Animals and DNA samples

DNA samples were obtained from five pig breeds, including Duroc (DU), Pietrain (PI), Landrace (LR), Large White (LW) and Wild Boar (WB). For each breed, a single DNA pool containing equal DNA amounts from all of the samples was prepared. The number of animals per pool was 34, 23, 29, 36 and 36 for DU, PI, LR, LW and WB, respectively. The DNA samples for the commercial breeds were representative of the worldwide distribution of the breeds (samples originated from the USA, Netherlands and Denmark), whereas the WB samples were collected mainly in Europe, with five samples originating from Japan.

Construction of the reduced representation libraries

For each breed, a total of 25 µg from each DNA pool, divided in five aliquots of 5 µg each, was digested with each of three restriction enzymes AluI, HaeIII and MspI (Fermentas GmbH, St. Leon-Rot, Germany), following the manufacturer's recommendations. For each restriction digest, 10 units of enzyme were used in a final volume of 10 µl and incubated at 37°C for 16 hr. Upon completion of each digestion, fragments were fractionated with a 10% non-denaturing polyacrylamide Criterion TBE gel (Biorad, Veenendaal, Netherlands) at 100 V for 190 min. DNA fractions were stained by immersing a gel for 15 min in a TBE 1× solution containing ethidium bromide. Fragments in particular size ranges were excised from the gel. The size ranges for fragments excised differed between libraries and included fragments of 160–200 bp (AluI short), 200–240 bp (AluI long), 160–200 bp (HaeIII) and 100–200 bp (MspI). After excision, the gel pieces were placed in a 0.5 ml microcentrifuge tube that had been pierced in the bottom, and then placed in a collection tube and centrifuged for seven min at 16000×g. A total of 400 µl of recovery buffer was added to the disrupted gel pieces, and the samples were kept overnight at 4°C. DNA was purified from the gel slurry by passing the samples through Sample Filter Cups from the DNA Gel Extraction Kit (Millipore, Billerica, MA, USA) by centrifugation at 4965×g for 10 min at room temperature. Next, 200 µl of recovery buffer was added and the samples were centrifuged at 16000×g for 15 min, followed by an ethanol precipitation. DNA was dissolved in 40 µl of Puregene DNA Hydration Solution (Gentra Systems, Valencia, CA, USA). Samples were quantified by spectrophotometry, and at least 300 ng of purified DNA was used to construct a RRL (reduced representation library) for sequencing. A total of 19 breed specific RRL were prepared, including 4 libraries each for DU, PI, LR and LW, but only three libraries were available for WB, because the AluI short library was not prepared.

Sequencing

All 19 libraries were sequenced on a 1G Genome Analyzer (Illumina, San Diego, CA, USA). The length of all sequences generated in this study was 36 nucleotides. In addition, two pooled samples derived from each of the AluI RRLs were prepared and sequenced using a 454 FLX system (Roche Applied Science, Indianapolis, IN, USA) on the GS FLX platform. For the Genome Analyzer, library constructions followed the protocol supplied with the Genomic DNA Sample Prep Kit (Illumina, “Preparing Samples for Sequencing Genomic DNA”, version 1003806_Rev. B, March 2008) with minor modifications. Library construction started with the addition of an adenosine base to the 3′ end of the DNA fragments, except for the MspI digested samples which were initially end repaired. The volumes of adapter reagent used in the adapter ligation reactions were titrated accordingly to the amount of input DNA to maintain a 10∶1 molar ratio of adapter to DNA. The concentrations of the libraries were determined using a NanoDrop ND-1000 Spectrophotometer (Saveen and Werner AB, Limhamn, Sweden) and the size and purity were determined using an Agilent 2100 Bioanalyzer in combination with the Agilent DNA 1000 Kit (Agilent Technologies, Nærum, Denmark). The libraries were diluted in buffer EB (QIAGEN) to 10 nM, denaturated with 2 N NaOH to a final DNA concentration of 0.5 nM, diluted to 2–4 pM with pre-chilled Hybridization buffer (Illumina) and loaded in individual lanes into a 1.0 mm flowcell together with a single lane of a 2 pM PhiX control library (Illumina). Following cluster amplification, linearization, blocking, and primer hybridization, 36 cycle sequencing were conducted on an Illumina Genome Analyzer (version I) using the Genomic DNA Sequencing Primer in combination with clustering and sequencing kits supplied by Illumina. The resulting images were transferred to the pipeline computer and analyzed using the Genome Analyzer Pipeline Software (version 1.0, Illumina) generating the raw fastq files. For FLX sequencing, the blunt fragments produced by AluI digestion were ligated directly to the adapters provided in the library preparation kit without the shearing, sizing or end-polishing steps. Single stranded libraries were then prepared and sequenced according to the manufacturer's instructions. The AluI short and AluI long libraries were sequenced using 5.5 and 4.5 full machine runs, respectively, using the two-region gasket.

Processing of the GA sequences

The raw sequence data were filtered according to different criteria. Each sequence was first evaluated for the presence of the expected sequence motif for each restriction enzyme; sequences not containing the expected sequence were discarded. The acceptable sequences began with CT, CC or CGG for the AluI, MspI, and HaeIII restriction digests, respectively. Sequences that contained the same nucleotide at more than 18 continuous positions were also eliminated. The average quality score was next calculated for each read by averaging the individual score for each of the 36 base positions. Previous results obtained in a smaller pilot study [17] indicated that a minimum average quality score of 12 was an acceptable threshold, and, therefore, all sequence reads with an average quality score<12 were removed from the dataset. Finally, sequences were also filtered for the presence of over-represented reads. For this purpose, we calculated the sequence depth for each RRL as the average of the total number of reads that aligned to each unique position in the reference genome. Reads that were present more than 5× the estimated sequence depth were removed from the dataset.

Reference genome

The reference genome against which the 36 bp sequences were aligned included all the porcine autosomes and the X chromosome. The state of completion of the assembly of each chromosome varied because the sequencing of the pig genome has not yet been completed. This genome sequencing and assembly effort is being performed under the direction of the international Swine Genome Sequencing Consortium [11] using a hierarchical shotgun sequencing strategy in which BAC clones from a minimum tiling path of clones are selected from the highly contiguous physical map of the pig genome [18]. These BAC clones are being sequenced to a minimum of 4-fold sequence coverage. The tiling path was selected from clones from the CHORI-242 BAC library generated from a single Duroc sow. Sequence data have been assembled into contigs on a BAC-by-BAC basis and have immediately been released into the EMBL/Genbank/DDBJ DNA sequence databases. BAC contig sequences have been assembled on a genome-wide basis at regular intervals. The unmasked “PreEnsembl” version of the Sus scrofa assembly 7 of the pig genome available on July 28^th 2008 was used for this study, and that resource was obtained from: ftp://ftp.sanger.ac.uk/pub/S_scrofa/assemblies/PreEnsembl_Sscrofa7/.

To represent regions of the genome that had not yet been assembled, two pseudo chromosomes were built from 454 sequence data generated from the two AluI libraries. For each library, all sequences were first aligned to the Sus scrofa assembly with BLAT [19]. Sequences that did not significantly align to that assembly were next used to create consensus sequences (contigs). Our strategy was to employ a simpler version of the approach used by TIGR to assemble large shotgun sequencing projects [20]. A Perl script was developed to cluster sequences which had the same motif in the first 32 bp. Batches of clusters were then assembled using CAP3 [21], producing a set composed of contigs and singletons (unique sequences) which were used to build a pseudo chromosome in which sequences were separated by blocks of 100 Ns.

SNP discovery

Sequences were aligned against the assembled reference genome and initial SNP detection was performed using MAQ [22]. For SNP discovery, only reads that aligned to a single unique location of the genome were considered. Because MAQ calls a SNP as being any difference between the reads or between the reads and the reference genome, the initial MAQ SNP prediction output is large and must be filtered. Several criteria were used to exclude the less reliable SNPs from the dataset. Thresholds were established for a number of MAQ values that were useful in predicting reliable SNPs. Specifically, MAQ's minimal map quality for the read, minimal consensus quality and minimal map quality of the best mapping read for each predicted SNP position were used as criteria to select reliable SNPs by setting the thresholds for all three parameters at 10 (SNPs with any values<10 were discarded). Moreover, we required the minor allele at each SNP be represented in at least three reads and that the total number of reads per SNP was lower than 120 (SNPs with higher read depth were discarded). The final SNP set comprised the de novo identified SNPs that had passed all filters. Finally, SNP MAFs were estimated by directly counting the number of reads for each allele. All de novo identified SNPs have been submitted to dbSNP (accession numbers from ss131027063 to ss131629651). The flanking sequences for the SNPs used to build the genotyping assays were derived from the ongoing sequencing of the porcine genome by the Sanger Institute and these sequences are regularly uploaded to Genbank.

In addition to these de novo identified SNPs, the final porcine SNP dataset also comprised previously identified SNPs that were available from different public sources, including SNPs detected in a pilot study conducted at Wageningen University using a DraI RRL [17], SNPs derived from the 7K porcine Illumina Custom Infinium Bead SNP Chip, SNPs identified by high throughput pyrosequencing [23], SNPs discovered through Sanger sequencing at INRA and publicly available SNPs from dbSNP, including a collection of porcine SNPs, derived from several sources, collected by the University of Cambridge.

Initial SNP validation

Information on 44,236 SNPs derived from the HaeIII library with flanking sequence information was available for validation of the SNP calling procedure. These markers were sorted by predicted location within the genome, and every fortieth sequence was processed using the Sequenom Assay Design software. From the output, assay groups were designed to have 32 multiplexed SNPs per well. Critical features to be tested were the position of the SNP within the GA sequence read, average quality score of the most frequently called base, number of alleles identified, as well as position within the genome. Nine assay groups (288 individual SNPs) were selected to represent SNPs in all tested categories. Reactions were set up per manufacturer's instructions for the Sequenom iPLEX GOLD chemistry and the 128 boars that contributed to the commercial breeds RRLs were individually genotyped to allow the determination of the true SNP allele frequencies in the discovery population.

Selection of the final SNP list

A list which included the final porcine SNP dataset comprising 549,282 de novo identified and publicly available SNPs, was submitted to Illumina for design score calculation, which was performed with Illumina's Assay Design Tool for Infinium. Type II Infinium SNPs were prioritized for selection in the design of the assay because they require only one bead type on the chip. In addition to the Illumina design score and the type of Infinium assay, we also considered other parameters, such as the estimated SNP MAF, spacing of the SNPs along each chromosome, genome-wide coverage (number of SNPs selected for each chromosome), presence of other SNPs within 40 bp of each target SNP and available information concerning the conversion rate of the SNP (i.e., had the SNP been assayed successfully before?). Finally, any SNP known to be linked to a patent was excluded from the list. Information for these parameters was collected for all available SNPs and was used to assign SNPs to waves [15]. A total of 22 waves were formed and used to conduct rounds of SNP selection. The first wave contained only type II Infinium SNPs with a design score≥0.8 and MAF≥0.25. In the second wave the MAF threshold was lowered to 0.15, while in the third wave the design score threshold was lowered to 0.7 while the MAF was again set to 0.25. As far as possible, the SNPs included on the Beadchip were selected from the waves with high stringency followed by additional rounds of selection in which the selection criteria were gradually relaxed. Upon completion of the selection process, the SNP list was manually inspected for the presence of large gaps (>250 kb) between SNPs. The identified large gaps were visually inspected and the size of the gaps decreased by manually selecting SNPs to reduce the size of the gap, whenever possible.

Because the available porcine genome sequence (build 7) only covered 70% of the genome, 70% of the SNPs selected to be included on the assay came from those for which position on genome build 7 was known. To ensure a complete coverage of the porcine genome, the remaining 30% of SNPs were selected from the list of SNPs that could not be mapped to porcine genome build 7.

Validation of the Illumina Porcine 60K+SNP iSelect Beadchip

To evaluate the performance of the Illumina iSelect BeadChip, which is now commercially available as the PorcineSNP60, the panel of 64,232 markers that passed the assay design tool informatics screen was used to genotype all 158 individuals from the five breeds that had been included in the original discovery panel. Samples were genotyped in the Illumina services lab and data was evaluated for robust genotyping [24]. Markers were evaluated for signal intensity, robust cluster formation and cluster separation. The DNA sample set included 6 replicate pairs and 94 trios to aid in cluster evaluation. The 62,121 loci that passed validation comprise the PorcineSNP60 panel.

Correlation between MAFs derived from sequencing and genotyping

Since allele frequency data were available from both the sequencing and genotyping efforts we investigated the relationship between the allele frequency estimates derived from both approaches. To obtain the best possible estimate for the correlation we applied some filters to the dataset. We used only SNPs that had been identified in the four RRLs prepared for this study (AluI short, AluI long, HaeIII and MspI) and that had not been mapped to the sex chromosomes. Since the X chromosome contains pseudoautosomal regions that are largely unknown it would not be possible to correct for the effect of gender. For each SNP the reference allele was set as the major allele over all breeds as derived from sequencing. For each of the breeds the sequence derived frequency was matched with the genotype derived frequency. The genotyping was performed on the same individuals that were used to make the sequencing pools. Comparisons between sequence-derived and genotype-derived frequencies were done on a breed by breed basis since the number of alleles counted per breed could vary widely between breeds. Correlations were calculated using Pearson's product-moment correlation as implemented in R (www.r-project.org). The correlation between genotype-derived and sequence-derived frequencies was plotted against sequence depth and fitted using locally weighted scatter plot smoothing (LOESS function in R).

Sequencing of the reduced representation libraries

A total of seven restriction enzymes were tested for the construction of the RRLs, including AluI, HaeIII, MspI, PvuII, RsaI, ScaI and StuI. Restriction enzymes producing blunt end fragments were chosen because this would result in DNA fragments that could be directly ligated to the adaptors used in the preparation of libraries for sequencing on the Illumina Genome Analyzer. We selected AluI, HaeIII and MspI because these enzymes maximized the number of fragments detected in the desired size range and minimized the presence of repetitive elements [13].

The total number of generated sequences surpassed 370 million and after filtering based on the quality criteria described in the materials and methods, over 247 million reads remained for SNP discovery using MAQ (Table 1). The estimated genome coverage for the different RRLs varied from 1.5 to 3.5% and the combined 4 RRLs span about 10% of the porcine genome.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Number of Illumina Genome Analyzer reads generated, filters applied to the dataset and final number of reads used for SNP discovery from the four RRLs.

https://doi.org/10.1371/journal.pone.0006524.t001

To identify SNPs in the 30% of the porcine genome that was not represented in genome build 7, the two largest RRLs (AluI short and AluI long) were also sequenced on Roche's 454 platform. The total number of sequences generated on this platform was 2,121,435 and 1,922,150 for the AluI short and AluI long RRLs, respectively.

Reference genome

The genome of the pig is comprised of 18 autosomes and the X and Y sex chromosomes with an estimated total size of 2.7 Gb similar to that of human. The reference genome (build 7) against which the GA sequences were aligned for SNP discovery, represented approximately 70% of the porcine genome with the state of sequence completion of each chromosome differing from less then 50% to up to 99% for individual chromosomes. To discover SNPs in the 30% of the porcine genome not represented in the assembly, the 454 sequences from the two AluI libraries were used. All 454 sequences were first aligned against sequences from the genome build 7 and only unique 454 sequences not represented in build 7 were used. These remaining 454 sequences were used to generate unique sequence contigs, which were then concatenated into two pseudo chromosomes (AluIshort and AluIlong, respectively). The final extended reference genome used with MAQ and to which all GA sequences were aligned consisted of porcine genome build 7 and the two pseudo 454-AluI chromosomes. Build 8 of the pig genome, released in October 2008, was also used to compare the distances between SNPs calculated using builds 7 and 8 of the pig genome.

SNP discovery

All 247 million sequences that passed our selection criteria (Table 1) were aligned to the extended reference genome using MAQ. MAQ aligns all GA sequences to the reference genome and reports all variation detected within the GA sequences as well as any difference between the GA and the reference sequences. The initial raw output from MAQ consisted of over 9 million positions at which variation was detected. The majority of these do not represent SNPs but result from sequencing errors in the GA sequences, errors in the reference genome or the alignment of paralogous sequences. However, after applying the filters described in the materials and methods, 315,130 variations remained representing high confidence SNPs (Table 2). While only SNPs for which the minor allele was represented by at least 3 sequences were considered in the selection of SNPs for the Beadchip design, we also identified 57,756 SNPs for which the minor allele was represented by only two sequences (low MAF SNPs) (Table 2). Initial validation of putative SNPs from the HaeIII RRL showed that a good indicator of the likelihood of SNP failure was the quality of the nucleotide at the SNP (low quality bases were more often associated with non-converting or monomorphic SNPs). Hence, we examined the quality score at each of the SNP bases in the two reads and established a threshold of 20 to discover reliable SNPs in those cases where the minor allele was seen only twice. The final total number of high confidence SNPs identified by using these strategies was 372,886.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Summary of the SNP discovered from the four analyzed RRLs.

https://doi.org/10.1371/journal.pone.0006524.t002

To further evaluate the conversion rate of the SNPs discovered by our approach we looked at the ratio between the number of transitions and transversions. For those SNPs where the minor allele was seen at least three times, 69.4% were transitions and 30.6% were transversions. Almost the same values were found for the 57,756 additional low MAF SNPs identified using the alternative thresholds (69.6% transitions and 30.4% transversions). Finally, because the error rate in GA sequences increases towards the end of the sequences, we also plotted the distribution of the identified SNPs against the nucleotide position within the GA sequences (Figures 1, 2).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 1. SNP distribution on each of the GA read positions.

The distribution represents all de novo identified SNPs, from the RRLs generated. The number of transitions and transversions identified is also illustrated.

https://doi.org/10.1371/journal.pone.0006524.g001

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 2. Distribution of the low minor allele count SNPs on each of the GA read positions.

For these SNPs two reads were identified for the minor allele. The number of transitions and transversions identified is also illustrated.

https://doi.org/10.1371/journal.pone.0006524.g002

Initial SNP validation

Before selecting the final 64,232 SNPs to be included on the Beadchip, we decided to validate the selection criteria used during the SNP discovery phase by genotyping a representative sample of 288 SNPs in the individuals used to create the discovery panel. Of the 288 assays attempted on the Sequenom MassArray platform, 241 (83.7%) assays yielded a sufficient number of genotypes to draw useful conclusions from the data (>70% call rate) across all 128 DNA samples. Of the 241 assays, 20 were monomorphic in the DNAs of the boars used to generate the RRL libraries (8.3%). An assessment of the quality score of the predominant base indicated that when the most predominant base scored was of low quality the frequency of monomorphic assays was tenfold higher than when the predominant base called was of high quality (44.0% vs 4.2%). We, therefore, added this as an additional selection criterion for SNP discovery. Twenty-four successful Sequenom assays were designed for SNPs with 3 alleles detected in sequencing the RRL. Only one of the 24 assays was actually tri-allelic, while two of the assays were monomorphic, representing a total drop-out rate of 12.5%. The average read depth in the RRL for the 288 SNPs tested was 37.4. Assays designed for SNPs with over 2× of the mean read depth resulted in 23.5% monomorphic assays (4 of 17) and 11.8% of assays with excess heterozygotes (2 of 17), likely due to duplicated regions in the genome. While an initial assessment of position within the read indicated that as the SNPs were located towards the end of a read the percentage of monomorphic markers increased, removal of SNPs detected with low quality bases eliminated this trend.

Design of the 60K SNP Beadchip

Besides the de novo generated SNPs, an effort was made to assemble a collection of porcine SNPs from other public sources. The final combined SNP collection surpassed 549K SNPs and this set was used to select the SNPs that were included on the PorcineSNP60 Beadchip. The information regarding the results for the Beadchip design is summarized in Table 3. The total number of SNPs submitted to Illumina for assay design was 72,000 and a total of 64,232 SNPs passed decoding and were included on the Beadchip. The information regarding the SNP sources used, as well as the number of SNPs available for selection and included on the Beadchip for each source is summarized in Table S1. Of the total number of selected SNPs, 45,510 were mapped to a specific chromosome in porcine genome build 7, including 21 SNPs on chromosome Y, while 18,722 were unmapped (Table 3). However, we were able to predict the position of 4,193 of these SNPs by comparative mapping. A total of five mitochondrial SNPs were also included.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Summary of SNPs included on the Beadchip, assay conversion, Infinium SNP type, wave number and estimated distances between SNPs using builds 7 and 8 of the pig genome (total numbers and breakdown by chromosome).

https://doi.org/10.1371/journal.pone.0006524.t003

The selection of SNPs to be included on the Beadchip was based on specific wave numbers assigned to each individual SNP based on quality parameters, MAF, type of assay (Infinium type I or II) and location in the genome. Final SNP selection was performed in waves [15] resulting in only 832 or 1.3% of the total Infinium I SNPs being selected. Over 90% of the SNPs included on the Beadchip were selected in the first three waves, where SNP quality was the highest. With the exception of the X chromosome, the average distance between SNPs on build 7 is between 30 and 40 kb (Figure 3). Nevertheless, larger distances between SNPs were found on all chromosomes, with the largest distances being detected on chromosomes 15 (407.4 kb) and X (447.8 kb). There were 115 intervals with a gap size larger than 250 kb. Most of these large gaps are located on chromosomes 14 and X, where 36 (SSC14) and 27 (SSCX) large gaps were detected However, there were also chromosomes with no large gaps between the SNPs (chromosomes 2, 3, 6, 10, 12, 16 and 18). For most of these regions with large gaps there were simply no SNPs available for selection.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 3. Distances between the SNPs included on the 60K+porcine Beadchip.

The distances (x axis) were calculated using builds 7 (blue) and 8 (red) of the pig genome sequence assembly.

https://doi.org/10.1371/journal.pone.0006524.g003

After the initial SNP discovery and our design of the Beadchip a new build of the pig genome became available (build 8). To increase the number of SNPs mapped to a specific position within the genome, the flanking sequences for all 64,232 SNPs were aligned to the new build using BLAT [19]. This allowed us to position an additional 4,215 previously unmapped SNPs onto the porcine genome and allowed us to verify the accuracy of our predictions based on the comparative mapping approach. Of the 4,193 SNPs with a predicted map location, 1,030 specifically aligned to porcine genome build 8. Only 68 of these SNPs mapped to a location that differed from our prediction, with the majority of the SNPs (93%) mapping in accordance to the predicted location. There were 42 SNPs with conflicting map positions on genome builds 7 and 8, and 411 SNPs positioned on genome build 7 that did not align to build 8. The number of large gaps between adjacent SNPs increased when build 8 was used, with 207 gaps>250 Kb identified using this build. This increase was observed for all chromosomes except for SSC14, where the number of large gaps decreased to zero, indicating that there were unmapped SNPs from build 7 that mapped to SSC14 in build 8, filling the observed large gaps.

Validation of the 60K SNP Beadchip

The panel of 64,232 porcine SNPs was used to genotype the 158 individuals from the five pools that were used for SNP discovery. SNPs were validated for robust genotyping quality by evaluating signal intensity and cluster separation. From the initial panel of SNPs, 62,121 (97.5%) genotyped successfully, while 1,611 SNPs (2.5%) could not be reliably genotyped. Of the 62,621 validated loci, 58,994 were polymorphic which indicates that the SNP conversion success rate was 94%. Furthermore, for 57,109 SNPs, the MAF was>5% and for 58,737 SNPs the MAF was>1%. The average MAF for all scorable SNPs was 0.274 (Table 4). The DNA sample set used for assay validation included 6 replicate pairs and 94 trios. The reproducibility rate was 100% for 350,312 genotype comparisons, and the Mendelian consistency rate was 100% for 5,478,678 genotype comparisons.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 4. Description, by SNP source, of the number of working SNPs, SNPs by minor allele frequency and monomorphic SNPs.

https://doi.org/10.1371/journal.pone.0006524.t004

Correlations between sequence-derived and genotype-derived allele frequencies were approximately 0.8 at sequencing depths of 20 or higher (Figure 4, Figure S1). Even at lower sequencing depths, sequence-derived allele frequencies were highly predictive of the actual frequencies, as estimated from genotyping the individuals in the pools (Figure 5). At sequence depths greater than 20, little increase in correlation was observed. Some variation exists in correlations between libraries (e.g., the MspI RRL presented higher correlation values when compared with the other RRLs), although these differences were small (Table 5).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 4. Correlation between sequence-derived and genotype-derived allele frequencies.

The scatter plot was determined using the frequencies of the PorcineSNP60 SNPs derived from the RRLs generated.

https://doi.org/10.1371/journal.pone.0006524.g004

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 5. Relationship between sequence depth and the correlation between sequence-derived and genotype-derived allele frequencies.

This relationship was determined for the PorcineSNP60 SNPs derived from the RRLs generated.

https://doi.org/10.1371/journal.pone.0006524.g005

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 5. Values for the correlation between sequence-derived and genotype-derived allele frequencies for different read depths and SNP sources.

https://doi.org/10.1371/journal.pone.0006524.t005

Conclusions

It is generally accepted that SNPs are the most common source of variation in vertebrate genomes and that significant numbers of SNPs will be needed for future studies in all species. It is now possible to conduct studies aimed at the molecular dissection of complex traits of economic importance. For this purpose, a high density porcine SNP genotyping Beadchip is now available to the community. This resource was developed using the four breeds primarily used in modern pig production and wild boar samples, which should make the Beadchip a very valuable resource for most types of pig genetic studies. The PorcineSNP60 Beadchip was tested using 158 samples and results confirm the high utility and importance that this resource will have, both in terms of working SNPs (62,621) and polymorphic SNPs (over 57K SNPs with MAF>0.05). This study also confirmed the utility of next generation sequencing technologies for the mass identification of genetic variation in any genome, including the identification of SNPs in regions of the genome that have not been previously sequenced. In addition, a very large number of additional SNPs were also identified, and even though they were not included on the Beadchip they are now publicly available for pig researchers worldwide.