Identification of genome-wide single nucleotide polymorphisms in allopolyploid crop Brassica napus

Background Single nucleotide polymorphisms (SNPs) are the most common type of genetic variation. Identification of large numbers of SNPs is helpful for genetic diversity analysis, map-based cloning, genome-wide association analyses and marker-assisted breeding. Recently, identifying genome-wide SNPs in allopolyploid Brassica napus (rapeseed, canola) by resequencing many accessions has become feasible, due to the availability of reference genomes of Brassica rapa (2n = AA) and Brassica oleracea (2n = CC), which are the progenitor species of B. napus (2n = AACC). Although many SNPs in B. napus have been released, the objective in the present study was to produce a larger, more informative set of SNPs for large-scale and efficient genotypic screening. Hence, short-read genome sequencing was conducted on ten elite B. napus accessions for SNP discovery. A subset of these SNPs was randomly selected for sequence validation and for genotyping efficiency testing using the Illumina GoldenGate assay. Results A total of 892,536 bi-allelic SNPs were discovered throughout the B. napus genome. A total of 36,458 putative amino acid variants were located in 13,552 protein-coding genes, which were predicted to have enriched binding and catalytic activity as a result. Using the GoldenGate genotyping platform, 94 of 96 SNPs sampled could effectively distinguish genotypes of 130 lines from two mapping populations, with an average call rate of 92%. Conclusions Despite the polyploid nature of B. napus, nearly 900,000 simple SNPs were identified by whole genome resequencing. These SNPs were predicted to be effective in high-throughput genotyping assays (51% polymorphic SNPs, 92% average call rate using the GoldenGate assay, leading to an estimated >450 000 useful SNPs). Hence, the development of a much larger genotyping array of informative SNPs is feasible. SNPs identified in this study to cause non-synonymous amino acid substitutions can also be utilized to directly identify causal genes in association studies.


Background
It is estimated that approximately 70% of angiosperms have experienced one or more chromosome doubling events during their evolutionary history [1]. Many crop species are also polyploid, including Brassica napus (rapeseed, canola), Triticum aestivum (wheat), Solanum tuberosum (potato), Gossypium hirsutum (cotton), Avena sativa (oat) and Saccharum officinarum (sugarcane). Brassica napus is an allopolyploid species originating from interspecific hybridization between diploid progenitors Brassica oleracea and Brassica rapa [2], which are themselves derived from ancient polyploidy events resulting in genome triplication 13 to 17 million years ago [3,4]. These recent and ancient polyploidy events resulted in numerous duplicated segments and homoeologous regions within the genome of B. napus [5]. Hence, discriminating between a) two homologous sequences and b) two nearly-identical homoeologous sequences is complex and difficult in B. napus [6].
SNPs (single-nucleotide polymorphisms) are singlenucleotide substitutions of one base for another in DNA sequences. SNPs are quite abundant throughout the entire genome of most organisms and every SNP in low copy DNA is a potentially useful marker [7]. SNP markers have been applied in studies of genetic variation, construction of genetic maps, population structure analysis, association genetics, map-based gene isolation, and other plant breeding applications [8]. In contrast to traditional SNP detection techniques, detecting SNPs using next generation sequencing (NGS) technologies (such as Illumina sequencing, Roche 454 sequencing, Applied Biosystems SOLiD Sequencing and Helicos Biosciences Corporation Heliscope Sequencing) is high-throughput, low cost and high efficiency. Hence, next generation sequencing has been used in the development of thousands of molecular markers in many species [9][10][11], such as Oryza sativa (rice) [12,13], Helianthus annuus (sunflower) [14], Zea mays L. (maize) [15], Triticum aestivum (wheat) [16], Manihot esculenta (cassava) [17] and Arabidopsis thaliana (thale cress) [18]. Mass SNP information has already been successfully used for genome-wide association studies [9,[19][20][21], and SNP markers are increasingly becoming the optimal marker system.
In recent years, many SNPs have been discovered in B. napus [22,23] and B. oleracea [24]. However, these SNPs are inadequate for large-scale applications [23]. Trick et al. [23] used Solexa sequencing to generate approximately 20 million expressed sequence tags (ESTs) from two B. napus cultivars. They obtained 23,330-41,593 (two accessions) putative SNPs through alignment to a publicly available set of approximately 94,000 Brassica species unigenes [23]. However, 87.5-91.2% of the putative SNPS were 'hemi-SNPs' , amplifying two or more different genomic loci. In comparison with 'hemi-SNPs' ,'simple SNPs' are derived from allelic differences at a single genomic locus. Similarly, Bancroft et al. [25] used transcriptome sequencing to construct two B. napus linkage maps from 21,323 and 1,714 SNP markers, but discovered that the first map comprised 16,800 (78.8%) hemi-SNP types and only 4,124 (19.8%) simple SNP types, and that the second map comprised 1,266 hemi-SNPs and only 409 simple SNPs [25]. Later, Bus et al. [26] used eight different B. napus germplasm types to identify genome-wide restriction-site associated DNA (RAD) fragments, and obtained over 20,000 SNPs [26]. Hence, usability and availability of SNPs in B. napus is still limited, and development of a large set of simple SNP markers is highly desirable.
At present, both the A and C progenitor genomes (B. rapa and B. oleracea) of B. napus have been sequenced. The B. rapa A genome was released in 2011 [27], and the B. oleracea C genome has been sequenced by a collaboration between the OCRI (Oil Crops Research Institute of Chinese Academy of Agricultural Sciences) and other research communities and hence could also be used for BLAST analysis using the BRAD database [28] (http://brassicadb.org/brad/), although the C genome sequences have not yet been released. Based on these available reference sequences, large-scale identification of simple SNPs in B. napus could be implemented. The objective of the present study was to develop a set of genome-wide and evenly spaced SNPs through genome re-sequencing of ten B. napus varieties, and to validate the use of these SNPs on high-throughput genotyping platforms.

Resequencing, SNP calling and SNP verification
Samples for resequencing were chosen from B. napus accessions which were parents of reference mapping populations or elite cultivars. After removing low quality and contaminant sequences, a total of 1600 million (M) paired-end reads of 75-bp or 100-bp read length (about 126 Gb total) were retained. The sequencing depth for each variety averaged 10.7 ×, ranging from 5.3 × to 37.5 × depth ( Table 1). All sequence reads were aligned against the reference B. rapa [27] and B. oleracea sequences using SOAP2. Prior to alignment, all reference sequences were masked for repetitive elements using a comprehensive Brassica repetitive element database (unpublished). This step served to minimize the data set to low-copy DNA. SNPs were then extracted from SOAP2 alignments after a filtering scheme that (i) excluded 521 million reads with redundant hits to the reference genomes, retaining 585 million reads uniquely matched to the reference sequences; (ii) excluded 78 million SNPs supported by less than four reads in each line (at least four reads support the genotype of a single line; the SNP error rate was 1/10000); (iii) excluded 6,331,887 SNPs that were heterozygous in at least in one individual and (iv) excluded 7,224,690 SNPs with a minor allele frequency more than 0.2. After applying the filters described above, the total number of remaining SNPs was 892,803, including 892,536 bi-allelic SNPs and 267 tri-allelic SNPs. The SNP flanking sequences and the mutated sites have been deposited in NCBI dbSNP databases (ss647660101-ss657954846) [29]. Transition-type SNPs comprised 57.5% of the total SNPs, while transversion-type SNPs accounted for 42.5%. The transition/transversion SNP ratio was 1.35. A total of 108,270 A/T type SNPs were detected. There were 79,533 A/T SNPs with a 'G' or 'T' base in the 3 rd nucleotide upstream, or a 'C' or ' A' base in the 3 rd nucleotide downstream: these SNPs can be efficiently genotyped using the allele-specific PCR method [30]. The largest number of SNPs were identified between PY-1 and 73290 (385,432 SNPs, 43.2%), while the smallest number of SNPs were identified between XY15 and 09CB01 (30,950 SNPs, 3.5%) ( Table 2). On average, there were 210,516 SNPs between any two of the ten accessions. A total of 758,454 SNPs were distributed on the pseudochromosomes of Brassica rapa and Brassica oleracea. On average, 119 SNPs were found per 100 kb in the A genome, and 89 SNPs were found per 100 kb in the C genome. SNP frequency in the A genome was about 1.3fold SNP frequency in the C genome (Table 3). Figure 1 shows the distribution of SNPs in the A and C genomes.
To test empirically the quality of the SNPs identified between the ten B. napus accessions, ten random loci containing SNPs were validated by sequencing: PCR primers were designed from reference sequences and used to amplify DNA fragments from the ten B. napus accessions. Of 100 high-quality reads that aligned to reference sequence, 93 contained SNPs that matched the predicted results. Hence, the predicted false positive rate of SNP discovery was 7%.

Non-synonymous SNP identification and enrichment analysis
Non-synonymous SNPs that lead to an amino acid change in the protein product are of major interest. Non-synonymous variations are more likely to lead to functional mutations ('drivers') which may further affect phenotype. A total of 36,458 non-synonymous SNPs were identified and were located in 13,552 predicted genes. A total of 479 non-synonymous SNPs transformed stop codons to amino acid codons, whereas 505 non-synonymous SNPs transformed amino acid codons to stop codons. GO enrichment analysis predicted that the genes containing non-synonymous SNPs were involved in binding and catalytic activity more often than predicted by chance (p = 2.52E-13, Figure 2).

High-throughput genotyping testing via the GoldenGate Array
A total of 110 candidate SNPs were submitted to Illumina Inc. to evaluate the designability of these SNPs. The rank score ranged from 0.56 to 0.99, with an average of 0.89. Fourteen SNPs with rank score 0.85 or lower were excluded, and the remaining SNPs were included in the OPA (oligonucleotide pool assay). The 96-plex OPA included 42 A-genome SNPs and 54 C-genome SNPs that were evenly distributed genome-wide.
Two mapping populations (DH and F 2 ) comprising a total of 130 B. napus lines were genotyped using the GoldenGate assay. Clustering of Cy3 and Cy5 normalized intensities in a polar coordinate plot was used to infer the SNP genotypes. Genotype calls for all SNPs could be grouped into one or more groups depending on whether a SNP site was monomorphic or polymorphic. There were 49 SNPs which were polymorphic in at least one population. In the DH population, there were 32 polymorphic SNPs between the two parents. Six of these 32 SNPs segregated in a 1:1 ratio (p > 0.05, χ 2 test). In the F 2 population, there were 44 polymorphic SNPs between the two parents. Twelve of these 44 SNPs segregated in a 1:2:1 ratio (p > 0.05, χ 2 test).
In order to evaluate the reproducibility of the SNPs, three repeats of one sample were conducted for all SNP assays. Of the 96 SNPs, only one SNP showed variable results across the three repetitions. The average call rate was 92% for the 130 B. napus samples. In order to evaluate the reliability of the developed SNPs, the two populations were mixed and clustered with the Geno-meStudio Data Analysis Software. Most SNPs were still clustered into three groups ( Figure 3). There were only 3 SNPs which were clustered into four groups ( Figure 4).

Discussion
Large polyploid genomes such as B. napus and wheat present a challenge for SNP discovery because of the presence of multiple homoeologous sequences [25,31]. Allelic variants need to be distinguished from non-allelic (paralog) variants (nucleotide polymorphisms between paralogs/homoeologs or between the A and C genomes) which present as false SNPs. In addition, the repetitive nature of the polyploid genomes has been one of the major obstacles to SNP discovery. In this study, three conditions were utilized to identify putative "simple SNPs". Firstly, low-copy DNA regions were identified by uniquely-aligned reads that were excluded from repetitive DNA regions and mapped to only one place in the B. rapa and B. oleracea reference sequences. Sequenced reads were classified into three categories: 'uniquely aligned' , 'repeatedly aligned' and 'unaligned'. Here, the 'repeatedly aligned' category represents duplicated loci across the allopolyploid B. napus genome. The 'unaligned' category may be partially derived from novel sequences induced by such events as genome rearrangements or transposon activity. Hence, only the uniquely aligned single-hit reads were selected from the aligned results for further analysis. Secondly, only homozygous loci were selected for subsequent analysis in each individual. Heterozygous loci could be unambiguously attributed to polymorphism between homoeologous chromosomes rather than to allelic heterozygosity. Thirdly, only reads with depth ≥ 4 were used for SNP discovery, in order to exclude SNPs generated by sequencing error. Generally speaking, minimum recommended read depth is ≥ 3 per genotype [8].
A total of 892,803 SNP polymorphisms were identified among the ten accessions of B. napus, using a stringent filtering approach favouring high quality SNPs over exhaustive SNP sampling to provide a resource of immediate value for crop improvement. Therefore, the actual frequency of SNP polymorphisms between these accessions is likely to have been underestimated, due to the stringent filtering methods used and due to exclusion of duplicated DNA.
In the present study, approximately 55% of SNPs were distributed on the A genome, and 45% of SNPs were distributed on the C genome. However, Bancroft et al. [32] identified 15559 SNPs on the A genome and 5675 SNPs on the C genome [32]: the bias towards A-genome SNPs was far more significant than in the present study. The genetic distance between the Ningyou7 and Tapidor C genomes is likely narrow, although these two genotypes were selected on the basis of their genetic dissimilarity, contrasting trait characteristics and different cultivation ranges [25]. However, the results of the present study agree that the A genome appears more variable than the C genome. Uneven distribution of SNPs throughout the genome is common, and has also been observed in Brassica relative Arabidopsis thaliana [33]. A total of 36,458 SNPs predicted to cause non-synonymous amino acid substitutions were identified in this study. These SNPs may represent causal genetic variation contributing to phenotype variation. Using this SNP set to perform genome-wide association in B. napus would be more efficient than using a general SNP set to identify causal gene mutations. GO analysis in the present study suggested that the genes predicted to contain nonsynonymous SNPs were more commonly associated with binding and catalytic activity than with other functionality. This may suggest that proteins with the function of binding and catalytic activity may play a significant role in adaptive evolution.
A 96-SNP GoldenGate assay can be used successfully for SNP genotyping in B. napus, despite the high number of paralogous sequences in this polyploid species. Figure 3 shows an example of a putative simple SNP (SNP RP13) in the two mapping populations. If the SNP was a hemi-SNP with one homozygous locus, the genotyped samples of mixed populations would cluster into four or more groups [34] (Figure 4). In the present study, when the results from the two populations were pooled for SNP chip analysis, only 3/96 (3%) of SNPs showed four genotype clusters, suggesting these were hemi-SNPs with genotype-specific heterozygosity in the additional amplified region. However, validating these SNPs over a wider range of accessions would be valuable in determining what proportion of SNPs are simple SNPs, and what proportion are hemi-SNPs.   and 'Zhongshuang11' × '73290' populations. The normalized R (y axis) is the normalized sum of intensities of the two channels (Cy3 and Cy5) and normalized theta (x-axis) is ((2/π)Tan-1 (Cy5/Cy3)) where a normalized theta value nearest 0 is a homozygote for allele A and a theta value nearest 1 is homozygous for allele B [42].
Although 892,803 SNPs have been developed, there are still some limitations to this work. Ten accessions was a productive number for the SNP discovery. However, increasing the number of resequenced accessions will enhance efficient, polymorphic SNP discovery. As well, eight of the accessions used were semi-winter-type B. napus, and therefore the effect of these SNPs in spring-type and winter-type B. napus needs to be further validated. Ten SNPs were randomly selected for sequencing and validation in ten lines: 97% of the sequenced SNP loci matched the prediction. The high validation ratio may have resulted from the stringent filtering conditions. Ninety-six SNPs were tested on a genotyping platform, and most polymorphic SNPs showed segregation distortion. The segregation distortion may have resulted from selection bias for particular alleles during the process of population construction (e.g. microspore culture to produce the DH population). It is also possible that for some of these SNPs genotyping using the GoldenGate assay resulted in theincorrect grouping of multiple genotype clusters together (e.g. AAAB and AABB), which would result in distorted segregation ratios. However, multi-locus genotypes are usually clearly identifiable in Genome Studio by the presence of additional separate clusters, so it is more likely that the segregation distortion observed was due to selective pressure for one or the other parental allele under population growth conditions. Future work could include validation of the genomic location of these SNPs by designing and using arrays in large mapping populations originating from diverse B. napus parent genotypes.

Conclusions
A total of 892,536 bi-allelic SNP markers were developed for allopolyploid B. napus. The average number of SNPs per 100 kb was 119 and 89 in the A genome and C genome respectively. Transition-type SNPs accounted for 57.5% of all SNPs, and transversion-type SNPs accounted for 42.5%. A subset of developed SNPs was tested through sequencing of PCR amplification products and the GoldenGate genotyping technique, and it is predicted that the majority of the SNPs identified in this study (>450,000) can be applied in the development of much larger arrays of informative SNPs, such as Infinium II assays.

Plant materials
A total of ten representative accessions were chosen for SNP marker development. These comprised eight semi-winter type accessions: 'Zhongshuang11' , '73290' , '08-806-2' , '09CB01' , 'Xiangyou15' , '09CB03' , 'PY-1' and 'PY-2'; one winter-type accession, 'Tapidor'; and one spring-type accession, 'Westar'. Two B. napus populations were used to validate the SNPs for high-throughput genotyping via GoldenGate Array. The first set comprised 92 lines of a DH population generated from crossing parents 'zy036' and '51070' [35], and the second set comprised 250 lines of a RIL population generated from crossing parents 'zhongshuang11' and '73290'. A total of 50 lines from the DH population and 80 lines from the F 2 population were genotyped using the GoldenGate Array.

Genomic DNA preparation and sequencing
Seeds of ten B. napus accessions were germinated at 25°C on MS medium in a dark chamber. After five days, etiolated seedlings were collected for genomic DNA extraction using a standard CTAB (cetyl trimethylammonium bromide) protocol [36]. Sequencing libraries were constructed according to the manufacturers' instructions (Illumina). Short reads were generated by applying the base-calling pipeline Solexa Pipeline-0.3 (Illumina). The Illumina sequence data have been deposited in the NCBI Sequence Read Archive (GenBank: SRA057227).

Sequence analysis, SNP detection and verification
The Brassica rapa (Brassica rapa Release v1.01) [27] and Brassica oleracea assembled scaffolds (unpublished) were combined and repeat sequences were masked using RepeatMasker software. Short Oligonucleotide Alignment Program 2 (SOAP2) [37] was used to map raw pair-end reads on to reference sequences with suitable parameters for sequence similarity (−v 5), seed size (−l 32) and minimal alignment length (−s 40). On the basis of the mapping results, reads were classified into three categories: 'uniquely aligned' , 'repeatedly aligned' and 'unaligned'. SNPs detection comprised four consecutive steps. Firstly, 'uniquely aligned' reads (only one hit to the reference genome sequences) were selected, in order to avoid paralogue interference. Secondly, genotype data which had more than four reads of each line mapped to the reference genome was extracted. Thirdly, any alleles heterozygous within an accession were eliminated. Fourthly, SNPs where the less common allele was present in at least two accessions were selected. Custom perl scripts were used to perform these steps according to the published documents [8,25]. Finally, 10 SNPs were randomly selected throughout the genome and SNPflanking PCR primers (Table 4) were designed using Primer3plus [38] for fragment amplification. Primers fulfilled design parameters of 18-22 bp length, 50%-60% GC content, and 58-62°C Tm. Genomic DNA (50 ng) was used as the PCR template for amplification with the following reagents and conditions: 1× Pfu buffer, 0.2 mM dNTP mix, 1 μM primer mix, 2.5 units of Taq, and 0.2 mU of Pfu for 35 cycles under the appropriate annealing temperatures using a DNA Engine Peltier Thermal Cycler. The amplified products were sequenced using an ABI3730 sequencer by BGI.

SNP annotation and enrichment analysis
The localization of SNPs in coding regions was based on annotation of gene models as provided by the Brassica Genome Database (http://www.ocri-genomics.org/bolbase/). Gene families were annotated using hmmer3 software [39] via the Pfam gene family database (Pfam26.0) [40]. Enrichment analysis for the supplied gene list was carried out based on the algorithm presented by GOstat [41], with the whole set of genes from B. rapa and B. oleracea as the background. All genes with non-synomyous SNPs were extracted via custom Perl script. The GO annotations of these genes were extracted from the Brassica Genome Database. The p-value was approximated by Pearson's chi-squared test. Fisher's exact test was used when the expected value of any count was below 5.

High-throughput genotyping via GoldenGate Array
A total of 110 SNPs were randomly selected from the identified SNP set. SNP-containing sequences were extracted and screened with RepeatMasker software (http:// www.repeatmasker.org/) using the repeat databases. Repeats in the SNP-containing sequences were replaced with lowercase letters prior to submission to Illumina Inc. to undergo a preliminary design phase of the custom oligo pool assay (OPA), which contains the allele-specific oligoes and locus-specific oligoes for all SNPs included in the assay. A designability rank score was given to each SNP by Illumina. Scores ranged from 0 to 1.0, where a rank score of <0.4 indicated a low success rate, 0.4 to 0.6 indicated a moderate success rate, and >0.6 indicated a high success rate for the conversion of a SNP into a successful GoldenGate assay. The GoldenGate assay was performed according to the manufacturer's protocol and as described in Fan et al. [42]. Genomic DNA was extracted from leaf tissues of 130 individuals. A NanoDrop spectrophotometer was used to ascertain that DNA quality and quantity met the requirements for the genotyping assay. Genotyping was performed using the Illumina GoldenGate Assay platform [43] and the resulting data were visualized and analyzed with the GenomeStudio Data Analysis Software package (1.0.2.20706, Illumina Inc.). Samples with a call rate lower than 0.8 and loci showing poor clustering were excluded. Accepted SNPs were manually re-clustered, to correct errors in allele calling due to inappropriate cluster identification.