The functional roles of MSC in response to different Staphylococcus spp were explored through gene expression profiling in the divergent sheep lines successively infected with Se during the first lactation, and Sa during the second lactation. The multifactorial ANOVA applied probe by probe identified 5,573 probes as differentially expressed according to the challenge effect (FDR of 5% and absolute fold-change (aFC) > 1.5; the FC is the ratio between Sa and Se challenges). Among these, 261 probes had an aFC > 5 (Additional file 1). The DE probe list enabled a perfect segregation between animals challenged by Se and Sa as shown by the hierarchical clustering in Figure 3. This probe list corresponded to 210 annotated genes with 95 and 115 genes that were more expressed in Sa and Se infections, respectively. The main functions of the genes expressed at a higher level in Sa than in Se infections were associated with the immune response: hematopoiesis (p = 0.010), cell-mediated immune response (p = 0.009), cell death (p = 0.001), immunological disease (p = 0.001) and inflammatory response (p = 0.010). Genes whose expression was higher after Se challenge were linked to cellular growth and proliferation (p = 0.048), infectious disease (p = 0.032), lipid metabolism (p = 0.042), molecular transport (p = 0.049) and small molecule biochemistry (p = 0.048) (Ingenuity Pathway Analysis - IPA - data, not shown).

Network analysis with IPA produced two networks. In the first network (Additional file 2A), which is characterised by cell-to-cell signalling and interaction and cell-mediated immune response, the T cell receptor signalling pathway (with the genes: cd247, cd3d, cd3e, cd3g, ctla4, itk, ppp3cc, rasgrp1, tra@, zap70) and the major histocompatibility complex - MHC - pathway (with six sub-units: dma, dqa1 (three probes), dqa2 (two probes), dra, drb1 (two probes) and drb3) are highly represented. In a second network, defined by inflammatory disease and response and haematological disease, both the IL1 receptor pathway and the TNF pathway are well represented with il1r1, il1rap and irak4, and traf2-3-5 respectively (Additional file 2B).

Real-time qPCR was used to confirm the gene expression differences between the two challenges. The most expressed gene in Se challenges was cpb2. We also examined the differential expression of chemokine (C-C motif) receptor 3 (ccr3) and interleukine 1 receptor type II (il-1r2) for their role in pathogen detection, as well as myxovirus resistance 2 (mx2) and granzyme H (gzmh) for their role in immune response. Hence, five genes were selected, three of which were more expressed in the Sa condition (ccr3, gzmh and mx2) and two in the Se condition (cpb2 and il-1r2). Chemokine (C-X-C motif) ligand 10 (cxcl10) was added to this gene list since it is known to induce T cell recruitment in inflamed tissue, and because it had been discarded from the microarray analysis due to missing data (data not shown). The differential expression between the two challenges was significantly (p < 0.05) confirmed for five out of the six genes (Table 1). Among those genes, the expression of cxcl10 was 60-fold higher in Sa than in Se challenges. Although not significant (p = 0.12), the expression of gzmh was higher in the Sa than the Se condition (Table 1).

Using a multifactorial ANOVA model applied probe by probe, 57 probes were shown to be significantly DE between the resistant and susceptible lines (FDR 5% and aFC > 1.5) (Table 2). These probes corresponded to 52 annotated genes, with 33 and 19 genes expressed at a higher level in the resistant and susceptible line, respectively. This list of 52 genes is further named as the "main list". Out of the 33 genes with higher expression in the resistant line, eight had a FC ≥ 2 (cryl1, tp53, a non-classical mhc-I, slc40a1, eif4ebp1, ppapdc1b, slc46a3 and loc784517). Only three out of the 19 genes that were expressed at a higher level in the susceptible line had an aFC > 2 (gtpbp4, mapre1, tmem87b).

As the gene expression of MSC response differs between challenges with two different Staphylococcus species, the line effect was also analysed independently within each challenge. The lesser amount of data in separate analyses made it necessary to loosen the significance threshold to find differentially expressed genes within Se or Sa challenges. Accordingly, a total of 152 probes (138 annotated genes) was considered to be DE between the lines after Se challenge and 235 probes (204 annotated) after Sa challenge (p-value ≤ 0.01 and aFC > 1.5) (Additional file 3). The latter two lists were compared to each other and also to the main list (57 probes) by generating a Venn diagram (Figure 4). Nine genes (bola-nc1, ccdc125, eif4ebp1, kdm4b, mapre1, ppapdc1b, ppil3, timm8a, tmem87) and also loc784517 and an unannotated probe were common to the three lists. Forty other genes belonged to the Main List and one of the two single-challenge lists (Figure 4).

To gain further insight into the biological signification of the differences between the resistant and susceptible lines, information from the separate analyses in Sa an Se was added to the main list. The three probe lists were pooled together (Additional file3), resulting in a super-list of 335 annotated genes (380 probes). In this list, further named as the "pooled list", 209 and 169 genes were expressed at a higher level in the resistant and the susceptible animals, respectively.

To confirm the line effect, seventeen genes were analysed by qPCR. Genes were chosen because they belonged to the main list (cryl1, eif4ebp1, gtpbp4, mapre1, ppapdc1b, rarα, slc40a1, tmem87b and tp53), or had been identified in Sa (ccl5, itgb6, s100a2, saa2 and tlr2) or Se conditions (capn3, psmd4 and st3gal4) (Table 3). Real-time qPCR confirmed the significant (p < 0.05) differential expression of eight genes (capn3, cryl1, itgb6, psmd4, rarα, saa2, st3gal4 and tp53). Six others (ccl5, gtpbp4, ppapdc1b, s100a2, slc40a1 and tmem87b) were close to signification (p < 0.10). The genes tlr2, eif4ebp1 and mapre1 were not significantly confirmed by qPCR (0.20 < p < 0.30), however the relative gene expression between resistant and susceptible was in accordance with microarray analysis (Table 3).

Principal component analysis (PCA) of the probes from the pooled list revealed that the first two principal components, that represented 19% of the total variations, could separate the sheep samples into three clusters: resistant animals infected by Sa, susceptible animals infected by Sa and animals infected by Se (Figure 5A). The principal component 1 (PC1) explained 11.2% of the total variations and clearly discriminated the Staphylococcus spp within the differentially expressed genes between the lines (Figure 5A). The gene expression of capn3 was mainly associated with Se challenge whereas the gene expression of ccl5 and cd36 was linked to Sa challenge (Figure 5B). The PC2 explained 7.8% of the total variations and generally tended to separate the resistant and susceptible animals whatever the challenge (Figure 5A). Gene expression of tp53, tlr2, map3k3, selplg and bola-nc1 was associated with the resistant animals whereas gene expression of plekhb2, tmem87b and csf3 was linked to the susceptible ones (Figure 5B).

Additional biological information was obtained using the Database for Annotation, Visualization and Integrated Discovery (DAVID v6.7) with human ortholog gene names - HUGO nomenclature (n = 306 out of 335 genes recognized by DAVID). First, Gene Ontology (GO) analysis of the biological process terms was performed. The up-regulated genes in the resistant line showed a significant enrichment of leukocyte adhesion and activation, cell death regulation, intracellular signalling cascades and negative regulation of macromolecule metabolic processes and catalytic activities (Additional file 4). On the contrary, the down-regulated genes showed a significant enrichment in positive regulation of inflammatory response (p = 0.029) (Additional file 4). Both up and down-regulated genes showed an over-representation of genes involved in the regulation of transcription and RNA maturation and in cell motion (Additional file 4).

Two BIOCARTA pathways were well-represented: apoptotic signalling in response to DNA damage (p = 0.029, with akt1, eif2s1, cycs, tp53), and adhesion and diapedesis of granulocytes (p = 0.087, with csf3, selp, selplg). Four KEGG pathways were also identified: lysosome (p = 0.010, with tcirg1, sgsh, cd68, smpd1, ctsd, ctsa, ctsb, fuca1), adipocytokine signalling pathway (p = 0.012, with akt1, cd36, mapk8, acsl4, acsl3, acsl5), hematopoietic cell lineage (p = 0.032, with csf3, cd36, itga5, cd59, itga2, csf1r) and focal adhesion (p = 0.054, with akt1, lama3, ccnd3, actn4, itga5, fyn, itgb6, itga2, mapk8).

Systemic identification and grouping of line-associated genes into biological networks was performed with IPA using HUGO names (n = 331 out of 335 genes, recognized by IPA). Five networks were obtained with a score superior to 30. Twenty-two to twenty-eight genes were involved in each network. The two first networks are presented in the Figure 6. Network 1 is characterised by lipid metabolism, molecular transport and small molecule biochemistry (Figure 6A). It presents tp53, the second most up-regulated gene in the resistant line (Table 2 and Additional file 3), as a hub which means that tp53 regulates or is regulated by a large proportion of the identified DE genes (Figure 6A). Network 2 is defined by cellular movement, haematological system development and function, and immune cell trafficking (Figure 6B). It highlights a central position for tlr2, linked to cd36, that are of major importance for pathogen recognition. Other genes were included in this network eif4ebp1 and itga5, nfkb (immune response), ccl5 (diapedesis) and rarα and ahr involved in the retinoic acid pathway (Figure 6B).

To enhance the understanding of the genetic basis of the DE genes, the potential transcription factors that can regulate these genes were investigated by looking for transcription factor-binding sites (TFBPS) that were over-represented in the promoter sequences of the DE genes. Using 260 genes recognised in InnateDB, significant enrichment was demonstrated for 11 transcription factors with a p < 0.05 (CEBPB, PAX5, E4F1, CREB1, ETV7, HIF-1, SMAD1, FOXO4, NRF-2, NRF1 and NFIC), and nine other transcription factors with p < 0.10 (AP-1, ETS1, 120-kDa CRE-binding protein, FOXO1A, NKX2-5, ATF, ATF3, XBP1 and PBX1). Furthermore, the transcription factors AHR, which showed lower expression levels in the resistant animals, and CEBPA, which showed higher expression levels, could bind the promoters of thirty-three and seven genes, respectively, that were more expressed in the susceptible line (p = 0.07 and p < 0.1, respectively). TP53, which showed higher expression levels in the resistant animals, could bind fourteen up-regulated genes (p = 0.16).

Some soluble factors such as chemokines form gradients from the blood towards infected sites and can enhance neutrophil attraction and migration. In our study, we found no expression differences for major chemokines such as il-1β, tumour necrosis factor alpha (tnfα), il-6 and il-8 (also named cxcl8, one of the main chemoattractants for neutrophils 4 ), contrary to Griesbeck-Zilch in MEC of mastitis resistant cows selected on QTL alleles 31 . However, integrin beta 6 (itgb6), serum amyloid A 2 (saa2), a marker of acute inflammation 32 33 , S100 calcium binding protein A2 (s100a2), a member of the S100 family that is highly correlated with somatic cell count 10 , and the chemokine C-C motif ligand 5 (ccl5, also named rantes) that triggers adhesion and transmigration of blood monocytes to/through endothelial cells were expressed at higher levels in susceptible animals. On the contrary, the expression of calpain 3 (capn3) was higher in resistant animals. This protein was shown to play a role in resting neutrophils and to operate as a negative regulator of protrusion and migration 34 . In the present study, although pro-inflammatory molecules have been widely shown to play an important role in neutrophil recruitment and IMI outcome 7 14 28 , their expression patterns at 12 hours post-challenge were contrasted in resistant and susceptible lines and not correlated to differences in milk cell concentrations.

Furthermore, a group of genes related to cell adhesion and movement, including ST3 beta-galactoside alpha-2,3-sialyltransferase 4 (st3gal4), but also activated leukocyte cell adhesion molecule (alcam), integrin alpha 5 (itga5), selectin platelet (selp) and its ligand (selplg), were over-expressed in the resistant line. st3gal4 is involved in the synthesis of selectin ligand 35 . The selectins are fundamental for the attachment of neutrophils to the endothelial cell surface 4 36 before diapedesis. These results suggest that neutrophil diapedesis might be more efficient in the resistant line.

The genes cd36 and tlr2 that were expressed at a higher level in resistant animals and that are central in the network analysis collaborate together in the Toll-like receptor (TLR) signalling pathway 37 . TLR2 is dedicated to Gram-positive bacteria recognition 38 and it may enhance the activation of neutrophil phagocytosis 39 . Griesbeck-Hilch et al. 31 also found that tlr2 was up-regulated in mastitis resistant cows selected on markers for one QTL. The TLR2 signalling pathway may activate a cascade of intracellular events that may initiate the transcription of pro-inflammatory cytokine genes through the ERK/MAPK signalling pathway with consistent up-regulation of eif4ebp1 and itga5 in the resistant animals. This pathway is involved in oxidative stress which has been shown to contribute to the variability of susceptibility to IMI in cattle 6 . Altogether, the modification of TLR2 and MAPK signalling pathways might be responsible for a higher uptake of bacteria by phagocytes, and therefore might contribute to the more efficient clearance of the infection in the resistant line.

A noteworthy fact is that considerable differences between lines were related to transcriptional activity within MSC. The activator protein 1 (AP-1) transcription factor is considered as an immediate-early response gene and is thought to be involved in a wide range of transcriptional regulatory processes linked to cellular proliferation and differentiation. RARα, TP53, and AHR were also transcription factors of interest.

RARα, the alpha receptor of retinoic acid, a compound derived from vitamin A metabolism, is a nuclear receptor. It can affect several aspects of innate immunity by enhancing the function of neutrophils, macrophages and natural killer cells 40 . It has also been shown to play a role in helping the development of T helper cells, B-cells and, thus antibody-mediated response 40 , and more recently in modulating antigen-presenting cells 41 . Retinoic acid metabolism has been shown to play a part in IMI outcome, and dietary vitamin A supplementation has been reported to have a protective effect against experimental S. aureus mastitis in mice 42 . Moreover, we found that the aryl hydrocarbon receptor (ahr) was under-expressed, which is consistent with several studies that have demonstrated its interaction with RA-signalling pathways 43 . The involvement of AHR in the control of inflammatory responsiveness has been reported previously 44 . To our knowledge the role of TP53 in the response to bacterial infection has yet to be studied whereas it has been shown elsewhere to be associated with mastitis infected quarters 45 . tp53 was the gene that displayed the second highest ratio between lines and was over-expressed in the resistant line; it was highly associated with resistant animals in PCA and was represented as a hub in IPA networks since it interacts with numerous other DE genes. tp53 can be regulated by S100A2 proteins 46 and JNK 47 , and in vitro it was shown to regulate TNFα and other cytokines 48 . TP53 has been largely studied in the cancer context for its role in cell cycle arrest, apoptosis, DNA repair and production of antioxidants 49 . Furthermore, TP53 has already been shown to play a pivotal role in determining cellular response to stress via NFκB 50 and TNFα 48 ; it may also enhance transcription of the complement regulator CD59 51 . These observations suggest an important role of TP53 during bacterial infections.

In addition, particular cell functions such as granulopoiesis, cell proliferation and apoptosis seemed to explain some of the differences between the lines. Strong increases in neutrophil efflux from the bone marrow are followed by intense granulopoiesis and efflux of band cells that will later be recruited at the infection site: these cells may show higher transcriptional activity than mature neutrophils. Gene expression analysis demonstrated that the transcription factors CCAAT/enhancer binding protein (C/EBP), alpha (CEBPA) and RARα were up-regulated in the resistant line. CEBPA is crucial for the differentiation of granulocytes 52 and RARα has been shown to be directly involved in some aspects of the immune response by enhancing granulopoiesis 53 54 . Collins et al. (60) showed that RARα can enhance granulocytic differentiation through a molecular pathway that is independent of CEBPA 55 suggesting that two different pathways stimulating granulopoiesis were up-regulated in the resistant line.

Furthermore, a number of genes differentially expressed between the sheep lines was related to cell proliferation and apoptosis, e.g. mapre1, znf259, gzmh cryl1 and tp53. These genes exhibited amongst the highest expression differences between lines. The genes encoding tp53, and the lambda-crystallin protein (cryl1) were expressed up to five times more in the resistant line. Recently, Cheng et al. (61) evidenced association between the expression of CRYL1 and inhibition of cellular proliferation and cell growth 56 . This is in agreement with the decreased expression of both microtubule associated protein RP/EB family member 1 (mapre1) and zinc finger protein (znf259) in resistant sheep. Indeed, ZNF259 (homologous to ZPR1) has previously been shown to accumulate in the nucleus of proliferating cells 57 and MAPRE1 has been associated during mitosis, with the centrosomes and spindle microtubules. Apoptosis is a critical process necessary to limit or terminate inflammation 6 58 and has previously been shown to be of importance in the response to S. aureus by Lutzow et al. 10 . The gene expression of tp53 has been shown to increase in neutrophils during apoptosis 59 . These results suggest that cell proliferation is limited and apoptosis increased in resistant sheep as compared to susceptible sheep. Altogether data support the hypothesis that, as a consequence of apoptosis/granulopoiesis, the cell turn-over may be enhanced in the resistant line when compared to the susceptible line.

Collectively, these findings highlight leukocyte adhesion and cell migration, pathogen recognition through the TLR2 signalling pathway, and cell turn-over with the balance between apoptosis and granulopoiesis as possible mechanisms to explain a higher susceptibility or resistance to intramammary infection. Nevertheless they probably give only a partial view as other cell types and compartments may be involved (epithelial cells, dendritic cells, lymphocytes, etc). Also, other conditions (bacteria strain, time point) might have provided some different results. However, these results advocate the use of our differentially expressed gene list as a benchmark to more detailed genetic studies, including genome co-localisation of resistance to mastitis using QTL analysis as reviewed previously 19 22 23 24 25 and polymorphism studies.

Total RNA was extracted from the 24 cell samples - six ewes from two lines for two challenges - using a typical phenol/chloroform extraction method with Trizol reagent (Invitrogen). Extracted RNA was further purified on Qiagen RNeasy columns (Mini kit, Qiagen). RNA quality was assessed using an Agilent 2100 BioAnalyzer and the RIN (RNA Integrity Number) index was above 7 for all samples.

For each of the 24 samples, 200 ng of RNA was converted into double-stranded cDNA using the Amino Allyl Message Amp II aRNA amplification procedure (Ambion kit). cDNA was then labelled with Cy3 and Cy5 to obtain 48 dyed samples. Samples were hybridised in a two-colour dye-switch experimental design (GenoToul, France, http://biopuce.insa-toulouse.fr/Maquette/en/) on 22 microarrays of the ovine oligonucleotide 019921 Agilent slide (Agilent Technology).

Chips were hybridised with labelled cDNA at 65°C for 17 hrs and then washed according to the Agilent Technologies protocol. Intensity values were recorded with a 4000B Axon scanner. Two channel images were imported into the Agilent Technology Feature Extraction software for feature spot finding and alignment, and data were normalised with a Loess procedure.

A total of 15,008 different probes are present on the Agilent ovine slide, but only 1,656 genes were annotated by Agilent (version available in January 2010). Moreover, 8,847 genes were identified as Human ortholog Gene Nomenclature Committee (HGNC) by SIGENAE (http://www.sigenae.org/ sheep oligo annotation version 5 of 2009/11/10) 61 . More information about unannotated probes was obtained through the Basic Local Alignment Search Tool programme on the NCBI website (http://blast.ncbi.nlm.nih.gov/Blast.cgi) and the ENSEMBL website (http://www.ensembl.org/index.html). After this annotation phase, only a few focus probes remained unannotated.

The Feature Extraction result files (.txt) were imported into GeneSpring^® GX 11 as single-channel values. Gene expression was analysed probe by probe using an intensity-based model - i.e. working on the intensity of spots and not on the ratio between conditions. This way of analysing two-colour-microarray data was shown to enhance the reproducibility of results and the sensitivity of the detection of DE genes 62 . Data were filtered according to spot intensity, saturation and uniformity. Genes were flagged individually for each of the four conditions: low SCS Sa, low SCS Se, high SCS Sa and high SCS Se. Only probes that were positively flagged in all samples for at least one condition were conserved for further analysis in order to keep only genes that were very representative of one condition (n = 9,098). Then data were normalised across arrays with the GeneSpring^® "scale to median" procedure.

Initially, a two-way ANOVA was performed for each probe to identify DE genes between the two Staphylococcus challenges (Sa and Se) and between the two divergent sheep lines (Low SCS and High SCS). The interaction between the Challenge and the Line effects was tested but was not significant; therefore, it was removed from the statistical model. The p-values of the tests were corrected with a 5% false discovery rate (FDR) with Benjamini-Hochberg 63 and genes with an absolute fold-change (aFC) superior to 1.5 were considered as differentially expressed. For the Line effect, FC represents the ratio between Low and High SCS; for the Challenge effects, it symbolises the ratio between Sa and Se. The experiment was deposited in GEO at the identifier number GSE24925 (BioArray Software Environment - version SIGENAE).

As 5,573 genes were differentially expressed for the Challenge effect, a second analysis was performed considering only samples infected by one Staphylococcus strain to focus on genetic differences between the lines. The whole raw data set was divided into two subsets: MSC from Sa and Se infections respectively. Data filtering and normalisation were performed as previously for the ANOVA model and 7,452 and 8,561 probes were retained for statistical analysis of the Se and Sa challenges, respectively. Then, an unpaired Mann Whitney test was performed probe by probe. Considering a 5% FDR no probe was identified as significantly differentially expressed between the lines probably because of the weak number of animals (six in each line). However to further explore the differences of mastitis susceptibility, the p-value threshold of statistics was relaxed. The p-value was rounded to two decimal places, then, the genes with a p-value ≤ 0.01 and an aFC greater than 1.5 were considered as differentially expressed.

Expression profiles for DE genes were classified using the hierarchical clustering algorithm in GeneSpring^® based on Pearson-centred gene distances to visualise the differences between the two conditions. They were also represented in principal component analysis (PCA) with the centred data in R (v. 2.9.0) to identify the most important genes to explain mastitis resistance or susceptibility.

Three software programmes were used to interpret the lists of focus genes obtained from statistical analysis: Ingenuity Pathway Analysis (IPA), Database for Annotation, Visualization and Integrated Discovery (DAVID v6.7) 64 65 and Innate Data Base (InnateDB) 30 .

IPA software (version 7.5, http://www.ingenuity.com/) was used to generate biological networks from a list of selected genes and to document the functions of these genes and the canonical pathways in which they are involved.

Gene Ontology analysis was performed using DAVID (http://david.abcc.ncifcrf.gov/) and led to the establishment of relationships between genes with similar biological functions. Transcription factors that potentially regulate several focus genes were identified with InnateDB (http://www.innatedb.com).

The expression of some differentially expressed genes was verified by qPCR. Primer pairs were designed using Primer3 66 based on the relevant ovine sequences and verified using Primer Express^® software. Their specificity was checked with BLAST (http://blast.ncbi.nlm.nih.gov/Blast.cgi). Primers were synthesised commercially by Eurogentec. For genes for which no ovine sequence was available, a comparative gene alignment of bovine, human, rat and mouse sequences was made and primers were then designed on the most conserved regions between the species. Absence of primer dimers was verified using melting curve analysis and the efficiency of the amplification was measured before use. The couples of primers used in qPCR experiments are listed in Additional file 5. qPCR reactions were performed on a 7300 Real-Time PCR System (Applied Biosystems). To validate genes differentially expressed between resistant and susceptible animals, qPCRs were performed either on both Sa and Se samples or only on Sa samples. All assays were carried out in duplicate and each reaction contained 5 μl of diluted cDNA (1:50) with 2.5 μl (0.5 μm) of each forward and reverse primer along with 12.5 μl of Power Syber Green PCR Master Mix (Applied Biosystems).

Specific amplification of each target was confirmed by melting curve analysis. Measured Ct values were exported from SDS software to Excel for data analysis. RT-qPCR technical replicates of samples were averaged. The stability of 7 housekeeping genes, previously cited in the literature, was checked in the 24 samples and data was analysed using GeNorm software 67 . The four most stable genes (rp19, hprt, sdh and gapdh) were selected for normalisation of RT-qPCR. Fold changes were calculated by the delta delta Ct method normalised to the four housekeeping genes 67 with R (version 2.9.0). Statistical analysis was performed using an exact non parametric Wilcoxon test with SAS (version 9.1).