Assignment of SNP allelic configuration in polyploids using competitive allele-specific PCR: application to citrus triploid progeny

Abstract

Background

Polyploidy is a major component of eukaryote evolution. Estimation of allele copy numbers for molecular markers has long been considered a challenge for polyploid species, while this process is essential for most genetic research. With the increasing availability and whole-genome coverage of single nucleotide polymorphism (SNP) markers, it is essential to implement a versatile SNP genotyping method to assign allelic configuration efficiently in polyploids.

Scope

This work evaluates the usefulness of the KASPar method, based on competitive allele-specific PCR, for the assignment of SNP allelic configuration. Citrus was chosen as a model because of its economic importance, the ongoing worldwide polyploidy manipulation projects for cultivar and rootstock breeding, and the increasing availability of SNP markers.

Conclusions

Fifteen SNP markers were successfully designed that produced clear allele signals that were in agreement with previous genotyping results at the diploid level. The analysis of DNA mixes between two haploid lines (Clementine and pummelo) at 13 different ratios revealed a very high correlation (average = 0·9796; s.d. = 0·0094) between the allele ratio and two parameters [θ angle = tan⁻¹ (y/x) and y′ = y/(x + y)] derived from the two normalized allele signals (x and y) provided by KASPar. Separated cluster analysis and analysis of variance (ANOVA) from mixed DNA simulating triploid and tetraploid hybrids provided 99·71 % correct allelic configuration. Moreover, triploid populations arising from 2n gametes and interploid crosses were easily genotyped and provided useful genetic information. This work demonstrates that the KASPar SNP genotyping technique is an efficient way to assign heterozygous allelic configurations within polyploid populations. This method is accurate, simple and cost-effective. Moreover, it may be useful for quantitative studies, such as relative allele-specific expression analysis and bulk segregant analysis.

INTRODUCTION

Polyploidy is a major component of eukaryote evolution, particularly in angiosperms (Gant, 1981; Soltis and Soltis, 1993; Wendel and Doyle, 2005). Many plant species result from autopolyploidization or allopolyploidization events, and polyploidization is considered the most common sympatric speciation mechanism (Otto and Whitton, 2000). Despite the indisputable importance of polyploid plant species, the genetics of these plant species are less well known than those of their diploid counterparts. Indeed, the estimation of molecular marker allele copy number has long been considered a challenge for polyploid species with polysomic inheritance, while it is essential to assign the allelic configuration for different types of heterozygotes for accurate population genetic studies. In segregating polyploid progeny, the population genetic structure can provide relevant information about the underlying meiosis mechanisms that take place in the formation of these progeny, which also greatly affect character segregation (Hutten et al., 1993; Tai and DeJong, 1997; Douches and Maas, 1998; Barcaccia et al., 2003; Brownfield and Kohler, 2011). Moreover, allelic dosage can affect gene expression and phenotype. Therefore, the determination of allelic dosages is particularly important for marker–trait association studies (De Jong et al., 2003; Sjoling et al., 2005). When parents of a polyploid progeny share one allele, only the dosage allele estimation allows knowledge of the alleles transmitted by each parent to heterozygous progeny. Therefore, knowledge of allelic dosage in polyploids appears to be essential for studies using single nucleotide polymorphism (SNP) markers, most of which are biallelic.

Several techniques have been used to estimate allele dosage in polyploid genotypes or tissues. When analysing microsatellite markers [simple sequence repeats (SSRs)], the microsatellite allele counting–peak ratios method (MAC-PR; Esselink et al., 2004) is especially useful. However, SSR analysis remains relatively costly and time consuming compared with actual SNP genotyping methods. Moreover, with the increasing availability of expressed sequence tag (EST) databases and whole-genome sequences, SNPs have become the most abundant and powerful polymorphic markers that can be selected throughout the entire genome (Edwards and Batley, 2010).

Several SNP genotyping methods have been developed. Some methods are based on electrophoretic separation after PCR amplification, including allele-specific primer extension (Kwok, 2001) and temperature-switch PCR (Tabone et al., 2009). High-throughput genotyping can be obtained using array methodologies (Sapolsky et al., 1999; Ishikawa et al., 2005); other techniques are based on pyrosequencing™ (Ronaghi et al., 1998; Ahmadian et al., 2000). However, the application of SNP markers has been limited primarily to diploid organisms, while the application of these markers to polyploid organisms for allele dosage estimation remains limited. Rickert et al. (2002) reported the use of pyrosequencing™ in polyploid potatoes, with some sequence-specific limitations. The usefulness of SNPlex™ (Berard et al., 2009) and Illumina Golden Gate™ assays (Akhunov et al., 2009) for the genotyping of polyploid wheat has been demonstrated. For genotype calling in tetraploid species with SNP analysis using the Illumina GoldenGate™ array, Voorrips et al. (2011) developed an algorithm using mixture models, but they assumed Hardy–Weinberg equilibrium within the population, which does not occur in all segregating polyploid progeny. Microarray data (Kirov et al., 2006; Meaburn et al., 2006; Steer et al., 2007) have also been used to estimate allelic frequencies in bulk populations or DNA pools, i.e. to perform genome-wide association scans. Array analysis is more suitable for genotyping large numbers of samples over numerous markers than for performing small-scale analysis, as array analysis lacks flexibility in terms of the numbers and panels of SNP loci that can be analysed. Targeted pyrosequencing™ (Gruber et al., 2002; Neve et al., 2002; Wasson et al., 2002; Lavebratt et al., 2004) can be useful for performing allele frequency estimation for a few genes in pooled DNA, but this technique remains relatively costly and time consuming. It is, therefore, important to develop alternative methods that offer a wider spectrum of genotyping possibilities to infer SNP allelic configurations in polyploid plants in small- to larger scale projects.

The KBiosciences Competitive Allele­Specific PCR SNP genotyping system (KASPar) is a homogeneous fluorescent endpoint genotyping system (Cuppen, 2007) that utilizes a unique form of competitive, allele-specific PCR and combines the use of a highly specific 5′–3′ exonuclease-deleted Taq DNA polymerase with two competitive, allele-specific, tailed forward primers and one common reverse primer. This system is simple and cost-effective compared with other SNP genotyping assays and is well adapted to low- to medium-throughput genotyping projects (Chen et al., 2010). This technology has been successfully applied to the study of humans, animals and plants (Nijman et al., 2008; Bauer et al., 2009; Cortes et al., 2011; Rosso et al., 2011).

Citrus is mainly diploid. However, many modern breeding projects for the production of seedless mandarins based on the production of triploid hybrids (Ollitrault et al., 2008; Aleza et al., 2010, 2012a, b) and tetraploid rootstocks are promising (Saleh et al., 2008; Dambier et al., 2011; Grosser and Gmitter, 2011). Triploid populations in citrus can arise from unreduced gametes in crosses between diploid parents or from interploid (diploid × tetraploid or tetraploid × diploid) crosses. Sexual polyploidization resulting from 2n mega gametophyte production is routinely exploited for triploid citrus breeding (Aleza et al., 2010). In such crosses, segregation of a marker depends on the parental genetic structure, the relative distance to the centromere and the mode of restitution [first division restitution (FDR) or second division restitution (SDR)]. The MAC-PR method has been successfully applied to demonstrate the SDR origin of the 2n gametes arising from the ‘Fortune’ mandarin cultivar and to locate the centromere in one chromosome (Cuenca et al., 2011). For interploid crosses, marker segregations are almost exclusively dependent on the parental genetic conformation and preferential chromosome pairing. SSR markers were also used to analyse the meiotic behaviour of a tetraploid interspecific somatic hybrid of C. deliciosa + C. lemon (Kamiri et al., 2011), with the authors concluding that there was predominant tetrasomic segregation. However, the low availability of SSR markers displaying a favourable parental allelic structure that can be used to differentiate male and female contributions to the hybrids limits such studies to just a few areas of the genome. Conversely, large SNP resources have become available from extensive sequencing projects (Terol et al., 2007, 2008; Gmitter et al., 2012; Ollitrault et al., 2012b).

The objective of the present work was to evaluate the potential of the KASPar method to assess SNP allelic configurations in polyploid plants. Citrus was chosen as a model system because of its economic importance, the worldwide ongoing polyploidy manipulation projects for cultivar and rootstock breeding, and the increasing availability of SNP markers.

The quantitative value of the KASPar assay was estimated by pooling DNA from two haploid lines at several relative concentrations, simulating, among others, triploid and tetraploid heterozygous progeny. A method was developed for semi-automated polyploid genotype calling and applied for allelic configuration analysis of 170 triploid hybrids from two families arising from both sexual polyploidization and interploid crosses.

MATERIALS AND METHODS

Sample preparation

DNA pool preparation

Genomic DNA from two haploid lines, Citrus maxima (Burm.) Merr. (pummelo) cv Chandler and C. clementina Hort. ex Tan. (clementine; Aleza et al., 2009a), was isolated using a Plant DNAeasy Kit from Qiagen Inc. (Valencia, CA, USA) following the manufacturer's protocol. DNA concentrations were estimated with PicoGreen^® and adjusted to 30 ng μL⁻¹. DNA from the two haploid lines was pooled at ratios of 9:1, 5:1, 3:1, 2:1, 3:2, 1:1, 2:3, 1:2, 1:3, 1:5 and 1:9. Five samples (replications) of each haploid line and pool were prepared and used to test the accuracy of the technique.

Simulation of triploid and tetraploid hybrid samples by pooling DNA from haploid lines

Two sub-sets of the haploid DNA pool, one corresponding to 2:1 and 1:2 ratios that simulated heterozygous triploid genotypes and the other corresponding to 3:1, 1:1 and 1:3 ratios that simulated tetraploid heterozygous genotypes, were jointly used with the haploid genotypes to test the capability of the technique to discriminate among different types of heterozygotes within triploid and tetraploid populations.

Natural triploid populations: 2x × 2x crosses

‘Fortune’ (C. clementina × C. tangerina hort. ex Tan.) and ‘Willowleaf’ (C. deliciosa Ten.) diploid mandarins and 39 triploid hybrids segregating from this cross (Aleza et al., 2010) were selected to test the accuracy of the technique by analysing two replicates of each sample. Moreover, 86 triploid hybrids from a ‘Clementine’ (C. clementina) × ‘Nadorcott’ (C. reticulata Blanco) population (Aleza et al., 2010) were also analysed as individual samples to perform genotype calling.

Natural triploid populations: 4x × 2x cross

Tetraploid ‘Clementine’ (C. clementina 4x) and diploid ‘Pink’ pummelo (Citrus maxima 2x) and 88 triploid hybrids segregating from this cross (Aleza et al., 2012b) were also analysed. Tetraploid ‘Clementine’ was obtained by treating buds of the diploid ‘Clementine’ with colchicine (Aleza et al., 2009b); therefore, this genotype should be duplex (aabb) for all heterozygous loci.

SNP selection

SNPs for the analysis of signal–dosage correlation

To validate the quantitative value that was obtained from the KASPar assay using pooled DNA, seven SNPs differentiating the two haploid lines (C. maxima and C. clementina) were selected from previous genotyping data obtained on the Illumina GoldenGate™ platform (Ollitrault et al., 2012a). These SNPs were also used to test the accuracy of the technique in genotyping repetitions of the same sample over the ‘Fortune’ × ‘Willowleaf’ triploid population.

SNPs for triploid population analysis

Three out of seven SNPs used for the previous analysis, and eight other SNP markers, were selected to test the capacity of the technique for differentiating between heterozygous genotypes within two triploid populations, one arising from a 2x × 2x cross, and the other from a 4x × 2x cross. These SNPs are heterozygous for ‘Clementine’ and homozygous or heterozygous with the null allele for ‘Nadorcott’ and ‘Pink’.

SNP genotyping

All samples were genotyped for the SNP markers using KASPar technology by KBioscience^® (http://www.kbioscience.co.uk/). The KASPar™ Genotyping System is a competitive, allele-specific dual Förster resonance energy transfer (FRET)-based assay for SNP genotyping. It uses two FRET cassettes where fluorometric dye [FAM (6-carboxy-fluorescein) or VIC^®] is conjugated to primer but quenched via resonance energy transfer; ROX dye (6-carboxy-X-rhodamine, succinimidyl ester) is used to normalize the data. Sample DNA is amplified with a thermal cycler using allele-specific primers, leading to the separation of fluorometric dye and quencher when the FRET cassette primer is hybridized with DNA. Primers were designed by KBioscience^®, based on the SNP locus-flanking sequence (approx. 50 nucleotides on each side of the SNP). Two 40-mer allele-specific oligonucleotides and one common 20-mer oligonucleotide were defined for each locus. Detailed information for all SNP markers can be found in Supplementary Data Table S1. Additional details about this genotyping method can be found in Cuppen (2007).

Data analysis method

Normalized signals from each SNP allele (x and y) were provided by KBioscience^® services, and two-dimensional plot representations were obtained using SNPViewer software (http://www.kbioscience.co.uk/software/SNP%20viewer%20intro.html). From the x and y normalized values, the theta angle [θ = tan⁻¹ (y/x); 0 º ≤ θ ≤ 90 º] and the relative y allele signal [y′ = y/(x + y); 0 ≤ y′ ≤1] of each sample were calculated. Further analyses were carried out that considered the y′ parameter, as this parameter was found to provide better clustering and genotype calling of the samples.

Data from all haploid lines and DNA pools with different allele configurations (9:1, 5:1, 3:1, 2:1, 3:2, 1:1, 2:3, 1:2, 1:3, 1:5 and 1:9) were tested for correlations between doses, and both the theta angle and y′ values that were obtained. Cluster analysis (MacQueen, 1967) using the farthest-neighbour method with standardized squared euclidean distances and analysis of variance (ANOVA) were performed from the normalized allele signals (x, y) jointly and from the y′ parameter data for each SNP.

Data from triploid and tetraploid simulated populations were also analysed separately by cluster analysis and ANOVA. Replications of the same samples were used to test the precision of the technique by genotype calling.

All statistical data were analysed using Statgraphics^® Plus v5.1 software (Rockville, MD, USA).

RESULTS

Marker design and data acquisition

Primers for the KASPar assay were successfully designed by KBioscience^® for all 15 of the submitted SNP-surrounding sequences. Data acquisition for x and y allele signals allowed successful allelic calling for 2535 out of 2563 marker–genotype combinations (98·91 %). The validity of the genotyping results was verified by comparing the results for 24 diploid varieties with previous data obtained with an Illumina GoldenGate™ array. Complete conformity was observed (data not shown).

Analysis of the correlation between relative allele signals and relative allele frequencies in the DNA pools

To confirm the value of the KASPar assay for producing semi-quantitative data, equimolar DNA extracts from two haploid lines (Clementine and pummelo) were mixed at 13 different relative concentrations, and five replicates were analysed for each of seven SNP markers. The correlations between relative allele signals and relative doses were analysed.

An example of correlation analysis between relative allele dosage and signals is shown in Fig. 1 for the CiC2840-01 SNP marker. From x and y signal values (Fig. 1A), theta angle [θ = tan⁻¹ (y/x); Fig. 1B] and the relative y allele signal [y′ = y/(x + y); Fig. 1C] of each haploid line sample and DNA pool were calculated. High values of correlation coefficients between both parameters and relative allele dosage in the DNA pools were obtained for all analysed SNP markers, with an average of 0·9796 and a standard deviation of 0·0094 for the y′ parameter and an average of 0·9710 and a standard deviation of 0·0176 for the theta angle. Correlation values obtained for the y′ parameter were slightly superior to those obtained by the theta angle for six of the seven SNP markers that were analysed (Table 1).

Fig. 1.

Correlation study of allele doses and x, y signals for the CiC2840-01 SNP marker. (A) Plot of normalized x, y allele signals. (B) Correlation between relative haploid Chandler doses and theta angle [θ = tan⁻¹ (y/x)]. (C) Correlation between relative haploid Chandler doses and the y′ parameter [y′ = y(x + y)].

Table 1.

Correlation coefficients between relative allele dosage and allele signals from DNA pools at intermediate proportions for the theta angle and y′ parameter for the seven SNP markers analysed

SNP marker	Correlation coefficient for angle θ	Correlation coefficient for y′ parameter
CiC2840-01	0·9941	0·9919
CiC5089-06	0·9753	0·9779
CiC5785-01	0·9580	0·9747
DXS-M618	0·9881	0·9923
F3H-M309	0·9788	0·9803
FLS-M400	0·9535	0·9717
TRPA-M593	0·9492	0·9684
Average	0·9710	0·9796

For some markers (CiC5785-01, F3H-M309, FLS-M400 and TRPA-M593; Fig. 2), the linear regression established from the mixed sample did not fit with the signals from the pure sample. For all these markers, the relative signals corresponding to the haploid ‘Chandler’ allele in the DNA mixes appear to be lower than expected in relation to the relative DNA dosages. This can probably be attributed to PCR allele competition between the ‘Clementine’ and ‘Chandler’ alleles in the DNA mixes.

Fig. 2.

Correlation study between the relative haploid Chandler doses and the y′ parameter for six SNP markers.

However, in the DNA mixes, the correlations between allele signals and allele doses remained high for these markers (between 0·9684 and 0·9803), testifying to a very good linear regression between relative signals of the two alleles and relative allele dosages.

These data indicate that the KASPar technique, using either the y′ parameter or the theta angle, can be useful for a quantitative analysis of the relative allele frequency in a genotype or DNA pool. Because y′ produced a slightly higher correlation coefficient, this parameter was employed in further analyses.

Cluster analysis and ANOVA for simulated triploid and tetraploid allele dosage

Separated cluster analyses and ANOVA from relative y allele signals (y′ parameter) in triploid and tetraploid simulated populations were performed. With diallelic markers, for a triploid heterozygous genotype, there are only two allelic configurations to distinguish: aab and abb (duplex and simplex of a-allele). For a heterozygous tetraploid genotype, three allelic configurations may be differentiated: aaab, aabb and abbb (triplex, duplex and simplex of a-allele). With higher ploidy levels, the number of possible allelic configurations becomes even larger (n–1 configurations for n ploidy).

The ANOVA (Table 2) revealed a complete and correct classification of the average value of the different configurations that were simulated. An example of the x and y allele signals, the frequency histogram for the y′ parameter and the l.s.d. intervals for the mean from ANOVA for simulating triploid and tetraploid populations is provided for the CiC2840-01 SNP marker in Fig. 3.

Table 2.

Homogeneous groups formed, and F-values from ANOVA of SNPs from DNA pools simulating triploid and tetraploid populations showing the percentage of correctly classified replications by cluster analysis based on the y′ parameter

SNP marker	Simulating 3n populations						Simulating 4n populations
	1:0	2:1	1:2	0:1	F-value from ANOVA	% 3n correctly classified	1:0	3:1	1:1	1:3	0:1	F-value from ANOVA	% 4n correctly classified
CiC2840-01	a	b	c	d	2975·34	100	a	b	c	d	e	1644·48	100
CiC5089-06	a	b	c	d	1045·50	100	a	b	c	d	e	673·68	96
CiC5785-01	a	b	c	d	5110·86	100	a	b	c	d	e	6084·71	100
DXS-M618	a	b	c	d	2938·98	100	a	b	c	d	e	3013·55	100
F3H-M309	a	b	c	d	2063·52	100	a	b	c	d	e	1445·66	100
FLS-M400	a	b	c	d	10046·43	100	a	b	c	d	e	10459·43	100
TRPA-M593	a	b	c	d	11140·59	100	a	b	c	d	e	6666·05	100
Average						100							99·43

Fig. 3.

Study of simulating triploid and tetraploid populations for the CiC2840-01 SNP marker. (A) Plot of normalized x, y allele signals. (B) Frequency histogram for the y′ parameter. (C) Cluster analysis. (D) Least significant difference intervals for the mean obtained from ANOVA.

Moreover, all expected homogeneous groups were formed by cluster analysis using the farthest-neighbour method with standardized squared euclidean distances. All of the triploid sample replications and 99·43 % of the tetraploid ones were correctly classified; only one replication for the CiC5089-06 SNP marker was classified into an incorrect cluster (Table 2).

Allelic configuration of triploid populations

Accuracy of genotype calling for duplicated triploid samples

Thirty-nine triploid hybrids arising from ‘Fortune’ 2n gametes in the ‘Fortune × Willowleaf’ hybridization were analysed for seven SNPs, with two technical replications. All SNPs had the following allelic configuration: ‘Fortune’, ab; and ‘Willowleaf’ mandarin, aa. Therefore, depending on the origin of the diploid gamete, three genotypic clusters were expected: aaa, aab and abb. Samples with replications that were classified in the same cluster and thus genotyped with the same allelic conformation reached 97·44 %. Errors in classification were observed in five of the seven SNP markers analysed. Considering a replicate for the same DNA sample (with different allele calling between replicates) to be classified correctly, the average error rate for further routinely genotyping without replicates was estimated to be 1·28 %.

To perform the genotype calling of triploid progeny, cluster analyses were performed according to the expected genotypes for each population and the parental-specific allelic configuration of each marker.

2x × 2x triploid progeny

When crossing a heterozygous female parent (ab) with homozygous parents (aa), maternal heterozygosity restitution (HR) is reflected in the duplex (aab) triploid hybrids. Under the SDR mechanism, HR is directly linked to the distance from the locus under consideration to the centromere and, therefore, the frequency of HR can be estimated from this distance, as proposed by Cuenca et al. (2011). To validate genotype calling for triploids resulting from a 2x × 2x cross, three SNP markers (CiC3440-07, CiC5785-01 and CiC6278-01) mapped in chromosome II were selected. Indeed, Cuenca et al. (2011) located the centromere position for the corresponding linkage group at 59·6 cM of the current reference ‘Clementine’ genetic map (Ollitrault et al., 2012a) using the Cx(Co)⁴ partial interference model. The expected HR for the three considered markers (also mapped in the ‘Clementine’ map) was estimated using the same partial interference model.

Cluster analyses were performed from y′ parameter values of each hybrid over 11 analysed SNPs (including the three markers on chromosome II) within the ‘Clementine × Nadorcott’ population to carry out genotype calling. Figure 4A,B shows an example of cluster analysis for the CiC2840-01 SNP marker.

Fig. 4.

Plot of normalized x, y allele signals and histogram representing genotype calling from cluster analysis over 85 triploids from the ‘Clementine × Nadorcott’ cross. (A, B) CiC2840-01 (ab × aa); (C, D) CiC1749-05 (ab × a0).

The cluster analysis allowed the detection of null alleles in the male parent for two markers. Indeed, if the supposed homozygous parent in fact had heterozygosity for a null allele, five clusters should be obtained (ab × a0: aab, ab0, aaa/aa0, abb and bb0), where one cluster contains both ab0 triploids and ab diploid genotypes. Such a cluster configuration was observed for the CiC0610-01 and CiC1749-05 SNP markers (Fig. 4C,D).

On average, over all of the markers, 99·37 % of the samples were assigned to a cluster and, therefore, could be accurately called.

For the three markers of LGII, the observed HR values were not significantly different from those estimated from the espective markers and centromere locations in the ‘Clementine’ map (Table 3). This provides additional validation of the accuracy of polyploid genotype calling using the method presented in this study. For the eight remaining markers, the observed HR allowed us to estimate the relative distances of these markers from the centromere (Table 4), revealing markers with centromeric locations (<5 cM distance from the centromere: CiC1380-05, CiC2840-01 and CiC4581-01).

Table 3.

Results of genotype calling for three SNP loci in the ‘Clementine × Nadorcott’ triploid population from 2n gametes for markers heterozygous for ‘Clementine’ and homozygous for ‘Nadorcott’, including the conformity (χ² test) of the observed %HR with the theoretical one calculated from the distance from each locus to the centromere on chromosome II

SNP marker	Map position (cM)	NI	aaa	aab	abb	%HR (aab) observed	%HR estimated from map position (centromere at 59·6 cM)	χ2 (P-value)
CiC3440-07	67·22	86	37	10	39	11·63	13·59	0·282; NS (P = 0·5954)
CiC5785-01	44·73	84	32	17	35	20·24	28·23	2·648; NS (P = 0·1037)
CiC6278-01	57·01	86	46	0	40	0·00	4·21	3·780; NS (P = 0·0519)

NI, number of individuals genotyped; aaa, aab and abb: number of individuals of each genotype; NS, non-significantly different at α = 0·05).

Table 4.

Results of genotype calling for eight SNP loci in the ‘Clementine × Nadorcott’ triploid population from 2n gametes for markers heterozygous for ‘Clementine’ and homozygous or heterozygous with a null allele for ‘Nadorcott’, showing the estimated marker–centromere distance

SNP marker	NI	aaa	Aab	abb	ab0	bb0	% HR (aab + ab0) observed	Estimated distance to centromere (cM)
CiC0610-01	86	22	22	14	22	6	51·16	27·42
CiC0868-01	86	39	14	33	–	–	16·8	9·23
CiC1380-05	86	43	1	42	–	–	1·16	0·75
CiC1749-05	85	11	33	7	24	10	67·06	40·61
CiC1757-02	86	42	12	32	–	–	13·95	7·80
CiC2840-01	85	46	2	37	–	–	2·35	1·49
CiC4581-01	86	37	6	43	–	–	6·98	4·15
CiC5089-06	84	27	40	17	–	–	47·62	25·23

NI, number of individuals genotyped; aaa, aab, abb, ab0 and bb0, number of individuals of each genotype.

4x × 2x triploid progeny

Triploid genotyping was also performed for progeny arising from a cross between doubled diploid ‘Clementine’ (Aleza et al., 2009b) and ‘Pink’ pummelo. When crossing a duplex tetraploid parent (aabb) with a homozygous diploid parent (aa), three clusters can be expected (triplex-aaa, duplex-aab and simplex-abb), where maternal HR is reflected into the duplex (aab) triploid hybrids. On average, over all of the markers, 97·08 % of the samples were assigned to a cluster and, therefore, could be accurately called.

The genetic structure of the triploid progeny arising from ‘Clementine 4x’ × ‘Pink’ is shown in Table 5, which indicates the percentage of HR. For the other loci, HR values varied, ranging from 52·94 % for CiC4581-01 to 69·41 % for CiC5089-06. The average HR value over all loci was 59·64 %.

Table 5.

Results of genotype calling for seven SNP loci in the ‘Clementine 4x × Pink’ triploid population, indicating the heterozygosity restitution in Clementine 4x (%HR) at each locus

SNP marker	NI	aaa	aab	abb	%HR (aab) observed
CiC3440-07	88	17	50	21	56·82
CiC0868-01	87	11	58	18	66·67
CiC1380-05	87	19	53	15	60·92
CiC4581-01	85	19	45	21	52·94
CiC5089-06	85	15	59	11	69·41
CiC5785-01	80	22	43	15	53·75
CiC6278-01	86	22	49	15	56·98

NI, number of individuals genotyped; aaa, aab and abb, number of individuals of each genotype.

DISCUSSION

The KASPar method is a powerful technique for assigning SNP allelic configurations in polyploid progeny

Several techniques have been used to estimate allele dosage in polyploids, such as the MAC-PR method (Esselink et al., 2004) for SSR markers, and techniques for SNP genotyping, including allele-specific primer extension (Kwok, 2001), temperature-switch PCR (Tabone et al., 2009), array methodologies (Ishikawa et al., 2005) and targeted pyrosequencing™ (Ahmadian et al., 2000). Our study demonstrates that the KASPar technique (Cuppen, 2007) is an alternative method to infer SNP allelic configurations in polyploid plants that offers a wider spectrum of genotyping possibilities. The KASPar method is simple and cost-effective compared with other SNP genotyping assays and is well adapted to low- to medium-throughput genotyping projects. In addition to the markers published herein, 51 KASPar markers were successfully developed to analyse triploid and tetraploid citrus populations (Aleza et al., 2012c, d; Cuenca et al., 2012). KASPar markers were also successfully developed (41 over 42 tested) and transferred in the true citrus group (Citrus, Fortunella, Poncirus, Microcitrus and Eremocitrus genera) from SNP mining by sequencing within a Citrus collection (Garcia-Lor et al., 2013). When SNPs are mined in a large discovery panel, this offers the opportunity to select markers without additional variation in the flanking DNA sequence used as template for the competitive PCR of the KASPar assay and therefore to have a high degree of success in marker development. KASPar markers were successfully developed in a large range of plant (Cortes et al., 2011; Rosso et al., 2011; Byers et al., 2012) and animal (Nijman et al., 2008; Murad et al., 2009; Luciano et al., 2010) species, demonstrating its universal applicability.

The SNP genotyping and data analysis method presented in this study is simple and effective for genotyping triploid and tetraploid progeny and can also be used in the quantitative analysis of allele-specific expression. Allele signals (x, y) obtained from KBioscience^® can easily be transformed into y′ [y′ = y/(x + y); 0 ≤ y′ ≤ 1], which is a very useful parameter to cluster analysed samples. Theta angles [θ = tan⁻¹ (y/x); 0 º ≤ θ ≤ 90 º] can also be used to analyse data, but the y′ parameter offers better clustering results. Quantitative analyses for correlation of the allele signals, and the allele doses and sample clustering carried out in this work, were powerful techniques for assigning allelic configurations in simulated triploid and tetraploid citrus genotypes for all SNP markers that were analysed (100 % of the triploids were correctly classified as well as 99·43 % of the tetraploids). The analysis of concrete triploid hybrids with technical replications confirmed the high degree of accuracy of the technique (error <1·5 %). This SNP genotyping and data analysis method allowed us to distinguish among very close allele ratios, and it can also be efficiently employed for the analysis of higher ploidy levels. Moreover, the segregations observed with this technique have allowed us to identify heterozygous null alleles in one parent for some of the markers. Diploid progeny genotyping confirmed these conclusions for null alleles in Pink pummelo (Ollitrault et al., 2012a).

PCR drift can affect allelic configuration inference in natural polyploid germplasm

Interpretation of relative allele dosage for markers based on relative PCR product intensities has been reported for various plants (Buteler et al., 1999; Julier et al., 2003; Landergott et al., 2006; Martins et al., 2009) and animals (McQuown et al., 2002). The limits of evaluation of such direct allele doses are associated with PCR selection caused by differential primer affinity and PCR drift resulting from random events during early cycles of PCR (Wagner et al., 1994).

In this study, such PCR drift has been observed for some markers, displaying incongruence between the linear regressions established from the mixed DNA pools and the pure sample. However, the correlations between allele signals and allele doses in the DNA pools remained high for these markers. Therefore, as linear regression appears to offer a good approximation of the doses/relative signal relationship, a control with two dosage points should be sufficient to establish a function that correlates both parameters.

Heterozygous diploid genotypes are suitable for determining the 1:1 ratios that are used as a baseline for calculations of allele quantification in the other heterozygous genotype. In the analysis of citrus triploid progeny, the location of different clusters relative to the heterozygous diploid parent allowed us to assign the alternative theoretical triploid heterozygous allelic configuration.

The situation is much more complicated when analysing polyploid germplasm of unknown origin. Indeed, the variability in the flanking regions of the SNPs that were studied (where the primers were defined) should result in different levels of relative PCR competition and, therefore, should avoid proper allele dose identification from relative x/y signals. This is inherent in all PCR genotyping methods. Perhaps, as suggested by Landergott et al. (2006) for the MAC-PR method, the KASPar assay may be very useful for determining the allelic configuration within crossing families, but it would not be generally applicable for estimating allelic dosage in polyploid germplasm without previous verification of the stability of relative allele amplification. An approach to limit the PCR drift associated with variations in the flanking area of the studied SNPs should be to select SNPs flanked by conserved sequences. Such information is available in SNP mining studies where large discovery panels are used, while there is generally no information on flanking sequences of microsatellite markers. This should be an important advantage of using SNPs rather than microsatellite markers for assignment of allelic configuration in polyploids.

Potential of KASPar for semi-quantitative estimation of allele-specific expression analysis or allelic frequency estimation in DNA extracted from pools

Many genetic variants resulting in phenotypic differences are mediated through changes in gene expression. Variation in gene expression can be due to polymorphisms either at the gene locus (cis) or in other genes that influence gene expression (trans) or cis/trans interactions (Rockman and Kruglyak, 2006). Allele-specific expression (ASE) studies have introduced a creative method to uncover the respective contributions of cis- and trans-regulatory variation (Ronald et al., 2005; Main et al., 2009). Allelic imbalance in non-imprinted genes has been shown to be common in humans, maize and arabidopsis (Lo et al., 2003; Guo et al., 2004; Zhang and Borevitz, 2009). Moreover, ASE analysis should enable the integration of potentially differential allelic functionality in association models between gene expression and phenotype. Therefore, gene expression analysis is a critical step for better understanding of genotype–phenotype relationships.

Analysis of allele-specific expression in relation to genomic structure requires the assessment of DNA and RNA allele dosage. This can be done using different methods: northern (Guo et al., 1996), RNA-FISH (Herzing et al., 2002), SNP-specific array-based (Bjornsson et al., 2008), Solexa (Main et al., 2009) or RNA-seq (Rozowsky et al., 2011).

Furthermore, the estimation of allelic frequencies on pooled DNA is of great interest both in ecological studies of plants (Ritland, 2002), animals (Shaw et al., 1998; Coop et al., 2010; Grant, 2010) or micro-organisms (Brauer et al., 2006; Wenger et al., 2010), and in bulk segregant analysis to locate genes involved in phenotypic variation (Quarrie et al., 1999; Tabor et al., 2000; Yang and Fann, 2007).

The high correlation coefficient values between relative allele dosage and SNP allele signals obtained with the KASPar technique, and the ability of this technique to distinguish between close relative allele dosages at the DNA level, has been demonstrated in this study. Moreover, we were able to detect a 0·1 allele frequency within DNA pools. This technique is therefore a promising method for performing semi-quantitative analysis of relative allele-specific expression by analysing cDNA compared with genomic DNA, to complement global gene expression studies performed by real-time PCR. The KASPar technique may also be useful for allele frequency estimation in populations from DNA pools as mentioned before. For such studies, it should be interesting to extend the range of relative allele dosages to estimate the lowest differences distinguishable with this technique.

Application for citrus genetics and breeding

Triploid citrus breeding is one of the most efficient techniques for the production of seedless mandarins (Ollitrault et al., 2008; Aleza et al., 2010, 2012c, d), and tetraploid rootstocks are promising tools that enable plants to adapt to various abiotic stresses (Saleh et al., 2008; Dambier et al., 2011). Triploid populations in citrus can arise from 2x × 2x crosses or from interploid crosses. Discriminating between different types of heterozygotes within triploid progeny is especially useful for population genetic structure studies and marker–trait association analysis.

Knowing the allelic configuration in triploid and tetraploid progeny is also necessary to identify the mechanism of 2n gamete formation. The maternal HR values of <50 % obtained in this study, which were estimated from a ‘Clementine × Nadorcott’ progeny for nine markers (CiC0868-01, CiC1380-05, CiC1757-02, CiC2840-01, CiC3440-07, CiC4581-01, CiC5089-06, CiC5785-01 and CiC6278-01), confirm the conclusion of Luro et al. (2001) that the 2n gamete in Clementine arose from SDR, as in the Fortune mandarin (Cuenca et al., 2011), while Chen et al. (2008) proposed FDR for sweet orange. Moreover, this study allowed us to identify several centromeric markers that should be very useful for further analyses of the origin of 2n gametes in different cultivars and genotypes, as was done for potatoes (Douches and Quiros, 1988; Werner et al., 1992).

Most tetraploid citrus germplasm arose from chromosome duplication of nucellar cells (Aleza et al., 2011) or were obtained by bud chemical treatment (Aleza et al., 2009b) of diploid genotypes. These tetraploids are, therefore, doubled diploids with the same aabb genomic structure at each heterozygous locus (ab) of the parental diploid line. For such tetraploids, the parental restitution (PR) of the heterozygosity to the diploid gamete depends on preferential pairing between chromosomes. In the case of total preferential pairing (disomic segregation), parental heterozygosity is transferred to all gametes (PR = 100 %). In the case of total random pairing (tetrasomic segregation), the PR ranged from 55 to 66 %, depending on the double reduction frequency (Marsden et al., 1987). In this study, the PR results for the tetraploid (doubled diploid) Clementine ranged from 52·94 % for the CiC4581-01 marker to 69·41 % for the CiC5089-06 marker, which is in agreement with the expected PR values under tetrasomic segregation (Kamiri et al., 2011).

In the case of triploid and tetraploids obtained by somatic hybridization (Dambier et al., 2011; Grosser and Gmitter, 2011), the assignment of allelic configuration will be useful for revealing genome regions acquired from each parent, as well as potential chromosome fragment elimination or duplication.

Conclusions

This work demonstrates that the KASPar SNP genotyping technique, combined with the cluster analysis method we proposed, enables the efficient assignment of heterozygous allelic configuration within polyploid populations. This method is accurate, simple and cost-effective. It has been successfully applied to two citrus triploid populations arising from 2n gametes and interploid crosses. Moreover, correlation studies, cluster analysis and ANOVA support the usefulness of this method for performing relative quantitative studies, such as relative allele-specific expression analysis or, eventually, bulk segregant analysis.

SUPPLEMENTARY DATA

Supplementary data are available online at www.aob.oxfordjournals.org and consist of Table S1: SNP information and genotyping of the parents used in this study.