Six porcine gene expression chips were hybridized with day 25 placental total RNA of occidental (n = 3) or Chinese Meishan (n = 3) swine origin. A linear-mixed model was fit to the dataset to test for probe-by-breed interaction effects. This model is adapted from the Rostoks et al. method with modifications in the correction for the average breed effect. In figure 2, the volcano plot provides a visualization of significant changes in probe-by-breed expression values between the two swine populations. Using Storey's q-value procedures 34 to correct for multiple testing, 4,635 probes were identified at an estimated false discovery rate (FDR) of 5%. Additionally, a threshold of a 2-fold change was imposed to examine the effect on reducing the number of false positives detected. After imposing this filter, which is represented by the vertical red lines in the volcano plot, 857 unique probes at a FDR of q < 0.05 and with greater than 2-fold difference in estimated corrected probe-by-breed interaction effects were identified.

To help interpret the probe-by-breed interaction effect, individual probe plots were constructed for each probe set, represented by 11 unique probes per transcript, using the Affymetrix porcine arrays (see Supplemental Files 2, 3 and 4). Specifically, each of the significant probes was plotted to visually inspect probe behavior within the respective probe set versus its normalized intensity. Representative plots are depicted in Figures 3a and 3b. In Figure 3a Meishan RNA for the TPT1 gene (Affymetrix probe set Ssc.14343.1.S1_at; rank 2) did not hybridize to probe number 2 in all three Meishan arrays (see Table 1); conversely, the occidental breed did hybridize. A rational interpretation of these probe dynamics is that Meishan RNA could be polymorphic with respect to the array probe and is essentially behaving as a mismatch probe instead of a perfect match probe. These efforts were repeated to sample more candidates, including PEG10 (Affymetrix probe set Ssc.13476.1.A1_at, probe 1; figure 3c rank 732). Interestingly, a non-breed specific polymorphism was revealed in one of the Meishan placentas profiled for PEG10 (see Table 1). In other words, because one of the probes with Meishan RNA behaved similar to the three occidental samples, a similar genotype could be inferred for the single Meishan sample. Indeed, the linear-mixed model was sensitive enough to identify a non-breed specific polymorphism as demonstrated by pyrosequencing data in the following section (See Table 1, Figure 4; Table 2).

For the purposes of initial method validation, a subset of 27 probes were arbitrarily selected for reverse transcription-PCR using universally-tagged primers compatible with pyrosequencing technology was performed 35. We opted to use pyrosequencing on cDNA as this technology has high accuracy in the immediate vicinity of the sequencing primer unlike conventional dye-termination chemistry, and the target sequences were readily available as compared with paucity of swine genomic sequence. Short amplicons (100–200 bp) were designed using the exemplar sequences available from Affymetrix NetAFFX36 to amplify a region surrounding the 25-mer probe. All primers used for these assays to amplify specific probes were designed using Primer3 37 or MPrime 38. Probe behavior as seen in the intensity plots in Figures 3a and 3b was explained by genotyping with pyrosequencing technology (see Table 2). A transversion mutation (G/C = S) in TPT1 was confirmed in position 10 of probe number 1 (Table 1). A forward pyrosequencing assay was generated for PEG10 to confirm the non-breed specific (G/A = R, forward) SNP and is illustrated in Figure 4, respectively. Of the twenty-seven probes examined by pyrosequencing in the absence of the 2-fold cut-off, 22 out of 27 probes were confirmed as containing SNPs. Imposition of the 2-fold expression difference criteria, improved the overall success rate of SNP detection (19 of 22 probes). Validation of SNP discovery efforts is summarized in Table 2.

These results also confirmed that this method tends to identify SFPs that are located near the center portion of the probe sequence rather than at the ends. This is consistent with what one would predict and with the distribution observed in Rostoks et al., 2005 31.

In order to obtain an unbiased confirmation rate of our approach, we randomly selected SFPs for sequencing from the entire pool of 857 candidate SFPs that match a q < 0.05, and |log₂ fold change| > 1 criteria. As the complete porcine genome assembly is not yet available, these sequences were derived by comparing Affymetrix Porcine target sequences by BLAST to the latest sequence available from the porcine high throughput genomic traces from NCBI's trace archive, and represent 44 predicted SFPs and 222 non-SFPs. Of these, 31 probes were confirmed to be true SFPs, with an unbiased confirmation rate of 70% (false discovery rate (FDR) of 0.30, sensitivity of 0.65, and specificity of 0.94; see Table 3 and Table 4).

The method of Rostoks et al 31has been successfully applied to a number of species including arabidopsis, rice, and barley; we were therefore interested to determine whether how our method performed in comparison. In order to directly compare method performance in a species with a larger number of known SNPs, we downloaded barley microarray and SNP data 31 and the respective researcher's website39. Both the method described herein and the method described by Rostoks et al. were executed on the six-array GEM subset, and performed within 1–2% of each other as measured by sensitivity and false discovery rate (see Supplemental File 5).

To answer the question of whether we could detect SFPs in the presence of confounding effects such as different tissue types, we tested whether gene profiles obtained from two different tissue types (placenta and fetal fibroblasts) could be used to effectively to detect SFPs. This question is particularly pertinent in the situation where one might download gene expression data online from two different breeds of animals where commonly different treatment or tissue types confound the breed effect. Our test of Spearman's rank correlation coefficient indicates (ρ = 0.1365, p < 0.0001) that there is a significant relationship between the breed-by-probe interaction p-values of a comparison between day 25 Meishan versus occidental placenta and day 25 Meishan versus day 30 occidental fibroblast. 588 SFPs are predicted by both approaches (q < 0.05, |log₂ fold change| > 1) where 16 are expected by chance indicating that there is a significant overlap (Table 5, chi-squared = 21,205, degrees of freedom (df) = 1, p < 0.0001). Taken together, these statistics suggest that SFPs identified by using divergent tissue types are real.

For this study, day 25 porcine fetuses were produced from natural matings of purebred white composite gilts 41 or purebred Meishans 6 months of age or older. Pregnant gilts of each breed were sacrificed on the morning of gestational day 25. Immediately following slaughter, the reproductive tracts were excised from the gilts, and fetuses and their placentas were collected from each uterine horn. Physical measurements were taken, including fetal body weight and wet placental weight. Following measurements, whole placental tissue was minced briefly, placed in cryovials and snap-frozen in liquid nitrogen, then stored at -80°C until they could be processed further. Porcine fibroblast lines were established as previously described 42. Procedures for the handling of animals complied with those specified in the Guide for the Care and Use of Agricultural Animals in Agricultural Research and Teaching (1999) [1st rev. ed. Savoy, IL: Federation of Animal Science Societies; 1999]. All animal protocols involving the use of swine for generation of fetal fibroblast lines were approved by the Institutional Animal Care & Use Committee at North Carolina State University.

Total RNA was extracted from each animal's placenta using RNAqueous Kit (Ambion, Austin, TX) following the instructions of the manufacturer with only slight modifications. Briefly, 200–300 mg of placental tissue was pulverized in a mortar and pestle under liquid nitrogen to homogeneity. Powdered tissue was weighed on a microbalance and lysis buffer was added directly to each sample in a weigh-boat (600 μL lysis buffer: 50 mg tissue), vortexed vigorously 60–90 seconds, pre-cleared by centrifugation for 5 min at 3,000 × g. In order to remove bulk genomic contamination, 600 μL of the pre-cleared supernatant was applied to an Agilent mini-prefilter column and centrifuged at 10,000 × g for 1 min. The filtrate was precipitated with an equal volume of 64% ethanol and the RNAqueous protocol was continued according to the manual.

Integrity of total RNA were crudely assessed by loading 2 μg per well for denaturing agarose gel electrophoresis and quantitated spectrophotometrically by the NanoDrop^® ND-1000 spectrophotometer (NanoDrop Technologies, Wilmington, DE). Total RNA (10 μg aliquots in nuclease-free H₂O) was divided for transcriptome analysis or first-strand cDNA synthesis and then stored at -80°C until they could be processed further. Fetuses were genotyped and selected for all females.

Three samples from each breed of swine were hybridized to an Affymetrix Porcine GeneChip microarray.

Gene expression microarrays were hybridized as previously described 42. The following procedure was performed by Expression Analysis (Durham, NC) in accordance with methods specified by the manufacturer for target preparation and hybridization. Before target production, the quality and quantity of each RNA sample was assessed using a 2100 BioAnalyzer and a RNA 6000 Nano LabChip Kit (Agilent Technologies, Palo Alto, CA). Total RNA (10 μg) was converted into cDNA using SuperScript III Reverse Transcriptase (Invitrogen, Carlsbad, CA) and a modified oligo(dT)₂₄ primer that contains T7 promoter sequences (GenSet, San Diego, CA). After first strand synthesis, residual RNA was degraded by the addition of RNase H and a double-stranded cDNA molecule was generated using DNA polymerase I and DNA ligase. The cDNA was then purified and concentrated using a phenol:chloroform:isoamyl alcohol extraction followed by ethanol precipitation. The cDNA products were incubated with T₇ RNA polymerase and biotinylated ribonucleotides using an In Vitro Transcription kit (Enzo Diagnostics, New York, NY). One-half of the cRNA product was purified using an RNeasy column (Qiagen, Valencia, CA) and quantified with a spectrophotometer. The cRNA target (20 μg) was incubated at 94°C for 35 minutes in fragmentation buffer (tris base, magnesium acetate, potassium acetate). The fragmented cRNA was diluted in hybridization buffer (MES; NaCl, EDTA, tween 20, herring sperm DNA, acetylated BSA) containing biotin-labeled OligoB2 and Eukaryotic Hybridization Controls (Affymetrix, Santa Clara, CA). The hybridization cocktail was denatured at 99°C for 5 minutes, incubated at 45°C for 5 minutes and then injected into a GeneChip cartridge. The GeneChip array was incubated at 42°C for at least 16 hours in a rotating oven at 60 rpm. GeneChips were washed with a series of non-stringent (25°C) and stringent (50°C) solutions containing variable amounts of MES buffer, tween 20 and SSPE buffer. The microarrays were then stained with streptavidin phycoerythrin and the fluorescent signal was amplified using a biotinylated antibody solution. Fluorescent images were detected in GeneChip^® Scanner 3000 and expression data was extracted using the MicroArray Suite 5.0 software (Affymetrix, Santa Clara, CA). All GeneChips were scaled to a median intensity setting of 500. The data discussed in this publication have been deposited in NCBIs Gene Expression Omnibus (GEO)43 and are accessible through GEO Series accession numbers GSE10446, GSE10447.

Log₂-transformed perfect-match (PM) intensities for all observations were fit to a linear mixed model that broadly corrected for breed and array effects 44. A gene-specific mixed model was fit to the normalized intensities (residuals from first model) accounting for fixed breed (B), probe (P), and breed-by-probe interaction effects (BP), and a random array effect.

The model used in PROC MIXED in SAS is given as follows (SAS, Cary, NC):

y_gijk = B_i + P_j + (BP)_ij + A_k + ε_gijk

g^th gene

i^th breed

j^th probe

k^th array

y_ijk is the normalized perfect match intensity of the j^th probe of the l^th individual of the i^th breed of the g^th gene hybridized on to the k^th array. We used a perfect-match only model, because it had been suggested that incorporating the mismatch probes increases noisiness of the data when estimating differential expression 4546.

We then tested the effect of breed (B) for each level of probe (P), resulting in one test for each of approximately 265,000 probes on the array. We corrected for multiple testing by converting the resulting p-values to q-values using Storey's method for controlling FDR 34. This procedure has been automated and is provided in Supplemental file 1.

The mixed model analysis described above was repeated using our gene expression data from day 25 Meishan placenta and data from day 30 normal occidental fibroblasts previously generated by Tsai et al 42. This represents the situation where gene expression data is downloaded from GEO from two different experiments involving different breeds, using the same short oligonucleotide platform. In this particular model, the fixed breed effect term is confounded by tissue and time. We asked the question of whether we could detect SFPs in the presence of these types of confounding effects by calculating Spearman's correlation coefficient for the p-values of the comparison between day 25 Meishan versus white composite placenta, and day 25 Meishan placenta versus day 30 occidental fibroblasts. This statistic indicates the degree of similarity between the ranks of detected putative SFPs as sorted by p-value.

In order to validate a subset of probes identified by the microarray experiments to harbor SNPs, we performed reverse transcription-PCR. The first strand cDNAs were synthesized from 5 μg of day 25 porcine placental total RNA using a 1.5 μM random pentadecamer priming method 47 and SuperScript^® III First-Strand Synthesis SuperMix reagents (Invitrogen, Carlsbad, CA USA) in a final volume of 20 μL. Reaction conditions were [50°C, 60 min; 85°C, 5 min]. First-strand product was diluted 1:3 with nuclease-free H₂O.

Aydin et al. 2006 35 established the feasibility of a universal PCR primer approach compatible with pyrosequencing. The initial PCR was performed with 25 μL reactions containing a final volume/concentration of the following components: 3 μL of reverse transcription product (50 ng cDNA template), 50 nM each gene-specific primers synthesized with the 5'-universal overhangs, 250 nM each of universal primers (with a terminal 5'-biotinylated moiety on one primer based on directionality of PSQ assay design), 3.5 mM MgCl₂, and 20.5 μL Platinum^® Blue PCR SuperMix (Invitrogen, Carlsbad, CA USA). The PCR thermocycling conditions were the following: [95°C, 15 sec, denaturation; 60°C, 15 sec, annealing; 72°C, 30 sec, extension]_{n = 8 cycles} followed by [95°C, 15 sec, denaturation; 58°C, 15 sec, annealing; 72°C, 30 sec, extension]_{n = 40 cycles} and a final extension at 72°C for 30 sec. Short amplicons (100–200 bp) were designed using the exemplar sequences available from Affymetrix NetAFFX to amplify a region surrounding the 25-mer probe. All primers used for these assays to amplify specific probes were designed using Primer3 37 or MPrime 38, were chemically synthesized with only desalting purification (IDT, Coralville, IA). A RT-negative control was used to check for DNA contamination and all such controls were negative on PCR amplification. Biotinylated amplicons were visualized on 8% PAGE prior to pyrosequencing.

Pyrosequencing was used as a quick and accurate method for SFP discovery in RT-PCR products amplified from cDNA generated from placental tissues of informative subjects (purebred Meishan or white composite porcine embryos).

For template preparation of each sample, 200 μg of streptavidin-coated sepharose beads (GE Biosciences) were washed twice in washing buffer (Biotage, Foxboro, MA) using a vacuum pump manifold and finally re-suspended in 45 μL of binding buffer (Biotage, Foxboro, MA)3548. An equivalent volume (25–45 μL) of biotinylated PCR product was agitated with re-suspended beads for efficient immobilization [15 min, 25°C], then washed briefly in 70% ethanol [10 sec, 25°C]. To remove non-biotinylated DNA strand, immobilized duplexed DNA was melted using a NaOH-containing denaturation buffer [5 sec, 25°C] (Biotage, Foxboro, MA) and subsequently washed once in wash buffer [10 sec, 25°C] (Biotage, Foxboro, MA). Sequencing primer was added in 40 μL annealing buffer at a final concentration of 0.4 μM and hybridized to template by incubating [95°C for 2 min] in a heat block and slowly cooled to ambient temperature.

The pyrosequencing reaction was conducted via an automated PSQ 96MA machine, which uses a disposable cartridge (PSQ 96 Reagent Cartridge, Biotage, Foxboro, MA) to deliver enzymes, substrates, and each of the four nucleotides (PSQ 96 SNP Reagent Kit 5 × 96, Biotage, Foxboro, MA) into the wells of a transparent 96-well micro-titer plate. Successive nucleotides are dispensed in each well at 1 min intervals. Data acquisition of bioluminescence was captured and transcribed into spectra (pyrogram) for analysis.

Post-pyrosequencing, each contig was aligned with MUSCLE 49 and the consensus sequence harboring the polymorphic nucleotide were displayed in the form of IUB codes 50.

Alternatively, fluorescent terminator sequencing was used as previously described 6 to assess SNP true confirmation rate, Genomic DNA from identical animals hybridized to expression arrays was phenol-chloroform extracted.

For the probesets predicted to contain SFPs with a q-value of <= .05 and greater than 2-fold expression difference, the Affymetrix Porcine array target sequence was compared by BLAST to the latest available set of sequences from the porcine high throughput genome sequencing deposited into NCBI's trace repository51, yielding 584 putative matches. Primer3 37 was used to generate primers from these targeting an area +/- 200 bp around the probe of interest, with a target amplicon length of 450–500 bp. Primer min, opt, and max Tm were 57°C, 62.5°C, and 65°C respectively; primer_max_poly_x was limited to 3. A list of all primers used in this study is freely available upon request