The most significantly differentially expressed gene involved in phase I oxidation reactions was cytochrome P450 2C49 (CYP2C49). This gene is a member of the CYP2C family 23 and genes of this family encode monooxygenases that catalyse several reactions involved in the metabolism of drugs and endogenous compounds. Different CYP2C isoforms show some cross reactivity towards substrates which makes it difficult to differentiate CYP2C activities, and substrate specificity for CYP2C49 is not known 24. In our study, CYP2C49 transcripts were shown to have a significant up-regulation in boars with high levels of androstenone in both D and NL lines (DH and NLH) but differences were clearly most prominent in the D breed.

Another member of the cytochrome P450 superfamily, that was differentially expressed in our study, is cytochrome P450 2E1 (CYP2E1). In contrast to expression of the CYP2C49 gene, CYP2E1 showed a down-regulation in DH and NLH boars. CYP2E1 encodes an enzyme that metabolises many endogenous and exogenous substrates such as alcohols, ketones and drugs 24. Androstenone has been shown to block skatole induced expression of CYP2E1 in pig hepatocytes and to inhibit activity of CYP2E1 in porcine liver microsomes 1025. Liver metabolism of skatole also involves this gene and enzyme 1314262728, and the observed pubertal increase in skatole levels has been attributed to the inhibition of CYP2E1 by the sex steroids androstenone and 17β-estradiol 25. This is the first study, however, to show gene expression differences for CYP2E1 in relation to androstenone.

Cytochrome P450 member 2A19 (CYP2A19), which is the pig ortholog of human CYP2A6, catalyses 7-hydroxylation of coumarin in pigs 26, in addition to being involved in skatole metabolism 29. In this study, CYP2A19 was significantly down-regulated in DH boars. Gene expression differences in animals with high and low androstenone levels have not previously been reported for porcine CYP2A19, however its protein expression has been shown to be inhibited by androstenone 30. Diaz and Squires 26 found that both CYP2A6 protein content and its enzymatic activity were negatively correlated with skatole levels in adipose tissue. In contrast, results presented by Terner et al. 28 indicate that the CYP2A6 enzyme is not important for metabolism of skatole in primary cultured porcine hepatocytes.

Two other members of the cytochrome P450 family, CYP27A1 and CYP2C33, were also up-regulated in DH. CYP27A1 is a multifunctional enzyme that oxygenates cholesterol, bile acids and vitamin D. It has not previously been identified as a candidate gene associated with androstenone, but it is regulated by other sex steroids in human cells 31. CYP2C33 belongs to the same sub-family as CYP2C49, but its function has yet to be described in the literature. Like the CYP2C49 gene, CYP2C33 was found to be up-regulated in DH and NLH boars. The differential expression of CYP2A19, CYP27A1 and CYP2C33 in D and not NL boars might suggest that more genes are involved in phase I reactions in D. This is supported by a high number of oxidoreductase pathways significant in D (figure 1). Monooxygenase reactions are, however, clearly important in both breeds according to our gene ontology results (figure 1). Down-regulation of CYP2E1 and CYP2A19 might suggest a different role for these genes compared to the up-regulated CYP2C18, CYP2C33 and CYP27A1.

Our study implicates another class of genes involved in phase I oxidation reactions, namely the flavin-containing monooxygenases (FMOs). The FMO family of enzymes converts lipophilic compounds into more polar metabolites and decreases activity of the compounds, a similar activity to that of the cytochrome P450s 32. Microarray expression results show that flavin-containing monooxygenase 1 (FMO1) was significantly up-regulated in DH in the microarray results. The expression pattern in DH boars was confirmed by rcPCR and the rcPCR analysis also revealed that FMO1 was up-regulated in NLH boars. As shown in figure 2, the fold change in NLH was only half of DH, which might explain why this gene was not found to be significantly differentially expressed in the microarray results. Regulation of FMO involves sex steroids. In male mice, castration was reported to increase FMO1 expression 33, while in rats the opposite effect has been shown, with positive regulation of FMO by testosterone and negative by estradiol 34. The monooxygenase activity of FMO1 in addition to its regulation by steroid hormones makes it an interesting candidate gene for boar taint.

Phase I oxidation reactions of androstenone by enzymes including cytochrome P450s and FMOs are often followed by phase II conjugation reactions that are catalysed by glucuronosyltransferases, sulfotransferases, acetyltransferases and glutathione S-transferases 22. Through the addition of polar moieties, these enzymes increase substrate solubility. Conjugation reactions are an important means of excreting steroid hormones and other compounds, these reactions have also been proposed to keep inactive steroids easily available in cells. The hydroxysteroid sulfotransferases SULT2A1 and SULT2B1 have been associated with androstenone levels in previous studies 12151617. These genes were not found to be significantly differentially expressed in our study, however the estrogen sulfotransferase STE (SULT2E1) was found to be up-regulated in DH boars and down-regulated in NLH boars. A previous microarray study performed by our group showed an up-regulation of CYP19 gene expression in the testis of D and NL boars with high androstenone levels 17. These genes encode enzymes biosynthesising estrogens and might explain why an estrogen conjugation enzyme is differentially expressed in livers of the same pigs. Furthermore, boars secrete large quantities of estrogens from the testis 35, and both fat and plasma estrogen levels have previously been shown to be highly correlated with levels of androstenone in adipose tissue 3637.

Phase II conjugation by glucuronidation is another major pathway of liver elimination of endogenous and exogenous compounds, and in D this molecular function was significantly overrepresented. It is catalysed by uridine diphospho-glucuronosyltransferases (UGTs) which transfer glucuronic acid to substrates to increase solubility 38. The UGTs have been divided into two subfamilies, the UGT1s and the UGT2s, and in this study the family member UGT1A5 was found to be up-regulated in DH boars. Furthermore, a transcript similar to UGT2B15 was up-regulated in D and a transcript similar to UGT2A1 was down-regulated in NLH. UGT2B15 has been shown to conjugate several androgens in humans 38, whereas UGT1A5 has been found to be catalytically active against some exogenous compounds 39. UGT2A1 contributes to glucuronidation of steroids and phenolic compounds in olfactory tissue and has been found to eliminate odourants 40. The UGT family of conjugation enzymes have previously been associated with the androstenone metabolites α-androstenol and β-androstenol 4142.

N-acetyltransferases (NATs) are another family of conjugation enzymes and act by transferring the acetyl group of acetyl coenzyme A to aromatic amines to increase their water solubility 22. NAT type 12 (NAT12) was found to be up-regulated in NLH based on the microarray results and this was verified by rcPCR. The rcPCR results also revealed an up-regulation in DH boars. Androgens have been shown to increase expression of NAT type 1 in human cancer cells 43, but the enzyme family has not previously been associated with levels of androstenone. Tryptophan, the precursor of skatole, contains an aromatic, suggesting a role for this family of conjugation enzymes in the control of skatole levels, possibly through regulation by steroid hormones.

Glutathione S-transferases (GSTs) are functionally diverse enzymes mostly known to catalyse conjugation reactions of endogenous substances, haem, fatty acids, xenobiotics and products of oxidative processes 44. They have also been implicated in the intracellular transport of steroids to their site of action 44. In this study, a glutathione S-transferase gene was found to be down-regulated in NLH. The gene is described in the database as LOC396850. We have previously described up-regulation of two GST genes, GSTO1 and MGST1, in association with testicular androstenone levels 17, and here we additionally propose a role for glucuronidation in liver metabolism. Expression of phase II metabolic genes in D and NL pigs suggest breed specific mechanisms. We found differential expression of a GST variant (LOC396850) in NL but not D boars, up-regulation of STE in DH but down-regulation in NLH, and up-regulation of two UGT genes in DH but down-regulation of another family member in NLH. Breed specific phase II mechanisms for these breeds are in agreement with our previous finding for the protein SULT2B1 12.

17β-hydroxysteroid dehydrogenases (17β-HSDs) regulate the availability of androgens and estrogens in tissues by catalysing interconvertion of active and inactive forms of steroids 45. They do so by regulating the occupancy of steroid nuclear hormone receptors like androgen receptor, estrogen receptor and progesterone receptor 46. Gene expression of 17β-HSD in porcine liver has been shown to be negatively associated with levels of androstenone in adipose tissue 37. In a previous study, we found that the HSD17B4 gene was significantly up-regulated in testicle samples from DH and NLH 17, and in this study HSD17B4 was up-regulated in the liver of DH boars. Additionally, we found that HSD17B2 was down-regulated in DH and that the isoform HSD17B13 was down-regulated in both DH and NLH. Differential expression of both HSD17B2 and HSD17B13 was verified by rcPCR, which also showed significant down-regulation of HSD17B2 in NLH boars. The HSD17B2 enzyme catalyses the interconversion of testosterone and androstenedione, as well as estradiol and estrone 47. The function of isoform HSD17B13, also known as short-chain dehydrogenase/reductase 9, has yet to be described but its expression has been characterised in human liver 48. The short-chain dehydrogenase/reductase 8 gene (DHRS8), also known as HSD17B11, was found to be down-regulated in NLH boars. HSD17B11 is involved in androgen metabolism 49 and we previously found this gene to be down-regulated in the testis of DH boars 17. The HSDs belong to two superfamilies: the short-chain dehydrogenase/reductases (SDRs) and the aldo-keto reductases (AKRs) 46. The four HSDs described above all belong to the SDR superfamily but a gene from the AKRs was also found to be differentially expressed. The aldo-keto reductase family member 1D1 (AKR1D1) was down-regulated in NLH animals. AKR1D1 is a liver specific enzyme that regulates the hormone levels of several steroids 46. The enzyme has a 5β-reductase activity and is necessary for the catabolism of aldosterone, cortisol and androgens 50. Aldosterone, cortisol and androgens are three of four metabolic compounds formed from pregnenolone, with the fourth compound being androstenone 5. A role for AKR1D1 in androstenone catabolism might therefore be possible. We previously reported up-regulation of the family member AKR1C4 in the testis of DH and NLH boars 17, but AKR1D1 has not formerly been associated with porcine androstenone levels.

Another class of enzymes that regulate the availability of steroids is the plasma proteins. Binding of plasma protein is a reversible reaction that has no physiological effect, but it controls the amount of free drugs and hormones available to tissues by protecting them from metabolism 51. Alpha-1-acid glycoprotein (AGP) is an acute phase serum protein synthesised in the liver and secreted to plasma where it binds and carries drugs and steroid hormones 51. The protein has been shown to interact with CYP3A and CYP2C19 in humans and to inhibit cytochrome P450 activity 5253. It is suggested that serum protein interaction is an important factor for cytochrome P450 mediated metabolism, and that the interaction is isoform specific 53. AGP has also been studied in pigs as a binding factor of pheromones, but has not been found to bind progesterone in pig nasal mucosa 54. AGP has two variants in humans: orosomucoid 1 and 2 (ORM1 and ORM2) 51. In this study we found that porcine transcript homologous to bovine AGP and human ORM1 were significantly down-regulated in DH and NLH boars. Significant expression of these genes might explain the inflammatory response ontology term in figure 2 as these genes have mostly been described in relation to inflammatory response. AGP has not been investigated as a binding protein and transporter for androstenone, but different levels of androstenone could be related to differences in its availability through plasma binding by AGP.

One of the most significant molecular function terms in NL was nucleic acid binding. This term includes transcription factor binding and several transcription factors were found to be differentially expressed in this study. The GATA factor 4 (GATA4) was found to be down-regulated based on the microarray data, but we were not able to confirm this result with rcPCR. Other transcription factors identified in this study should be further investigated, including RAR-related orphan receptor A (RORA), transcription elongation factor B3 (TCEB3), nuclear factor I/X (NFIX), transcription factor 8 (TCF8) and heat shock transcription factor 1 (HSF1). We have previously identified iron ion binding, ferric ion binding and electron transport as being associated with levels of androstenone 17, possibly through interaction with the haem-containing cytochrome P450s. In this study we also found iron ion binding and electron transport, in addition to haem binding, to be important pathways, supporting our previous findings. Cytochrome b5 (CYB5) is involved in electron transfer to cytochrome P450s and has been proposed as a candidate gene for androstenone through its interaction with cytochrome P450 c17 (CYP17) 55. In this study we found CYB5 to be down-regulated in NLH boars, however we previously identified CYB5 as being up-regulated in the testis of both DH and NLH 17. Both CYP2E1 and CYP2A6 have been found to be activated by CYB5 in humans 56, supporting the down-regulation of CYB5 together with CYP2E1 and CYP2A6 in this study. Up-regulation in testis and down-regulation in liver might also support a key regulatory role of CYB5 in both tissues.

Breed differences in boar taint candidate genes have previously been shown between D and NL 1217. Consistent with these findings, we found breed differences in the expression profiles of genes involved in both phase I metabolism and phase II metabolism. However, rcPCR results might suggest that genes appearing in one breed can be differentially expressed also in the other breed. Additional and more comprehensive rcPCR studies are planned to clarify this. In general, the D breed showed higher levels of significance compared to the NL breed, potentially due to the larger contrasts between high and low androstenone groups in this breed. A number of new candidate genes have been identified in this study, however, we do not know their effect on performance and sexual maturation in pig. Additional studies are needed to investigate potential roles of the genes identified for phenotypes related to fertility.

Pigs in this study were selected based on extreme androstenone values. However, levels of androstenone and skatole have previously been found to be highly correlated, showing genetic correlations of 0.62 for D and 0.36 for NL 21. Some of the genes found differentially expressed in this study have previously been found to be associated with tryptophan or skatole and this could be explained by high correlations with androstenone. This might also be the reason for the significance of GO terms such as aromatic amino acid metabolism and catabolism, since the skatole precursor tryptophan belongs to this family of amino acids. The tryptophan 2, 3-dioxygenase gene (TDO2), which oxidises tryptophan, was differentially expressed in DH. A gene of the FMO family, kynurenine 3-monooxygenase (KMO), was found to be down-regulated in NLH. KMO is involved in tryptophan degradation and might therefore be interesting in regards to skatole 57. The aldehyde oxidase (AOX1) gene was significantly down-regulated in high androstenone animals of both breeds and this gene has not previously been associated with androstenone levels. It has, however, been shown to play an important role in skatole metabolism in several species including pigs 585960.

Animals used in this study were Duroc (D) and Norwegian Landrace (NL) boars from NORSVIN's three boar testing stations. The D boars were on average 156 days old at 100 kg live weight compared to the NL boars that were on average 143 days old at 100 kg live weight. The boars were slaughtered on average 14 days later. Tissue samples from liver were frozen in liquid N₂ immediately after slaughter and stored at -80°C until used for RNA extraction as described below. Samples from adipose tissue were collected from the neck at slaughter and stored at -20°C until used for androstenone measurements. Androstenone levels were measured by a modified time-resolved fluoroimmunoassay at the hormone laboratory, Norwegian School of Veterinary Science (NVH) 61 using an antiserum produced at NVH 62. The androstenone measurements were performed on more than 2500 boars, and statistical power calculations showed that selecting animals from each tail of the androstenone distribution would yield sufficient power to detect differentially expressed genes with a limited number of arrays. The 30 most extreme boars from each tail of the androstenone level distribution were therefore selected. Due to poor RNA quality for two of the samples, 29 samples were subsequently used from each group. 42 of the animals used were the same individuals as those used in our previous study examining gene expression in boar testis 17, with 16 being new animals as liver samples were not available for all of the previously studied animals. Average androstenone levels for the selected boars were 1.17 ppm and 3.22 ppm for NL and D, respectively. (See additional file 5: Androstenone levels). Average values for the groups were 5.95 ± 2.04 ppm for NL high (NLH), 0.14 ± 0.04 ppm for NL low (NLL), 11.57 ± 3.2 ppm for D high (DH) and 0.37 ± 0.17 ppm for D low (DL). In order to reduce family effects, a maximum of two and three half sibs were chosen from NL and D, respectively. The selected animals were used for expression profiling by microarrays and for the following verification of selected genes by rcPCR.

The present work utilises and extends methods described in our previous microarray experiment 17. The porcine cDNA microarrays were produced at the Faculty of Agricultural Sciences, University of Aarhus and contained 27,774 features printed in duplicates. 26,877 features were PCR products amplified from cDNA clones produced by the Sino-Danish Porcine Genome Sequencing project 6364, and 867 were control features. The 26,877 features represent approximately 20K gene transcripts. Additional information about the porcine cDNA microarray can be found at NCBIs Gene Expression Omnibus (GEO, 65) using the platform accession number GPL3585. This is a different batch of microarrays compared to the one we used in the testis experiment.

Total RNA was extracted from liver tissue using Qiagen's RNeasy midi kit according to manufacturer's instruction (Qiagen, CA, USA). RNA quantity was measured on a NanoDrop ND-1000 Spectrophotometer (NanoDrop Technologies, DE, USA) and RNA quality was evaluated by the 28S:18S rRNA ratio using a RNA 6000 Nano LabChip^® Kit on 2100 Bioanalyzer (Agilent Technologies, CA, USA). The microarray experiment was conducted as a common reference design using RNA purified from liver tissue sampled from an unrelated Danish Landrace × Hampshire pig as a reference. Aminoallyl-cDNA was synthesised from 15 μg of total RNA using the SuperScript indirect cDNA labelling system (Invitrogen Corporation, CA, USA) and labelled using ARES Alexa Fluor labelling kits (Molecular Probes, OR, USA). Amino-modified and fluorescently labelled cDNA was purified using NucleoSpin 96 Extract II PCR Clean-up kits (Macherey-Nagel, Düren, Germany). The individual samples were labelled with Alexa Fluor 647 and the reference was labelled with Alexa Fluor 555. "Green" spike-in RNA from the Lucidea Universal ScoreCard (Amersham Biosciences) was added to the individuals and "red" spike-in RNA was added to the reference. Hybridisation was performed in a Discovery XT hybridisation station (Ventana Discovery Systems, AZ, USA), followed by manual washing and drying by centrifugation. The microarrays were scanned using a ScanArray Express scanner (Perkin Elmer Inc., MA, USA) and image analyses were conducted using GenePix Pro 6 software (version 6.0.1.26, Molecular Devices Corp., CA, USA). Statistical analyses were carried out in R version 2.3.1 66 using the software package Linear Models for Microarray Analysis (Limma version 2.7.2) 676869 which is part of the Bioconductor package 70. Log-transformed ratios of mean foreground intensities (not background corrected) were print tip loess normalised. The duplicate correlation function in Limma was used to consider duplicate printing of each feature. To evaluate the analyses, MA-plots (M = log₂594/log₂488, A = (log₂594 + log₂488)/2), image plots and box plots were constructed using the Limma tools for visualisation both before and after normalisation (see additional file 6: Boxplots). To assess differential expression, Limma uses linear models in combination with an empirical Bayes method to moderate the standard errors of the estimated log-fold changes 68. The nominal p-values were corrected for multiple testing by false discovery rates (FDR) using Benjamini and Hochberg approach 71. Each of the groups DH, DL, NLH and NLL animals were hybridised in individual batches, causing some confounding between androstenone levels and hybridisation batch. However, as neither boxplots nor MA-plots were found to differ between hybridisation batches, it was assumed that the impact of hybridisation batch on the obtained data can be ignored. The top 1% of the genes was considered for further analyses. The array features were mapped to a LocusLink identifier and an annotation package was built using the Bioconductor package AnnBuilder (version 1.9.14). GO terms (p < 0.01 and more than 10 significant genes) were analysed for overrepresentation using the GOHyperG function of the Bioconductor package GOstats (version 1.6.0) 7273. More detailed descriptions of the microarray experiments are available at the GEO database through the series accession number GSE 11073.

A real competitive PCR (rcPCR) gene expression analysis was used to verify a subset of the results from the microarray study. Quantitative Gene Expression (QGE) was performed using MassARRAY methodology and the iPLEX protocol (Sequenom, CA, USA). Total RNA was isolated from liver using an automatic DNA/RNA extractor (BioRobot M48 workstation; Qiagen; CA, USA). Total RNA was treated with TURBO DNA-free™ (Ambion, Huntingdon, UK) for removal of contaminating DNA and first strand cDNA synthesis was conducted on 0.5 μg total RNA using SuperScript™-II Rnase H^- Reverse Transcriptase (Invitrogen, Carlsbad, CA, USA). Assays for the genes included in this study (See additional file 7: Gene transcripts included in the rcPCR analyses) were designed and multiplexed into a single reaction using MassARRAY QGE Assay Design software (Sequenom, CA, USA). The competitor, a synthetic DNA molecule matching the targeted cDNA sequence at all positions except for one single base, served as an internal standard for each transcript. A 10-fold competitor dilution was initially used over a wide range of concentrations to determine an approximate equivalence point. Following this, a 7-fold dilution of competitor from 4.04 × 10^-11 to 1.43 × 10^-19 was used to achieve accurate quantification. The cDNA and competitor were co-amplified in the same PCR reaction with the following conditions: 95°C for 15 minutes, 45 cycles each of 95°C for 20 second, 56°C for 30 seconds and 72°C for 1 minute, and 72°C for 3 minutes. A clean-up step was performed to remove unincorporated nucleotides. The iPLEX reaction cocktail mix and PCR conditions were according to manufacturer's instructions 74. Parallel PCR reactions were performed for all samples and the products were printed with two replicates on a SpectroCHIP. The primer extension reaction uses PCR products as templates and generates distinct mass signals for competitor and cDNA-derived products. Mass spectrometric analysis generated signals from which peak areas were calculated. Gene expression levels were analysed using TITAN software version 1.0–13 7576 that runs in the R statistical environment. Raw data from the Genotype Analyzer Software (Sequenom) were imported into TITAN and analysed using the default values of linear least squares polynomial regression and 4000 bootstrap replicates. cDNA concentrations were corrected with respect to the housekeeping gene (TFRC), and p-values and confidence intervals for fold changes were calculated.