SNP mining in C. clementina BAC end sequences; transferability in the Citrus genus (Rutaceae), phylogenetic inferences and perspectives for genetic mapping

Design of the Citrus Illumina GoldenGate SNP set

Polymorphism and allele call for the different SNPs; selection and classification of valid SNPs

For all SNPs, the genotyping was visually confirmed, taking advantage of the distribution of the CxN progenies relatively to 'Nules' clementine and 'Chandler' pummelo positions. The SNPs were assigned to different categories based on the quality of the polymorphism detected, the detection of null alleles, and the type of segregation observed for 'Nules' Clementine in the CxN progeny (additional file 1). The first category (C1) consisted of 230 markers exhibiting very low technical quality which did not allow for clustering. Among the other categories:

• 608 (C2) displayed the expected segregation for clementine in the CxN progeny. However, for 80 of these loci the clustering between the three classes of genotypes was not totally clear, leading to missing data.

• 85 (C3) with validated heterozygosity in Clementine presented an unexpected segregation, revealing heterozygous (Figure 1a) or homozygous (Figure 1b) null alleles in 'Chandler' pummelo or some others germplasm accessions.

To validate the genotyping data, 24 of the 54 Citrus accessions that were genotyped with the GoldenGate array were sequenced (Sanger) for 15 SNPs from five genes. Among the 360 genotype × SNP data, 357 (99.2%) were in agreement with the GoldenGate genotyping and Sanger sequences. In two cases (PSY-C-246 for 'Eureka' lemon and LCY2-P-75 for 'Sunki' mandarin), the GoldenGate genotyping concluded heterozygosity, while the Sanger sequencing inferred homozygosity. The opposite case was obtained with PSY-C-497 in Mexican lime.

The origin of the unexpected polymorphisms displayed by several SNP markers from the Clementine BES, such as null alleles, no heterozygosity for Clementine and 'fixed heterozygosity', was analyzed using Sanger sequencing of the amplicons of four accessions: 'Nules' Clementine, haploid Clementine, 'Chandler' pummelo and Corsican citron.

For 7 loci (CiC0002-01; CiC0049-02; CiC0063-12; CiC0074-09; CiC0091-09; CiC0113-01; CiC2553-04) genotyped homozygous for Clementine in Golden Gate, despite previously being labeled as heterozygous based on BES sequencing, the homozygosity was confirmed by Sanger sequencing. For the analyzed locus of apparent fixed heterozygosity in the PxC progeny (CiC4252-10), the haploid Clementine line also displayed a heterozygous-like pattern, thus confirming the hypothesis of a duplicated locus. For the loci with null alleles, fragment amplification was observed for all of the genotypes. Additional SNPs in the GoldenGate primer area were observed for CiC3064-07 and CiC3275-02 in 'Chandler' pummelo and Corsican citron. However, for another marker of this class (CiC2151-02), no polymorphisms or explanation for the null alleles was found.

Citrus Germplasm diversity for markers without null alleles (WONA loci)

Structuration between basic taxa: the analysis of the Fstat parameter in considering the accessions of C. maxima, C. medica and C. reticulata confirmed the higher differentiation between species with CHom markers (table 3). Indeed, when the Fis value indicated no significant deviance from HWE with the two classes of markers, the significant Fit value (mainly due to between species differentiation with a high positive Fst value) is close to double for the CHom markers (0.80/0.40 for Fit and 0.80/0.45 for Fst). Very similar Fst values are estimated when considering only the differentiation between C. maxima and C. reticulata (table 3). NJ representations with the two set of markers (Figure 2) clearly shows an increase in inter-specific differentiation with the CHom marker set compared to the CHet set. Particularly, with the CHom markers, C. medica is strongly differentiated from the two other species. Moreover, within C. reticulata, the dissimilarities between 'Cleopatra', Sun Chu Sha', 'Mediterranean', 'Imperial', 'Ponkan', 'Fushu' and 'Dancy' genotypes appears significantly smaller with CHom than with CHet markers.

	3 basic taxa (C. maxima, C. medica, C. reticulata)			C. maxima /C. reticulata
	Fis	Fit	Fst	Fst
All loci	-0.09+/-0.02	0.42+/-0.03	0.47+/-0.03	0.46+/-0.03
Het. Cle.	-0.10+/-0.02	0.40+/-0.03	0.45+/-0.03	0.45+/-0.03
Hom. Cle.	0.012+/-0.14	0.80+/-0.10	0.80+/-0.08	0.70+/-0.11

Neighbor joining (NJ) and principal component analysis (PCA) of the whole germplasm sample: NJ trees are not well adapted to describing the genetic relationships of hybrid structure with the parental gene pool. However, NJ trees allow a global synthetic representation of dissimilarities between genotypes. In factorial analyses such principal component analysis, the hybrid positions between the parents is very clear in the axes that differentiate the parental genotypes. However, representation in a two- or three-axes space may result in a biased view of the global similarities between genotypes. The two representation types (NJA and PCA) are thus complementary when analyzing the genetic relationships and potential parentage of hybrid genotypes. The genetic organization around C. reticulata, C. medica and C. maxima appears very clearly both with the PCA (Figure 3a) and NJA (Figure 4). Forty-eight percent of the entire diversity is represented by the first two axis of PCA. The first axis discriminates C. reticulata from the other species while the second axis separates C. medica and C. maxima. Very few within species differentiations are observed for the accessions of C. aurantium, C. sinensis and C. paradise. By contrast, C. aurantifolia and C. limon display more intra-specific polymorphism. Two accessions of C. aurantifolia, 'Alemow' and 'Mexican lime', exhibit an intermediate position between C. medica and a papeda cluster including C. ichangensis, C. micrantha and C. hystrix. The other two C. aurantifolia accessions (Calabria and Palestine sweet lime) are more related to the C. limon accessions. The C. limon accessions are subdivided into three close clusters: Meyer lemon is the more isolated, while 'Eureka', 'Lisbon' and 'Marrakech lime' are clustered to one side and 'Rangpur' lime, 'Rough lemon' and 'Volkamer' lemon cluster in the other side. This last cluster presents higher coordinates in the second axis of the PCA(a) and lower coordinates in the first axis, compared to the first lemon cluster. This suggests a higher contribution of C. reticulata and very few or no C. maxima contribution. Interestingly, a sub cluster of acidic mandarin ('Cleopatra', 'Sunki', 'Depressa' and 'Sun Chu Sha') also present higher coordinates in the second axis relative to the other C. reticulata accessions. Sweet and sour oranges are strongly differentiated by the second axis. Clementine is very close to sweet orange in this axis and is intermediary between Mediterranean mandarin and sweet orange in the first axis. To analyze the contribution of C. maxima and C. reticulata to the genome of their supposed deriving secondary species (C. sinensis, C. aurantium and C. paradisi), a second PCA was generated using only C. maxima and C. reticulata as active individuals to define the new axes (Figure 3b). Of the total diversity, 52.1% was supported by the first axis opposing the two species. The contribution of the SNP loci to the definition of this axis (estimated by the cos² of the coordinate in the new components) presents a very high correlation with the Fst value for the C. maxima/C. reticulata differentiation (r² = 0.948, additional file 4, figure S1). This confirms the validity of this axis for the synthesis of the relative contribution of the two basic species to the secondary ones. C. sinensis and C. aurantium display similar positions in this axis at a closer distance to the C. reticulata gene pool than the C. maxima one. C. paradisi has an intermediary position between these two secondary species and C. maxima.

Citrus Germplasm diversity displayed by markers with null alleles (WNA loci)

The within species discrimination of accessions obtained with the whole set of WNA markers is respectable, with the exception of C. sinensis and C. paradisi. C. reticulata, the most polymorphic species, displayed a genotypic diversity of 0.361 and an observed heterozygosity of 0.234. The observed heterozygosity of all of the secondary species is strongly reduced when compared with the WONA loci, suggesting frequent but not observed heterozygous null alleles.

The differentiation of the three basic species (C. maxima, C. medica and C. reticulata) was confirmed by the NJA based on WNA loci (Figure 5). However, the global picture displayed by this analysis varies significantly from the previous one without null alleles (Figure 4). Indeed, C. maxima, C. medica, most wild Citrus species (except C. amblicarpa), as well as two C. aurantifolia accessions (Mexican lime and Alemow) are strongly clustered. The numerous loci sharing null alleles between C. medica, C. maxima and some wild species may explain this strong clustering as well as the position of supposed hybrids between these taxa (Mexican lime and Alemow). Moreover, C. sinensis, C. paradisi and several C. limo n and C. aurantifolia accessions are integrated within the C. reticulata cluster. These secondary species are suspected to be hybrids between C. reticulata and C. maxima and/or C. medica. Due to the recessive nature of null alleles, these secondary species appear artificially closer to their C. reticulata parent because of the lower frequency of null alleles in this gene-pool compared with the other ancestral species. An interesting point is that the recessive nature of a high proportion of alleles from the other species allows the approximation of the sub gene pool within C. reticulata at the origin of some interspecific hybrids. Clementine is clustered with 'Mediterranean' mandarin. The parentage of sweet orange and grapefruit is also clearly revealed. Volkamer and Rough lemons, as well as Palestine Sweet lime, are clustered with a group of acidic mandarins (Sunki, Sun Chu Cha, Cleopatra and Depressa). C. amblycarpa is also associated with this cluster.

Markers for interspecific comparative genetic mapping

SNP mining in Clementine and the unexpected segregations in Chandler × Clementine progeny

Among the 6,617 SNPs mined in silico, 1,457 markers were selected for the GoldenGate assay based on their distribution on the different BACend contigs, as well as their inclusion in or their vicinity to the coding region. Thirty seven SNP loci found in 10 candidate genes were added for the analysis. Low technical quality was displayed by 230 markers, preventing any clustering. A total of 693 markers presented segregations that conformed to Clementine heterozygosity (C2+C3). The 'Chandler' × Clementine (CxN) progeny revealed heterozygous null alleles in Clementine for 28 markers (C4). Moreover, 481 markers appeared to be homozygous for Clementine (C5+C6+C8) while 472 of them were supposed to be heterozygous from the Clementine BACend sequencing [31]. This homozygosity was confirmed using Sanger sequencing for the seven tested markers. SNP analysis on BAC ends was carried out by analyzing nucleotide variation within assembled reads in one contig, each reading originated from different E. coli clones. A base miscall in one of the BAC end reads, or even a mutation introduced by the E. coli replication machinery in the BAC sequence, could create a false SNP that is not present in the genomic sequence, as it was confirmed by direct Sanger sequencing of the genomic DNA amplified by PCR. For 62 markers, potential locus duplication (C7) was suspected and confirmed for the tested locus by the heterozygous profile for the haploid Clementine line selected for whole genome sequencing [29]. Moerover multiple blasts in the reference citrus whole genome sequence (http://www.phytozome.net/clementine.php) of the corresponding sequences comfort this hypothesis (data not shown). Five hundred ninety six of the SNPs mined in Clementine BES and 26 from targeted gene sequencing were finally selected for genetic analysis. The validation of the SNP genotyping for 15 SNPs × 24 genotypes was conducted by Sanger sequencing and a validity rate of 99.2% was obtained.

GoldenGate SNP marker transferability and loci with null allele usefulness

Marker transferability was estimated by null allele identification and dispersion in the gene pool. The null alleles may result from unexpected polymorphisms affecting the allele amplification/hybridization on the GoldenGate array. This may result from deletions spanning a polymorphic site [33, 34], secondary polymorphisms interfering with genotyping at the primary polymorphic target (as was observed for two markers with Sanger sequencing of the C. maxima and C. medica accessions), and even unexpected alleles at the primary polymorphism (such as triallelic sites; [35]). All of these are important potential sources of reproducible, but inaccurate, genotypes for population genetic studies because heterozygous null alleles are indistinguishable from the expected homozygotes on most genotyping platforms. In this study, among the high quality markers, 506 WONA and 116 WNA loci were selected. The transferability of SNPs to the whole Citrus genus can thus be estimated as 506/622 = 81.4% if all loci are considered, or as 480/596 = 80.5% considering loci identified in the Clementine BES. The null alleles were primarily found in C. medica, C. maxima, and wild Citrus with an average homozygosity frequency for null alleles being 0.64, 0.61 and 0.47, respectively, but only 0.11 in C. reticulata. As these markers were identified from the Clementine BES and GoldenGate primers defined from Clementine sequences, these results are reasonable based on the strong genetic relationship between Clementine and C. reticulata and the important inter-specific differentiation between C. reticulata and the others basic taxa [28]. For the secondary species resulting from hybridization between the C. reticulata gene pool and the other basic taxa, the WNA loci present the advantage of frequent recessivity for the other parental gene pools. Therefore, it may allow identification of the C. reticulata subset that contributed to the secondary species genesis. Moreover, heterozygous null alleles should be useful for genetic mapping.

SNPs mined in a single heterozygous genotype provide a distorted view of the gene pool diversity but confirm the high stratification of the Citrus genus

The selection of heterozygous markers in Clementine primarily affects the estimation of two components of genetic diversity. The first is the differentiation between the basic taxa that appears to be underestimated with CHet loci when compared with CHom loci. The second is the intra-specific diversity. C. reticulata within diversity (both intercultivar and heterozygosity) and C. sinensis heterozygosity appear to be overestimated using the CHet loci when compared with the CHom loci. This is in contrast with the results obtained for C. paradisi, C. aurantifolia and C. limon. Previous studies have shown that Clementine is highly related to C. reticulata with a limited introgression of C. maxima[24, 28, 36]. Therefore, it can be inferred that the majority of the Clementine heterozygosity arose from the C. reticulata gene pool diversity. This may explain the overestimation of C. reticulata diversity when compared to C. maxima and C. medica, as well as the underestimation of the Citrus gene pool stratification. The genetic constitution of C. sinensis (mainly issued from the C. reticulata gene pool; see below for more detail) may also explain its higher heterozygosity for the CHet markers. According to their supposed origin (see below), the heterozygosity of C. limon and C. aurantifolia is based on the interspecific differentiation between the basic taxa. On the other hand, C. paradisi arose from C. maxima/C. reticulata differentiation and C. maxima within diversity. Thus, the underestimation of interspecific differentiation and the underestimation of C. maxima within diversity with the CHet markers explain the underestimation of diversity and heterozygosity of the above mentioned secondary species. In the present study, very low intraspecific polymorphism was identified for C. maxima and C. medica, regardless of the heterozygosity of the markers in Clementine. However, previous SSR studies reported similar within species diversity in C. maxima and C. reticulata[27, 28]. It is highly probable that the set of markers used in this study target primarily within C. reticulata polymorphisms from one side and interspecific polymorphisms from the other side. Overall, the results reported here illustrate the limit and bias of the SNP array approach for large diversity analysis on a highly stratified population when the SNP discovery is based on a very limited panel. The SNP ascertainment bias has been widely discussed in humans [10, 12] and animals [11, 13–15] in relation with geographical stratification. This study reports bias associated with 'racial' differentiation in cultivated plants. This bias is enforced when taking into account the WNA loci. Future accurate analysis of the interspecific mosaic structure of secondary species and intra- and interspecific polymorphism analysis should be based on a non-biased pangenomic set of markers. The availability of a reference citrus genome sequence [29, 30] and the new sequencing methodologies will soon allow these objectives to be re-visited by the resequencing of several accessions of the basic taxa and secondary species.

Despite the distorted view of the gene pool diversity, the global organization around the basic taxa is still clear in both the PCA and NJ representation based in the WONA loci. The analysis of Fstat parameters on the subset of the genotypes of the three basic taxa (C. reticulata, C. medica, C. maxima) with a non-significant Fis value but high Fit and Fst values confirms this important organization of the allelic diversity between these taxa. Moreover, a very high proportion of loci pairs display significant linkage disequilibrium in the germplasm sample. The majority of these locus pairs with significant LD in the germplasm sample appear in equilibrium within the segregating 'Chandler pummelo × Nules Clementine' (CxN) population, testifying for very extended LD in the Citrus genus. Similar results were observed by Garcia-Lor et al. [28] for SSRs and InDel markers with significant LD for loci situated in different linkage groups. Breeding systems and domestication history are determinant factors of the LD structure in the germplasm of cultivated plants [37, 38]. The extent of LD is generally greater for species that possess a selfing mating system [39–41] than for outcrossing ones [42–44]. The heterozygous deficit and generalized linkage disequilibrium observed in the Citrus genus indicates a strong population subdivision and thus a low gene flow between C. medica, C. reticulata, C. maxima and wild Citrus. The differentiation between these sexually compatible taxa may be explained by foundation effect in three geographic zones and by an initial allopatric evolution. C. maxima originated in the Malay Archipelago and Indonesia, C. medica evolved in north-eastern India and the nearby region of Burma and China, and C. reticulata diversification occurred over a region including Vietnam, southern China and Japan [20, 45]. Secondary species arose from the hybridization of the basic taxa. The partial apomixis of most of the secondary species has certainly been an essential element in the limitation of gene flows after that human activities have put into contact the differentiated gene pools of the basic taxa. Apomixis may also explain that, in agreement with previous molecular studies [27, 28, 46], very few polymorphisms were found between the analyzed genotypes within C. sinensis, C. aurantium and C. paradisi although they were highly heterozygous (Ho of 0.65, 0.55 and 0.42, respectively, with the whole set of WONA markers). This confirms that most of the intra-specific polymorphism of these secondary species arose from punctual mutation, transposable element movement [47] or epigenetic variation.

Some parentage hypotheses for secondary species are strongly comfirmed

The parentage hypothesis of some very important commercial species or cultivars was checked using their position in the PCA and NJA representations and the loci count for which the hybrid genotype disagrees with the supposed parent ones (data not shown based on the 506 WONA loci). The synthesis of the parentage hypothesis is given in Figure 6.

Sweet orange (C. sinensis) and sour orange (C. aurantium): these two species are believed to derive from hybridizations between the C. maxima and C. reticulata gene pools [24, 27, 48]. Their positions in the PCA analysis with C. reticulata and C. maxima used as active individuals confirm that a predominant portion of their genome arose from the C. reticulata gene pool, as previously deduced from SSR markers [28, 49].

Clementine: it is generally agreed that Father Clement selected, a little more than one century ago in Algeria, Clementine as a chance seedling from 'Mediterranean' mandarin. The mandarin female parentage was confirmed by mitochondrial genome analysis [50]. 'Granito' sour orange was initially considered to be the male parent. However, previous molecular studies suggested that Clementine was more likely a mandarin × sweet orange hybrid [24, 36]. The position of Clementine relative to sweet orange, sour orange and 'Mediterranean' mandarin in the PCA analysis supports this hypothesis. The hypothesis of a 'Mediterranean' mandarin × sweet orange is definitively authenticated with only one locus out of 506 suggesting incompatible genotypes, while 86 loci disagree with the hypothesis of an hybridization between 'Mediterranean' mandarin and sour orange.

Grapefruit (C. paradisi): the origin of grapefruit is attributed to a natural hybridization between pummelo (C. maxima) and sweet orange (C. sinensis). This hybridization may have occurred in the Caribbean more than 200 years ago [21, 51–53]. In this study, grapefruit had an intermediary position between the sweet orange and pummelo gene pool in the PCA representation. Moreover, the NJA for the WNA markers clearly demonstrates the relationship of grapefruit and sweet orange. According to the sweet orange/pummelo combination, 96.3% to 98.0% of the 506 analyzed markers agree with this hypothesis. The best fitting is found with Tahiti pummelo. When searching for a potential sour orange × pummelo origin, the percentage of loci in disagreement varies between 12.5% and 14.5%. C. maxima is a polymorphic non-apomictic species. Therefore, due to the limited number of pummelo accessions analyzed, the absence of a 100% fit is reasonable. Moreover, as previously discussed, null alleles are relatively frequent in pummelo and it is likely that some of them have not been identified due to their heterozygous status. Therefore, the absence of some parental alleles in grapefruit, although they seem homozygous in one of the parents, may be explained by heterozygous null alleles for this parent. Upon looking for alleles present in grapefruit but absent in the two potential parents, only one to four loci disagree according to the considered pummelos. Thus, the data strongly confirm the hypothesis of the sweet orange × pummelo hybridization.

'Eureka' and 'Lisbon' lemon: Based on RFLP, RAPD and CAPS data, Nicolosi et al. [24] were the first to propose that lemon arose from a hybridization between C. aurantium and C. medica. This hypothesis was supported by nuclear SSR analysis [28]. In the present study, 'Eureka' and 'Lisbon' lemon varieties were highly heterozygous and very similar. These lemons are very likely two somatic mutants of the same ancestor. In PCA, their position was between the C. aurantium and C. medica group in each factorial axis. According to the citron accession, 96.0% to 97.8% of the 506 analyzed markers agree with this hypothesis. Moreover, null alleles are relatively frequent in citron and pummelo (contributing to sour-orange genesis). As for grapefruit, a search for alleles present in lemon but absent in the two potential parents reveals only one locus (CiC4841-04) out of 506 that disagrees with this hypothesis.

Mexican lime and Alemow: these two citrus were considered two distinct species, C. aurantifolia and C. macrophylla, respectively, by Tanaka [54]. However, Swingle and Reece [19] joined the two in C. aurantifolia. In all of the analysis reported here, these two were very close and intermediary between the citron cluster and a papeda cluster including C. hystrix and C. micrantha. For Mexican lime, this position is in agreement with the hypothesis proposed by Nicolosi et al. [24]. These authors proposed that Mexican lime was a hybrid between C. micrantha x C. medica. The maternal phylogeny was recently confirmed by Froelicher et al. [50]. According to the citron cultivar, 97.4% to 98.5% of the 506 analyzed markers agree with the C. micrantha x C. medica hypothesis with 'Humpang' citron providing the best fit. Moreover, no allele observed in Mexican lime was absent in the two potential parents. A C. hystrix x C. medica hypothesis produced very close results (97.3% to 97.7%). Very similar results were observed for 'Alemow' with 96.7% to 97.4% and 96.0% and 96.9% of loci in agreements with C. micrantha x C. medica and C. hystrix x C. medica origins, respectively. The papeda maternal parentage of Alemow was demonstrated by mitochondrial genome analysis [50]. Mexican lime and Alemow clearly have similar origins, and the papeda × C. medica hypothesis was confirmed by this data. An enhanced study of papeda germplasm will be necessary to definitively conclude C. micrantha or another papeda as the female parent.

C. amblycarpa is native to Indonesia where it is called Djerook leemo (http://www.ars-grin.gov/cgi-bin/npgs/html/taxon.pl?10679). It is generally considered to be a mandarin hybrid, and its common English name is Nasnaran mandarin. However, Froelicher et al. [50] showed that it has a papeda mitotype, identical to C. micrantha and C. hystrix. In PCA, C. amblycarpa displays an intermediary position between the two mentioned papedas and the acidic mandarin group ('Depressa', 'Sunki', 'Cleopatra' and 'Sun Chu Cha' mandarins). Its relationship with these mandarins is confirmed by the sharing of the same cluster in the NJA based on the WNA loci. A search for a potential direct papeda × C. reticulata origin was conducted. The best observed fit was a C. amblycarpa x 'Sun Chu Cha' mandarin hybridization with 92.5% of loci in agreement. For only 0.4% of the loci, one allele absent in the two parents was found in C. amblycarpa. Therefore, the hypothesis of papeda × acidic mandarin is proposed for C. amblycarpa.

A very useful set of SNPs markers for the comparative genetic mapping in Citrus and association genetic studies in C. reticulata

Compared with other crops, genetic mapping in citrus is still undeveloped. The recent genetic maps based on codominant markers (primarily SSRs) [55–57] integrated around 150 markers, while maps based on dominant markers such as AFLPs [58], SRAPs, ISSRs, and RAPDs [59] included a little more than 200 markers. The markers mined in Clementine BES will be very useful for genetic mapping and association genetics in C. reticulata (554 polymorphic loci). The usefulness of these markers is more limited for the other basic taxa with 189, 99 and 17 polymorphic markers within C. maxima, C. medica and C. micrantha, respectively. For the secondary species, the marker number varies from 221 for C. paradisi to 460 for C. limon. For C. sinensis, C. aurantium and C. paradisi, where very little inter-cultivar diversity was found, the numbers of polymorphic loci are very similar to the number of heterozygous loci in a concrete genotype (and thus, directly mappable from a segregating population having such a genotype as a parent). For the other species, a consensus map should be established from several progenies to map all of the indicated markers. Moreover, it is probable that secondary species, such as C. paradisi, C. sinensis, C. aurantium and C. limon, have numerous heterozygous null alleles (inherited from the C. maxima or C. medica gene pools) for WNA loci in phylogenetic heterozygosity (C. reticulata/C. maxima or C. reticulata/C. medica). Therefore, the real number of mappable loci should be higher for these species. The biallelic nature of SNP markers limits the possibility to establish two anchored maps (male and female) from a single cross. This is because the allelelic phase of the markers heterozygous for the two parents can only be inferred for homozygous progenies. Multiallelic markers like SSRs are more powerful for such application [46].

The 547 markers heterozygous in Clementine are currently mapped in the framework of the International Citrus Genome Consortium (ICGC; [60]) and will contribute to the assembly of the reference citrus whole genome sequence. Interestingly, at least 346 of these markers should also be mapped on sweet orange in order to anchor the sweet orange genetic map developed by an US consortium [57] and the Clementine map to establish a saturated consensus citrus genetic map. Furthermore, it is also notable to mention that a large proportion of the analyzed SNPs are located in or close to putative coding regions [31]; therefore, these 'functional SNPs' may provide an important resource for the identification of genes associated with specific trait loci.

Plant material

In addition to Clementine cv 'Nules' (whose BES were used for SNP mining), 53 varieties from the citrus germplasm bank of IVIA (Spain) and INRA/CIRAD (France) were used for the transferability and diversity study of SNPs within the Citrus Genus. According to the Swingle and Reece classification [19] and the Nicolosi et al. [24] hypothesis on the origin of cultivated citrus species, 29 belong to the three primary species (12 C. reticulata, 10 C. maxima and seven C. medica), 19 represent secondary species (two C. aurantium, four C. sinensis, two C. paradisi, seven C. limon and four C. aurantifolia), and five are wild species (additional file 6). Among the C. medica accessions, the present work confirmed the previously doubted classification of two cultivars as true citron ('Damas' and 'Rhob el Arsa' cultivars). These cultivars were thus excluded from the evaluation of within species diversity and between species organization of diversity.

A segregating population of 52 interspecific hybrids of 'Chandler' pummelo × 'Nules' Clementine (CxN, developed by Cirad in Corsica) was used to confirm Clementine heterozygosity. This population was helpful in making the genotypic assignments of the germplasm samples and in comparing the linkage disequilibrium (LD) distribution of the germplasm (depending on the evolutionary history of the gene pool and marker linkage) and the segregating population (depending only on the marker linkage).

DNA extraction

Total DNA was extracted from fresh leaves according to Doyle and Doyle [61].

SNP mining from Clementine BACend sequences

As described in Terol et al. [31], in silico SNP mining was performed from 46,339 C. clementina cv. Nules BACend sequences (BESs) covering 28.1 Mb of genomic sequences. Assembly of BESs that did not contain repetitive sequences was performed using CAP3 [62]. A total of 6,461 contigs, including 19,057 reads and covering 6.14 Mb of sequence, were produced. The SNPs were mined in these contigs using POLYBAYES software. A total of 6,617 putative SNPs (1.08 SNPs per kb) were found. A total of 4,904 SNPs were in silico tested for their potential use in the Illumina Golden Gate array following the Illumina procedure.

SNP mining in candidate genes

In an effort to identify SNPs within the Citrus genus, two cultivars of C. medica (Corsican and Budha's hand citrons), two cultivars of C. reticulata ('Cleopatra' and 'Mediterranean' mandarin), two cultivars of C. maxima ('Chandler' and 'Pink' pummelos) and one C. micrantha accession were selected. Primers (additional file 2) were defined from EST sequences available in the public databases for six genes implicated in primary and secondary metabolite biosynthesis pathways involved in determining citrus fruit quality (sugars, acids, flavonoids and carotenoids: Chalcone isomerase -CHI-, Vacuolar citrate/H+ symporter -TRPA-, Phosphofructokinase -PKF-, Lycopene β-cyclase -LCY2-, Phytoene synthase -PSY-, Lycopene β-cyclase -LCYB-) and four candidates genes linked to salt tolerance (Cation/H+ antiporter -CAX-, Ascorbate oxydase -AOC-, High-affinity K+ Transporter 1 -HKT1- and Tréhalose-6-Phosphate Synthase -TS-). PCR amplifications of the samples were performed using a Mastercycler EP Gradient S thermocycler (Eppendorf) in a 100 μL final volume containing 0.025 U/μL of Pfu DNA polymerase (Fermentas), 2 ng/μL of genomic DNA, 0.2 mM of each dNTP, 2 mM MgSO₄, 75 mM Tris-HCl (pH 8.8), 20 mM (NH₄)₂SO₄, and 0.2 μM of each primer. The following PCR program was used: denaturation at 94°C for 5 min; 35 repetitions of 30 s at 94°C, 1 min at 55°C or 60°C (according to primer T_m), and 45 s at 72°C; and a final elongation step of 4 min at 72°C. PCR product purification was performed using the QIAquick^® PCR Purification Kit (Qiagen). Amplicons of the seven selected genotypes were sequenced by the Sanger method from the 5' extremity using dideoxynucleotides marked with fluorescence (Big Dye Terminator Cycle Sequencing Kit v3.1). Sequences were aligned with BioEdit [63] and SNPs were detected along the sequences. Of the identified polymorphisms, 121 were in silico tested for their potential use in the Illumina GoldenGate assay.

SNP genotyping with an Illumina GoldenGate microarray

For each selected SNP locus, three primers were designed using the Illumina Assay Design Tool (https://icom.illumina.com). Sequence and primer information for the selected SNPs are listed in the additional file 1. The DNA of the samples to be genotyped was quantified according to Illumina specifications using PicoGreen (Molecular Probes) and a Gemini XPS Fluorescence Microplate Reader (Molecular Devices). The genotyping reactions were performed according to the standard Illumina GoldenGate assay instructions (http://www.illumina.com). In brief, 250 ng of template DNA was used per sample. SNP-specific oligonucleotides containing both detection specific sequences and universal primer sequences were hybridized, extended and ligated to a common oligonucleotide containing a universal primer sequence. Ligated products were amplified using a universal primer set. Genotypes were determined by hybridizing the amplified products to a bead array which was complementary to the sequence specific tags. The fluorescence of the bead array was determined using a Bead Array Reader (Illumina). Two genotype controls ('Nules' Clementine and 'Chandler' pummelo) were repeated twice in each plate. The data were collected and analyzed using the Genome Studio software (Illumina). The automatic allele calling was visually checked and corrected if necessary, taking advantage of the segregating Pummelo × Clementine population.

SNP genotyping validation by Sanger sequencing of amplicons

Of the 54 Citrus accessions genotyped with the GoldenGate array, 24 were used to validate the genotyping data for 15 SNPs from five genes (LCY2, LCYb, PKF, PSY, TRPA). This subset included 'Nules' Clementine, seven accessions of C. reticulata, five C. maxima, four C. medica, two C. aurantium, one C. sinensis, one C. paradisi, one C. limon, one C. aurantifolia, and one C. micrantha (additional file 6). The primers and PCR amplification, purification and sequencing were the same as that used for the SNP mining in the candidate genes.

Study of the origin of unexpected SNPs by Sanger analysis

The origin of unexpected polymorphisms displayed by several SNP markers from the Clementine BES, such as null alleles, no heterozygosity for Clementine and 'fixed heterozygosity', was analyzed by Sanger sequencing of the amplicons of four accessions: 'Nules' Clementine, haploid Clementine, 'Chandler' pummelo and Corsican citron. Primers flanking the SNP site were defined from the contig sequences obtained from the BES [31] to produce amplified fragments ranging from 200 to 620 bp (additional file 7). PCR amplification, purification and sequencing were performed in the same manner as the SNP mining in the candidate genes.

Data analysis

with d_i-j, the dissimilarity between units i and j; L, the number of loci; m_l, the number of matching alleles for locus l. Weighted neighbor-joining trees were computed from the dissimilarity matrix with 1000 bootstraps to test branch robustness. Principal Component Analyses (PCA) were computed using XLSTAT on the matrix of the frequencies for each allele. Genetic population parameters (Ho, observed heterozygosity; He, expected heterozygosity equivalent to Nei diversity index [66]; and Fstats - Fis, Fit and Fst - based on the parameters of Wright [67] and Weir & Cockerham [68] were calculated with GENETIX v. 4.03 software.

with G indicating the total number of observed genotypes and g_i indicating the frequency of each observed genotype.

Linkage disequilibrium (LD) was estimated by r² (chi square Pearson's correlation coefficient). The significances were estimated with the exac-test pvalue using PowerMarker software v. 3.25 [69].