Molecular genetic markers describe genetic variations and provide a link between observed phenotypes and the underlying genotype [1,2,3]. The development of high-throughput methods for the detection of single nucleotide polymorphisms (SNPs) has led to a revolution in their use as molecular markers [4,5,6,7]. SNPs may be considered the ultimate genetic marker as they represent the finest resolution of a DNA sequence, are generally abundant in populations and have a low mutation rate [8]. The principal challenge in SNP discovery remains the discrimination between true genetic polymorphisms and the often more abundant sequence or mapping errors. SNP discovery is further confounded in polyploid species where multiple related genomes are present within each nucleus. The identification of sequence errors can be based on three methods: sequence quality score, redundancy of the polymorphism in a sequence alignment and presence of multiple haplotypes at a locus [9,10]. SNP redundancy provides an effective means for estimating confidence in the validity of SNPs independently of sequence quality scores and has been demonstrated to be an accurate method for SNP discovery in a range of species [11,12,13].

Many plant genomes are large and complex. The bread wheat (Triticum aestivum) genome is 17 Gbp in size, around 6 times larger than the human genome [14], consists of 75%–90% repeats [15,16] and is hexaploid, containing of the A, B and D genomes, each with 7 homoeologous chromosomes. The presence of multiple genomes, large size and abundant repeats make any genetic and genomic analysis a challenge [17]. The recent shotgun sequence assembly of isolated chromosome arms for the group 7 chromosomes provides a reference for SNP discovery across these chromosomes [18,19,20]. When combined with the recently produced whole genome shotgun sequence data for several wheat cultivars [21], there is a potential to identify large numbers of cultivar specific SNPs.

The rapidly expanding genome datasets, driven by advances in second generation DNA sequencing, present a challenge for their management and application [22]. The majority of SNP discovery software is designed for human or simple bacterial genomes, and they are not well adapted for polyploid plant genomes, especially crop genomes which are often highly homozygous [11,23,24,25]. Because of this, we have established a novel method for SNP discovery from complex genomes called SGSautoSNP, extended from original concepts developed in autoSNP, SNPServer and autoSNPdb [11,12,26]. Rather that attempting to identify all possible SNPs across a genome, SGSautoSNP aims to identify as many highly confident SNPs as possible with the acknowledgement that not all biologically present SNPs will be identified. Here, we present the application of SGSautoSNP to wheat chromosomes 7A, 7B and 7D to identify more than 800,000 SNPs with an accuracy of greater than 93%, and present these polymorphisms within a GBrowse genome viewer at the wheatgenome.info web site [27].

We have developed a novel pipeline for the discovery of SNP polymorphisms in complex genomes. The SGSautoSNP (Second-Generation Sequencing AutoSNP) pipeline calls single nucleotide polymorphisms (SNPs) between different individuals using Illumina paired read data aligned to a reference. SGSautoSNP uses BAM (Binary Alignment/Map) format in order to save memory and space. These SNPs can be viewed using a broad range of visualization tools using GFF3, VCF and Flapjack output files. There is often a requirement to generate a consensus sequence based on the reads mapped to the reference and so SGSautoSNP can generate a consensus sequence as well as marker design files Illumina GoldenGate of Infinium assay designs.

There are many SNP discovery tools available. The main difference between our approach and most other SNP callers such as ACCUSA [28], AGSNP [29], NGS-SNP [30] and Atlas-SNP2 [31] is that the SGSautoSNP method does not consider the reference for SNP discovery. Instead, the reference is used to assemble the reads, and SNPs are then called between these assembled reads. Another difference is that other SNP callers are designed for SNP discovery from heterozygous populations. However, crop plants are frequently homozygous. In SGSautoSNP, mismapped reads produce a heterozygous genotype call at a locus, allowing their distinction from true homozygous SNPs. The SGSautoSNP method does not consider read quality score because these scores are not very reliable, with erroneous nucleotide calls often having high quality scores.

We demonstrate the potential of SGSautoSNP by searching for SNPs between four wheat cultivars using whole genome shotgun paired read Illumina sequence data. The reference consisted of bread wheat chromosome arm shotgun assemblies representing chromosomes 7A, 7B and 7D, as well as 4AL [32]. All three of the group 7 chromosome homoelogs were included in the mapping reference to ensure reads mapped to their correct homoeologous location. In the absence of one of the homoeologs, cultivar specific reads from the missing homoeolog would likely map to one of the other homoeologous genomes, confounding SNP discovery. An assembly from chromosome arm 4AL was included as this arm contains a reciprocal translocation with 7BS [18]. If 4AL was absent from the reference, 4AL specific reads which correspond to the 7BS translocation may map to the orthologous region on 7AS or 7BS, again confounding SNP discovery.

Cultivar specific reads were mapped to the reference sequences using SOAP [33]. The −r 0 option was applied which removes reads where they match multiple positions equally well. This option aims to increase SNP calling accuracy by ignoring read pairs that cannot be accurately positioned on the reference. Similarly, only reads that mapped as a pair were used for SNP discovery. Due to the short length of the reads, one read could match at many positions, but two reads separated by a gap of defined insert size provides a greater confidence of specific and accurate read mapping. The calling of SNPs between reads aligned to a reference while ignoring the reference allele allows this pipeline to be applied to accurately call SNPs between individuals using a reference from a divergent species. While this pipeline does not attempt to call all biological SNPs, the very large numbers of SNPs identified are valuable for genetic studies and the association of traits with candidate agronomic genes.

Between 661,600,411 and 899,700,085 genome sequence read pairs where generated for each of the four wheat cultivars (Table 1). Of these, between 4.70% and 3.10% mapped to the group 7/4AL reference as read pairs. As the reference is estimated to cover 14% of the complete genome, the number of mapped reads is less than predicted. This is likely due to many read pairs mapping to multiple locations in this highly repetitive genome and subsequently being ignored due to the SOAP –r 0 option.

SGSautoSNP identified a total of 881,289 SNPs across the group 7 chromosomes. These consisted of 68% transitions and 32% transversions. This bias in transition/transversion ratio is commonly observed in SNP discovery and reflects the high degree of methlyl C to U mutation in genomes [34]. It may be expected that the bread wheat genome is highly methylated due to the two rounds of polyploidy and high repeat content. The observed transition/transversion bias provides a level of confidence in SNP prediction accuracy since erroneously called SNPs caused by sequence read errors or mismapping would be unlikely to display such a bias.

Validating individual SNPs in a hexaploid species is a challenge as the amplification of loci requires the design of homoeolog specific PCR primers. Of 40 SNPs selected for validation, 12 did not produce clean PCR amplification products or Sanger sequence. This reflects inefficiency in validation rather than SNP calling errors and so these SNPs were ignored. Of the 28 SNPs that did produce clean Sanger sequence data, 26 (93%) produced the expected genotype. One SNPs was homozygous across cultivars and not a true SNP, while one appeared to be heterozygous, suggesting a SNP between the homoeologous genomes rather than between cultivars.

All predicted SNPs have been included in a public wheat genome GBrowse database hosted at the wheatgenome.info web site [27]. This resource represents a substantial expansion over other recent wheat SNP discovery projects and provides a much greater density of SNPs than recently described methods for wheat. Allen et al. [35] recently identified 14,078 putative SNPs in 6,255 distinct reference sequences with Illumina GAIIx data from wheat lines Avalon, Cadenza, Rialto, Savannah and Recital. The validation rate from a subset of 1,659 was 67%. In a separate project, Lai et al. [36] identified a total of 38,928 candidate SNPs from bread wheat Roche 454 transcriptome data with an accuracy of 78% and presented these in an online database [37]. A pipeline package called AGSNP has also been applied to identify SNPs between two accessions of one of the diploid progenitors of bread wheat, Aegilops tauschii. Roche 454 sequencing of A. tauschii accession AL8/78 was combined with Applied Biosystems SOLiD sequencing of genomic DNA and cDNA from A. tauschii accession AS75 using AGSNP to identify a total of 497,118 candidate A. tauschii SNPs [29]. Currently, genome wide identification of hexaploid bread wheat SNPs using our pipeline is limited by the lack of publically available chromosome sequences. However, the identification of 881,289 SNPs across the group 7 chromosomes suggests that genome wide discovery would identify a total of more than 6 million SNPs.

The SNPs identified using SGSautoSNP can be used for genotyping by sequencing by low coverage skim sequencing of segregating wheat populations and calling genotypes where the low coverage sequence data aligns to a predicted polymorphic position. The successful application of the SGSautoSNP method to hexaploid wheat demonstrates that this approach should work for SNP discovery in other large and complex genomes.

We have developed a pipeline for the identification of single nucleotide polymorphisms in complex genomes using second generation DNA sequence data. The application of this pipeline to wheat identified more than 800,000 SNPs across the three homoeologous group 7 chromosomes 7A, 7B and 7D with a validation rate of greater than 93%. This provides an unprecedented resource for wheat genomics and diversity analysis and demonstrates the potential to expand this application for SNP discovery in other large and complex genomes. SGSautoSNP is freely available on request for academic use.