High levels of genetic diversity and population structure in an endemic and rare species: implications for conservation

Petunia secreta is a rare and endemic species, that was found in two different landscapes, approximately 21 Km apart from each other. In this study we showed that P. secreta presented high genetic diversity that was equivalent to or even higher than that of widespread Petunia species. Two evolutionary lineages were found and they are correlated to the different landscapes where P. secreta grows: open areas in conglomerate sandstone towers at an elevation of approximately 300-400 m or along the road growing in an open vegetation flat area. Therefore the major risk to P. secreta maintenance is its rarity, suggesting the necessity of a preservation program.


Introduction
Many species are rare or can be rare at some point during their existence. Various species maintain this rarity over the course of their existence; thus, one question is how can rare species maintain their population sizes when demographic challenges appear? Ecological and genetic explanations to rarity have been suggested. Rabinowitz (1981) proposed that rarity should be considered in three ways: geographic range (wide or narrow), habitat specificity (broad or restricted) and local abundance (somewhere large or everywhere small). Further, species could be rare in different senses, having many forms of rarity. Only one possible combination (wide range, broad habitat specificity and somewhere large local abundance) is classified as common species and all other combinations are different forms of rarity.
The hypothesis of niche breadth is one of several suggested mechanisms to explain species commonness and rarity; species that maintain populations in more varied environments may have a wide geographic distribution (Brown 1984;Slatyer et al. 2013). Niche and geographic distribution can be correlated with genetic variability. Species that present small population sizes and a restricted geographic range are expected to have reduced levels of genetic diversity (Hamrick and Godt 1989) because of natural selection or demographic phenomena believed to be more pronounced in these populations (Ellstrand and Elam 1993;Gibson et al. 2008). Insufficient variability would lead species to restrict their geographic range and be more vulnerable to extinction under novel selection pressure (Sheth and Angert 2014). However, whether and how the population would be affected by demographic processes, particularly genetic drift, depend on gene flow within and among populations (Choo et al. 2012), mating systems, pollen and seed dispersal, and effective population sizes (Kettle et al. 2007). Small populations might also experience higher levels of inbreeding, which might maximize the effect of genetic drift (Oleas et al. 2014).
Studies that analyse congeneric species account for phylogenetic relatedness (Felsenstein 1985) and are ideal for comparing genetic diversity of species with wide and restricted geographic distributions. Comparisons of different congeneric rare and widespread species could yield different patterns-in many cases, rare species maintain equivalent (i.e. Gitzendanner and Soltis 2000) or even higher (i.e. Ellis et al. 2006) levels of genetic diversity than do widespread congeners. However, in other situations, rare species have significantly lower levels of genetic diversity (i.e. Gibson et al. 2008). The lack of a constant pattern makes it important to study each rare species in order to critically evaluate its genetic variability and the drivers that shape it.
The genus Petunia (Solanaceae) is known worldwide through commercial garden petunias and 8 of its 14 species have a narrow geographic distribution. This genus is endemic to South America, showing a subtropical distribution ranging from 228 to 398S (Stehmann et al. 2009). Several studies at the molecular level have been conducted involving Petunia wild species, and a common result of these analyses is the short genetic distances that are observed between taxa, indicating recent diversification of the genus (Ando et al. 2005;Kulcheski et al. 2006;Chen et al. 2007). As well, genetic variability in both plastid and nuclear markers is low (Kulcheski et al. 2006;Lorenz-Lemke et al. 2010;Reck-Kortmann et al. 2014).
Petunia secreta is one rare, annual and heliophilous herb that is bee-pollinated and easily recognizable by its purple and salverform corolla ( Fig. 1A -C). This taxon was collected for the first time in 1995 and described as a new species belonging to the Petunia genus in 2005 (Stehmann and Semir 2005). Petunia secreta is morphologically similar to P. axillaris axillaris and the corolla colour is the unique consistent trait distinguishing them (Stehmann and Semir 2005). Phylogenetic reconstructions suggest that P. secreta is the sister group of other taxa growing in the same geographic area [P. axillaris subsp. axillaris (hereafter P. axillaris; Fig. 1D -F and J) and P. exserta ( Fig. 1G -J); Kulcheski et al. 2006] and shares with them a long corolla tube (Reck-Kortmann et al. 2014). Whereas P. secreta and P. exserta are distributed in a small geographical region, P. axillaris is widespread (Turchetto et al. 2014a, b).
Petunia secreta was originally portrayed as endemic to a small area called Pedra do Segredo municipality of Caçapava do Sul, Serra do Sudeste, southern Brazil ( Fig. 2), growing in open areas in conglomerate sandstone towers at an elevation of 300 -400 m. However, whereas P. secreta is restricted to Pedra do Segredo and its neighbourhood, P. exserta also has an endemic distribution and is found in a locality referred to as Guaritas in a rock formation, disjoint from Pedra do Segredo by 30 km Segatto et al. 2014a). There are no records of P. secreta in Guaritas region. Petunia axillaris species complex is widely distributed in temperate South America and the axillaris subspecies can be found growing in Guaritas in sympatry with P. exserta until 15 km from Pedra do Segredo but has never been observed within the geographic range of P. secreta. Since its discovery, few individuals of P. secreta have been seen in nature, and although it appears to be extremely rare, this species is not federally listed. Moreover, very little is known about the biology of P. secreta, and its genetic diversity has never been evaluated. One major question is centred on whether the few extant individuals in nature are sufficient to maintain adequate levels of genetic diversity.
In addition to their particular geographic range, these species present different floral syndromes and reproductive systems. Flowers of P. secreta are self-compatible (SC), non-fragrant and bee-pollinated (Stehmann and Semir 2005). The white flowers of P. axillaris are sweet and strongly fragrant after dusk (Gü bitz et al. 2009) and mainly hawkmoth-pollinated (Ando et al. 1995;Venail et al. 2010;Klahre et al. 2011). However, P. axillaris individuals have more complex reproductive systems over the species geographic range. This subspecies was initially described as self-incompatible (Ando 1996), with observations of various SC individuals (Ando et al. 1998Kokubun et al. 2006) at the edge of distribution. Recently, a study showed a mixed mating system for the P. axillaris axillaris with a high proportion of SC individuals in the sympatric AoB PLANTS www.aobplants.oxfordjournals.org & The Authors 2016 3 area with P. exserta (Turchetto et al. 2015a). Petunia exserta is a SC species that exhibits red flowers, exserted reproductive organs and hummingbird-pollination Lorenz-Lemke et al. 2006). With all Petunia species, the seeds fall close to the mother plant (Stehmann et al. 2009) and the plastid genome is maternally inherited (Derepas and Dulieu 1992). The goals of this study were to (i) describe the genetic diversity of the rare P. secreta; (ii) compare the genetic diversity of P. secreta with that of other rare (P. exserta) and widespread (P. axillaris) congeners and (iii) identify potential treats owing to rarity that may limit P. secreta persistence. The results of the work presented here have contributed to the knowledge on the relationships between the levels of genetic diversity and the geographic range size in clades of the Petunia genus, which contains rare and widespread plant species that share several morphological traits.

Sample collection
We searched for P. secreta individuals in the Pedra do Segredo locality and nearby beginning in 2006. Every year, only a few individuals were observed, and overall, 50 individuals from 12 occurrence sites were localized (Pop1 -12; This work was conducted under permit MP 2.186/16 of the Brazilian Federal Government to access plant genetic information to develop evolutionary or taxonomic studies. No specific collection permits were required because P. secreta is not federally listed as endangered or protected and because no collection sites were based in protected areas.

DNA extraction, amplification, sequencing and genotyping
Young leaves of 73 individuals of P. secreta were collected, such that injury and damage to the plants would be minimized, then dried in silica gel. We pulverized the leaves in liquid nitrogen for DNA extraction with cetyltrimethylammonium bromide as described by Roy et al. (1992). The quality and quantity of genomic DNA was evaluated by measuring the absorbance at 260 and 280 nm on a Nanodrop Spectrophotometer (NanoDrop 1000 spectrometer, Thermo Scientific Corp., USA).
The non-coding plastid trnH-psbA and trnS-trnG intergenic spacers of 65 individuals of P. secreta were amplified using universal primers as described by Hamilton (1999) and Sang et al. (1997), respectively, and following the amplification conditions according to Lorenz-Lemke et al. (2006). The polymerase chain reaction products were purified according to Dun and Blattner (1987) and sequenced in a MegaBACE 1000 (GE Healthcare Bio Sciences Corp., Piscataway, NY, USA) automatic sequencer according to the manufacturer's instructions and the DYEnamicET Terminator Sequencing Premix Kit (GE Healthcare). The sequences were deposited in GenBank [see Supporting Information- Table S1].
We also amplified 15 previously developed microsatellite markers [simple sequence repeat (SSR); Bossolini et al. 2011] that were scattered throughout the Petunia genome in 73 individuals of P. secreta. These markers were selected according to their location within the genome, their polymorphic index content (PIC) and their successful amplification in P. secreta samples. Polymerase chain reactions were conducted in a final volume of 10 mL, which contained 10 ng of genomic DNA as template following the protocol set forth by Turchetto et al. (2015b). The forward primers were FAM-, NED-or HEX-labelled. DNA amplicons were denatured and sizefractionated using capillary electrophoresis on a Mega-BACE (GE Healthcare, USA) with a GeneTab 550 internal size ladder (GE Healthcare). The manufacturer's software was utilized to determine alleles.

Genetic diversity and population structure based on plastid information
The 65 plastid sequences were manually aligned with GENEDOC software (Nicholas and Nicholas 1997).
Bayesian inference was performed as implemented in BEAST 1.6.1 (Drummond and Rambaut 2007) to estimate the phylogenetic tree using the different haplotypes of the three species. Two independent runs were employed consisting of 1 × 10 8 Markov chain Monte Carlo (MCMC) iterations, sampling every 1000 generations under the HKY (Hasegawa, Kishino and Yano) nucleotide substitution model with four gamma categories in a Yule tree prior. Convergence was looked for in the stationary distribution by visually inspecting the posterior distribution of independent runs and ensuring that all parameters had effective sample sizes . 200 in TRACER 1.6 (Rambaut et al. 2013), removing the first 10 % iterations as a burn-in. A maximum clade credibility tree was obtained and the posterior probabilities for each node (Rannala and Yang 1996) using the TREEANNO-TATOR software, part of the BEAST package. Finally, FIG-TREE 1.4.0 (Rambaut 2008) was utilized to edit the phylogenetic tree.
The p-distance was also estimated in MEGA 6 (Tamura et al. 2013) to compare the mean intra-and interspecific genetic distances among P. secreta, P. exserta and P. axillaris. To detect evidence for deviation from a neutral equilibrium model of evolution in P. secreta, Tajima's D (Tajima 1989) and Fu's F S (Fu 1997) neutrality tests were performed using the ARLEQUIN software. In these analyses, all haplotypes obtained for P. secreta were included and compared with those previously published for P. axillaris from Serra do Sudeste  and P. exserta ). In addition, changes in population size were verified over time for P. secreta by performing Bayesian skyline plot (BSP) analysis (Drummond et al. 2005) as implemented in BEAST package. For this analysis, a strict molecular clock model with a mean substitution rate of 2.8 × 10 29 per site per year (standard deviation 5.4 × 10 211 ) according to Lorenz-Lemke et al. (2010) and HKY nucleotide substitution model were used as priors. Markov chain Monte Carlo was performed for 100 000 000 steps, sampling every 10 000 steps. TRACER was employed to compute BSP and inspect for convergence. The program Alleles in Space 1.0 (Miller 2005) was made use of to associate the genetic and geographic distances between the two disjointed areas of P. secreta distribution according to Mantel's test (Mantel 1967). For this analysis, logtransformed geographic distance was utilized to compare the two P. secreta groups (Pop1 -12 and Pop13).
Genetic diversity and population structure based on SSR FSTAT 2.9.3.2 software (Goudet 1995) was used to evaluate the summary statistics, such as the number of alleles per locus, gene diversity, allelic richness and inbreeding coefficient (F IS ), for each locus. The frequencies of null alleles, PIC, levels of observed (H O ) and expected (H E ) heterozygosity and any significant deviations from the Hardy -Weinberg equilibrium (HWE) to evaluate the informativeness of the markers were estimated with CER-VUS 3.0.3 software (Marshall et al. 1998;Kalinowski et al. 2007).
To investigate the genetic similarity among P. secreta individuals, discriminant analysis of principal components (DAPC; Jombart et al. 2010) was carried out as implemented in ADEGENET (Jombart 2008;R Development Core Team 2011). Discriminant analysis of principal components relies on data transformation from principal component analysis (PCA) prior to discriminant analysis, maximizing the separation between the groups. This analysis is not based on predefined population genetic models and makes no assumption about HWE or linkage disequilibrium. Whereas PCA and principal coordinates analysis focus on the entire genetic variation, DAPC searches for linear combinations of the alleles, enhancing the differences between groups while minimizing variation within clusters as measured by F-statistics. Discriminant analysis of principal components was performed with 13 groups set a priori representing the collection sites and also without prior information on the individual spatial origin. The number of clusters was assessed using K-means clustering and the optimal number of groups according to the Bayesian information criterion was selected. The number of PCs was set during the calculation process.
The genetic structure of P. secreta was examined using the STRUCTURE 2.3 software package (Pritchard et al. 2000) to determine the most likely number of independent genetic clusters (K ). STRUCTURE was run without any prior information regarding sampling location. An admixture model was implemented using correlated allele frequencies (Falush et al. 2003) because gene flow is expected among natural populations. The best number of groups (K ) was evaluated from 1 to 15, with 10 independent runs per K value. Each run was executed using 2.5 × 10 5 burn-in periods, and 10 6 MCMC repetitions after burn-in for population clustering were used. The optimal K value was identified from the maximum value of DK (Evanno et al. 2005) as implemented in STRUCTURE HARVESTER 0.6.93 (Earl and vonHoldt 2012). CLUMPP 1.1.2 (Jakobsson and Rosenberg 2007) was used to summarize the results of the optimal K value based on the pairwise similarity average of individual assignments across runs through Greedy's method and G ′ statistics. DISTRUCT 1.1 (Rosenberg 2004) was employed to envisage the STRUCTURE results after processing with CLUMPP. One additional run of the STRUCTURE analysis took place to compare the 73 P. secreta with 25 P. axillaris from Serra do Sudeste and 24 P. exserta individuals, as previously studied by Turchetto et al. (2015b). Genetic differentiation among P. secreta, P. axillaris and P. exserta individuals was quantified, as well, through pairwise estimators of fixation index (F ST ) via ARLEQUIN with 10 000 permutations to assess the significance, as well as the genetic differentiation between the two disjointed areas of P. secreta distribution (Pop1 -12 and Pop13).
Mantel's test to compare the geographic (logtransformed) and genetic (F ST /1 2 F ST ) distances among populations was employed with statistical significance assessed via 10 000 permutations as implemented in ISOLDE as part of Genepop 4.2 package (Raymond and Rousset 1995).
A distance matrix was put together based on shared alleles among individuals [see Supporting Information- Table S2] and collection sites [see Supporting Information- Table S3] to depict the relationships among all of the P. secreta individuals using MSA 4.05 software (Dieringer and Schlö tterer 2003). PHYLIP software (http://evolution.genetics.washington.edu/phylip. html) was made us of to construct an unweighted neighbour-joining (N-J) tree (Saitou and Nei 1987) based on the matrix of shared alleles, including 15 SSRs and 73 P. secreta individuals. In addition, genetic distance to compare P. secreta, P. axillaris and P. exserta individuals was acquired.
ARLEQUIN was also employed to estimate the Analysis of molecular variance (AMOVA; Excoffier et al. 1992) within P. secreta. In the variance analysis, the length of the amplified product of SSR markers of each accession as the value of microsatellite alleles was used.
To test for evidences of population reduction, BOTTLE-NECK 1.2.02 software (Piry et al. 1999) was utilized with 12 sites from Pedra do Segredo as one population and Pop13 as other. BOTTLENECK assumes one population that has undergone recent bottleneck presents more heterozygotes than expected under equilibrium. This was excess of heterozygotes was tested using a Wilcoxon signedrank test (Cornuet and Luikart 1996) under a two-phase model (TPM) of microsatellite evolution (Di Rienzo et al. 1994) with a variance of 12 % (Piry et al. 1999) and different percentages of the stepwise mutation model in the TPM (70, 85 and 95 %) (Di Rienzo et al. 1998).

Demographic parameters
Effective population size based on SSRs and plastid markers was estimated to the two disjointed areas of P. secreta occurrence (Pop1 -12 from Pedra do Segredo and Pop13 from the road) using MIGRATE-N 3.6.11 software (Beerli 2009) with the Bayesian coalescent approach (Beerli and Felsenstein 2001;Beerli 2006). MIGRATE-N estimates Q ¼ 4Nem for SSRs and Q ¼ 2Nem for plastid markers, respectively (where Ne is the estimated population size and m the mutation rate per loci per generation). To analyse the SSR markers, the Brownian motion model was used, with starting conditions based on F ST and uniform prior distribution to estimate u (range: 0 -0.1). Our searches included one long chain with 10 7 steps sampling, 10 4 recorded genealogies and 10 5 chains as burn-in. We ran four independent chains with distinct temperatures (1.0, 1.5, 3.0 and 10 5 ) in a Markov Coupling (MCMCMC) procedure (Geyer and Thompson 1995) to ensure sampling over more genealogies. Estimates were computed per locus and summarized as weighted values over all loci. Estimates for plastid markers were performed using the sequence model summarizing the results of 10 replicates with 10 7 steps sampling, 10 4 recorded genealogies and 10 5 chains as burn-in. The MCMCMC procedure (Geyer and Thompson 1995) was also used with distinct temperatures (1.0, 1.5, 3.0 and 6.0). F ST was used to calculate starting conditions and a uniform prior distribution was used to estimate u (range: 0 -0.05). The same substitution rate used for the BSP analysis, 2.8 × 10 29 per site per year (Lorenz-Lemke et al. 2010), was used to estimate the number of individuals. We assessed stationary of the Markov chain checking the posterior distribution over all loci and running the programme several times with different starting points for both analyses.

Morphological characterization
The morphology of flowers of 3 individuals from Pedra do Segredo and 12 individuals from the BR290 site was examined, one flower per individual. In the field, digital calipers measured three flower traits: the length of the corolla tube, the diameter of the corolla and the distance between the anthers of the longest and medium-length stamens [see Supporting Information- Table S4]. Principal component analysis implemented in PAST software (Hammer et al. 2001) was able to determine whether the individual from the two disjoint locations could be distinguished by these three morphometric variables, as these morphological traits are important to discriminate subspecies in P. axillaris (Kokubun et al. 2006) and are associated with geographic distribution and genetic components (Turchetto et al. 2014b). The two principal components were plotted to evaluate the dispersion of the data.

Conservation
The criteria adopted by the International Union for the Conservation of Nature (IUCN 2012) were used to identify the threat category of P. secreta. Conservation status was based mainly on the distribution area, as estimated with the help of GeoCAT-Geospatial Conservation Assessment Tool software (Bachman et al. 2011). The distribution area was estimated from the value of two parameters: the extent of occurrence and the area of occupancy. The conservation status followed the nomenclature of IUCN: Least Concern (LC), Not Threatened (NT), Vulnerable (VU), Endangered (EN) and Critically Endangered (CR).

Genetic diversity and population structure based on plastid sequences
The plastids intergenic spacers, trnH-psbA and trnS-trnG, of 65 individuals of rare P. secreta were sequenced. Individuals were sampled in 13 different sites (Pop1 -Pop13) ( Table 1). The concatenated alignment was 1060 base pairs (bp) long (410 bp in trnH-psbA and 650 bp in trnS-trnG). These sequences possessed nine polymorphic sites (one transition and eight transversions), resulting in nine haplotypes (Fig. 3A). The evolutionary relationships among the nine haplotypes of P. secreta (Fig. 3A) did not exhibit a spatial correlation considering the different collection sites. Only at five collection sites, two or more AoB PLANTS www.aobplants.oxfordjournals.org individuals were found, and only in Pop7, Pop12 and Pop13, more than one haplotype was collected [ Table 1; see Supporting Information- Table S5]. The majority of individuals were of the H1 haplotype (35/65, from Pop1, Pop7 and Pop12), whereas H2 haplotype was found in 16/65 individuals from six sites (Pop2, Pop6, Pop9, Pop10, Pop11 and Pop13). The H9 haplotype was exclusive to individuals from Pop13 and at this collection site, only H2 was also found. Compared with P. axillaris and P. exserta haplotypes, six haplotypes were exclusive to P. secreta (H1, H5, H6, H7, H8 and H9), two were shared by the three species (H2 and H4) and one was shared strictly by P. secreta and P. axillaris individuals (H3) (Fig. 3B). One to three evolutionary steps separated these nine haplotypes of P. secreta (Fig. 3A). Haplotype diversity in P. secreta was h ¼ 0.648 + 0.051, and nucleotide diversity was p (%) ¼ 0.17 + 0.11 (Table 2).
The Bayesian inference tree comparing the three species haplotypes [see Supporting Information- Fig. S1] showed two fully supported clades, both including haplotypes of the three species. As expected, low support for the majority of the internal branches was observed.
Based on plastid sequences, the interspecific genetic p-distance values were very low (0.002) among the three species, as was the intraspecific genetic distance in P. secreta (0.002) when considering all of the analysed individuals of this species. The neutrality Fu's F S and Tajima's D tests (Table 2) to P. secreta were negative and not significant. The BSP (Fig. 4) suggested stability for P. secreta across time with a weak population growth signal beginning 60 kya, though this was not statistically significant given the size of the estimated confidence limits. The Mantel's test suggested a moderate association between genetic diversity and geographic distance  based on plastid sequences considering the two groups of P. secreta (r 2 ¼ 0.5513; P , 0.001).

Genetic diversity based on SSR
Seventy-three individuals of P. secreta were genotyped at 15 SSR loci and all markers were polymorphic. These markers yielded a total of 111 alleles, with an average of 7.4 alleles per locus (Table 3). The number of alleles per locus ranged from 2 (PM74) to 20 (PM177). The PIC values for these markers ranged from 0.06 (PM74) to 0.89 (PM177), with an average of 0.57, also demonstrating that these genetic markers were informative within this taxon (Table 3). Allele richness was, on average, 5.91 and ranged from to 1.81 (PM74) to 14.61 (PM177). The gene diversity was, on average, 0.63 and ranged from 0.06 (PM74) to 0.91 (PM177). The average H O (0.24) was lower than the H E (0.63) across all loci. Thirteen loci exhibited a significant departure from HWE expectations   (Table 3), indicating a deficit of heterozygotes. Heterozygote deficiency may be the result of biological factors, such as genetic drift or inbreeding, or indicators of null alleles. The frequency of null alleles was low (,1 %) across all loci. The genetic diversity estimates of P. secreta, as allele richness, H O , H E and F IS values were statistically compared with the same estimates obtained for P. axillaris (25 individuals from Guaritas) and P. exserta (23 individuals also collected in Guaritas) (see Table 1 in Turchetto et al. 2015b for details of locality of sample: P. axillaris Pop4 -7 and 9 -11, P. exserta Pop26 -32 and 34). The results were as follows: allele richness (P. secreta . P. exserta, P. secreta ¼ P. axillaris and P. axillaris . P. exserta), H O (P. secreta ¼ P. exserta, P. secreta , P. axillaris and P. axillaris . P. exserta), H E (similar for all comparisons) and F IS values (P. secreta ¼ P. exserta, P. secreta . P. axillaris and P. axillaris , P. exserta). The allele richness of a sample is affected by the size of a sample (large samples are expected to have more alleles than small samples) and can produce unbiased estimates. Because of this, we also estimated the allele richness in HP-RARE 1.0 software (Kalinowski 2005), on which the statistical technique of rarefaction compensates for sampling disparity. These results were similar to those described above (data not shown).

Population structure based on SSR
Petunia secreta individuals formed five genetically homogeneous groups according to DAPC analysis (Fig. 5A) with three of them grouping individuals from Pedra do Segredo (Pop1-12), whereas two clusters encompassed the individuals from Pop13. The Bayesian clustering analysis as implemented with the admixture model in STRUCTURE (Fig. 5B) revealed two genetic components to 73 P. secreta individuals according to the Evanno et al. (2005) method (best K ¼ 2), splitting the sample into two major groups relative to the collection location (Pop1 -12 or Pop13). The same results were obtained when prior information about the collection points was included in the STRUCTURE analysis (data not shown). Only a few individuals in each group showed some degree of admixture. According to Mantel's test, the genetic differentiation between the two groups in P. secreta was weekly associated with geographic distance (r 2 ¼ 0.1255; P , 0.001).
To identify the evolutionary relationships between P. secreta and other Petunia species based on genetic components according to SSR alleles, a STRUCTURE analysis was conducted based on 14 SSR loci, as described by Turchetto et al. (2015b) for P. axillaris and P. exserta, and the 73 P. secreta individuals. K values ranging from 1 to 15 were tested. The best number of groups was three (K ¼ 3), separating P. secreta individuals into two groups (all collection sites from Pedra do Segredo and Pop13) and the third group corresponding to P. axillaris subsp. axillaris Figure 5. Population structure and evolutionary relationships of P. secreta individuals based on 15 microsatellite loci as observed through clustering analyses in the DAPC scatter plot (A) and STRUCTURE (B). Different colours indicate groups (DAPC, K ¼ 5; STRUCTURE, K ¼ 2), and vertical lines correspond to individuals. Neighbour-joining tree (C) obtained from a distance matrix based on shared alleles among individuals and collection sites (Pop13 individuals are presented in grey). Individual numbers according to describe in Supporting Information- Table S1 [see  Supporting Information-Table S1]. and P. exserta individuals (Fig. 6A). When K ¼ 4, P. axillaris and P. exserta became separated (Fig. 6B), and two groups followed as observed for P. secreta.
The genetic distance based on shared alleles among the collection sites of P. secreta [see Supporting Information- Table S3] was the lowest between Pop7 and Pop12 (0.27) and the highest between Pop2 and Pop9 (0.83). Comparing P. secreta with the other Petunia species based on the 14 in-common analysed SSR [see Supporting Information- Table S6], P. secreta was closer to P. axillaris (0.42) than to P. exserta (0.50). Accounting for the samples from Pedra do Segredo as one group and Pop13 as another based on their geographic distances [see Supporting Information- Table S7], the genetic distance between these groups (0.63) was higher than that observed between P. axillaris subsp. axillaris and P. exserta (0.42), and Pop13 was more different from these species than the remaining populations were.
The N-J tree (Fig. 5C) presented a similar topology to that of the STRUCTURE results and was divided into two main groups. In this analysis, only one individual from Pop12 was closer to the group that was formed by Pop13 individuals, and no individuals from Pop13 were outside this group. Individuals from the same site from Pedra do Segredo did not form preferential groupings.
An AMOVA revealed that higher molecular variance is equally distributed among populations (34 %), within individuals (33 %) and among individuals within populations (34 %). The F-statistic over all loci was significant at the level of 0.001 (F ST ¼ 0.34, F IS ¼ 0.51 and F IT ¼ 0.67). When just the 12 sites from the Pedra do Segredo locality were considered, the most variation was found among individuals within populations (51 %) than among populations (23 %) or within individuals (26 %). The F-statistic over all loci was significant at F ST ¼ 0.229 (P , 0.001), whereas F IS (0.665) and F IT (0.742) were not significant (P . 0.01).
No excess of heterozygotes was detected with BOTTLE-NECK for the collection sites Pedra do Segredo (Pop1 -12) and Pop13 (lower P value in Wilcoxon test; P ¼ 0.92 and P ¼ 0.88, respectively), indicating that the populations did not undergo a founder effect. However, this result should be interpreted with caution as many loci were not in HWE.

Morphological variability
Principal component analysis showed that the percentage of variance was 84.5 for the first component and AoB PLANTS www.aobplants.oxfordjournals.org 9.5 % for the second component. The 95 % distribution included all of the individuals [see Supporting Information- Fig. S2] and confirmed that individuals from all of the collection sites corresponded to the morphological description of P. secreta previously put forth by Stehmann and Semir (2005). The corolla colour was also in agreement with the original description.

Conservation
According to IUCN (2012) criteria, the conservation status of P. secreta was classified as critically endangered [CR: B1ab(ii,iii,iv)] because of the species' extremely limited extent of occurrence (81 km 2 ). This species inhabits an area with high fragmentation, reducing the area of occupancy (20 km 2 ) and, consequently, decreasing habitat quality. Moreover, P. secreta has been found in just two different locations.

Discussion
Conservation biologists are interested in knowing whether there are generalities that can be made with regard to rare species, such as whether these species typically exhibit reduced genetic diversity or restricted gene flow between populations, as predicted by population genetic theory when the populations are small and isolated, with profound evolutionary and ecological consequences. Ecological and genetic explanations to rarity have been suggested. Both genetic variation in ecologically relevant traits and differences in selection pressures, in addition to neutral forces, such as the significant impact of genetic drift and inbreeding in small populations of rare plants, can drive the geographic distribution of a species (Sheth and Angert 2014). An important question in conservation genetics is how does genetic variability compare among rare plants and their widespread related species (Furches et al. 2013).
The IUCN criteria have been used to assess conservation status at a species level. Here, the status of P. secreta is classified as Critically Endangered. Additionally, as proposed by Rabinowitz (1981), P. secreta could also be considered as a rare species exhibiting geographic range and small populations, but it is broad because individuals grow in two different habitats. There are biological, ecological and evolutionary mechanisms that allow many rare plant species to persist. Species currently rare may have become so in recent history (Bekker and Kwak 2005). Moreover, Schwartz and Simberloff (2001) observed a trend for rare species to be included into species-rich clades, suggesting that speciation and extinction might be linked. Knapp (2011) made note of the fact that, for example in Cape flora region, threatened species are more common in clades that are young and diversifying quickly, signifying that risk is conserved in lineages but differs wildly at the tips of those lineages. These differences can be associated with the mode of speciation. For example, plants that speciate via small isolated populations at the edges of a large species range present rapid diversification and will have a larger numbers of threatened species. Moreover, plants diversifying through peripatry could have large variations in threats among the tips of phylogenetic trees (Davies et al. 2011).
With only 13 known collection sites (Fig. 2) scattered over a small area, P. secreta is a perfect example of a rare plant that has a restricted geographical range and small local populations. Given its ecological characteristics and despite the general expectation of reduced genetic variation in a rare species, P. secreta possesses high genetic variation, as denoted by the extensive microsatellite polymorphism and plastid haplotype numbers, and no evidence of having gone through a recent founder effect. The results ultimately show that the species persisted over time (Fig. 4) and was able to grow in varied habitats, occurring on the rocks in Pedra do Segredo and along the road in another kind of soil. Moreover, plants from Pop13 are higher and display bigger leaves than those collected from Pedra do Segredo probably because of soil richness and moisture.
Compared with other Petunia species, P. secreta displays one of the highest diversity values. Based on the same microsatellite set, Turchetto et al. (2015b) found 90 alleles with an average of 6 alleles per locus and an allele richness average of 6.3 in P. axillaris, and 68 alleles with an average of 4.5 per locus and an allele richness average of 4.8 in P. exserta. Comparing three species of the P. integrifolia complex based on seven microsatellite loci, Segatto et al. (2014b) observed lower values, especially in relation to the allele richness (3.73 in P. inflata, 3.29 in P. integrifolia subsp. integrifolia and 4.39 in P. interior). The geographic range of these P. integrifolia complex species and the number of individuals per occurrence site are much larger than those observed for P. secreta (except Pop13). Obtained Q values based on nuclear markers are bigger than those obtained for P. exserta (Q ¼ 0.057) and smaller than that estimated for P. axillaris (Q ¼ 0.43) by Segatto et al. (2014a). Considering plastid markers, the effective population size of P. secreta is comparable with that of P. axillaris (Q ¼ 0.0025).
Plants with restricted distributions are expected to exhibit low genetic diversity at the species level (Cole 2003). In addition, fragmentation into isolated and discrete demes composed of relatively few individuals contributes to a depletion of overall genetic diversity (Ellstrand and Elam 1993). The genetic diversity found in P. secreta is comparable with that in other plant species associated with narrow geographic distribution and small populations (Furches et al. 2013), or even those much more widespread (Koopman and Carstens 2010).
The extension of the geographic range alone either should not be used to reliably predict genetic dynamics (Premoli et al. 2001) or constitutes a strong enough reason to guide conservation actions. Indeed, there are several other factors that influence the levels of diversity within and among populations, such as ecological and biogeographical factors and mating systems (Wang et al. 2004;Pé rez de Paz and Caujapé -Castells 2013;Coppi et al. 2014). Additionally, a number of causes of rarity have been proposed, like ecological equivalency, frequency-dependent selection and narrow niche requirements, and these factors play an important role in the abundance of species (Markham 2014).
As demonstrated here, genetic variability and population structure descriptors suggest ancient variability and a stable founder population to P. secreta. Populations of P. secreta from Pedra do Segredo (Pop1 -12) presented a higher partitioned variation among individuals (51 %), suggestive of this species being part of a single panmictic group. Moreover, all of the analyses indicate that Pedra do Segredo and Pop13 constitute different units. Consequently, P. secreta is a narrowly distributed species that is primarily distributed into two large populations genetically differentiated and occurring in a fragmented area, 20 km from each other. The environmental conditions are different between the major localities where P. secreta was found, especially the soil substrate and associated vegetation. Whereas plants in Pedra do Segredo grow directly on the rock in shallow soil, close to cacti and other xerophyte plants, the plants along the road grow in deep and rich soil and are surrounded by grass in a completely modified environment. Though the floral traits did not have significant differences between individuals and the two groups (Pop1 -12 and Pop13) presenting morphological characteristics, as proposed in the P. secreta description, the genetic data of each population correspond to different evolutionary lineages both in the plastid haplotypes and in the microsatellites (see Figs 3 and 5). The results indicate a lack of connectivity by pollination or seed dispersion between Pop13 and the Pedra do Segredo group. Besides, Pop 13 has an effective population size about half of the Pop1 -12 when plastid markers are considered.
Genetic information may be important for conserving and recovering rare species. However, these rare species may not always be genetically depleted (Edwards et al. 2014). Though high genetic diversity is not a common occurrence for rare species, Gitzendanner and Soltis (2000) demonstrated that endangered species occasionally exhibit levels of diversity as high as or higher than that of a widespread congener. Petunia secreta presents indices of genetic variability that are similar to P. exserta that is also endemic to a small area and lower than P. axillaris, which is distributed in a larger range. The three species possess different floral syndromes and reproductive systems, possibly influencing the different levels of diversity seen. Although P. exserta and P. secreta are SC, higher allelic richness was observed within P. secreta. As a consequence of relying on just a few pollinator species for sexual reproduction, low density or narrow geographical range of the pollinator can contribute to rarity of the plants (Phillips et al. 2014). Despite the high level of genetic diversity, a decrease in heterozygosity was evident (13 loci deviated from HWE, and no significant null alleles were observed), suggesting a significant amount of inbreeding in P. secreta as a consequence of self-fertilization and/or preferential mating among relatives. A spatial genetic structure was observed across collection sites and groups formed in Bayesian inference tree, clustering and DAPC analyses, suggesting the presence of two evolutionary lineages in P. secreta with some gene flow between neighbouring sites. The individuals from Pop13 were less variable than the individuals from Pedra do Segredo (Pop1 -12) and had only a subset of the entire species' set of SSR alleles. This difference could be from a founder effect or genetic drift. The higher differentiation between Pop13 and other P. secreta populations compared with that between P. axillaris and P. exserta, as evidenced by STRUCTURE (Fig. 5), could be a consequence of the assumptions within this analysis.
One major risk to P. secreta maintenance is that it is composed of only two effective population or evolutionary lineages (Pop1 -12 from Pedra do Segredo and Pop13) because no other natural populations were found despite the exhaustive search for this species along the area for ca. 10 years. It is suggested that in order to protect against loss of genetic variability and extinction of species, seeds are collected from as many individuals as possible and stored in seed banks and living collections, thus promoting regular genetic monitoring to restore eventual stochastic losses. Moreover, P. secreta needs to be federally listed as a threatened species, and the two different groups should be treated at least as management units.
Further experiments are necessary to gather knowledge on the mating system and understand the complete evolutionary history of P. secreta.

Conclusions
The results from this work demonstrate that even within the same category of rarity, different species might exhibit a range of spatial and temporal characteristics, AoB PLANTS www.aobplants.oxfordjournals.org of which the outcome is differential genetic patterns. Here, it is shown that although P. secreta is intrinsically rare, this rarity is not a feature that negatively affects genetic diversity; the fragmented distribution preventing gene flow between major populations is what puts the species' persistence at risk.

Conflict of Interest Statement
None declared.

Supporting Information
The following additional information is available in the online version of this article - Table S1. GenBank accession numbers. Table S2. Genetic distance based on SSR alleles. Table S3. Genetic distance based on microsatellite per collection site. Table S4. Petunia secreta flower measurements. Table S5. Plastid haplotypes per individual per collection site. Table S6. Genetic distance based on SSR among Petunia species. Table S7. Genetic distance based on SSR among Petunia species separating P. secreta into two subgroups. Figure S1. Evolutionary relationships among plastid haplotypes. Figure S2. Principal component analysis based on morphology.
Literature Cited