Retraction Note: Comparison of traditional and new generation DNA markers declares high genetic diversity and differentiated population structure of wild almond species

Editor's Note: this Article has been retracted; the Retraction Note is available at https://www.nature.com/articles/s41598-020-72522-x.

Due to various distributions and evolutionary histories, there are many different wild relatives of crops. In species complexes related to crops, a number of clades may have occupied original regions fairly recently, for example after the last glacial period, and migration bottlenecks were also possible over this time 7 . These processes are inadequately understood in the majority of non-domesticated plants, although they may have a crucial impact on the conservation/breeding programs exploiting wild relatives. Introgression may serve as a bridge in the spatial variation in wild resources and cultivated genotypes. Finally, inherent diversity in non-cultivated species are adapted to specific environmental conditions, which might be utilized through introgression in crops 8 .
Knowledge on genetic diversity and relatedness is important to design strategies for preservation, as well as organizing accessions of wild species with appropriate attributes, designed for use as breeding material. Therefore, molecular characterization has become the most important tool for evaluating genetic diversity and similarity within and between populations or accessions, mapping of functional genes, assembly of genetic linkage maps, marker-assisted selection (MAS), and phylogenetic experiments in crop species 9 .
In addition, several molecular marker techniques based on retrotransposons (RTNs) have been developed [10][11][12] . Distributions of RTNs in plant genomes provides the basis for developing marker systems, and compare them with other markers RTN markers seem to be useful and polymorphic in many species 13 . The need for RTN sequence information is a major limitation to design family-specific primers. A number of studies have demonstrated that due to homology, RTN sequences between long terminal repeat (LTR)-RTN plant families can be used across species 14,15 .
Evaluation of genetic diversity of the genus Medicago using Tms1Ret1 LTR-based sequence-specific amplification polymorphism (S-SAP) had a high marker index for S-SAP compared to amplified fragment length polymorphism (AFLP) and selective amplification of microsatellite polymorphic loci (SAMPL) 16 . Conversely, S-SAP relies on digestion and the sensitivity to DNA methylation of generally-used enzymes, for example, PstI and EcoRI, is associated with the variable degree of CG and CXG methylation in plant DNA 17 , which means that a number of evident polymorphisms could be neither sequence-based nor heritable. Inter-retrotransposon amplified polymorphism (IRAP) and retrotransposon-microsatellite amplified polymorphism (REMAP) are two RTN-based markers that do not require DNA digestion and can be used in genetic differentiation 12,13,18 .
Due to a high degree of polymorphisms, simple-sequence repeats (SSRs) have been recognized as valuable markers in plants and animals, and therefore the majority of genetic studies in Prunus species have usually been based on SSR markers [19][20][21] . Although the majority of these SSRs have been developed in peach 22 , it has been established in numerous studies that these markers were robust in other Prunus species such as cherry, almond, or plum 21,22 . On the other hand, an important drawback of SSR analysis is the elevated price of the fluorescent labels. Conversely, SNP markers are present in large quantities in the plant genome 23 and thus can be used as genetic markers for cultivar recognition, assembly of genetic maps, measurement of genetic diversity, or marker-assisted breeding 24,25 . In addition, the recognition of SNPs and INDELs has been simplified by the current progress in sequencing technology. HRM analysis was developed to identify SNPs in PCR amplicons as it is a simple and low-cost method 25 . Recently, this advance has been used for genetic mapping 24,26 . HRM was also used to check mutation in large multi-exon genes to discover disease-related alterations in humans 26 and construct linkage maps in crop species 24 . Wu et al. 25 used HRM analysis to detect SNPs in EST sequences that have been retrieved from a database.
This study was carried out to evaluate the usefulness of formerly published (REMAP, IRAP and S-SAP) and newly designed (IRAP and REMAP) RTN-based markers compared to formerly published (ISSR, SSR, EST-SSR and AFLP) and newly designed (EST-SSR) non-RTN-based markers for the discrimination and phylogenetic analysis of Prunus species. In addition, the current investigation applied the HRM approach to identify SNPs in wild almond species based on publicly available Prunus ESTs and next generation sequencing (NGS) data, and established a technique for SNP detection and genotyping to characterize genetic variation in wild species that might be exploited in almond breeding programs. Data were also informative on demographic and evolutionary history of species that are close relatives of a long cultivated crop plant.

RTN activity in wild almond species and IRAP analysis.
To analyse genetic variations in 389 individuals of 18 wild almond species (Table 1) a total of 22 single, and 60 IRAP primer combinations ( Table 2) from Prunus and non-Prunus RTN families were used. All single IRAP primers (designed based on Sukkula, Daniela, Fatima and RTNs LORE1 and LORE2) generated discernible and polymorphic banding patterns (a representative banding pattern is shown in Fig. S1). Single IRAP primers (designed based on RTN Tps12a from Pisum sativum) also amplified scorable but not polymorphic banding patterns. IRAP amplification of RTNTps19 (from pea), showed a non-scorable banding pattern at low annealing temperature (less than 40 °C) (results not shown).
IRAP primer combinations of RTN families of Prunus and other origins created polymorphic and scorable banding patterns, except the primer Tps19. Ten out of the 19 IRAP primers produced 225 loci. Of these 225 amplified loci, 157 were polymorphic (69.9%). The primer combination Bare 1LTR-Sukkula amplified the most polymorphic loci. The average number of polymorphic loci was 9.32 per primer. The size of the amplified IRAP loci ranged from 75 to 2,000 bp (Table 2).
IRAP-based UPGMA dendrogram grouped populations into three major groups (Fig. 1). Group I integrated populations CAR, COM, ELA, FEN, KOR, ORI and TRI. The second group included REU, GLA, HAU, PAB and SCO (abbreviations of populations are explained in Table 1). Populations from the Lyciodes section were found in the third group. Similarly to the results of the cluster analysis, the populations EBU, ERI, URU, ARA, and LYE were distinct from the rest of the population on PCA biplot derived from IRAP markers (data not shown). To evaluate and partition the total genetic variation between and within populations, AMOVA was carried out based on the eight populations using IRAP data. Significant differences (P < 0.05) were evident within populations. The level of genetic variation was higher between populations (80%) compared to within populations (20%).

R E T R
The characteristics of amplified IRAP loci using 19 primers are displayed in Table 3. The percentage values of IRAP polymorphic loci in population ranged from 47.6 (FEN) to 89.6 (KOT), with an average of 80.44. The frequency of all IRAP amplified loci was greater than 5%. A genotype-specific locus was generated in the population CAR by IRAP primer combination Tms1Ret1-LORE1. Primer LORE2 amplified a specific locus in a genotype of population EBU. Mean heterozygosity varied from 0.208 (ORI and SCO) to 0.261 (CAR), averaging 0.234 (Supplementary Table S1).
The primer combination Tms1Ret1-A7 created a monomorphic banding pattern. No bands were amplified using REMAP primer combinations with RTN Tps19. Of the391 amplified bands, 300 were polymorphic (71.2%) (a representative banding pattern is shown in Fig. S2). The average number of polymorphic bands was 8.82 per primer. The size of the amplified fragments varied from 100 to 2,500 bp.
There were three well-known important clusters, based on REMAP data (Fig. 2). Group I consisted of two subgroups: CAR, COM, ELA, FEN, KOR and KOT in subgroup I, and ORI and TRI in subgroup II; the species belonging to the Lycioides section were grouped in group II. The third group consisted of the populations REU, GLA, HAU, PAB, and SCO. Associations between 18 populations using REMAP markers were well-established through PCA (data not shown). REMAP-based AMOVA was approved using the 18 populations for the basis of the analysis. Similar to the results acquired by the IRAP method, the level of genetic variation was higher between populations (92%) compared to within populations (8%), demonstrating clear discrimination of the studied populations. The characteristics of amplified loci by means of 34 REMAP primers are shown in Supplementary  Table S1.
The percentage of polymorphic loci in a population based on REMAP data varied from 63.2 (KOT) to 88.6 (KOR), with an average value of 78.9. Two loci with a frequency of less than 25% were detected (Supplementary Table S1), suggesting that most of the amplified loci are common in the studied populations. Nine genotype-specific REMAP loci were identified in populations KOR, KOT, TRI, ERI, URU, ARA, HAV, PAB and SCO. The mean of heterozygosity varied from 0.221 (HAU) to 0.275 (FEN), with an average of 0.267.
Combined data IRAP and REMAP analysis. The combined data of IRAP and REMAP markers were utilized to construct a dendrogram to calculate the influence of both techniques when taken together. Populations were assembled into three most important clusters (Fig. 3). Group I included COR, COM, ELA, KOR, KOT, ORI and TRI. Populations EBU, ERI, URU, ARA and LYC clustered in group II and REU, GLA, HCU, PAB, and SCO were classified into group III. A PCA of the IRAP + REMAP data was used to validate associations between eighteen populations.
The first (PCA1) and second (PCA2) principle coordinates accounted for 33% and 21% of the total variation, respectively. Relationships between populations on the biplot were similar to the results acquired from cluster analysis. AMOVA was carried out to investigate total genetic variation between and within populations. Similar to that shown by IRAP and REMAP markers, the level of genetic variation was higher between populations (83%) compared to within populations (7%). The percentage of polymorphic loci in populations varied from 78.5 (TRI) to 86.35 (HAU), with an average of 81.9. The mean of heterozygosity varied from 0.206 (SCO) to 0.272 (TRI), with an average of 0.242 (Supplementary Table S1). Due to high genetic diversity between populations, IRAP + REMAP-based cluster analysis was carried out using complete linkage algorithms based on simple matching (SM) similarity coefficient to recognize groups among all 389 wild almond accessions.  SSR and EST-SSR for fingerprinting Prunus spp. As there is a lack of a standardized set of microsatellite primers for almond, primer pairs of closely related species have been regularly used in molecular descriptions [27][28][29] . In this study, amplification of genomic DNA was performed for 219 SSR and EST-SSR loci designed from different Prunus species (peach, almond, plum, sweet cherry, and sour cherry). For every examined locus, calculated indices are shown in Supplementary Table S2. Data of nine primer pairs with either in distinguishable amplification (PS01H03, EPPISF027, UDP96-001, CPDCT027, BPPCT025, and CPPCT006), stutter bands (UDP96-019), or monomorphic bands (EPPISF014 and EPPISF025) were excluded from the analysis. The rest of the 210 primer pairs provided clear and polymorphic multi-allele bands for all accessions of almond and related Prunus samples.
The total number of examined alleles (na) was 16,948, ranging from 50 to 90, with an average of 81.1 alleles per locus. SSR primers derived from the non-coding DNA region had more alleles (17,141 alleles) than EST-SSR primers (14,189 alleles). The average number of alleles per locus were similar in both types of SSRs (80.8 and 81.1 alleles per locus for SSR and EST-SSR, respectively). PruAvest-3, PruMest-5, Prupest-2 and PruPest-36 had the highest number of alleles (90).
The average frequency of the major allele was 0.36, with a maximum of 1.10 in EPPISF018 and a minimum of 0.25 in the BPPCT039, PruArest-4, PruAvest-4, PruAvest-24, PruMest-12, PruDest-5, PruPest-15, PruPest-35, PruPest-55 and PruPest-75loci. The high allelic diversity of microsatellite loci revealed high genetic variation in the wild almond germplasm investigated in this study. The sizes of the amplified DNA fragments of all the loci ranged from 70 to 508 bp. The smallest DNA fragment (70 bp) amplified was in an accession of P. scoparia (accession number #350) in locus UDP98-412. Allele sizes in this locus ranged from 70 bp in P. scoparia (accession number #350) to 191 bp in P. orientalis (accession number #130), which was generally lower compared to other loci. The largest allele, 508 bp, was found in the pchgms5 locus in the accession of wild almond species of P. elaeagnifolia (accession number #66). The locus pchgms5 amplified a diverse range of alleles for different groups of accessions. The allele sizes of this locus ranged from 429 to 508 bp in cultivated almond and some related species (P. orientalis, P. spartioides, and P. lycioides), while in P. chameamygdalus, P. leptopus, peach and plum, the size of the amplified alleles was 139-162 bp. The greater difference in the size of the amplified alleles was related to the presence of most divergent samples (from different locations and species) used in this experiment.
In germplasm studies, the PIC was greater in the EST-SSR loci (0.919) than in the SSR loci (0.874), which showed that EST-SSR markers were more suitable than SSRs for recognizing genetic variation. The 174 EST-SSRs used in this experiment were developed from the Prunus genus. They had a range of polymorphism (PIC = 0.699-0.971) in the samples studied.

SNP identification and detection from EST and RNA-seq databases. All EST (111,699) sequences
of Prunus species, namely peach (P. persica; 80,797), apricot (P. armeniaca; 15,105), sweet cherry (P. avium; 6,035), Japanese apricot (P. mume; 4,589), almond (P. dulcis; 3,864), sour cherry (P. cerasus; 1,255) and European plum (P. domestica; 54) were downloaded from GenBank (ftp://ncbi.nlm.nih.gov/genbank/genomes/). To construct longer and less redundant sequences, publicly-available ESTs were assembled from CAP3. CAP3 is a commonly used program 29 that identifies overlapping sequences and creates contigs with consensus sequences. A total of 111,699 Prunus EST sequences were brought together into 12,159contigs, and 125 EST contigs were predicted to have SNPs with a redundancy score ≥2 according to Wu et al. 25 , and adequate flanking sequences for primer design. In addition, 100 EST contigs with putative SNPs were chosen from a Prunus SNP database at ESTree for HRM analysis. In total, 125 EST contig sequences were used for primer design to confirm and/or recognize SNPs by HRM analysis. The sequences of the SNP primers used in the study are shown in Supplementary Table S3.  61  58  55  64  48  53  56  64  50  47  49  68  66  42  41  53  56  60 Number        The value on the dendrogram gives the stability of nodes estimated with a bootstrap procedure (no number indicates support less than 10%). Bar graphs showing genetic diversity structure for the 389accessions of Prunus species as assessed using STRUCTURE software. Each group is represented by a different color (blue, red and green).

R E T R
accession numbers of the SNPs are available in the dbSNP database (http://www.ncbi.nlm.nih.gov/SNP/). All four classes of SNPs 25 were identified by HRM analysis, and the genotypes with different SNP alleles were distinguished by distinct melting profiles. A/T transversions were considered to be the SNP difference most difficult to resolve by melting analysis 31, 32 . Genetic relatedness of the accessions of Prunus spp. based on SNP-HRM. A total of 389 accessions of wild almond species were genotyped using HRM analysis, and data were used for phylogenetic analysis. The results demonstrated that apanel of 17 exon fragments as well as SNPs, INDELs and a microsatellite were able to determine the genetic differences between the accessions of studied wild almond species. At the hierarchy levels, where a small group of the accessions was formed, the bootstrap values were relatively high for most of the clusters, while at higher hierarchy levels, the bootstrap values became much lower (most frequently they reached the value 100). Therefore, the tree can only consist of groups of accessions closely related. Consequently, 389 accessions of wild almond species were divided into three groups.
Population genetic structure inference. From the outcome of population structure analysis using each marker system, we observed maximum value of ΔK at K = 2, followed by K = 3 and K = 4, while distribution of ln(K) stabilized at K = 3. Thus three genetic groups were identified in the population. Under the non-admixture model at K = 3, the three clusters (G1, G2, and G3) had a share of 40.9, 78.3, and 37.8% of the population, respectively. At K = 3, cluster 1 differentiated the wild germplasm accessions that including in Euamygdalus section. Under the admixture model, genetic admixtures was high at K = 3. At K = 3, cluster 1and cluster 2 exhibited high divergence (0.45). Cluster II differentiated the cultivars from Spartioides section and exhibited closer genetic relatedness with cluster 1. Genetic admixture was higher in the wild population. At K = 4, clear population structure differentiation could be identified. Cluster 3 included wild species from Lycioides section.
On the basis of the molecular data, the results from Bayesian clustering analysis using STRUCTURE software confirmed the groupings we observed in UPGMA dendrogram and PCA. The most likely value of K (as chosen by Evanno's DK method) in Bayesian clustering analysis was three, which indicates the division of variation into three clusters, indicating the most appropriate four main clusters within the wild Prunus species populations studied, confirming the clustering of UPGMA dendrogram and PCA. The first cluster (blue colour) consisted of population from Euamygdalus section. The populations from Spartioides section were placed into the second cluster (red colour), while, population of Lycioides section were placed into the third cluster (green colour).
Overall, we found an admixture model to be more effective than a non-admixture model for markers based population structure analysis. This approach identified more groups in the population for each marker (K = 3).

Population demography inference. Results of Tajima's D and Fu's Fs tests for the 18 populations, Euamygdalus
section, the Spartioides section, and the Lycioides section are shown in Table 4. Values of both tests were significant and negative for all three groups, suggesting a possible historical population expansion. Furthermore, mismatch distributions for the Euamygdalus, Spartioides and Lycioides sections were unimodal and supported the hypothesis of the sudden expansion model (Fig. 4). The low and non-significant raggedness index values (Table 4) suggested a significant fit between the observed and the expected distributions.

Discussion
Different PCR-based molecular markers were used for the assessment of wild almond species including RAPD and ISSR 33 , microsatellites 29,33 and AFLP 34 . In this study, three molecular marker methods (IRAP, REMAP, and S-SAP), based on insertion polymorphisms of Ty1-copia and Ty3-gypsy RTs were evaluated and compared to the AFLP, ISSR, SSR and SNP systems. We used a wide range of statistics to evaluate the performance of these molecular marker systems, including a measure of genetic polymorphism, the efficiency of polymorphism detection, and the capacity of different techniques to identify genetic relationships of accessions.
Of the 22 LTR primers investigated (Supplementary Table S5), Ty1-copiaF and Ty3-gypsy F primers were designed to amplify products including the LTR of copia and gypsy retrotransposon-created IRAP fragments. The IRAP, REMAP, and S-SAP marker systems allowed the differentiation of all Prunus species analysed, as did the AFLP, ISSR and SSR systems. The dendrogram obtained from the REMAP similarity matrix (Fig. 3) appeared to be the most representative of the effective relationship between all accessions of wild almond species populations. In the REMAP phylogenetic tree, all Prunus species of different sections are in separate clusters 6,34 . The EUA and LYC accessions were the most different, and the rest of the accessions were divided into two sub-clusters.
The high reliability of the REMAP tree is most likely associated with the large-scale alterations in the genome induced by transposable elements. By comparison, SSR and, to some extent, AFLP (and other markers based on DNA digestion) essentially distinguish single nucleotide changes, which are affected by homoplasy and are bi-directional because the number of bands increase or reduce reversibly, so that it is difficult to infer valid phylogenetic connections of distantly related accessions 12,15,34 .
The ISSR phylogenetic tree was similar to the REMAP tree (results not shown). In this tree, it was also possible to differentiate the accessions of wild almond species that belonged to EUA and LYC, although LYC was positioned between the EUA and SPA groups and not into a sub-cluster of the EUA and LYC species, as in the REMAP tree. The similarity between REMAP and ISSR trees is strongly supported by the fact that they showed the highest value of correlation between cophenetic matrices and between similarity matrices (r = 0.970 and r = 0.952, respectively; Table 5) of all the dendrograms. Furthermore, various polymorphic bands, PIC, MI and SI values (Table 6) and the CV trends were similar in both of these molecular marker systems. The accuracy of REMAP and ISSR trees is shown by the higher bootstrap values than those characteristic for other dendrograms: bootstrap values are significant for the majority of the clusters (Fig. 3). In the AFLP, IRAP, SSAP and SSR dendrograms, the sections of Euamygdalus and Lycioides species are not clearly separated from other species. Furthermore, the topology of each dendrogram is very different from the others and these differences are confirmed by low correlation values between cophenetic matrices and between similarity matrices (Table 5). Moreover, bootstrap values of the dendrograms are much lower compared to the REMAP and ISSR trees, and the SSR marker system required the highest number of bands to obtain a 10% CV and showed the lowest SI, MI and PIC values ( Table 6). The unpredictability of the SSR dendrogram may be due to the use of the binary matrix in scoring SSR data. However, a dendrogram achieved by combining all marker systems data was very similar to that based on the binary matrix constructed from each marker (Fig. 5).

R E T R
Retro-transposons are mobile genetic elements that replicate by reverse transcription, which contributes to the physical size of the host genome. Retro-transposons are ubiquitous and abundant, which has played a major function in the structure and evolution of the plant genome. Retro-elements are long and cause a large-scale genetic alteration at the point of insertion, so insertions of variable numbers and sizes of retro-transposons into the host genome results in detectable polymorphisms. Previous studies have shown that Ty-1 copia-like-retrotransposons occurred in plant genomes in the early stages of evolution, and had diverged into heterogeneous subgroups before modern plant orders arose. There is clear evidence that retro-elements in the Prunus genome are heterogeneous because of their variable number and size. Analysis of the repetitive fraction of the peach genome showed that LTR retrotransposons comprise 18.56% of the genome, with 8.6% of Ty1-copia.Ty-1 copia-like-retrotransposon is also present in the Citrus genome at approximately 1.8-7.2 × 10 5 molecules, and accounts for almost 17% of the genome. The IRAP and REMAP techniques are retrotransposon-based markers that are extensively used in plant breeding including genotyping and gene mapping 12,35 .
In the present study, the IRAP and REMAP banding patterns of the genotypes demonstrated a high level of polymorphism. In total, IRAP and REMAP analysis generated 93 scorable amplification products ranging from 250 to 2,000 bp. Of all the bands, 79 were polymorphic, with an average of 15 polymorphic bands per primer combination. The level of polymorphism was 84%, which was comparable between IRAP and REMAP.REMAP bands were polymorphic with all of the microsatellite primers investigated, and bands were not produced by amplification between microsatellite repeats (ISSR). The ISSR pattern established considerably less variation among wild almond species. The analysis of the genetic relationship of 389 wild accessions of almond showed that Prunus species are noticeably differentiated. This is the first report of IRAP-and REMAP-based evaluation of Prunus and other RTN activity and genetic diversity in wild almond species. Of the tested primers, 19 IRAP and 34 REMAP primers were shown to amplify visible banding patterns and applied to study the RTN activity and genetic diversity among 389 wild almond species. RTNs may be in corporate in two orientations into the genome, and hence, any two members of one or more RTN families could be found head-to-head, tail-to-tail, or head-to-tail 12,18 . Moreover, different RTN families may be integrated in each other. Therefore, in order to increase the likelihood of finding bands, we also combined primers from LTR end of different RTN families. A number of investigations have confirmed that primers based on LTR sequences of RTN families can be readily used across species, among closely related genera and even sometimes between plant families 12,14 . In this study, single IRAP primers Tms1Ret1 and LORE1 and   35 reported that LORE1, a low-copy-number TY3-gypsy RTN family in the model legume species, L. japonicus, was active. LORE2A is estimated to be 600,000 years old, yet active in L. japonicas genome 36 . Tps12 may be inactive in the Prunus L. spp. genome as it creates a monomorphic banding pattern; nevertheless it produced greater polymorphism in combination with Tms1Ret1, representing the insertion of these two RTNs near or into each other in the Prunus genome. The primer based on Tps19 amplified no bands, representing its absence or fast divergence in Prunus species. Pearce et al. 37 separated a large heterogeneous population of formerly uncharacterized Ty1-copia RTNs from pea (Tps) and demonstrated that each element group in pea is   Table 5. Correlations between cophenetic matrices (above diagonal) and similarity matrices (below diagonal) obtained with different markers systems. Bold typeset indicates statistically significant (P ≤ 0.05) values.

A R T I C L E
related to the more distantly-related Vicia species, demonstrating that heterogeneous populations of these elements were present throughout the evolution of the Pisum and Vicia genera from their common ancestor. It was shown that Tps12a and Tps19 have a high level of insertion polymorphism in pea and have been active during Pisum species evolution. Kalendar et al. 15 showed the activity and insertion polymorphism of Bare1-based IRAP markers in Hordeum, Triticum, and Aegilops species. Tam et al. 38 , in a study of comparative analyses of genetic diversity in tomatoes, stated that RTN sequences isolated from one species can be used in related Solanaceae genera. REMAP amplification of the RTN families used indicated the insertion of the Tms1Ret1, as a native RTN, near the different SSRs in the Prunus genome. Non-native RTNs amplified bands in combination with a few SSR motifs, most likely signifying their low copy number, divergence, and preferential insertion within SSRs in Prunus L. spp. genome. The insertions of the RTN families in the vicinity of microsatellites have been formerly reported in barley and wheat 39 . Polymorphisms detected by markers based on non-native RTNs from P. sativum were low compared to those based on non-native RTNs from L. japonicus.  In total, 125 PCR amplicons designed from 100 EST contigs or genomic DNA that contained SNPs, INDELs and microsatellites were investigated using HRM based on the putative SNP information acquired from the almond and the Prunus EST databases. During sequencing of the HRM amplicon and flanking regions, 100 SNPs, as well as single nucleotide INDELs, were established in the population. HRM profiles of the fragments and evaluation techniques were established and the resulting SNP data were used to cluster the accessions of wild almond species. HRM is a novel, homogenous and closed-tube post-PCR method that can be applied to analyse the genetic variations as well as SNPs, polymorphism length and methylations of DNA in PCR amplicons 40,41 . HRM requires an additional step, the melting process following cycling, and an additional reagent, a specific generic DNA fluorescence dye, to complete the assay compared to conventional PCR. Therefore, the time and costs of the analysis are similar to conventional PCR, but it omits the need for post-PCR separation required by many other assays. Therefore, an HRM assay has the advantages of speed, simplicity, and lower cost.
HRM was intensively used for detecting mutations in known human genes. This approach for SNP analysis in plant species is very limited 42 . This study extended the application of the HRM method to the development of SNP markers by designing an HRM assay based on the putative SNPs from EST databases. This approach took advantage of the existing EST database, but avoided unnecessary sequencing efforts for putative SNPs in amplicons with in variant HRM curves in the test population. In our investigation, homozygous and heterozygous genotypes of all four SNP classes were distinguishable. In addition, the assay was able to resolve other variations, including INDELs, microsatellites, and complex multiple SNP amplicons.
As expected, the SNP frequencies were the lowest in coding regions (1:157), moderate in introns (1:130), and the highest in UTRs (1:51). These results are consistent with the findings in other taxa 43,44 . The average ratio of transition-to-transversion was 1.84:1 in our analysis. Wu et al. 25 reported a slightly lower value (1.16:1) in almond; however, the genome of other species was characterized by considerably higher values (e.g. 2.45:1 in humans 43 , and 1.53:1 in maize 44 ). Transition bias over transversion has been fairly universal in the genome 45 , although differing results have been described in grasshopper pseudogenes 46 . Transition bias has been considered to be partially due to cytosine methylation 46 . Therefore, the low transition bias may reflect low methylation levels in the almond genome, and this could be significant because of the role of methylation in epigenetics and imprinting 41,46 .
HRM has been applied in SNP detection, mainly in diagnosis and scanning for mutations in genes causing human diseases. We have shown this approach was valuable in the detection and recognition of plant SNPs. While many high-throughput SNP detection approaches, such as SNP microarrays, are cost-efficient for whole genome scanning in the species where genome-wide SNP information is available, it is expensive to assay the small amount of SNPs using these methods. It was demonstrated that HRM is a feasible means for this assay. As expanding DNA sequence information becomes available for species such as almond, HRM will be a valuable method for SNP detection and genotyping. This is particularly useful in plant cultivar identification, genetic mapping, QTL analysis, diagnosis of pathogenic species, and gene discovery. Furthermore, data from HRM analysis is portable, and therefore not only feasible for inter-genotype comparison, but also for library-based database construction. This feature may facilitate international collaborative efforts using SNP-based genotyping, and therefore genetics and biodiversity studies by using HRM analysis.
Populations at demographic equilibrium or in decline present a multimodal distribution of pairwise differences, while populations that have experienced a sudden population expansion display a unimodal distribution 47, 48 . Tajima's D and Fu's Fs tests are expected to show significant negative values under population expansion and positive values under a population bottleneck and hence both of the parameters supported the sudden expansion of almond related species. In addition, growth in population size should result in elevated frequencies of rare alleles, explaining the unexpected 3-40% values found in the populations. It is also confirmed by the fact that most almond lineages differentiated during the Holocene 49 . Since wild almonds also have edible kernels human ancestors might have harvested their nuts as early as 780,000 years ago 50 . Therefore, it is reasonable to suppose that similarly to P. dulcis wild almond species were also influenced by human care. The number of accessions sampled per site varied from one to five, depending on the environment, diversity and accessibility at collection time 6 . A total of 389 accessions were sampled from the 18 studied species (Table 1).

Methods
The comprehensive procedure for field expeditions is presented in Sorkheh et al. 6 . Sites were selected based on literature 6 , indigenous information, or conspicuous presence. Collections were made from both wild and cultivated habitats, which were concentrated in two different regions in Iran. The first region (Azerbaijan and Kurdistan) is characterized by a relatively lush environment, a high biological diversity, and relatively under-developed agricultural activity. The second region (Shahrekord and Shiraz) is in a more xerophytic region with widespread agriculture (Supplementary Table S6; Fig. 6). Sampling sites and their geographic locations are reported in Table 1. The distance between samples was 200 m; the pairwise distance between the main regions was 100-500 km. Sampling was determined following the natural distribution of wild almond species in Iran, according to Sorkheh et al. 6 in the wild. For the locations (see Table 1 for details), a specific agreement was not necessary because these locations are exterior-kept areas, and leaf gathering did not correspond to a threat to the sampled individuals. The sampled stands were selected to most closely represent the natural environment of the region. Collected leaves were stored at −80 °C until DNA isolation.  Table S6). PCR amplification was carried out in a Bio-Rad thermocycler (Bio-Rad Laboratories Inc., Hercules, CA, USA). The PCR amplification profile was set according to Abdollahi Mandoulakani et al. 18 with minor modifications 34 .

SSR and EST-SSR analysis.
A set of 32 SSRs and 187 (12 of that were previously described, and 175 were developed in this study) EST-SSR primer pairs were selected based on previous information and new EST-SSRs were developed after database research (Supplementary Table S7 and Supplementary Table S8) on different

A R T I C L E
Prunus species and identifying 33, 7, 2, 1 and 1 SSRs in peach, almond, plum, sweet cherry and sour cherry EST sequences, respectively (Supplementary Table S9). This set of markers was primarily displayed to confirm amplification in the collected samples of Prunus species, and afterwards 25 SSR and 9 EST-SSR primer pairs, covering eight linkage groups (G1 to G8), were chosen for subsequent analysis (MWG, Biotech, Germany). The forward primers were labelled with either 6-FAM or HEX fluorescent dye for recognition in a capillary genetic analyser. PCR reactions were carried out in a 96-well block cycler (BioRad Ltd. USA), in a final volume of 10 µl consisting of 10x PCR Buffer (Takara, Japan), 0.5 µM of each primer pair, 0.05 mM of each dNTP (Cina clone. Co., Iran), one Unit of Taq DNA polymerase (Takara, Japan), and 1 µl (100 ng) of genomic DNA. Cycling conditions were 94 °C for 3 min; 40 cycles at 94 °C for 0.5 min, 50-62 °C for 0.5 min, and 72 °C for 1 min, followed by 20 min at 72 °C for the final extension. Details are given by Rahemi et al. 29 .
Forward and reverse primers on both sides of at least one assumed SNP were considered for HRM analysis using Primer 3 56 . Primer pairs were designed to have an annealing temperature of 60 ± 1 °C and to give an anticipated product size of 60-100 bp with few exclusions (Supplementary Table S3). Primers were analysed using NetPrimer to distinguish possible secondary structures, i.e. primer dimers, hairpins, palindromes and repeats (http://www.premierbiosoft.com/netprimer/netprimer.html, Premier Biosoft International, Palo Alto, CA), as secondary structures of primers are thought to influence the PCR amplification efficiency and, as a result, HRM accuracy. Secondary construction of the amplicons was analysed using the DINAMelt Server (http://www.bioinfo.rpi.edu/applications/hybrid/twostate-fold.php). The amplicons were considered suitable for HRM analysis when the DG value of the considered secondary structure was >−1 (Corbett Research 2006). The method for SNP detection and primer design is described in Wu et al. 25 , Koepke et al. 55 , and Salazar et al. 57 .
HRM analysis was based on the procedure of Wu et al. 25  Co., Japan). Touchdown PCR has been previously described by Wu et al. 25 . HRM curve analysis was performed using the HRM module of the Real-Time PCR System Software (Life Technologies, USA). The melting data were standardized by correcting the initiate and stop fluorescence signals of all samples analysed for the same levels, according to Wu et al. 25 . Genotypes of the individuals were scored automatically by the software and verified manually. Whilst microsatellite polymorphism was involved in an amplicon, PCR products were separated using an 8% polyacrylamide (Cinagen Co, Iran) gel and stained with silver nitrate (Scharlau, Ltd. Spain). Details are provided by Sorkheh et al. 34 .
Data analysis. The amplified fragments were achieved separately, equal to 1 or 0 for their occurrence or absence, respectively, and the acquired binary data were utilized for analysis. Each PCR product represented a single locus, and Shannon's index (SI), marker index (MI) and polymorphic information content (PIC) was calculated. The SI, according to Shannon and Weaver 58 is defined as: where p i is the frequency of the i th band in the sample. This formula was considered using the PopGene software version 1.32 59 . The MI was calculated according to Powell et al. 60 as the product of expected heterozygosity (He) and the effective multiplex ratio (E). The heterozygosity of a locus is defined as: where p i is the frequency of the i th allele (band). The effective multiplex ratio of a primer combination is defined as: where n is the number of loci detected per primer combination and P0.95 is the percentage of polymorphic loci 60 . Heterozygosity and P0.95 were calculated using GDA 1.0 software 32 . The polymorphic information content is commonly used as a representation of the expected heterozygosity 60 . Details are provided by Botstein et al. 61 . For the SSR markers, we considered expected heterozygosity (He), observed heterozygosity (Ho), the frequency of null alleles and the probability of identity (PI) using the IDENTITY 1.0 software 62 .
The calculated parameters were subjected to analysis of variance using SPSS 12.0. Different binary matrices related to different assays were imported into the NTSYS-PC2.01a package 63 for cluster analysis. Genetic similarity matrices between genotypes were determined based on the simple matching (SM) similarity index 64 for dominant multi-locus markers (AFLP, IRAP, ISSR, REMAP, and S-SAP), using the SIMQUAL routine, and according to the Nei coefficient 65 for codominant SSR markers, using the SIMGEND routine. Dendrograms were constructed using similarity coefficients by the UPGMA with the NTSYS-PC2.01a software package. Bootstrap analysis (1,000 replacements) was carried out using the WinBoot software 66 , and differences between dendrograms

A R T I C L E
were evaluated on the basis of correlations between similarity matrices and between cophenetic matrices calculated from the Mantel matrix correspondence test 67 .
The coefficient of variation (CV) trend from 5 to 120 polymorphic bands was evaluated on Nei coefficient matrices for SSR markers and on dissimilarity coefficient (1-SM) matrices for multilocus markers. For every molecular marker system, the CV trend was determined three times for three independent sets of 1,000 bootstrap analyses performed with the Phylip package 68 on three randomized matrices of presence/absence. The number of loci required to acquire a 10% CV was approximated and proposed to the analysis of variance using the SPSS 12.0 package.
For each retrotransposon family (Ty1-copia and Ty3 -gypsy), the two LTR sequences of each element were collected from the Prunus genome and evaluated. The main diverse LTR pairs for each retrotransposon family were used to calculate the number of nucleotide replacements by Kimura's two-parameter model 69 , and the corresponding insertion ages for the transposable elements were then estimated using the formula T = K/2r, where T is the time of insertion, K is the difference parameter and r is the means of substitution rate. For r, the average value (4.11 × 10 −9 subs/site/year) of the two values reported for woody perennial plants by Kay et al. 70 was used.
With the aim of partition, the total genetic variation between and within populations, the analysis of molecular variance (AMOVA), was approved based on IRAP, REMAP, AFLP, SSR, S-SAP and ISSR data using GenAlEx 6 71 . The number of loci, polymorphic loci (%), the number of alleles or loci with a frequency higher or the same as 5%, the number of private loci or alleles, the number of less common loci with a frequency lower or equal to 25% and 50%, the mean of heterozygosity 28 , Nm (Number of migrants between populations), and standard error of mean heterozygosity were also calculated for each population using GenAlEx 6 for AFLP, IRAP, ISSR, REMAP, SSAP and SSR data.
Principal component analysis (PCA) was used to easily visualize relationships among the individuals and determine optimum number of clusters. The EIGEN module was used to calculate Eigen values and two-dimensional plots based on the variance-covariance matrix calculated between each two pairs of the one hundred accessions of wild Prunus species. Population structure was analysed using a model-based approach, Bayesian method, in STRUCTURE ver.3.0 software 72 . Since dominant markers were used in this study, each class of the accessions of wild Prunus species was treated as a haploid allele 73 . Model-based cluster analysis was used to test the number of populations (K). The most appropriate K can be detected by which values of log Pr (X/K) reach plateaus after a major decrease. We used both approaches to estimate K. No admixture and correlated allele frequencies models were used. For each population (K), 1000 iteration and 1000 burn-in period options were used. For each number of K from 1 to 10, five independent calculations were performed, and likelihood values obtained from these 10 calculations were averaged for each K.
Inferences on demographic history were obtained by neutrality tests and mismatch distribution, based on all marker data. As for neutrality test, Tajima's D test and Fu's Fs test 74 were calculated using Arlequin 3.5 75 with 10,000permutations. Mismatch distribution was constructed for each geographic population to test a model of exponential population growth 76 . A goodness of fit test was performed to test the validity of the sudden expansion model, using a parametric bootstrap approach based on the sum of square deviations (SSD) between the observed and expected mismatch distributions. The raggedness index which measures the smoothness of the mismatch distribution was calculated for each distribution. The demographic expansion parameter (t) was calculated using Arlequin 3.5 75,77 .