AFLP Genome Scan to Detect Genetic Structure and Candidate Loci under Selection for Local Adaptation of the Invasive Weed Mikania micrantha

Why some species become successful invaders is an important issue in invasive biology. However, limited genomic resources make it very difficult for identifying candidate genes involved in invasiveness. Mikania micrantha H.B.K. (Asteraceae), one of the world's most invasive weeds, has adapted rapidly in response to novel environments since its introduction to southern China. In its genome, we expect to find outlier loci under selection for local adaptation, critical to dissecting the molecular mechanisms of invasiveness. An explorative amplified fragment length polymorphism (AFLP) genome scan was used to detect candidate loci under selection in 28 M. micrantha populations across its entire introduced range in southern China. We also estimated population genetic parameters, bottleneck signatures, and linkage disequilibrium. In binary characters, such as presence or absence of AFLP bands, if all four character combinations are present, it is referred to as a character incompatibility. Since character incompatibility is deemed to be rare in populations with extensive asexual reproduction, a character incompatibility analysis was also performed in order to infer the predominant mating system in the introduced M. micrantha populations. Out of 483 AFLP loci examined using stringent significance criteria, 14 highly credible outlier loci were identified by Dfdist and Bayescan. Moreover, remarkable genetic variation, multiple introductions, substantial bottlenecks and character compatibility were found to occur in M. micrantha. Thus local adaptation at the genome level indeed exists in M. micrantha, and may represent a major evolutionary mechanism of successful invasion. Interactions between genetic diversity, multiple introductions, and reproductive modes contribute to increase the capacity of adaptive evolution.


Introduction
Why some species become successful invaders is an important issue in invasive biology. When species are introduced into a new region, they face two fates. Some species quickly go extinct, whereas others persist and finally become highly competitive invaders, posing a serious threat to native diversity and ecosystems [1,2]. Successful invasions involve three major phases: introduction, naturalization, and invasion [3,4]. In the initial introduction phase, invasive species often contain low levels of genetic diversity due to bottleneck and founder effects [5][6][7][8]. Then, invaders produce pre-adapted genotypes in response to the abrupt environmental changes during naturalization [9][10][11][12]. Finally, exotic species become broadly invasive in the extended period [9,13]. The rapid population expansion of invaders is expected to promote adaptive evolution, since it has been shown that the rapidly increasing population size is conducive to withstanding (and responding to) strong directional selection [14,15]. A substantial time lag is involved during the transition from introduction via naturalization to invasion [4]. The occurrence of a lag phase allows populations to adapt to new environmental factors such as ecological niche, temperature, precipitation, soils, frost, and wind speed or growing season length [11,12,[16][17][18]. On the other hand, neutral or deleterious alleles, which become favored in new ecological contexts, will contribute to adaptive changes of invasive populations [19]. These changes may increase the survival rate of invasive species [12], making them become gradually dominant in the introduced range [20]. Therefore, preadaptation to novel environments is often counted as a premise for successful invasion [3,4,21].
Genomic scans are useful to identify potential adaptive loci under selection at the genomic level [22,23]. All loci across the genome are anticipated to possess similar demography and neutral evolution history of populations, including genetic drift and gene dispersal [24]. If variation of a locus is beyond the genomic pattern with an unusual frame of higher genetic differentiation, it is deemed an ''outlier locus'' under natural selection [24,25]. The outlier locus can be identified explicitly in the genes under selection and also in neutral flanking regions due to hitchhiking effects [26,27]. In model organisms for which whole genomic information is available, it is easy to track the ''outlier locus'' under selection [28]. However, for non-model organism such as invasive species, it becomes difficult to identify candidate genes and pinpoint the evolutionary and genetic factors involved in invasiveness because of restricted genomic resources [29,30]. New methods, especially those based on the polymerase chain reaction (PCR) to obtain amplified polymorphisms, have frequently been used to scan genomes in non-model organisms [28,31]. Among these, amplified fragment length polymorphisms (AFLPs) are reported to be the most efficient approach to identify candidate genomic regions under selection (candidate loci or outlier loci) [22,23,[31][32][33][34][35][36][37]. They can provide several hundred random loci scattered throughout the genome at less cost [37]. So far, the most commonly used analysis approaches for AFLP genome scans are Dfdist and the hierarchical Bayesian method (Bayescan). Dfdist, which was originally developed by Beaumont & Nichols [38], employs a classical Wright's island model to generate the expected neutral distribution of F ST estimates [39][40][41]. In contrast, Bayescan relies on a logistic regression model [42], representing an extended method devised by Beaumont & Balding [43]. To decrease false positives, Dfdist mainly depends on the trimmed mean of the empirical F ST distribution for the simulation [44], whereas Bayescan uses a likelihood ratio test to evaluate the most likely of two alternative models, one that includes the effect of selection and another that excludes it [41,42]. Bayescan has been suggested to be more efficient at detecting high selective loci with low false positives [45].
Mikania micrantha H. B. K. (Asteraceae) is one of the top ten worst weeds in the world [46]. It is a many-branched climbing perennial vine native to tropical Central and South America where it is a weed of minor importance [46][47][48]. M. micrantha was deliberately introduced into Asia as early as 1900s [49]. In humid sub-tropical China, M. micrantha is subjected to new selective pressures, including such abiotic and biotic stresses as new climate, soils, pathogens, herbivores, pollinators, and competitors [50]. By adaptive evolutionary changes, M. micrantha survived and reproduced in new environments using strikingly different strategies from those employed in its original niche [19]. It demonstrated the ability to outcompete native plants in utilizing such limited resources as soil nutrients and sunlight and releasing phytotoxic compounds to inhibit growth of neighboring plants [51][52][53][54]. Once established, M. micrantha became a dominant plant across most regions, influencing ecosystems, biological diversity, and natural communities [50]. By the late 1980s and early 1990s, it had spread extensively in southern China, colonizing agricultural land, orchards, nurseries, lawns, mangroves, secondary forests, scrubland, waste ground, ponds, and seashore [55,56]. Unlike in its native range, in southern China M. micrantha grows on dry soils as well as shady sites [56]. Its favorable growth conditions have changed to an average annual temperature higher than 21uC and soil moisture content over 15% [52]. Since spreading at accelerating rates, it is reasonable to postulate that adaptive evolution provides a key mechanism allowing the success of M. micrantha in new environments [31,57]. Such adaptive processes may leave specific signatures in the genome of M. micrantha [36]. Thus we expected to detect the ''outlier locus'' signature of such local adaptations under selection at the genomic level. It is particular important to find the ''outlier locus'' associated with the local adaptation because, firstly, the information is critical to understanding the introduction history and genetic consequences of introduced populations of M. micrantha [2,30]. Second, it may provide fresh insights into the evolutionary potential of M. micrantha populations, which is helpful to predict their adaptability in response to management practices [9,58]. Finally, the locus will offer a prime candidate for functional surveys targeting the linked gene and dissecting the molecular mechanism of invasiveness [36].
To test our prediction that outlier loci can be found in southern China populations of M. micrantha ( Figure 1; Table 1) and to detect genetic structure, in the present study we performed a genome scan analysis based on a large number of AFLP polymorphisms. The goals of the study were (1) to identify candidate loci under selection for local adaptation in M. micrantha and (2) to understand contributions of interactions between genetic diversity, reproductive modes, bottlenecks, and multiple introductions to the adaptive evolution of M. micrantha.

Population genetic variation
AFLP analysis of M. micrantha was optimized by determining which selective primer sets produced the clearest DNA fragments based on the eight primer pair combinations. The eight primer pairs were polymorphic in each population. A total of 483 clear polymorphic fragments were detected. The proportion of polymorphic loci within a population ranged from 10.57% to 73.47%, with an average value of 41.84% ( Table 2). The average number of fragments per individual was 338.39, and 67.75 per primer combination. At the regional level, either Hong Kong or Zhuhai was consistently found to maintain the highest level of variation according to all the genetic diversity statistics (Table 2). However, populations such as MA106, SZ72, NLD10, NLD26, and NLD30 growing in regions of Macao, Shenzhen, and Neilingding were observed to possess lower amounts of genetic variation than others in the same region ( Table 2).

Linkage disequilibrium and character compatibility
We estimated levels of linkage disequilibrium (LD) and character compatibility among the 483 AFLP loci. Each region was observed to possess significant LD and matrix compatibility obtained by testing the index of association (I A and r r d ) and the incompatibility excess ratio (IER) ( Table 2). Likewise, significant LD and matrix compatibility were also detected at the population scale ( Table 2).

Allele frequency distribution and bottleneck signature test
Alleles were defined as rare when they occurred at a frequency of less than 0.05 in the examined populations [59]. In total, 22 rare alleles were identified at the species level. Across regions, 14, 12, and 9 rare alleles were detected in Hong Kong, Shenzhen, and Neilingding, respectively, whereas no rare alleles were found in Macao, Zhuhai, or Dongguan. In addition, no rare allele was detected within any populations at the population level.
Using the stepwise mutation model (SMM) and the infinite allele model (IAM), we identified bottleneck signatures in each population with a heterozygosity excess/deficiency ratio that significantly deviated from the expected ratio (1:1) at mutationdrift equilibrium ( Table 3, P,0.05). Moreover, the entire range of M. micrantha appeared subject to a significant bottleneck (Table 3).

Population genetic differentiation and relationship
The genetic differentiation W ST measured by AMOVA analysis was 0.3335; 66.65% of the variation was partitioned within populations, 28.61% was attributed to differences among populations within regions, and only 4.74% of the variation was due to regional differences (all three hierarchical levels were significant with P,0.001) ( The f-free model chosen based on the smallest mean DIC was more suitable than other models for the AFLP data set of the 28 populations. Therefore, h B = 0.2927 (95% credible interval: 0.2772-0.3058) was determined to be an unbiased Bayesian estimate of all population genetic differentiation.
No significant associations between geographical distances and pairwise estimates of h B were observed in the full AFLP data set (r = 0.0032, P = 0.452). A similar pattern was derived at the regional level. Hence no evidence of ''isolation by distance'' [60] was revealed by Mantel tests at either the regional or entire range level.
The DK criterion of Evanno et al. [61] was applied to estimate the number of population clusters. The maximal value of DK was K = 24, indicating the need to divide the samples into 24 clusters. Bar plots showed varying extents of admixture among populations ( Figure 2). Ten clusters including populations NLD10, NLD26, NLD30, NLD31, SZ35, SZ72, HK87, HK90, MA105, and MA106 were detected, although each population also contained a very few individuals from different regions. The substructure of the rest populations was weak and could only be resolved when information about the sampling locations was included. A mix of individuals from other clusters was often observed in each cluster. The results suggested that M. micrantha populations in southern China maintained a relatively weak genetic structure.
In addition, the UPGMA tree rooted with M. cordata revealed that populations of M. micrantha from different locations usually were intermingled (Figure 3), suggesting a lack of geographic pattern. HK87, which was documented as the earliest introduced population in China [49], formed an independent branch, one that first diverged from the other M. micrantha populations.

Detection of signatures of positive selection
The AFLP data set of M. micrantha was used in a global analysis for outlier detection, both with Dfdist and Bayescan (Table 5). By using Dfdist on the 28 populations of M. micrantha, 23 out of 483 loci (4.8%) were identified as outlier loci under directional selection at the 99.5% probability level ( Figure 4A). Bayescan analysis produced high differentiation loci at a threshold of log 10 PO.2.0 (posterior probabilities higher than 0.99), corresponding   5B). Nevertheless, this lack of detection of any outliers may also be due to lower sample size.

Discussion
To the best of our knowledge, this study is the first report on detection of candidate loci under selection by genome scan in the invasive weed M. micrantha since its introduction to southern China. To avoid spurious outlier loci, several authors advocated employment of two or more outlier detection methods and a conservative significance level [25,28,[62][63][64][65]. In this study, the AFLP explorative genome scan has revealed 14 loci as under selection among a total of 483 loci in M. micrantha. These 14 loci are considered to possess high credibility because they were picked Currently, there is increasing interest in identifying genes or outlier loci that underlie local adaptations in invasive species [66][67][68]. Understanding the process of invasive adaptive evolution is critical to the introduction history and genetic consequences on introduced populations, the planning of management strategies of invasive species, and for identifying the possible risks of introduced non-native species in the future [2,30]. Because selection for local adaptation is unlikely to simultaneously occur at many loci [69], outlier loci are limited to a minor part of the genome, which has been suggested in other species [28,63,[69][70][71]. Our results indicate that only a small part (2.9%) of the genome of M. micrantha has been under directional selection during its invasive processes. The percentage of loci detected is slightly lower than the 5-10% reported in the generality of AFLP genome scans [72,73]. As AFLP loci are likely located in non-coding DNA, some of the outlier loci may only exhibit the signature of selection because they are linked to the actual target [26,37]. Although it is difficult to know the location and function of the loci involved in the adaptation to invasiveness, a genome scan of M. micrantha still offers unique opportunity to unravel the genetic basis of invasive adaptation without known phenotypes and whole genome sequences. In particular, the AFLP primers that were developed to amplify the outlier loci identified here can be directly used to construct a reduced representation library of the M. micrantha genome, which will allow efficient sequencing of the linked genomic regions by next-generation sequencing technology [74].
Genetic diversity is anticipated to increase the adaptive potential of invasive populations in the new environment [4,9]. In invasive species, the alleles from standing genetic variation have been supposed to control the adaptive evolution [29,30], because favorable alleles compared to neutral or deleterious alleles are immediately available and often occur at a greater frequency in populations [19,30]. Compared with other non-invasive and invasive weedy species (Table 6), we have detected relatively high levels of genetic diversity in the introduced populations of M. micrantha in southern China. This substantial genetic variation enables to provide a large pool of raw material for adaptive evolution [30,67].
High genetic diversity can be created by multiple introductions, which bring together large amounts of genetic variation and novel genetic combinations [8]. Such admixtures elevate the expansive ability and pace of proliferation of invasive species and are able to enhance adaptive evolutionary responses to novel environmental selection [10,12,19,75]. Based on the UPGMA and STRUC-TURE analysis results, closely related populations often are from different geographical locations (Figures 2 and 3), while individuals from the same location do not group together. Thus multiple introductions are inferred among populations of M. micrantha. In addition, the significant genetic differentiation detected among M. micrantha populations is also consistent with the expectation that each population has experienced severe bottleneck (Table 3). It is generally suggested that bottlenecks tend to increase genetic differentiation among populations by changing allelic frequencies [76,77] and they are perceived to reduce the potential for adaptive evolution as well [29,78]. Nevertheless, multiple introductions can ameliorate this effect and accelerate the speed of adaptive evolution [8]. Such phenomena have been recorded in the populations of other invasive species including Heracleum mantegazzianum [79], Hieracium lepidulum [80], and Alliaria petiolata [81]. Furthermore, no geographical signature is revealed in the AFLP variation pattern among the introduced populations of M. micrantha at either regional or entire range scale. Repeated or secondary introductions, which are induced by long-distance dispersals of copious amounts of fine and fluffy seeds, pollination or propagules mediated by humans, insects, and violent winds [82], may further complicate the geographic consequences of the population genetic variation of M. micrantha. Therefore, multiple introductions can add to the successful invasiveness of M. micrantha by increasing genetic variation, preventing genetic bottlenecks, and strengthening the capacity of adaptive evolution. Adaptive evolution of invasive species is also determined by the reproductive system such as the amount of sexual and asexual reproduction and the patterns of mating [19]. Mikania micrantha mainly propagates by seeds, but its local spread mostly results from vegetative propagation [56,83,84]. When reproducing vegetatively, M. micrantha generates shoots from small stem fragments and rosettes [83,84]. In this study, significant LD and locus compatibility have been identified in a majority of M. micrantha populations ( Table 2), indicating that asexual reproduction is the predominant propagation system. In addition, self-fertilization (selfing) has also been reported in the southern China populations of M. micrantha [85]. The results are in accord with the findings that most invasive plants reproduce asexually or by selfing [86][87][88], and reproductive systems may be subject to evolutionary modification during invasion such as switching to higher levels of inbreeding or vegetative reproduction [5,19,[88][89][90][91]. For species that are able to reproduce both sexually and clonally, sexual reproduction would promote allele recombinations that result in high genetic variation, whereas clonal reproduction preserves successful combinations and maintains genetic variation by occasionally creating new alleles through somatic mutations [92][93][94][95]. The specific reproductive modes (asexual and selfing) of M. micrantha are likely to enhance its adaptive potential in southern China.
In summary, this study is the first report to identify outlier loci under selection in the genome of the invasive weed M. micrantha without reference to coding vs. noncoding sequences, or specific models of selection by explorative genomic scans [25]. We find that 2.9% candidate loci are under selection, indicating that local adaptation has indeed occurred in M. micrantha since its introduction to southern China. This study also demonstrates high genetic variability, strong genetic differentiation, severe bottlenecks, and multiple introductions, as well as primarily asexual reproduction in the introduced populations. The interactions between genetic diversity, multiple introductions, and reproductive modes are revealed to be involved in the response of M. micrantha to the new local environment, allowing a more precise understanding of its successful invasion. In future studies, we will isolate and sequence the outlier AFLP bands to identify their genomic locations and neighboring genes to further detail the adaptive molecular mechanisms of M. micrantha.

Ethics statement
This study was conducted in accordance with all People's Republic of China laws. No specific permits were required for the described field studies. No specific permissions were required for the locations/activities described in this study. The location is not privately owned or protected in any way. The field studies did not involve endangered or protected species.

Plant material
For M. micrantha, individuals were sampled from 28 populations in six regions (Hong Kong, Macao, Shenzhen, Neilingding, Zhuhai, and Dongguan) representing different habitats ( Figure 1A; Table 1). In each population, we collected fresh leaf material from 12 to 19 randomly selected individuals (Table 1). These samples represented a large number of individuals of M. micrantha throughout its introduced range in southern China. To further confirm that the outlier loci under selection in M. micrantha were caused by novel environments, we tested whether AFLP loci can demonstrate evidence of selection in the congeneric species M. cordata, which is native to sub-tropical China. For M. cordata, five to 12 individuals were collected from its two populations (HN1 and HN2) located on Hainan Island ( Figure 1B; Table 1). Leaves were preserved in silica gel. Voucher specimens (M. micrantha: CGP1-409; M. cordata: CGP410-426) were deposited at the herbarium of Sun Yat-sen University (SYS), Guangzhou, China.

DNA extraction
Total genomic DNA was extracted from ground tissue, following the modified CTAB protocol [96]. The quality and quantity of the DNA were determined with a spectrophotometer (Pharmacia 2000 UV/Visible, Amersham Pharmacia Biotech, Piscataway, NJ) and on 0.8% agarose gels.

AFLP protocols
AFLP analyses were performed according to Vos et al. [97] with the following modifications: genomic DNA (50 ng) was digested with 2 U EcoRI and 4 U MseI (New England Biolabs, Ipswitch, MA) for 3 h at 37uC in 16NE buffer and incubated at 70uC for 20 min. To ligate the resulting fragments to the corresponding adapters, 10 ml of restriction products were added into a 20 ml reaction mixture containing 0.1 mM EcoRI adapter, 1 mM MseI adapter, and 60 U T4 DNA ligase (New England Biolabs). After being incubated at 20uC for 3 h, the samples were diluted 10 times with ddH 2 O, then 2 ml of each sample was used as a template to conduct the pre-amplification in a final volume of 20 ml that contained 16PCR buffer, 100 nM each of EcoRI+A and MseI+C primers, 0.24 mM dNTPs, and 1.5 U Taq DNA polymerase. The preamplification reaction was carried out for 20 cycles of 30 s at 94uC, 1 min at 56uC and 1 min at 72uC. After that, preamplified product was diluted 1:50 with ddH 2 O, and then used as template for the selective PCR amplifications to generate AFLPs.   . The selective PCR products were separated by electrophoresis on 6.5% polyacrylamide gels on an ABI 377 automated sequencer (Applied Biosystems, Carlsbad, CA). An ROX-500-labeled internal size standard (Applied Biosystems) was added to each sample to size fragments.

Data analysis
Software GeneScan 3.7 (Applied Biosystems) and genographer (version 1.6.0; http://hordeum.oscs.montana.edu/genographer) were utilized to collect and score raw fluorescent AFLP data. The presence (1) or absence (0) of data from unambiguous AFLP bands was used to establish the matrix of genetic identity of the sampled individuals. Genetic diversity statistics, including percentage of polymorphic loci and Nei's gene diversity [98], were calculated using POPGEN32 software [99]. Shannon's index of phenotypic diversity was quantified as S = 2g pilnpi where pi is the frequency of a given AFLP band in the population [100].
Analysis of Molecular Variance (AMOVA) based on a Euclidean squared distance matrix was hierarchically calculated to estimate the allocation of genetic variation among and within populations by ARLEQUIN 3.0 (available at http://cmpg.unibe. ch/software/arlequin3) [101]. In this analysis, the AFLP data set was partitioned at three levels: regional, among-population, and within-population. One thousand random permutations were used to infer the significance of the variance components [101]. Also using the same software, a Mantel test was performed to investigate the correlation between genetic differentiation and geographic distances (km) among populations. The matrix of genetic differentiation was composed by pairwise h B estimates.
Holsinger et al. [102] proposed a Bayesian approach to estimating genetic structure for dominant and co-dominant markers. The nearly unbiased parameter estimations of heterozygosity, genetic distance, and population differentiation can be obtained using the Bayesian method [102]. Its f and h B are equivalent to the inbreeding coefficient (F IS ) and the fixation index (F ST ) of F-statistics, respectively. The posterior distributions of f and h B were estimated through Markov Chain Monte Carlo (MCMC) methods by HICKORY v1.0, with a burn-in of 50 000 iterations and a sampling run of 250 000 iterations from which every fiftieth sample was retained for posterior calculations [102]. The analysis model was chosen based on the deviance information criterion (DIC). The ''f-free'' model, where f was not estimated but was chosen at random from the prior distribution, was decided due to its smaller DIC than for other models. Alleles were deemed rare if their frequencies were #0.05 in the sampled populations [59].
To gain further perspectives on genetic structure, we also employed the Bayesian clustering method to infer the pattern of population structure by implementing the STRUCTURE 2.2.3 program [103,104]. To determine the best number of clusters, 30 independent runs of K (K = 1 to 30) were performed with an admixture model at 100 000 MCMC iterations and a 10 000 burn-in period. We used DK, the second-order rate of change in ln P (X|K) for successive values of K to determine the number of clusters [61]. The distribution map of STRUCTURE was plotted according to K value at the highest log likelihood. Moreover, based on Nei's genetic distance [105], an unweighted pair group method with arithmetic mean analysis (UPGMA) was used to generate a dendrogram of the relationships among the populations by TFPGA 1.3 [106]. One thousand bootstrap replicates were permuted to assess the reliability of the UPGMA dendrogram. We further tested whether populations have suffered a bottleneck using the program BOTTLENECK [107]. The heterozygosity (Heq) expected at mutation-drift equilibrium was calculated based on both the stepwise mutation model (SMM) and the infinite allele model (IAM) [108,109]. The significance of heterozygosity excess was determined by the sign test [110].
Linkage disequilibrium, which is an association of alleles at different loci on chromosomes in a population, is mainly affect by population size [111]. The dependency of population size can be removed by multilocus linkage disequilibrium with both indices of association I A [112][113][114] and its modified measure r r d [115]. These indices and their significance by randomization were computed using the software Multilocus v1.2 (http://www.bio.ic.ac.uk/ evolve/software/multilocus/).
A character incompatibility analysis was carried out to probe the predominant mating system in the populations at the molecular level [116][117][118]. In a pair of binary character data, such as the presence or absence of AFLP bands at two loci, the presence of all four possible combinations of characters (0/0, 1/0, 0/1, 1/1) is more parsimoniously explained by sexual recombination than by three mutation events. Once all four character combinations arise, an incompatibility is said to occur, and can be used as a measure of recombination [116]. The incompatibility excess ratio (IER) was calculated using PICA 4.0 by comparing the observed incompatibility count for the original data and mean incompatibility count for randomly permuted data [118]. If none or a small fraction of incompatible loci occur, asexual reproduction is implied to have happened [80,119].
To detect outlier loci under selection for local adaptation of M. micrantha and M. cordata, two complementary methods were applied to our AFLP data set. We first used Program Dfdist, which is the most popular software for detecting candidate loci [120]. Dfdist software (http://www.rubic.rdg.ac.uk/,mab/stuff) was recently modified from Beaumont and Balding [43] to analyze dominant data, which is a hierarchical Bayesian approach based on summary statistics in a symmetrical island model [24,39]. Outlier loci were detected by comparing empirical F ST values for each locus against a null distribution of F ST values expected from a neutral drift model [34]. A potential defect of the Dfdist approach for detecting outlier loci is the possibility of false positives [121][122][123]. The statistical power was enhanced to avoid false positives by setting a conservative constraint [22]. A null distribution of F ST close to the empirical distribution was acquired by 50 000 coalescent simulations. Simulations were computed with a mean F ST similar to the trimmed mean F ST , which was calculated by excluding 30% of the most extreme F ST values observed in the empirical dataset [71,124]. The 4Nm parameter value was set to 0.04 in all simulations. A global analysis was done for M. micrantha and M. micrantha using 28 populations and 2 populations, respectively. The threshold for outliers was set to the more conservative 0.005, estimated from simulated F ST values to control for false positives [28,33,125]. In addition, Dfdist assumes that populations are at migration-drift equilibrium, which does not often occur in natural populations [124]. Hence, to cross-check the reliability of the outlier loci detected by Dfdist, we also ran Bayescan software (http://www-leca.ujf-grenoble.fr/logiciels.htm), which better handles dominant marker data by directly estimating the posterior probability of a given locus to be under selection [42]. Assuming that allele frequencies within populations follow a Dirichlet distribution [126][127][128], the Bayesian method not only permits for different demographic scenarios and different amounts of genetic drift between populations when estimating populationspecific F ST coefficients, but also considers all loci in the analyses [124]. The Bayesian approach also disposes of the problem of multiple testing of a large number of genomic loci through prior distribution [124]. In our genome scan, the log 10 PO.2.0 was considered a threshold value for determining loci under selection according to Jeffreys' interpretation [129], which is a logarithmic scale for model choice as follows: log 10 PO.0.5 (substantial); log 10 PO.1.0 (strong); log 10 PO.1.5 (very strong); and log 10 PO.2.0 (decisive support for accepting a model) [36]. We used 10 pilot runs of 5000 iterations to estimate model parameters. A burn-in of 50 000 iterations was employed to cover the MCMC. The sample size was set to 5000 and the thinning interval to 20, resulting in a total chain length of 150 000 iterations [45]. The loci were ranked according to their estimated posterior probability and all loci with a value over 0.993 were retained as outliers. This corresponds to log 10 PO.2.0, which provides decisive support for acceptation of the model. Outliers identified by both Dfdist and Bayescan are likely to be truly adaptive regions of the genome, because the two approaches differ in algorithms and assumptions [27].