A set of 312 pure lines derived from worldwide accessions of N. tabacum (Table 1) was investigated to detect the allelic variants at 49 SSR loci (Table 2). This panel of SSRs was selected based on technical reliability, uniqueness, and even distribution in the tobacco genome, as described in two papers by Bindler et al. 3133, and was used to infer phylogeny and genetic diversity in the set of accessions, eventually leading to the assembly of a core collection.

The total number of alleles amplified at the 49 SSR loci was 335, with an average call rate of 99%. The high level of polymorphism revealed for the 49 SSR supported their usefulness for applications in diversity analysis.

The mean number of alleles detected for each locus was 6.84 (s.d. = 2.57), ranging from 13 alleles for marker PT61336 to three alleles for PT54203 and PT52002 (Table 2). This value is about half that recorded in previous studies 332. The difference may be due to the choice of the marker loci as well as to the set of accessions analyzed. All tobacco accessions were genotyped as homozygous at the 49 SSR loci (H_o = 0 at all loci, Table 2), with a gene diversity (H_e ) per locus spanning from 0.013 to 0.841 (average 0.59), a value lower than those reported in similar investigations carried out on TI accessions of tobacco 3. The relatively low levels of Hd revealed by molecular markers in tobacco 34 can be due to relatively recent evolutionary and breeding bottlenecks, through which only a small proportion of the variability of the gene pools of the progenitor species was funneled through 2. The polymorphic index contents (PIC) was > 0.4 at almost all loci, with PT53216 and PT54203 having the highest and lowest values, respectively (Table 2).

Clustering of the 312 genotypes (Figure 1) revealed the relationships among accessions were distributed over six main clades. Accessions of "Oriental" clustered mainly in two different clades encompassing 88 accessions (green clades in Figure 1). Only 14 of the accessions were members of the heterogeneous group of tobacco accessions defined as "Other", whereas another 10 were classified as different tobacco types. "Flue-Cured" lines clustered mainly in one clade (yellow clade in Figure 1), although this also contained two, seven, four, and eight genotypes classified as "Oriental", "Dark", "Primitive", and "Other", respectively. Excluding six genotypes assigned to different clades, "Burley" accessions clustered in one clade (light-blue clade in Figure 1), which also contained nine members of the "Other" tobacco type and six lines classified as "Primitive", "Dark", "Oriental", and "Flue-cured". Non-group associated genotypes ("Other" in Table 1) clustered in two different clades (violet clades in Figure 1), one of which also included lines containing a large sub-set of the "Primitive" accessions (blue clade in Figure 1).

Different tobacco types originated as the early growers saved seeds for subsequent planting, before the initiation of science-based breeding 3. Tobacco growers selected plants for cultivation in different environments, for their agronomic performance using different agronomic practices, for the smoking characteristics of the leaf, and for adapting the leaf type to different leaf curing methods (i.e., the way the leaf is dried in a controlled way). The tobacco accessions we investigated clustered, based on molecular markers, according to their type, thus supporting the effectiveness of the breeding programs which have restricted the original breeding pool when selecting specifically for each market destination. The results are in agreement with previous data 31 supporting the correlation between type classifications and genetic distances 35. The accessions that were found "contaminating" the homogeneity of groups (for example, "Cigar" varieties interspersed among "Oriental" varieties, in Figure 2), may be the result of misclassifications, as reported for the TI accessions 36. In addition, inaccurate sampling procedures carried out during tobacco cultivation, or errors during varietal reproduction and conservation, can be the origin of the observed heterogeneity of major phylogenetic clades.

The distinct and homogeneous clustering of "Oriental" and "Flue-cured" tobaccos, the most outstanding tobacco types, is most likely due to the ~400 years of divergent selection in Europe/Middle East for the "Oriental" types 37, and to the adoption of a stringent conservative breeding strategy for "Flue-cured" tobaccos 38. In "Flue-cured" tobaccos, genetic variability decreased significantly with the adoption of an "advanced cycle pedigree breeding", i.e., the exclusive usage of elite materials to produce breeding crosses 32. Also "Burley" genotypes clustered together, although less homogenously than the previous two groups, as described also in 39, possibly because their selection has been traditionally performed on a wider geographic scale. Two phylogenetic clades were heterogeneous, containing most of the "Primitive" accessions, and the majority of "Cigar", "South American", and "Indonesian" tobaccos. This may represent the most ancient gene pool, making it particularly interesting for future breeding and mapping programs. According to the phylogeny, the "Primitive" genotypes should be genetically strongly related to "Cigar" tobaccos.

Principal coordinate analysis (PCoA) was carried out on the same SSR data set (Figure 2). The first principal component explained 40% of the genetic variance, and 71% was explained by the first three principal components, indicating that despite the high number of alleles detected at some SSR loci (Table 2), the collection was characterized by a narrow genetic basis. PCoA clustering indicated that molecular associations mainly reflected the physio-morphological characteristics associated with the tobacco types and their agronomic and commercial uses (Figure 2). A further Bayesian cluster analysis 40 identified the most probable number of K subpopulations present in the whole panel. The analysis of posterior probabilities supported the conclusion that four subpopulations had the highest likelihood (Figure 3). In the collection, a small number of genotypes were molecularly not aligned with their assignment to a tobacco type. Namely, while most of the "Burley", "Oriental", "Flue-cured", and "Cigar" genotypes clustered molecularly in four distinct subpopulations (yellow, violet, red, and green bars, respectively in Figure 3), "Primitive" and "Dark" genotypes were characterized by a more heterogeneous genome constitution. The close link between "Dark" and "Flue-cured" 2 was evidenced by the number of common alleles (red bars, Figure 3).

The levels of admixture (i.e., interbreeding between individuals of previously isolated populations) estimated by STRUCTURE appeared low in all lines considered, supporting the role of the conservative breeding to which the species was subjected.

Our PCoA results support the conclusion that the main tobacco types can be discriminated by molecular fingerprinting. In this sense, genetic distance and model-based analyses provide for the first time strong evidence of population substructure in tobacco.

To test the hypothesis of co-ancestry between tobacco accessions belonging to the same tobacco type, the pairwise kinship coefficients between accessions, as well as the population mean kinship (MK) among groups of tobacco accessions (Table 3), were calculated. The kinship coefficient is the ratio of the probability that, at a given locus, alleles of i and j individuals are identical by descent vs. the same probability of two random individuals. In this work alleles at one SSR locus were defined identical by descent if identical by state in the capillary electrophoresis analyses. Kinship coefficients are expressed relative to the average of the population and thus can assume negative values. The pairwise computations were used to calculate the MK coefficients in accessions of the same tobacco type, and in all possible pairwise combinations of the seven tobacco types (Table 3). The MK coefficient of the whole collection was -0.004326 revealing a generally low level of co-ancestry. When MK coefficient calculation was restricted to accessions of the same tobacco type, higher values of MK were obtained (Table 3). The highest value was obtained for the "Oriental" subset, while the lowest was obtained for the "Other" subset (both in bold in Table 3). The MK values calculated within types were positive, suggesting that a certain level of co-ancestry linked the accessions included in each tobacco type (Table 3).

The first core collection of tobacco was created that identified the minimum set of accessions capturing most of the genetic diversity at the microsatellite loci tested on the full set of accessions. Five different lists of accessions selected using different rationales were created. The first list identified the minimum set of accessions capturing all 335 alleles identified in the whole panel of tobacco, and allowed us to isolate 60 genotypes. The other lists (60 genotypes each) were manually created based on Bayesian clustering, PCoA scatter-plot, co-ancestry analysis, and phylogeny. The five sets of accessions were then merged and a core collection was produced (Table 4) composed of 12 "Burley" (including 1 "Maryland"), 20 "Flue-cured", 20 "Oriental", 14 "Cigar", 10 "Primitive", 8 "Dark", and 5 "Other". Twenty-one of the accessions included in the core collection corresponded, according to Moon et al. 3, to samples collected before 1938. They still represent the best available sampling of the genetic diversity existing before modern breeding. Some of the genotypes (<5%) were selected because of their potential for tobacco breeding and not because they were identified following the protocols described.

A total of 422 SSRs were used to scan the tobacco core collection at seven genomic regions located in different chromosomes (Table 5). The regions were selected based on marker density and their potential to harbor genes putatively important for crop improvement. The markers used had a density of 0.9 marker/cM, ranging from 0.6 on LG1 to 1.1 on LG17. The mean information index 41 varied from 1.68 for LG7 to 2.14 for LG17 (Table 5). Only 6.45%, 1.92%, and 6.06% of SSR markers, on LG1, LG7, and LG22, respectively, were found to be monomorphic. The lowest average number of alleles per locus was on LG7 (4.57 alleles per SSR), and the highest was on LG22 (7.8 per SSR; Table 5).

A total of 312 tobacco accessions (Additional file 1) maintained at the Philip Morris International collection, Neuchatel (CH), were investigated in this study. Accessions were classified as described in Chaplin et al. 36: "Burley" & "Maryland": 45 entries; "Flue-cured": 70; "Orientals": 77; "Cigar tobaccos" (filler, wrapper, binder): 36; "Primitives": 23; "Dark tobacco" & "Fire-cured": 22; "Other": 18 ("Perique", 1; "South American", 4; "Semi-oriental" 1; "Indonesian" and "other", 12). Twenty-one accessions were of unknown type.

The majority of accessions were originally obtained from the U.S. Nicotiana Germplasm collection in Oxford, NC (USA); the accessions used represent tobacco collected from or cultivated in 45 different countries. Seeds were germinated and grown under greenhouse conditions until plants reached a height of approximately 20 cm before DNA extraction.

Leaves from 5 plants were pooled and genomic DNA was isolated from 6 mg of lyophilized material in 96-well microtube plates using Macherey Nagel^® NucleoSpin Plant II kit and following manufacturer's instructions. The quality and the concentration of the genomic DNA were assessed using electrophoretic analysis and Picogreen^® technology (Invitrogen, San Diego, CA), respectively. Genomic DNA was normalized at 20 ng/μL before genotyping.

All SSR loci considered in this study were amplified using a three-primer system for indirect labelling PCR fragments 49. The amplification of SSR loci was carried out in 384-well plates (Applied Biosystems, Foster City, USA) in Eppendorf Mastercycler EPgradient thermalcyclers (Eppendorf, Hamburg, Germany). Each reaction was performed in 10 μl with the following mixture composition: 20 ng of DNA, 1.5 mM of MgCl₂, 0.4 μM of the first primer, 0.2 μM of the second primer with M13 complementary tail, 0.2 μM of M13 fluorescent labelled primer, and 0.25 U of Taq HotStart DNA polymerase QIAGEN (Valencia, USA). The reactions were subjected to the following thermal protocol: after an initial denaturation step at 95 C for 15 min, amplification reactions were subjected to 11 cycles at 95 C for 30 s, 58 C for 45 s and 72 C for 90 s, decreasing annealing temperature by 0.7 C in each cycle. The reactions were further subjected to 29 cycles of 95 C for 30 s, 50 C for 45 s, and 72 C for 90 s. A final elongation step of 10 min was applied. 0.25 μL of amplification products, each of which was labeled with the four ABI dyes, was mixed with 10 μL of formamide, loaded in a ABI3730 DNA analyzer (Applied Biosystems), and analyzed through capillary electrophoresis.

Fragment analysis was carried out with GeneMapper^® 4.0 software (Applied Biosystems, Foster City, USA) using stutter peaks of known sizes as internal controls. Automatic allele calls were subsequently assessed reviewing all electropherograms. Fragments of lengths not comparable to the control or with fluorescent intensities lower than 75 percent of the peak assumed as true allele were considered artefacts. Genotyping tables were exported as tab-delimited files and formatted in Microsoft Excel (Redmond, USA) to conduct phylogenetic and statistical analysis.

Basic statistics (number of alleles detected at each locus, allelic and genotypic frequencies, call rate, heterozygosity, and PIC) were calculated using the R 50 and GenAlEx packages 51.

PHYLIP package gendist software 52 was used to calculate pairwise Cavalli-Sforza's genetic distances among the 312 tobacco accessions. Triangle matrix of pairwise genetic distances was subsequently formatted in NEXUS file to cluster the tobacco accessions with Neighbor-Joining using the SplitsTree4 software 53. To better plot the resulting large clustering of tobacco accessions, a circular cladogram was generated with Dendroscope software 54.

To assess the population structure of the tobacco-sample accessions, a multivariate analysis and a heuristic method based on Bayesian clustering algorithms were utilized. Principal coordinate analysis (PCoA) was initially performed on the SSR data using the "ape" package in the R software. The clustering method based on the Bayesian-model implemented in the software program STRUCTURE 40 was used on the same data set to better detect population substructures. This clustering method is based on an algorithm that assigns genotypes to homogeneous groups in such a way that departure from neutral equilibrium is minimized among genotypes within each group, but it is absent among groups. The number of potential subpopulations varied from 2 to 10, and their contribution to the genotypes of the accessions was calculated based on 5x10⁵ iteration burn-ins and 5x10⁵ iteration sampling periods. Eventually, the most probable number (K) of subpopulations was identified following Evanno et al. 55.

Pairwise coefficients of kinship (F_ij ), a measure of relatedness between individuals i and j based on molecular markers, were calculated using SPAGeDi software 56). Mean kinship (mk) coefficients were obtained averaging the pairwise kinship coefficients of each single accession with all other accessions of the whole collection 40. In addition, the computation of mean kinship coefficients was restricted to pairwise kinship coefficients of accessions of the same tobacco type (mk_i ), as well as for all possible pairs of accessions of different tobacco types (mk_p ). In order to assess the higher level of relatedness of tobacco accessions of the same type, population kinship coefficients (MK) were calculated arithmetically averaging the mk_i coefficients of tobacco accessions sharing the same tobacco type. Similarly, the level of relatedness of accessions of two different tobacco types was assessed averaging mk_p coefficients calculated for all possible pairs.

An algorithm was developed and implemented that allowed us to identify the least number of accessions capturing all of the alleles that were unique in the set of tobacco accessions. A first screening with one SSR marker was performed on a random sub-set of the accessions, followed by a pairwise comparison with the remaining accessions. Only accessions showing at least one unique allele were used in the following iterative analysis, leading to a list of accessions that represented all of the alleles. Because the group and the number of accessions in the final list can change, depending on the original order of the list, the accessions were randomly re-ordered and this process was repeated 2 x10⁵ times. This method allowed for the selection of 60 genotypes. Additional methods were used to create three more lists of accessions, each with 60 genotypes showing the most extreme values of PCA, mk coefficients, and pairwise genetic distance. A fifth list of 60 genotypes was created by picking individuals with the highest values in the Q matrix of the STRUCTURE analysis. The five lists of genotypes were then merged and a consensus list of 89 genotypes was compiled.

The squared allele-frequency correlation r², was calculated for all possible combinations of alleles to estimate the extent of LD in the core collection of tobacco accessions, using the software package TASSEL 2.01 57. The weighted average of r² values was obtained by further weighting for the corresponding allele frequencies. The significance of pairwise LD (p-value) among all possible pairs was also evaluated by TASSEL with the rapid permutation test.

To avoid the bias imposed by the usage of the squared-allele-frequency correlation r² in the presence of rare alleles, only alleles having a frequency larger than 0.1 were considered.

The square root of each pairwise r² among allelic variants of physically unlinked SSR loci was calculated. The 95^th percentile of this approximate normal distribution was assumed as the threshold of the r² value to declare the presence of LD among molecular markers 58.