Population structure of Eupemphix nattereri ( Amphibia , Anura , Leiuperidae ) from Central Brazil

This study reports on 156 specimens of the amphibian Eupemphix nattereri, a widely distributed leiuperid, obtained from 11 municipalities of central Brazil. The extent of genetic variation was quantified by determining the mean number of alleles per locus and the proportion of polymorphic loci. An analysis of molecular variance (AMOVA) was performed on the random amplified polymorphic DNA (RAPD) haplotypes. The genetic distances obtained by calculating pairwise Φst among local samples were used to determine population relationships using the unweighted pair-group method (UPGMA) and non-metric multidimensional scaling (NMDS). The cophenetic correlation was calculated to confirm agreement between the genetic matrix and the unweighted pair group method with averages (UPGMA) dendrogram. To determine if genetic distances were correlated to geographical distances we constructed pairwise genetic distance and geographical distance matrices and compared them using the Mantel test. The AMOVA results indicated significant genetic differences (p < 0.001) between E. nattereri populations, representing 69.5% of the within population genetic diversity. The Mantel test showed no significant correlation (r = 0.03; p = 0.45) between the genetic and geographical distance matrices. Our findings indicate that the genetic variation of E. nattereri populations was randomly distributed in geographic space and that gene flow for this species is probably structured at spatial scales smaller than those between our samples


Introduction
Amphibian populations have been the focus of numerous studies that have contributed to the general understanding of ecological and evolutionary phenomena (Newman 1992;Wilbur, 1997;McDiarmid and Altig, 1999;Funk et al., 2005).Amphibians are good models for investigating the genetics of wild animal populations because they are widely distributed in most ecosystems, easy to sample in breeding assemblages and often philopatric to breeding sites.These characteristics can generate high levels of population genetic structure (McDiarmid and Altig, 1999;Beebee, 2005).
Although a global decline in amphibians has been reported since the 1980s, relatively little is known about the status of amphibian populations in South American countries due to insufficient data on species distribution and population dynamics, combined with high levels of species diversity (Myers et al., 2000;Young et al., 2001).Information on the population genetic structure of amphibians could help, at least in part, to understand the reason for such declines.Amphibian populations are often thought to have a metapopulation spatial structure (Alford and Richards, 1999), but few studies have actually assessed interpopulation movement, much less the effects of such movement on population dynamics and genetic structure (Newman and Squire, 2001).Amphibians are also thought to have low dispersal rates (Blaustein et al., 1994), although this may not apply to all species (Funk et al., 2005).Low numbers of breeding individuals and limited gene flow between populations can reduce genetic diversity in such a way that a species ceases to be viable (Blaustein et al., 1994;Kraaijeveld-Smit et al., 2005;Spear et al., 2005;Telles et. al., 2006).
Neutral genetic markers have been used in studies of amphibian population structures (Rissler et al., 2004;Spear et al., 2005;Telles et. al., 2006).DNA-based markers are capable of disclosing microevolutionary variation and PCR-based genotyping of molecular markers has greatly benefited areas such as population and conservation genetics, allowing the assessment of relatedness and genetic variability between individuals, populations, and species (Avise, 1994;Buso et al., 1998;Zhang and Hewitt, 2003).
Random Amplified Polymorphic DNA (RAPD) is one class of DNA markers used to study the structure of genetic diversity in many species (Wu et al., 2002;Almeida et al., 2003;Telles et al., 2006).This molecular marker can be used for rapid testing, due to its relatively low cost per reaction, the small amounts of DNA needed and the simple method of acquiring data on variation in genomic DNA (Welsh and McClelland, 1990;Hadrys et al., 1992).RAPD markers have been considered suitable for genetic analysis because they allow for examination of accumulated genetic differences that are important at various taxonomic levels (Hardy, 2003).Furthermore, in comparison to codominant markers, dominant markers such as RAPD markers, easily generate a genetic profile even for species to which no prior genetic information is available (Mueller and Wolfenbarger, 1999).
The genus Eupemphix belongs to the anuran family Leiuperidae a widely distributed family which occurs in almost all the Neotropics (Frost, 2007).The genera Eupemphix (Leiuperidae) was described by Steindachner (1863) to include the frog Eupemphix nattereri (Amphibia: Anura: Leptodactylidae), the type locality for which is Cuiabá, in the Brazilian state of Mato Grosso (Nascimento et al., 2005).It is known that E. nattereri is widely distributed in South America, ranging from the east of Paraguay to the mid-east and southeast of Brazil (Frost, 2007) but, however, little is known about the genetic structure of populations of this species or any other frog belonging to this family.
The objective of this study was to investigate the genetic similarity and diversity within and between natural populations of E. nattereri from eleven geographical sites in central Brazil using RAPD markers.Genetic and geographic distances were also associated to test whether nearby populations were more genetically similar than distant populations, thus providing the first insight into the microevolutionary processes underlying genetic variation.

Sample collection and DNA extraction
During the rainy seasons (October to March) of 2002-2004 we obtained 156 Eupemphix nattereri specimens from 11 localities in central Brazil, the number of specimens studied in each municipality and the geographic coordinates of the sampling areas are listed in Table 1.Voucher specimens are deposited in Brazil at the Zoological Collection of the Federal University of Goiás (ZUFG).This research was done according to the statutes of COBEA (Brazilian College of Animal Experimentation), once our institution does not have an ethics committee of animal experimentation.
The 156 E. nattereri specimens were humanly sacrificed and liver tissue was removed and stored at -20 °C until DNA extraction was performed.Genomic DNA was purified from 20 mg of frozen liver with the Wizard Genomic DNA ® purification Kit (Promega Corporation, USA), according to the procedures described by the manufacturer.To estimate the quality and amount of genomic DNA we 1162 Silva et al. subjected 5 μL samples of extracted product to electrophoresis on 1% (w/v) agarose gel at 5V cm -1 for 30 min.When necessary, the samples were diluted to 2.5 ng μL -1 .

Random amplified polymorphic DNA (RAPD) analysis
A total of 40 primers of 10 base pairs were tested to select those with the best amplification pattern.A preliminary screening was performed with DNA from two frogs from the studied populations and allowed the identification of eight RAPD primers (Table 2) that yielded distinct, well-separated and reproducible bands.These bands were subsequently chosen for final analyses.Band repeatability for each primer was confirmed by duplicated PCRs with DNA from at least two frogs from each population sampled.Amplification patterns obtained from two independent sets of PCR reactions using DNA from the Mambai population (n = 20) were scored twice by two different technicians to test the technical reproducibility and the reliability of the software Image Master 1D system for interpreting and analyzing RAPD markers.Loci that did not reproduce were excluded from the analyses, as well as unclear and undefined bands.The presence of a determined locus was confirmed by the Total Lab Master 1D software (Amersham Pharmacia Biotech, USA) by verifying the peaks of the bands produced during gel analyses.Loci with peaks smaller than 20 pixels, after two repetitions, were excluded from the analyses (Figure 1).A binary matrix was constructed from the gel readings, where the samples from the individual frogs were genotypically characterized for the presence (1) or absence (0) of bands.The percentage of polymorphism obtained with each primer was calculated from this matrix.

Data analysis
The extent of genetic variation in each population was quantified by determining the mean number of alleles per locus and the proportion of polymorphic loci, using the TFPGA (Tools for population genetic analyses) v 1.3 soft-ware (Miller, 1997).Analysis of molecular variance (AMOVA) was performed on the RAPD haplotypes and the Euclidean distances between all pairs of haplotypes were calculated (Excoffier et al., 1992).The AMOVA was used to partition the total genetic variation into that occurring within and among population, expressed by the Φst statistic and tested by randomization using 1,000 permutations.These analyses were performed with WINAMOVA software provided by L. Excoffier (University of Geneva).
The Φst statistic was also estimated between pairs of local populations, providing an explicit estimate of genetic divergence among local populations to be further used for spatial analyses.Another direct estimate of F st from dominant markers was obtained using the Bayesian approach proposed by Holsinger et al. (2002), which does not make explicit assumptions about inbreeding and Hardy-Weinberg equilibrium in local populations.An a posteriori distribution of the θ B estimator (an estimate of F ST ) was numerically approximated by a Markov Chain Monte Carlo (MCMC) simulation, and tended to converge to a beta distribution.The HICKORY v 1.0 software (Holsinger and Lewis, 2003) allows the estimation of four different models.The first is a full model, in which both θ B and f (similar to F IS , the inbreeding coefficient) are estimated.Alternatively, other two models can assume θ B or f equal to zero.Finally, because estimates of f based on dominant markers may often be strongly biased, especially for small sample sizes (n < 10), the HICKORY program allows the estimation of a final model in which f is free to vary.The sampler will not attempt to estimate f and instead it will choose values of f at random from its prior distribution while estimating other parameters during the MCMC run.These models were compared using the deviant information criterion   singer, 1999;Holsinger et al., 2002;Spiegelhalter et al., 2002;Holsinger and Wallace, 2004).The genetic distances obtained by calculating pairwise Φst values between local samples were used to determine population relationships using the unweighted pair group method with averages (UPGMA) (Sneath and Sokal, 1973).Cophenetic correlation using the NTSYS (Numerical Taxonomy and Multivariate Analysis System) software (Rohlf, 1989) was used to verify agreement between the genetic matrix and the UPGMA dendrogram.A high correlation coefficient with a cophenetic correlation > 0.8 indicates a dendrogram that well reflects the original data but for values lower than 0.8 the clustering result for individual samples should be carefully checked.
Despite the fact that the UPGMA method is one of the methods most frequently used to represent multivariate genetic distances, some authors suggested that hierarchical structures do not always appear in genetic similarity since evolutionary processes tend to generate continuously clinal or reticulate patterns (Lessa 1990;Rodrigues and Diniz-Filho, 1998).In this case, ordination techniques, such as non-metric multidimensional scaling (NMDS), could also be useful for describing population divergence.Starting from the multidimensional genetic distances, the NMDS program uses iteration to produce the best possible representation of the relationship among populations in a space with a lower number of dimensions (in this paper, a twodimensional (2-D) solution was used), by minimizing the stress (S) value, a measure of "badness-of-fitness".
In order to determine whether genetic distances, based on pairwise Φ st values, were correlated with geographical distances Mantel tests were performed using the NTSYS statistical package program (Manly, 1997;Rohlf, 1989).Matrix correlation was tested using 1,000 random permutations.To refine the Mantel statistics and detect short-distance spatial patterns, analyses were also performed by comparing genetic distances with the binary model matrix (1/0) connecting sampling sites situated only at distances of less than 50 km or 100 km, forming a spatial correlogram at small geographical distances.

Results
The eight primer sets amplified a total of 82 scorable RAPD fragments in 156 frogs from the 11 populations studied, with 81 (98.8%) of the 82 loci being polymorphic and only one locus being monomorphic.The variation in band size ranged from approximately 100 to 2000 bp and the number of RAPD fragments per primer varied from 06 to 14 considering all populations (Table 2).Primer set OPA13 produced the highest number of fragments among all the primers used, with an average of 14 loci (Figure 2).On the other hand, primer set OPB18 produced the lowest number of fragments with an average of 6 loci.All primers resulted in polymorphic banding patterns both between and within populations.
The AMOVA results indicated significant genetic differences (p < 0.001) between E. nattereri populations (Table 3).The Φ ST, analogous to F ST (Excoffier et al., 1992), was equal to 0.3, indicating a large amount of interpopulational variation.Of the four models obtained using the Bayesian approach (Table 4), the model with the smaller DIC value (3303.4) was the full model, in which the θ B value was equal to 0.36.However, in this model, the f value was high (f = 0.87), probably the result of an artifact due to the small sample sizes and high number of loci (Holsinger and Lewis, 2003).The free f-model was the second best one according to the DIC value (3350.1),with 1164Silva et al. θ B = 0.34 and f = 0.51, but this free f model was only slightly better than the f = 0 model, in which θ B was equal to 0.29 (DIC = 3357.2).The worst of all models was the one in which θ B was equal to zero (DIC = 7255).So, the Bayesian approach supports results from AMOVA, suggesting a significant between-population component of genetic variation of about 30%.
Pairwise Φ st values ranged from 0.23 to 0.42 (Table 5) and its representation using the two multivariate procedures was not very satisfactory considering the relatively low UPGMA cophenetic correlation coefficient (Figure 3) and the high NMDS stress value for a bi-dimensional solution (S = 0.36).However, increasing the NMDS stress value to a 3-D solution did not improve the stress, so a 2-D visualization was shown (Figure 4).The 2-D NMDS space did not reveal a clear correspondence with the spatial distri-bution of local populations.Local populations 2 and 11 were in opposed genetic space, these populations being in the extreme positions along the northeast-southwest geographical axis.On the other hand, populations such as 1 and 7, which were separated by larger distances, were close in genetic space.Some groups of populations close in geographic space (such as 1 and 8; 3 and 7) were also close in genetic space.
Thus, no clear spatial structure was apparent either from the UPGMA or NMDS methods.Indeed, the Mantel  test showed no globally significant correlation between pairwise Φ st values and geographical distances (r = 0.03; p = 0.45) (Figure 5).Moreover, no spatial structure appeared when comparing genetic distances with short-distance connectivity matrices, situated at distances of less than 50 km (r = 0.05, p = 0.37) or 100 km (r = 0.02; p = 0.45).

Discussion
The RAPD markers indicated that there is a significant population structure of E. nattereri in Goiás State, as found for many other species of anurans worldwide (Lampert et al., 2003;Palo et al., 2004;Smith andGreen, 2004 Telles et al., 2006).Although both AMOVA and Bayesian analyses showed a significant interpopulational variance component of about 30%, the observed population differentiation was not structured in geographic space, as shown by the Mantel tests and NMDS analyses.According to these results, for the relationship between genetic and geographic distances, patterns of population differentiation would approximate to Wright's island model.In our case, allele frequencies in each population are allowed to drift independently without relation to the geographic distances separating them.This pattern probably arises because natural populations are of finite size and dispersal is usually constrained to some extent by geographic distance (see Hutchinson and Templeton, 1999).This is indeed to be expected for E. nattereri, for which high levels of population differentiation could arise because this species reproduces only in the beginning of the rainy season, a fact that contributes to the limited dispersion of individual frogs and high site fidelity.
Our data was in partial agreement with a recent RAPD marker study involving 18 populations of the barker frog Physalaemus cuvieri also sampled in Goiás State (Telles et al., 2006).In that study, only short-distances (> 100 km) structure was found, but not an overall significant correlation between genetic and geographical distances.These results confirmed that amphibian populations tend to be relatively isolated from each other, due to reduced gene flow and absence of migration, even at a fine scale, tending to generate and maintain high levels of genetic population structure (Rowe et al., 1998;Lampert et al., 2003;Burns et al., 2004;Palo et al., 2004).
Our E. nattereri results were very similar to those for the British marsh frog Rana ridibunda in the study conducted by Zeisset and Beebee (2003) using RAPD and microsatellite markers.These authors found that 24.7% of the total variance was between populations and 72.6% was within populations and the Φ st coefficient estimated significant genetic differentiation between most of the R. ridibunda populations studied.A microsatellite marker population study was carried out on Bufo calamita, sampled at distances between 0.5 and 9.0 km, by Rowe et al. (2000) who found an average Φ st value of 0.11, lower than the value found by us for our E. nattereri study.However, other genetic structure studies have reported even higher Φ st values, e.g. the Geocrinia rosea allozymes study by Driscoll (1998), which reported a Φ st of 0.6, and the Bufo canorus single-strand conformation polymorphism study by Shaffer et al. (2000), which found a Φ st value of 0.2.All the reported studies using amphibians have indicated that ecological and behavioral factors can also create genetic structures at local scales according to the species studied (Palo et al., 2004;Rissler et al., 2004;Rowe and Beebee, 2004).In fact, the Φ st value found in our group is higher when compared to other groups, because our geographical range varied from 71 km to 916 km.Thus, some amphibian species also presented high genetic structuring, even when populations sampled were less than 5 km apart (Rowe et al., 2000;Lampert et al., 2003;Burns et al., 2004).Significant F st and Φ st values   found in amphibian populations indicate that dispersal is low, even between pools in close proximity (Kraaijeveld-Smit et al., 2005) and, in our case, the genetic structuring of E. nattereri could be also explained by local extinction and recolonization by a few individual frogs from one or a few source populations.However, we can make no further comments on past fragmentation or other types of range fluctuations since we only sampled a small subset of the species range.Our results presented in this paper provided a preliminary indication of the population structure of E. nattereri, and while RAPD markers are insufficient to infer phylogeographic processes mitochondrial DNA analyses could be used to test these predictions.
Our results have revealed that E. nattereri presented high levels of genetic differentiation.However, the genetic variation between E. nattereri populations was randomly distributed in geographic space and, probably, gene flow was structured at spatial scales smaller than those between our samples.As mentioned above, in the absence of migration-drift equilibrium mediated by geographic space, populations are weakly connected and a large amount of random differentiation between populations is to be expected.In conclusion, this study has greatly improved our knowledge on the genetic diversity of Eupemphix nattereri.Sampling more sites and the use of different markers would be necessary to test for historical evolutionary forces that shape species population structure.

Population structure of Eupemphix nattereri 1163 Figure 1 -
Figure1-Random amplified polymorphic DNA marker analysis using the OPA13 primer and one Eupemphix nattereri specimen from Palmeiras municipality.The analysis was made using the Image Master 1D software (Amersham Pharmacia Biotech, USA).The amplicons varied from 1104 to 3274 bp and band intensity ranged from 20 to 90 pixels for eleven amplified loci.

Figure 3 -
Figure 3 -Unweighted pair group method with averages (UPGMA) dendrogram showing the genetic distance between eleven Eupemphix nattereri populations based on genetic divergence (Φ st ).

Figure 4 -
Figure 4 -Two-dimensional plot showing the relative position of the 11 local Eupemphix nattereri populations in pairwise genetic space (Φ ST ) reduced by non-metric multidimensional scaling (NMDS).

Figure 5 -
Figure 5 -Graphical representation of the Mantel test results between the geographic and genetic distances for Eupemphix nattereri populations in central Brazil.

Table 1 -
The 11 local Eupemphix nattereri populations analyzed, sample sizes, municipalities in the Brazilian state of Goiás and the geographical coordinates.

Table 2 -
Random amplified polymorphic DNA (RAPD) primers, primer sequences and the number of polymorphic loci used in the population study of Eupemphix nattereri from Central Brazil.
(DIC), with smaller DIC values indicating the best model.Estimates of θ B and other parameters obtained in this way incorporate all of the uncertainty in the prior f value (Hol-

Table 3 -
Analysis of Molecular Variance (AMOVA) based on 82 random amplified polymorphic DNA (RAPD) loci for all populations of Eupemphix nattereri from Central Brazil.

Table 4 -
Bayesian analysis of population divergence for 11 local populations of Eupemphix nattereri using the HICKORY version 1.0 program including the tested models, the estimates of f and θ B and the deviant information criterion (DIC).

Table 5 -
Matrix of Φ ST values for each pairwise combination of 11 Eupemphix nattereri populations based on 82 random amplified polymorphic DNA (RAPD) loci.