After completion of one single GS 20 run, a first bioinformatics analysis was performed using the GS 20 software (i.e. “on-rig” software). A total of 754,199 reads (“totalRawWells”) were obtained. The number of reads after the first filtering for Key sequences (“totalKeyPass”) was 739,042. Of these, 399,252 GS 20 raw sequencing reads remained after the final filtering by the GS 20 software. This number of sequence reads is higher than the specifications of the GS 20 but in line with other runs we performed earlier (data not shown) as well as results reported by others [25]. Their average read length was 103 nt (Table 1).

Further bioinformatics analysis took place “off-rig” (i.e. on a separate server) using the CRoPS pipeline (Fig. 1). The GS 20 raw sequencing reads were trimmed (adapter removal) and 383,566 (96%) sequences remained (i.e. sequences for which a significant match with a tagged AFLP primer was found). The reasons for rejection of the remaining 15,686 (4%) reads (classified as faulty reads) were three-fold: 1) AFLP adapter not found, 2) conflict in adapter position (concatamers), and 3) sample identification tag conflict, i.e. a sequence with sample identification tag of one sample at one end and with sample identification tag of the second sample at the other end of the sequence read (so called “mixed fragments”, see further below).

Using the TIGR Gene Indices clustering tool (TGICL) [26], the remaining 383,566 sequences were clustered and assembled. Among these were two very large clusters (119,717 and 23,608 reads respectively) containing heavily repeated sequences. Homology searches using the Basic Local Alignment Search Tool (BLAST) revealed that the sequences within these two clusters were in fact chloroplast sequences. These two clusters were excluded from further processing. Subsequently, sequences within the remaining clusters were assembled into multiple sequence alignments (MSAs) (Table 1). In addition to the 18,989 MSAs containing 211,100 sequence reads, 29,141 (7.6%) singletons were found, i.e. sequences that were not assembled into an MSA.

Finally, SNPs were mined between the reads contained in an MSA. Parameters for SNP mining were set to include only SNPs for which both alleles were observed at least twice and SNPs not being part of homopolymers larger than 3 bases. The threshold for minimal distance to a neighboring SNP was initially set at one base, i.e. all SNPs were selected irrespective of their distance to (a) neighboring SNP(s). In addition, and importantly, SNPs were mined according to sample origin, i.e. only SNPs “segregating homozygously” between the two maize lines were included. By doing so, a strong filter was created to select against “false” SNPs resulting from alignment of highly homologous duplicated sequences as opposed to genuine SNPs derived from single-copy sequences in the sequenced genome fraction (Fig. 2). As a result, 1262 putative SNPs, including 37 putative indels were mined (Table 1).

To investigate the relationship between the number of putative SNPs and SNP mining parameters, SNPs were mined under different parameter settings regarding the minimal available sequence information flanking the target SNP (1, 2, 3, 4, 6 or 12 bases), and the minimal interval of flanking sequence that must be devoid of additional SNPs (1, 2, 3, 4, 6 or 12 bases). SNP mining was performed by varying these two parameters in all 36 (6 times 6) possible combinations, while keeping the other SNP mining parameters, including minimal representation of both alleles at least twice, homopolymer settings and segregation according to sample origin constant. As expected, the number of SNPs mined according to these more stringent criteria decreased to less than 50% (from 1262 to 591; Fig. 3). This selection of 591 SNPs was available for subsequent assay design.

Small-scale validation of putative SNPs was carried out using the SNPWave® technology [24]. From the selection of 591 putative SNPs mined according to the stringent criteria mentioned above (including a minimal of 12 bases flanking sequence surrounding the target SNP and minimal interval of 12 bases devoid of additional SNPs), 30 SNPs were randomly selected. Two 15-plex SNPWave assays were designed and tested using two parental lines and 94 recombinant inbred lines (RIL) offspring of the ISU (B73×Mo17) maize mapping population. For 23 out of 30 tested loci (77%) clear SNPWave reactions products were observed for both alleles, while for the remaining 7 loci one or both alleles were not observed (conversion failure). For all 23 SNP loci functioning properly in the SNPWave assay, the parental lines B73 and Mo17 were polymorphic and segregation was observed among RIL lines (Fig. 4), indicative of a high proportion of mined SNPs being derived from single-copy regions in the genome.

Total genomic DNA was isolated from leaf material of the two parental lines (i.e. B73 & Mo17) of the ISU mapping population (www.maizegdb.org), using a modified CTAB procedure [29]. These 2 parental lines were chosen to be able to validate and map the discovered SNPs in the ISU mapping population.

AFLP templates were prepared as described previously [20]. In short, 100–500 ng total genomic DNA was digested using 5 units HpaII and 2 units MseI for at least 1 hour at 37°C. After digestion, the mixture was heated at 80°C for 10 min. Next, AFLP adapter ligation using HpaII and MseI adapters was carried out for 3 hours at 37°C. The restriction-ligation (RL) mixture was subsequently diluted 10-fold with T₁₀E_0.1 and 5 µl diluted mix was used as a template in a selective pre-amplification step, the so-called +1/+1 pre-amplification. Primer sequences for the +1/+1 pre-amplification were 5′-GTAGACTGCGTACACGGA-3′ (HpaII site, including 1 selective nucleotide “A”) and 5′-GATGAGTCCTGAGTAAC-3′ (MseI site, including 1 selective nucleotide “C”). Twenty µl PCRs were performed containing 5 µl diluted RL mixture, 30 ng HpaII primer, 30 ng MseI primer, 0.2 mM dNTP, 0.4 U AmpliTaq® (Applied Biosystems) and 1× AmpliTaq buffer. PCR was performed for 20 cycles with the following cycle profile: 30 sec 94°C, 60 sec 56°C, 60 sec 72°C, followed by cooling down to 4°C.

The +1/+1 pre-amplification reaction was diluted 20-fold with T₁₀E_0.1, and used for the second selective amplification step, the so-called +1/+2 selective amplification. Primer sequences for the +1/+2 selective amplification were 5′-‘P-ACACGTAGACTGCGTACACGGA-3′ (HpaII site, including 1 selective nucleotide “A”) and 5′-‘P-ACACGATGAGTCCTGAGTAACT-3′ (MseI site, including 2 selective nucleotides “CT”) for sample B73. The four most 5′ bases of these primers serve as sample identification tag (KeyGene™ SeqTag technology). These 4-nt sample identification tags were selected from a collection of 4-nt sequences differing by at least 2 nt to exclude the possibility that a single nucleotide substitution error could cause incorrect assignment of the sequence to a sample. Similarly, primer sequences for the +1/+2 selective amplification of the Mo17 sample were 5′-‘P-AGCTGTAGACTGCGTACACGGA-3′ (HpaII site, including 1 selective nucleotide “A”) and 5′-‘P-AGCTGATGAGTCCTGAGTAACT-3′ (MseI site, including 2 selective nucleotides “CT”). Fifty µl PCRs were performed containing 5 µl diluted +1/+1 pre-amplification mixture, 75 ng HpaII primer, 75 ng MseI primer, 0.2 mM dNTP, 1 U AmpliTaq (Applied Biosystems) and 1× AmpliTaq buffer. PCR was performed for 30 cycles with the following cycle profile: 30 sec 94°C, 60 sec 56°C, 60 sec 72°C, followed by cooling down to 4°C.

Next, 100 µl of PCR products of each sample were purified using the QIAquick PCR Purification Kit (Qiagen). Concentrations of both samples were determined using the Nanodrop ND-1000 (Nanodrop Technologies), after which equal amounts of the two samples were pooled and further treated as one fragment library sample. This saves costs and prevents relying on physical compartmentalization to separate both samples. Furthermore this approach provides flexibility regarding processing multiple samples.

3.45 µg of the fragment library sample (i.e. pooled, purified and tagged AFLP products) were used as input for GS 20 library construction. The use of tagged and pooled PCR products, however, necessitated several adaptations in the published GS 20 library construction protocol [23]. First, no shearing was carried out. Second, the end-polishing step was omitted, and modified A and B adapters were used as follows: adapter A-upper strand: 5′-CCATCTCATCCCTGCGTGTCCCATCTGTTCCCTCCCTGTCTCAGT-3′, adapter A-lower strand: 5′-CTGAGACAGGGAGGGAACAGATGG-3′, adapter B-upper strand: 5′-BIO-TEG-CCTATCCCCTGTGTGCCTTGCCTATCCCCTGTTGCGTGTCTCAGT-3′ and adapter B-lower strand: 5′-P-CTGAGACACGCAACAGGGGATAGGCAAGGCACACAGGGGATAGG-3′. Finally, the Bst DNA polymerase fill-in step of the published protocol was left out.

After library construction, a titration run was carried out using 16, 64, 256 and 512 copies per bead. The copies per bead ratio to be used in the titration run is estimated based on the concentration of sstDNA (single stranded AB library). Therefore, the outcome of the titration run determines the ratio which needs to be applied in the actual sequencing run. Based on the titration run carried out for this experiment, a 48 copies per bead ratio was selected, founded on the “Predicted Recovery” of approximately 2.10⁶ enriched beads, > 60% PassFilter, <20% Mixed+Dots and approximately 6000 Keypass reads.

Emulsion PCR and bead enrichment were carried out according to the standard GS 20 protocol (Roche Applied Science). One full picotiterplate (PTP) (70×75 mm) with two regions was used. Enriched beads were divided over both regions. Sequencing was performed according to the manufacturer's instructions (Roche Applied Science).

The basecalled reads from both regions were added together in one file and further processed for SNP mining using a fully automated pipeline (Keygene N.V.). The web based pipeline was written in Perl 5.8.0 and runs via an Apache web server on a Linux platform. Microsoft Internet Explorer was used as client. An Oracle 10g relational database served as the central repository for all raw and processed data and the material and process definition.

The SNP discovery process consisted of four parts, namely (1) GS 20 data processing, (2) CRoPS pre-processing, (3) the CRoPS analysis, and (4) CRoPS SNP mining (Fig. 1).

GS 20 data processing was performed on-rig using the standard GS software. This process resulted in the “GS 20 raw sequence reads” that were directly used for further processing in the CRoPS pre-processing step. During pre-processing, the origin of the reads was identified according to their four base sample identification tags. The implementation of this step was based on the internal BLAST function in Oracle 10g (Oracle). Furthermore, the AFLP adapter sequences were trimmed. Pre-processed reads were saved to the database. In the CRoPS analysis step, reads were clustered and assembled using the TGICL tool [26]. Clustering was performed using the following variable parameters: minimum percent identity for overlaps (94%), minimum overlap length (30 nt) and maximum length of unmatched overhangs (30 nt). Again, all data obtained were subsequently saved to the Oracle database.

Polymorphisms were selected during the SNP mining step. For each putative SNP a number of features were recorded, including relative position in the consensus sequence, sample count in the MSA, allele count per sample, the MSA depth (number of reads at the SNP position), distance to flanking SNPs, flanking size around SNP and the presence of a homopolymer stretch in which a SNP may occur. Mining rules were created from these features and defined as follows: 1) each allele should be present at least twice in a MSA, 2) SNPs should not be part of homopolymers larger than 3 bases, and 3) SNPs should be segregating according to sample origin. SNPs that passed the filters were selected as the best candidates for conversion into genotyping assays.

The pre-processed sequence data will be deposited at the NCBI Short Read Archive (SRA) as soon as this archive is ready to accept the data (expected at the end of 2007). Until then, the data can be requested from the authors.

Probes were designed for 30 putative SNPs in two multiplex (15-plex) SNPWave assays using ProbeDesigner software (Keygene N.V.) as described previously [24]. SNPWave reactions were carried out as described previously [24]. In short, ligation reactions were carried out in 10 µl volume containing 200–400 ng total genomic DNA, 1×Taq DNA ligase buffer [20 mM tris-HCl, 25 mM KAc, 10 mM MgAc₂, 10 mM dithiothreitol (DTT), 1 mM NAD, 0.1% triton X-100; pH 7.6 at 25°C; New England Biolabs Inc], 2 U Taq DNA ligase (New England Biolabs Inc) and 0.5 fmol of each of the ligation probes. Next, 10 cycles of repeated denaturation, probe hybridization and ligation were performed in a Perkin Elmer 9700 thermal cycler (Applied Biosystems) using the following profile: initial denaturation for 2 min at 94°C, followed by 10 cycles of 15 s at 94°C and 60 min at 60°C, and storage at 4°C. After ligation, the mixture was diluted with 30 µl of 1×Taq DNA ligase buffer to 40 µl.

Ten µl of diluted ligation reaction was subsequently amplified in a 20 µl mixture containing 1×GeneAmp® PCR buffer (Applied Biosystems), 0.2 mM of each dNTP, 0.4 U AmpliTaq Gold DNA polymerase (Applied Biosystems), and 30 ng unlabeled forward primer (5-GACTGCGTACCAATTC-3) and 30 ng FAM-labeled reverse primer (5-GATGAGTCCTGAGTAA-3). The amplification profile was 12 min at 94°C, followed by 13 cycles of 30 s at 94°C, 30 s at 65°C with a reduction of 0.7°C per cycle to 56°C in cycle 13, followed by 1 min at 72°C. This was followed by 23 cycles of 30 s at 94°C, 30 s at 56°C and 1 min at 72°C, and storage at 4°C.

Purification of diluted SNPWave PCR products and subsequent detection on the MegaBACE 1000 (Amersham Biosciences/GE Healthcare Life Sciences) were as described previously [24].

SNPs and flanking sequences can be found in the supplementary file (Table S1). Probe sequences are available upon request from the authors.