It has proved to be difficult to reliably analyse in vivo the time course of the pathogen-specific immune response in the udder . Several groups have designed and standardized experimental conditions to reproducibly infect lactating udders from cows [6, 30, 31] and ewes  to generate meaningful results describing molecular mechanisms of host-pathogen interactions. However, infection of animals suffers from among-individual variation, especially regarding dairy ruminant species. In addition, the udder is a complex organ made of several cell types contributing differentially to its immune response . Therefore, to characterize the MEC response, relevant model cells such as primary bovine MEC have been proposed for the molecular analysis of immune defense mechanisms in the udder [12, 34, 35]. Their potential to express genes of immune defense in response to stimulation with heat-killed bacteria was recently characterized showing pbMEC is capable of secreting large amounts of mediators of inflammatory response, namely IL-8 and TNFα in a pathogen dose dependent fashion . Comparing the inflammation dependent regulation of factors contributing to the complement system between the udder and MEC to those from fully inflamed udders after infection with live E. coli pathogens also underlines some regulatory limitations of the pbMEC cell model, which does not properly reflect the mastitis induced regulation of chemokines and the complement system in the udder .
The present study is, to our knowledge, the first report describing the in vivo transcriptional host response of MEC in dairy ruminants, early after an IMI challenge with S. aureus. We assessed the contribution of MEC to initiate a pro-inflammatory response and to participate in the recruitment of neutrophils and macrophages at the beginning of staphylococcal infection. Usually, in vitro models are used to get an inside view of MEC function, but it remains a simplified and artificial model. A surrogate and less reductionist approach than the in vitro cellular and invasive biopsy methods is the use of “cytoplasmic crescents” enclosed in MFG, first proposed by Maningat et al.  in humans and recently validated in goats . However, RNA from MFG, which in theory only contain RNA from MEC, could be contaminated by RNA from infiltrating immune cells, whose population increases in milk during the infection process, and which can be trapped in the cream layer along MFG isolation. Our results show that up to 18 hpi, RNA extracted from MFG is uncontaminated by RNA from immune cells (Figure 3). The expression level of neutrophil and macrophage specific markers was 17 times less than in a non infected mammary gland whose immune cell population is not more than 5% in the cow mammary tissue . Conversely, at 24 hpi, data should be interpreted with care, and we considered it wiser not to treat MFG as fully representative of MEC due to the level of contamination of MFG by immune cells which can then not be negligible. RNA extracted from MFG reflects the specific physiological metabolic process in functional epithelial cells within the mammary gland, only until 18 hpi. However, regarding some genes known to be specifically and highly expressed in MEC, such as genes encoding milk proteins, RNA extracted from MFG remains representative of this cell type. As shown by qPCR experiments, LCM allows obtaining MEC enriched material with low contamination by MMC. In such a manner we were able to complete until 30 hpi our study of MEC response in their physiological context, even though this contamination was slightly higher than that previously reported by Bevilacqua et al.  on healthy tissue. This is probably due to tissue damage associated with inflammation leading to a loss of the secretory epithelium cohesion (Figure 4A). However, since MMC were not shown to be able to induce the production of chemoattractants for neutrophils , we can assume that this slight contamination of MEC by MMC has in fact little impact on changes in the expression profile induced by the infection and therefore does not interfere with the contribution of MEC in innate immune gene expression.
In addition, throughout the time course of the challenge we confirmed that mRNA molecules encoding proteins involved in the primary function of the udder, i.e. milk synthesis, were not affected by the experimental infection, thus demonstrating that there is no significant loss of function in the mammary tissue, at least during the first 24 h. Therefore, the experimental model measured changes in immune gene responses that are likely to be appropriate to study the initial stages of infection. On the contrary, it was clearly demonstrated in cows that signals are transmitted to the surrounding tissue and to the neighbouring uninfected quarters, very early (within the first 24 h) after contact with E. coli.
Transcriptome analyses provided here, even though performed with an incomplete repertoire of gene probes, revealed that gene expression profiles clearly changed at 18 hpi with a low number of differentially regulated genes (n = 39), concomitantly with a 5 × 106-fold increase of S. aureus bacterial count in milk from infected udders. If similar changes occurred in response to the bacterial infection in the neighbouring uninfected udder, as reported by Jensen et al.  in bovine, then such changes may have masked some of the differences between the infected and uninfected udder and this could be a reason why the list of differentially expressed genes is relatively short.
The rather late change in gene expression, only 18 hpi, could be explained by the delay to reach sufficient bacterial concentrations in milk. Sensing the presence of bacteria is an important component of innate immunity . Indeed, kinetic studies of experimental clinical mastitis induced by the intra-mammary infusion of S. aureus in ovines showed that the inflammatory response (leucocyte reaction) is not initiated until bacterial concentrations reach 4.104 bacteria/mL of milk . Differentially expressed genes, identified in the microarray screening, of which 15 were confirmed by qPCR, indicated that MEC respond to an IMI challenge with S. aureus by substantial increase in abundance of mRNA coding for a range of pro-inflammatory cytokines, chemotactic proteins, antimicrobial factors and acute phase proteins. Our results clearly demonstrate that MEC, which form the first line of defense against invading pathogens, are actually able to induce an up-regulation of immune-associated genes involved in the inflammatory response, thus substantiating previous in vitro studies [9, 10, 12, 13, 42]. We note that, as pointed out by Gunther et al.  in a comparison of primary culture of bovine MEC and udder, such a MEC model, though useful, is generally not considered to reflect what actually happens in vivo. It seems, however, that MFG provides more accurate information to analyze the MEC response, at least during the very first steps of an IMI challenge.
Furthermore, some of the genes found up-regulated in MEC play a key role in orchestrating the regulation of other immune effectors. For instance, the temporal regulatory network, built from the expression data using the Empirical Bayes Dynamic Bayesian Network (EBDBN) algorithm, underlines the first role of MEC in the immune response. This role consists of the recruitment of neutrophils through the up-regulation of PDE4B and TNFα and in the differentiation and proliferation of lymphocyte T through the up-regulation of GADD45α. More precisely, it was observed that the gene encoding the phosphodiesterase PDE4B, which is expressed in the bovine mammary gland , plays a key role in immune cell recruitment. Ariga et al.  showed that neutrophil recruitment to the lung is impaired in PDE4-deficient mice, by inactivating the second messenger, the cyclic adenosine monophosphate (cAMP), thus abrogating its negative effects on neutrophil recruitment and activation. Jin et al.  provide evidence that the gene encoding PDE4B, which is expressed in mouse peritoneal macrophages, is involved in the control of TLR signaling, significantly reducing TNFα expression.
The multifunctional pro-inflammatory cytokine TNFα was shown to induce the expression and release of IL-8 by bovine MEC . Our results which are consistent with this, show that genes encoding TNFα and IL-1β and, to a less extent, IL-6 which are typical pro-inflammatory mediators produced in response to TLR stimulation, are up-regulated at 18 hpi. IL-8, a potent chemokine capable of initiating an acute inflammatory response, plays a primary role in the recruitment of neutrophils into the gland . Thus, upon recognizing the pathogen, MEC are able to send out rapidly a strong signal to recruit cellular factors of immune defense (macrophages, neutrophils) from the blood stream into the infected mammary gland. Indeed, this change in gene expression was then followed, 24 hpi, by a massive arrival of cells in milk (Figure 2B), mainly represented by neutrophils .
Our results also suggest that the gene suppressor of cytokine signalling 3 (SOCS3) occupies a crucial position in the network arising from the temporal analysis. This gene, important for the mammary tissue homeostasis [50, 51], encodes an intracellular inhibitor of cytokine signaling that acts in a classical negative feedback loop [52, 53]. This is consistent with the fact that the up-regulation of different cytokine encoding genes such as IL-1β, TNFα and CCL4 by MEC is restricted at the very early stages of infection, before a decrease at 30 hpi. This short-lived up-regulation was also observed during in vitro studies . Nevertheless, in comparison with in vitro studies, after challenging primary bovine MEC cultures (pbMEC) with heat-inactivated preparation of S. aureus, the induction of cytokine-encoding genes such as IL-1β and IL-8 occurred later on in our study. Gunther et al.  also observed the same results with a stronger and earlier increase in mRNA abundance in pbMEC challenged with E. coli than that found in the udder as caused by acute mastitis. This ability to respond more quickly and strongly could be due to the fact that MEC in culture does not properly reflect the mastitis induced-regulation of cytokines in the udder and rapidly lose their functional features. The immune activity of MEC is regulated and balanced within the udder by factors locally delivered from other cell types . These observations reinforce the interest of studying MEC biology in its physiological context, which can be done using RNA extracted from MFG. This non-invasive technique allowed analyzing the in vivo contribution of MEC during the very first steps of infection on the same individual, thus allowing a study of time-course variations while accounting for individual variability.
On the contrary, increased expression of genes encoding molecules with bactericidal functions, such as S100A12 and PTX3, is the second very clearly identified immune function of MEC. PTX3 together with S100A12 that could assist defense of the mammary gland against chronic and subclinical infections, have also been reported to be up-regulated in milk somatic cells in response to S. aureus infection in goats  and the resulting proteins were shown to be present in bovine milk, helping to resolve the mammary tissue infection as well as potentially contributing to the maturation of the newborn calf epithelium and establishment of the newborn gut microbial population . Concomitantly with the early induction of cytokine and chemokine-encoding genes, we identified an up-regulation of two genes coding for acute phase proteins (APP): pentraxin 3 (PTX3) and α-1 anti-proteinase (SERPINA1) which were also found in bovine milk at the beginning of mastitis [39, 42]. We demonstrate here from LCM experiments that MEC contribute significantly to the increases of PTX3 and SERPINA1 transcripts, even after the burst of immune cells in milk. Therefore, as previously proposed for SERPINA1 [55, 56] which is a potent inhibitor of IL-8-induced hematopoietic stem cell mobilization , these two acute phase proteins could be useful markers in the early diagnosis of inflammation. We also observed a slight up-regulation of the gene encoding antimicrobial peptide S100A12 which is known to be induced by inflammatory cytokines such as IL-1β  and which demonstrates chemotactic activity, attracting circulating leucocytes in inflammation conditions . This protein, which probably amplifies the inflammatory response via a recruitment of neutrophils , was reported to occur during all but the early stage response . This is in agreement with our results. Indeed, it appears that MEC do not greatly contribute to its expression in the early stages of infection. However, the time-course of its response showed a clear tendency towards up-regulation from 24 hpi. Conversely, when the infection is well established, MEC are able to specifically express the acute phase protein SAA3 which can act directly on pathogens . The gene encoding SAA3 is mainly induced by IL1-β and TNFα . Studies have shown a growing interest for this protein and propose SAA3 as a good marker for mastitis. Interestingly, a high level of SAA has been observed in milk, in clinical  as well as in sub-clinical  bovine mastitis. Furthermore, whatever the dose and the invading pathogen, in vitro  as well as in vivo  studies reported high levels of mRNA molecules encoding SAA3 in response to infection or experimental challenge with E. coli or S. aureus cell wall components. However, 30 hpi we observed a very weakly (2.7 to 3-fold) enhanced expression of SAA3. This discrepancy could be explained, either by the fact that MEC in culture do not properly reflect MEC in their physiological context and/or that the udder is a complex tissue in which the dynamics of inflammation are different from that observed with a cellular model or even because the reading windows are not comparable. In previous studies (unpublished results), in agreement with the results reported by Eckersall et al.  in cattle, we observed that increases in acute phase SAA in milk of experimentally infected ewes occurred within 12 h after S. aureus infusion reaching a peak concentration at 72 h. Thirty hpi can still be considered as the onset of the IMI challenge with S. aureus, and consistent with our results, Eckersall et al.  who recorded at the early stage of infection (48 h) the expression of SAA3 (mRNA), reported a slight increase (3.6-fold) in abundance of mRNA molecules. To substantiate our results, it is worth noting that most of these proteins were found in proteomic studies performed from E. coli, S. aureus and S. uberis mastitis milk whey. In addition, it should be underscored that the pattern of the response to IMI is basically similar and the qualitative differences observed between species most likely reflect different basal conditions and the type of invading pathogen .
Finally, as previously demonstrated in an in vitro study , we confirmed in vivo that the expression of TLR2, which is a key component in immune recognition of gram-positive bacteria by host cells, recognizing a wide spectrum of microbial components , was slightly affected by the infection (2.9-fold up-regulated 30 hpi) and essentially expressed by MEC (80% contribution). Thus, it is likely that MEC contain a fully functional and constitutively active Toll-Like Receptor signalling pathway that is slightly induced by the bacterial challenge but immediately responsive. Therefore, MEC have an intrinsic role in innate immune surveillance of the mammary tissue. The primary function is recruiting immune cells and the recognition likely occurs via the MyD88-dependent TLR signaling pathway. Stimulation of this pathway is considered the main mechanism enhancing expression of pro-inflammatory cytokines and activation of the innate immune response . TLR2 stimulation triggers intracellular signalling cascades leading to the activation of NFκB which, in turn, leads to the activation of several genes encoding pro-inflammatory mediators such as TNFα, IL-1β and IL-6. In addition, TLR2 stimulation engages the production and release by MEC of chemokines such as IL-8, a potent chemo-attractant and activating factor of neutrophils, and, to a less extent, of cytokines CCL2 and CCL4 which display chemotaxis activity for monocytes and macrophages. Interestingly, due to a considerable among-goat variation, a number of genes possibly involved in MEC response to IMI challenge with S. aureus, has not emerged from the microarray analysis. Therefore, we analyzed some relevant genes, known to play crucial roles in the immune function, by comparing their gene expression profiles between goats, during the course of the infection. Thus, MyD88, an adapter protein involved in the Toll-Like Receptor and IL-1R signaling pathway, with a sharp increase at 24 h seemed to follow the same dynamic trends (Figure 9) as S100A12, whereas NFκB and IRAK4 (essential in the activation of NFκB), showed a strong increase at 18 hpi, before rapidly declining at 24 hpi, at least for 3 of the 5 goats analyzed in this study.
These data collectively demonstrate that we can easily access MEC biology in its physiological context via MFG. This non-invasive technique has allowed assessing the contribution of MEC during the first steps of infection, in vivo. Our results suggest that the MEC response to IMI challenge involves both genes likely to affect pathogen function as well as genes (e.g. cytokines) that alter the behaviour of other cell types. Consistent with others, the picture emerging from this study is that the mammary epithelium is not simply a mechanical barrier but rather a functionally complex tissue capable of responding effectively to the intrusion of pathogens by altering its own gene expression profile and possibly that of other cell types to favour recruitment of immune cells and to synthesize bactericidal molecules. We report here a novel strategy of sampling to monitor the dynamics of gene expression in MEC which allows going further into the understanding of MEC immune capacity. Furthermore, modulation of these roles could be of importance in determining the outcome of an infection. The chronologically induced synthesis of cytokines at the inflammation site is important for pathogen clearance, wound healing and return to normal conditions. This approach could allow a better understanding of MEC functions in animals showing different levels of genetic predisposition to mastitis, as tested with an in vitro study on sheep . From a diagnosis point of view, several candidates for an early detection of mastitis were found.
Despite incomplete and imperfect annotation (the probe with the highest differential expression unfortunately remains unknown), which may explain why some effectors known to be induced in response to infection were not identified in this study (e.g. epithelial β-defensins such as LAP of which the expression is delayed  as well as other antibacterial molecules), the repertoire of gene probes used (Sheep Gene Expression Microarray, 8 × 15 K) has nevertheless confirmed the involvement of a number of master cytokines and chemokines (IL-8, IL-1β). Regarding IL-1β, which is first synthesized as biologically inactive pro-IL-1β, its processing into mature, biologically active pro-inflammatory cytokine supposes activation by caspase-1 which remains to be reported for MEC, before being released in the extracellular milieu following a non-classical secretory pathway . One cannot take for granted that mRNA detection guaranties protein expression. The application of an original algorithm to construct gene networks of temporal regulation revealed the involvement of several factors known to play an important role in the inflammatory response, but for which we did not suspect that the MEC could be providers. In a recent review  and two proteomic studies [75, 76], increasing of low abundant proteins such as IL-8, CD14, SAA, S100A12 and PTX3, were reported in mastitis milk from ewes and cows experimentally infected by E. coli, S. aureus and S. uberis or challenged with LPS, thus corroborating our transcriptional findings. Finally, the non-invasive sampling method (RNA extraction from MFG) provided an opportunity to perform a dynamic study of IMI, and this in spite of the significant individual variability observed which can in such a manner be partly bypassed. It also gives the opportunity to achieve a large-scale validation of the results in a significant number of individuals, which we plan to do in the near future.