010, detailed fieldwork by some of the authors (YMS and WHC) estimated the number to be significantly lower, with 630 mature plants in Yunnan,150 in Guangxi, and 60 in Guizhou (Table 1). Incidentally, approximately 70-80% of the seedlings and young individuals died, and 50-60% of mature individuals sustained damage due to an extreme drought that occurred from 2010 to 2011 in the karst region in Yunnan where P. mileense grows (Fig. 2). This might partially explain the discrepancy in reported plant numbers, and illustrates the great vulnerability of the species to even short term climate fluctuations. Until recently, only the occurrence point in Guangxi was located in a provincial nature reserve (established in 2005), while the other two were outside protected areas. In 2011, however, the Chinese government set up a small reserve in Shilin, Yunnan, to protect the species at this primary rediscovery point (Fig. 2). The species is also one of five selected key plants to be used to establish a new category of protected areas in China for plant species with extremely small populations (PSESP) [4].
Until 2011, Paraisometrum W.T.Wang was regarded as a monotypic genus, but was then included in an enlarged genus Oreocharis Benth. as O. mileensis (W.T.Wang) Mich.Möller & A.Weber [8]. Irrespective of this recent inevitable taxonomic change, the species represents a highly threatened taxon. Furthermore, under its old name, it has received considerable attention as a strong flagship species for plant conservation in China, and we therefore use its original name here in this context, for consistency with current conservation initiatives, such as the establishment of the above new conservation category in China [4].
To be able to devise meaningful conservation strategies, knowledge of the taxonomic status, closest congeners and the level and distribution of genetic diversity within a taxon is essential [9][10][11]. Even though P. mileense has been included in phylogenetic analyses previously, its exact phylogenetic affinities in Oreocharis are still unclear. While it was clearly shown in the phylogenetic analyses that the species has evolved from within the enlarged Oreocharis, it fell on a polytomy with species of the hitherto genera Ancylostemon Craib, Briggsia Craib, Isometrum Craib, Opithandra B.L.Burtt and Tremacron Craib [8,12,13]. The generic characters to establish Paraisometrum are the presence of four upper corolla lobes, and one lower lobe of the pentamerous flowers [2]. Since this characteristic also occurs in other species of the newly defined Oreocharis (e.g. O. saxatilis (Hemsl.) Mich.
Möller & A.Weber = Ancylostemon saxatilis Hemsl.) [14][15][16][17], the present work will address the species delineation of P. mileense. Furthermore, the plants in Guangxi and Guizhou appear to possess some floral features different from those in Yunnan (pers. obs. YMS and WHC). However, only two samples of P. mileense, one from Yunnan and one from Guangxi, were included in the most comprehensive study of Oreocharis to date [12]. This is not enough to test the taxon coherence in the light of modern approaches such as DNA barcoding [18][19][20]. Clearly, more molecular work was needed to include samples from all three occurrence points and from diverse species within the enlarged Oreocharis. Figure 1. Map of localities of Paraisometrum mileense in Yunnan, Guangxi and Guizhou. A. Pie charts for haplotypes (above) and nucleotypes (below) are given for the three main geographical distribution areas. B. Median-joining networks of chloroplast haplotypes based on trnLF and matK sequences (above) and of ITS nucleotypes (below) for P. mileense samples rooted on its closest relative Ancylostemon hekouensis. doi:10.1371/journal.pone.0107967.g001 There are many Gesneriaceae species endemic to the karst region in China, typically with small populations [7], such as Primulina tabacum Hance. This species occurs in only four populations distributed in Guangdong and Hunan with less than 1,000 plants in each [21]. Despite the relatively low plant number, a surprisingly high genetic diversity was found within the populations. Additionally, because of the long inter-population distances, a disruption of gene flow resulted in high population differentiation [10]. From a conservation perspective, the situation in P. tabacum was seen as 'a window of opportunity' to preserve a high level of extant genetic variation in the species. To determine whether a similar situation is present in P. mileense, we generated population genetic data using amplified fragment length polymorphisms (AFLPs) for individuals from the three localities in Guangxi, Guizhou and Yunnan. AFLP is a powerful tool for generating data from multiple loci for the detection of genetic variation without the need for pre-existing knowledge of genomic sequences [22,23], and have been successfully applied to small and relict populations (e.g. [10,[24][25][26][27]), and a range of suitable analytical packages are available for these dominant markers (e.g. [27]).
Thus, our main aims were fourfold, to phylogenetically analyse sequence data, from the chloroplast intron-spacer sequences of trnLF, and the nuclear ribosomal internal transcribed spacer regions (nrITS), to a) test the taxonomic status of the species, b) to determine the phylogenetic position and relationships of P. mileense within the enlarged Oreocharis, c) to reconstruct the phylogeographic history of the P. mileense populations using trnLF, matK, and ITS sequence data for individuals from 12 populations (or sub-populations) from the only three localities in Guangxi, Guizhou and Yunnan, and to acquire and analyse AFLP data for population samples to d) determine the levels of genetic diversity and differentiation of this species to assess its conservation requirements.

Ethics Statement
Of all sampled species, the target species Paraisometrum mileense is listed as one of five selected key plants to be used to establish a new category of protected areas in China for plant species with extremely small populations (PSESP). Approvals and permission for field studies were obtained from the Yunnan Forestry Bureau, China (permit no. [2011]-115). The GPS data are sensitive and are not provided for the protection of the plants according to the Yunnan forestry administrative request in the permit. However, we gave their approximate location in Fig. 1.

Study Materials
For the phylogenetic study, we expanded the ingroup sampling from 38 Oreocharis species plus one variety (41 samples) [12] to 52 species plus 2 varieties (73 samples), among which 11 species covered at least two population samples each. Of P. mileense, we included three samples each from Guangxi and Guizhou and two samples from Yunnan (Table S1). Thus, the Oreocharis ingroup sampling covered about 2/3 of the genus (53 out of about 80 species) [12]. The outgroup included 14 samples across six genera of straight-fruited advanced Asiatic and Malesian Gesneriaceae of subtribe Didymocarpinae (sensu [28]), with two Didymocarpus Wall. samples to root the trees [8,12].
The Paraisometrum materials for the phylogeographic analyses included 43 individuals from Yunnan (from 10 populations or subpopulations), 15 from the Guangxi population and 15 from the Guizhou population, and for the AFLP analysis 68 individuals Table 1. Detailed locality information for the 12 populations of Paraisometrum mileense used for AFLP analysis.  Table 1).

Molecular Methods
Genomic DNA was extracted from silica-gel-dried leaves collected in the field following a modified CTAB method [29,30]. The PCR primers for ITS were 'ITS5' and 'ITS4' [31], for trnLF 'C' and 'F' [32], and for matK '3F' and '1R' [30]. PCR was performed in reactions containing 30-50 ng genomic DNA, 0.3 ml of each primer (5 mM/ml), 10 ml 26Taq PCR MasterMix (Tiangen Biotech Co., Ltd, Beijing China: 0.1 U Taq polymerase/ ml, 0.5 nM of each dNTP, 20 mM Tris-HCl (pH 8.3), 100 mM KCl, 3 mM MgCl 2 ) and ddH 2 O to make up 20 ml. PCR amplifications were conducted under the following profile: 95uC for 3 min followed by 35 cycles at 94uC for 30 s, at the annealing temperature specific for each primer pair for 30 s (ITS: 55uC; trnLF and matK: 52uC), 72uC for 1 min, and a final extension step at 72uC for 5 min. After PCR amplicon purification, Sanger sequencing was carried out in 6 ml reactions containing 0.5 ml PCR product, 0.3 ml primer (5 mM/ml), 1.05 ml SeqBuffer, 0.3 ml BigDye Terminator Mix (Applied Biosystems, Foster City, USA) and 3.85 ml distilled H 2 O. Sequencing reactions were cycled under the following conditions: 32 cycles at 96uC for 10 s, 50uC for 5 s, and 60uC for 3 min, and the products analysed on an ABI 37306l sequencer (Applied Biosystems, Foster City, USA). The newly acquired sequences have been submitted to GenBank.
AFLP was performed based on Vos et al. [23]. Restriction endonuclease enzyme digestion and link reactions were performed in 20 ml reactions containing 4 ml template DNA (50 ng/ml), 1 ml Adaptor, 2 ml PstI/MseI, 2.5 ml 106 reaction buffer, 2.5 ml ATP (10 mM), 1 ml T4 Ligase, 7 ml distilled H 2 O. The solution was centrifuged for a few minutes after stirring, and incubated at 37uC for 5 h, then at 8uC for 4 h and then at 4uC overnight. Preamplification reactions were performed containing 2 ml template DNA, 1 ml Pre-ampmix, 0.5 ml dNTPs (Table 2), 2.5 ml 106 PCR buffer, 0.5 ml Taq polymerase, and 18.5 ml ddH 2 O. PCR amplifications were conducted with the profile: 94uC for 2 min, followed by 30 cycles at 94uC for 30 s, 56uC for 30 s, 72uC for 80 s, with a final extension at 72uC for 5 min, and terminated at 4uC. Eight selective amplification primers were used for each PstI and MseI ( Table 2). The selective amplification reactions were performed in 25 ml reactions containing 2 ml 1:20 diluted preamplification product, 2.5 ml 106 PCR buffer, 0.5 ml dNTP, 1 ml PstI primer, 1 ml MseI primer, 0.5 ml Taq polymerase, and 17.5 ml ddH 2 O. PCR amplifications were conducted as follows: 94uC for 30 s, 65uC for 30 s, 72uC for 80 s for the first cycle, followed by 12 cycles with progressively decreasing annealing temperature by 0.7uC each cycle starting from 65uC to 56.6uC, which was followed by 23 cycles of 94uC for 30 s, 55uC for 30 s, 72uC for 80 s, with a final extension step at 72uC for 5 min, and then cooled to 4uC. The AFLP primers, Taq polymerase and dNTPs for PCR were purchased from Dingguochangsheng Biotechnology Co. Ltd. (Beijing, China). The amplified fragments were separated and detected with an ABI Prism 377 sequencer (Applied Biosystems, Foster City, USA). Due to ambiguous banding patterns or obvious PCR failures, 19 additional samples (GX11a-s) collected exclusively from Guangxi province were included ( Table 1). The scoring error rate was about 1.3%, determined by replicating AFLP runs for 16 individuals (ca. 15% of all samples analysed) (cf. [33]).

Phylogenetic Analyses
The phylogenetic analyses were conducted on a matrix containing 86 ITS and trnLF sequences. ITS and trnLF sequences for 37 samples were newly acquired (Table S1), and for 49 samples retrieved from GenBank. The newly acquired DNA sequences were assembled and trimmed in Sequencher 4.1.4 (Gene Codes, Ann Arbor, MI, USA), and added to, and manually aligned with, the existing matrices. Maximum parsimony (MP) and Bayesian inference (BI) analyses followed Möller et al. [8,12] and Weber et al. [34], using PAUP* 4.0b10 [35] and MrBayes 3.2.2 [36,37], respectively.
The partition-homogeneity test in PAUP* (incongruence length difference test [38,39], on 1,000 replicates of repartitioning with tree bisection-reconnection (TBR) indicated no significant incongruence between the two datasets (P = 0.28) and the ITS and trnLF data sets were analysed combined. Parsimony was implemented on unordered and unweighted characters and through heuristic tree searches on 10,000 random starting trees with both TBR swapping and MulTrees on. Clade support was obtained as bootstrap indices with 10,000 heuristic replicates of random additions with TBR on and MulTrees off.
The most suitable substitution model for the BI analysis was obtained under the AIC criterion in MrModeltest [40], and was GTR+G for both, trnLF and the ITS spacers, and SYM+G for the 5.8S gene. Two independent runs of four MCMC chains of two million generations were run, sampled each 1,000 th generation. The first 200 trees (10%) were discarded as burn-in (generations prior to stationarity of likelihood values), and the posterior probabilities (PP) obtained from 50% majority rule consensus trees obtained using the 'sumt' command in MrBayes. A low value of the average standard deviation of split frequencies (0.009025), the high correlation of the PP support values and the parallel distribution of variance values found between the two parallel runs of the Bayesian analysis, and the level runs of the split posteriors indicated a good convergence of the MCMC runs (Table S2; Figs. S1-S3).

Population Genetic Analyses
The ABI trace files were analysed in GeneScan 3.1 (Applied Biosystems), and only intensive bands between 70 and 500 base pairs in size converted to a binary matrix (1 = band presence, 0 = band absence) (Table S3). Genetic diversity parameters were obtained in GenAlEx 6.5 [41], and included percentage of polymorphic loci (P), number of different alleles (Na), number of effective alleles (Ne), Shannon's Information index (SI), expected heterozygosity (He) and unbiased expected heterozygosity (uHe). To investigate for patterns among the Paraisometrum populations, we used STRUCTURE 2.3.4. [42] to assign individuals to genotypically distinct groups. We used the admixture model and Table 2. Name and DNA sequences of primers and adaptors used in the AFLP experiments on Paraisometrum mileense samples. the option for correlated allele frequencies as recommended by Falush et al. [43]. The program was run 14 times for each cluster (K) from K1 to K10. Each run of 100,000 iterations was preceded by 10,000 iterations as burn-in when convergence was achieved. We plotted the mean likelihood for each cluster L(K) against the cluster number (K). To establish the optimal number of clusters, the relationship between K and DK, the second order rate of change of the likelihoods, was plotted [44]. The distribution of genetic variation within and between populations and between regions was analysed in a hierarchical AMOVA in GenAlEx.
To test an isolation-by-distance scenario, a Mantel test was performed on the geographic and ln geographic distance versus Nei's genetic distance (D) in GenAlEx. For a 3D illustration of spatial relationships of the populations, a principal coordinate analysis (PCoA) was performed using the Jaccard distance in Rpack Le Progiciel R.4.0d10 [45]. To illustrate the genetic relationships between the populations, an unrooted Neighbor Joining (NJ) tree was reconstructed using Nei and Li's restriction site distances in PAUP*. Branch support was obtained from 1000 NJ bootstrap replicates in PAUP*.

Phylogeny
The MP analysis recovered 12 most parsimonious trees of 1538 steps length with a CI of 0.5566 and RI of 0.6982. The genus Oreocharis in its new definition formed a strongly supported clade in the strict consensus tree (BS = 100%). Paraisometrum was nested deeply within this clade in a derived position with the P. mileense samples forming a strongly supported clade (BS = 100%) (Fig. S4). Sister to this clade were two samples of A. hekouensis, though with no branch support.
The BI trees showed the same phylogenetic positions with view to the Paraisometrum and A. hekouensis samples (Fig. 3), the sisterrelationship of the two clades received only low support (PP = 0.63). The BI tree was more resolved both within the enlarged Oreocharis and within the Paraisometrum clade. Paraisometrum was nested deeply in a derived position within the clade dominated by species with yellow tetrandrous flowers (Fig. 3). The populations from Guizhou and Guangxi appeared intermixed in two groups not reflecting their origins (PP = 0.62), while the samples from Yunnan were most similar to each other (PP = 1.0), and somewhat distant to those from the other provinces.

Phylogeography
Among the 1676 bases in the combined trnLF and matK matrix, three positions were variable among the Paraisometrum samples, resulting in three chloroplast haplotypes ( Fig. 1; Table 3). Haplotype I was the only one in Yunnan and present in the other two provinces, while haplotype II was unique to Guangxi, and haplotype III was a private haplotype in the population from Guizhou (Fig. 1A). The median-joining network, rooted on A. hekouensis, placed haplotype I in a central position with haplotypes II and III as peripherals (Fig. 1B).
Among the 643 bases of the ITS region, eleven positions were variable, giving six different ITS nucleotypes among the samples with types 1 and 2 exclusive to Yunnan, type 4 and 5 shared among the Guangxi and Guizhou populations, while type 3 was private in Guangxi, and type 6 private in Guizhou ( Fig. 1A; Table 3). The median-joined network placed type 4 present in Guangxi and Guizhou in the centre, from which first types only present in these two provinces have evolved; the ITS types 1 and 2 present in Yunnan were in a derived position (Fig. 1B).

Genetic Diversity
At the population level, the AFLP data suggested a low level of genetic diversity of the populations, particularly in Yunnan (Table 4). This is likely a result of the low number of individuals included per populations. At the region level, the data suggested that the Yunnan populations harboured an overall higher level of diversity. For example, the percentage of polymorphic loci was 90.88% in Yunnan, compared to 61.18% in Guangxi and 66.99% in Guizhou. Other genetic diversity indices were also slightly higher, such as the Shannon's Information index (Yunnan: 0.247, Guangxi: 0.219, Guizhou: 0.235), though the number of effective alleles and heterozygosity indices were very similar in the three provinces.
The optimal number of clusters for the STRUCTURE analysis was determined as K = 4 ( Fig. 4 A and B). The analysis revealed a strong separation between the Yunnan populations 1-4 and 5-10 (Fig. 4C). The Guangxi population (Pop 11, Gx) was very homogeneous and distinct. The Guizhou population (Pop 12, Gz) was a mix of two types, one shared with the Guangxi population and the other individuals similar to a great degree to samples of the Yunnan populations 8 and 9. With K = 3, the patterns did not change greatly, except that the latter genotypes were clustering with the populations 1-4 from Yunnan. With K = 5, no further change in clustering was observed, the fifth genetic cluster scattering in small proportions among all populations. The Mantel test was significant using the geographic distance (r = 0.398; P = 0.01) and ln geographic distance (r = 0.383; P = 0.03).
The PCoA (1 st axis 12% variance; 2 nd axis = 7.7%; 3 rd axis = 6.6%) separated the samples of the three regions strongly in the first axis, including the Yunnan populations 1-4 and 5-10 ( Fig. 5A). Samples from Guangxi clustered very tightly together, while those from Guizhou and Yunnan were widely scattered. The samples from Guizhou were also scattered and almost overlapped with individuals from Guangxi at one end of their distribution, and with those from Yunnan, populations 5-10, at the other (Fig. 5A).
In the unrooted NJ tree, the samples clustered into four groups according to their province of origin including the split between Yunnan populations 1-4 and 5-10 (Fig. 5B). The Guangxi and Yunnan (populations 1-4) clusters received a high bootstrap support of 93%. These clusters were also characterized by short and uniform branch lengths. The same was the case for the cluster of Yunnan populations 5-10, except for three individuals, Yu8e, Yu9c and Yu9e. These had extremely long branches and were the same samples that seemed to be categorized as belonging to the Guizhou population in the STRUCTURE analysis (see Fig. 4C). The Guizhou population cluster was characterized by a loose clustering and relatively uniformly long branches (Fig. 5B).

Taxonomic Status and Phylogenetic Affinities
The present study aimed to address several important issues surrounding this enigmatic, and until recently, monotypic genus that was thought to be extinct for 100 years. It was recently reported to occur only in one population in Shilin, Yunnan [4], but we were able to sample from two further populations, from two different provinces. Since all Paraisometrum samples in the phylogenetic study fell in a single strongly supported clade separated from other Oreocharis taxa by a long branch, they can be regarded as a single taxonomic unit (Figs. 3, 4). Unlike previous phylogenetic analysis where relationships of this genus to other congeners were unresolved [8,12,13], we found indications that the closest species to Paraisometrum mileense is Ancylostemon hekouensis, a recently described species endemic to the karst region in Southwest China [46], with a similar leaf and flower morphology to Paraisometrum, but without the distinct 4-lobed upper corolla lip. The species with the most similar corolla shape to P. mileense, Ancylostemon saxatilis Hemsl., was found not closely related to P. mileense (Figs. 3, 4). This morphological homoplasy, is in line with previous findings of high levels of parallelism in the evolution of floral morphology in Oreocharis [12].

Phylogeographic History
Both, haplotype and nucleotype distribution, suggest a scenario of a close relationship between the Guangxi and Guizhou populations (shared nucleotypes), and a derived Yunnan population (lower haplo-and nucleotype diversity, derived nucleotypes) (Fig. 1, Table 3), and suggests a migration of Paraisometrum westward into Yunnan [9].
The significance in the Mantel test on the AFLP data indicates an isolation-by-distance scenario [47], and suggests a limited dispersal across the landscape. This may be a consequence of Paraisometrum being pollinated by bees, which have a relatively limited flying distance (e.g. [48,49]), and the seeds of Paraisometrum, that, though small (ca. 0.6 mm long), are not dust-like, as in orchids [50,51], to be dispersed far by wind, and do not have special dispersal aids, e.g. hooks or appendages (e.g. [52]) for long distance dispersal [53]. The significant genetic structure detected among the populations and regions (Tables 5, 6), further indicates a breakdown of their genetic connectivity. The ITS data may be used to give an indication of the divergence time between the Paraisometrum localities. Taking the average rate of evolution of the ITS spacers of 11 herbaceous plants (4.13610 29 substitutions per site per year) [54,55], the Yunnan populations might have separated from the other populations around 1.6 million years ago (60.26 SE), at the beginning of the Pleistocene. This is a period that would cover repeated glacial-interglacial cycles during which limited secondary contact may have occurred (see below). This could explain the higher levels of within-population genetic variation observed here.
Our study provides a first insight into the history of a plant considered extinct in the wild. To fully address the evolution of the species, and to untangle historic from contemporary events, an increased sampling is required, additional markers employed such as microsatellites, combined with population demographic analyses (e.g. [56,57]), and, with view to conservation, ecological niche modelling (e.g. [58]).

Genetic Diversity and Differentiation
The STRUCTURE analysis indicated several noteworthy aspects of the Paraisometrum populations; firstly, that the Yunnan populations represented two distinct lineages with at least three gene pools and can be divided into two main populations, populations 1-4, and populations 5-10 with some admixture (Fig. 4). This bipartition of the Yunnan populations was also seen in the PCoA clustering analysis (Figs. 5A, S5) and the NJ tree (Fig. 5B). Both groups have the same haplotypes which indicates Table 3. Haplotypes and nucleotypes found among the samples of Paraisometrum mileense.   Phylogenetics and Genetics of Paraisometrum W.T. Wang their common ancestry, and the same composition of ITS nucleotypes (Table 3), which suggests fragmentation of a previously more continuously distributed population. Populations 1-4 occur near villages, in the foothills of Mt. Guishan in secondary open forests with frequent disturbance due to human activities, while populations 5-10 grow mostly undisturbed further away from villages, above the foothills almost to the summit of Mt.
Guishan, and in more dense and primary forests (Fig. 2B, Table 1). Whether the bipartition is linked to the condition of the habitats surrounding the population groups, being disconnected from each other by disturbed forest, or has an older origin, would require further research. Secondly, the population in Guangxi was distinct and very homogeneous, also seen in the tight clustering in the PCoA (Fig. 5A) and the uniformly short branches  in NJ tree (Fig. 5B). This might indicate that while the plants contain similar genetic diversity levels as other regions, the plants were closer related to each other possibly due to consanguineous matings, perhaps as a side effect of their small population sizes and limited area of distribution of less than a hectare (Table 1), and long distance to the other distribution localities with no apparent intervening populations known (Fig. 1). Thirdly, on the contrary, the internal and terminal branches of the population from Guizhou were long, and the STRUCTURE analysis suggested that some plants had a genetic makeup greatly similar to plants of the Yunnan populations 8 (plant Yu8e) and 9 in particular (plants Yu9c, Yu9e). This was not reflected in a mixed clustering of the respective Yunnan samples among the Guizhou samples, though their branches were unusually long (Fig. 5B). This might reflect genetic links and allele exchanges between these populations, perhaps due to habitat expansion during glacial-interglacial cycles that brought the populations in closer proximity (e.g. [58]). They are currently only some 120 km apart (Fig. 1). The genetic diversity found at the population level was very low, although this might have been a consequence of the low number of individuals (often 5) included for analysis of the Yunnan populations. However, even when calculated within the three regions, the AFLP diversity levels were relatively low, with no marked difference between the regions (Table 4). Even at the species level, the genetic diversity of Paraisometrum was comparatively low (He = 0.144), and much lower compared to Primulina tabacum (He = 0.339), another herbaceous Gesneriaceae species from China [11]. Primulina tabacum and P. mileense have similar distribution sizes though in different provinces (P. tabacum in Guangdong and Hunan, Paraisometrum in Guangxi, Guizhou and Yunnan), but both occur in limestone karst areas, though at different altitudes (P. tabacum below 300 m, Paraisometrum between 1,180-2,500 m) and in different aspects (P. tabacum grows around entrances of limestone caves, Paraisometrum in limestone forests). The high genetic diversity levels found in P. tabacum populations were explained by their refugial history and/ or breeding system [11]. The more restricted habitats in cave entrances of P. tabacum, as opposed to the more open forest habitats of Paraisometrum could be a factor that may have allowed a more continuous distribution and genetic connectivity of Paraisometrum populations. However, it is more likely that the greater population sizes of P. tabacum (1,000 individuals per population, as opposed to 30-150 individuals in Paraisometrum populations) has allowed P. tabacum to retain higher diversity levels. The similarly low levels of AFLP diversity compared to Paraisometrum found in other herbaceous perennials, such as Trollius europaeus L. (He = 0.158-0.229) [24], Silene otitis St.-Lag. (He = 0.167-0.240) [25], and Draba aizoides L. (He = 0.07-0.15) [26], were linked to effects of population fragmentation. Thus, the low genetic diversity in Paraisometrum is likely a result of a combination of fragmentation and small population size. Intriguingly though, in Draba L., the low genetic diversity levels were not correlated with a limited reproductive fitness, as suggested by their high germination rates. This may have important conservation implications and could be a field of study in Paraisometrum in the future.

Implications for Conservation
Our work provides an example of an integrated study for endemic species in the karst regions in South China with extremely small populations. It is well known that the karst region is characterized by a limestone topography with an abundance of endemic plants with extremely small populations [7,59,60]. In these regions, localised endemics occur often restricted to one or few limestone hills, which are usually isolated by non-limestone topography and the plants effectively occur on islands with high risks of extinction [7,14]. The three distribution points of Paraisometrum are isolated in different limestone forests, and the occurrence nearby villages is not unusual for endemic plants of the karst region [59,61].
The villages are commonly closely associated with characteristic forests on the limestone hills, the Feng Shui Forests. These are important for the water supply of the villages, but are recently influenced by activities of local people, such as free range poultry and goat keeping [62], which negatively affect the forest habitats.
With a view to conservation of Paraisometrum, the only three distribution points known to date appear to contain a significant amount of diversity within their populations with only a limited level of differentiation. Though, analysing a larger number of individuals per population might show some more resolution. Furthermore, the STRUCTURE analysis shows that they seem to have experienced some gene flow between Yunnan and Guizhou populations and are thus not genetically greatly isolated (Table 1). However, the calculated levels of genetic diversity were quite low, even when estimated across the regions including Yunnan (and not based on the small 'sub-populations' there). This is likely linked to the relatively small population sizes. The low genetic diversity might hinder adaptation to rapid climatic changes, as shown in the exceptionally dry season of 2011 during which they contracted significantly, due to the death of many immature individuals, and hardly any flower was produced in the Yunnan populations, their inflorescences withered (SYM, pers. observation). The relatively uniform distribution of genetic diversity in the three provinces of Paraisometrum does not immediately allow a prioritization of regions or populations for conservation efforts. Even though little differences in the genetic make-up were detected between the two Yunnan lineages, the one including populations 1-4, may deserve stronger protection due to their close proximity to human habitation and the negative effects these can bring (Fig. 2). For all of the reasons given above, conservation efforts should thus include all populations and include in situ as well as ex situ measures to safeguard this enigmatic species.
In situ measures relevant to Paraisometrum have already been implemented in Yunnan on a small scale at Shilin populations and involved the removal of Ageratina adenophora (Spreng.) R. M. King & H. Robinson, an invasive plant alien to China [63], the exclusion of goats through fencing in the plants, and by involving the local villagers to safeguard these plants that have a high potential for eco-tourism. Future in situ measures may also involve forest restoration [64], particularly for populations near villages. However, these may not be sufficient and ex situ conservation measures are required alongside in situ measures in the light of the observed strong detrimental effects of recent rainfall variation on the populations. This may, for the moment preclude transfer of plants from threatened habitats to new localities in the wild. Though, the plants are easily cultivated and propagated vegetatively and through seeds. Currently, about 56 plants are cultivated at the Kunming Botanical Garden, Yunnan, China, and seeds of three populations have been deposited at the Germplasm Bank of Wild Species, Plant Germplasm and Genomics Center, Kunming Institute of Botany, Yunnan, China. Thus, the immediate and medium term survival of the species seems ensured. However, seeds in germplasm banks are prone to genetic erosion during seed rejuvenation cycles even of inbreeding species [65]. Currently research is underway studying the pollinators to elucidate the reproductive strategy and breeding system to better devise a tailored ex situ strategy for Paraisometrum mileense.

Conclusions
The erstwhile monotypic genus Paraisometrum mileense (now Oreocharis mileensis) represents an isolated and independent taxonomic unit, and appears closest related to the erstwhile Ancylostemon hekouensis (now Oreocharis hekouensis). The populations in Yunnan seem to have derived from those in Guangxi and Guizhou with some evidence of limited gene transfer between Yunnan and Guizhou. The occurrence points in Yunnan were divided into two quite distinct lineages, while the population in Guangxi is relatively homogeneous. Overall, the populations of P. mileense contain relatively low levels of genetic diversity with no apparent gradient across the species' range. Several conservation measures are currently implemented, but additional actions are needed. From a conservation genetic perspective, all populations seem equally important for in situ and ex situ protection.