Escherichia coli infection induces distinct local and systemic transcriptome responses in the mammary gland

Establishment of intramammary infection

Prior to infection, the udder secretions of all animals had repeatedly been tested negative for major and minor mastitis pathogens. Following administration of 500 cfu of E. coli 1303 to a single udder quarter and the same volume of physiological saline into a neighboring quarter of each of five cows (Figure 1), all inoculated animals suffered from acute mastitis and showed a strong cellular response. Major clinical symptoms, such as udder swelling and elevated counts of somatic cells in milk, were restricted to pathogen-inoculated udder quarters (Figure 2A). All milk samples from the inoculated quarters were positive for E. coli bacteria whereas neighboring quarters and all quarters of control animals remained bacteriologically negative during the observation period.

Body temperature, milk yield, differential blood cell counts

All infected cows developed fever (>39.2°C) within the first 14 h after infection with maximal values (41.1 ± 0.2°C) between 15 h and 18 h after inoculation (Figure 2B). This was accompanied by significant leukopenia, reaching minimal blood leukocyte counts 12 h after pathogen inoculation (Figure 2C). Milk yield dropped progressively within the 24-h period after infection (Figure 2D), with a stronger decrease in infected quarters than in neighboring quarters. In infected quarters the number of somatic cells in milk increased starting between 6 and 12 h after infection (Figure 2A). In mock inoculated and untreated neighboring udder quarters of infected animals no significant increase in the number of somatic cells was observed.

Microarray analysis

Mammary tissue samples were used to generate four sets of transcriptome profiles from: i) inoculated quarters 6 h post infection; ii) inoculated quarters 24 h post infection; iii) noninfected quarters neighboring the inoculated quarters 24 h post infection; and iv) quarters of untreated healthy control animals. Samples of sets iii and iv were generated by combining tissues of two udder quarters from each cow in order to equalize potential quarter specific effects. Based on somatic cell counts there was no indication for a different behavior of mock inoculated and untreated udder quarters of infected cows. Therefore tissues of two quarters with the lowest somatic cell counts were combined (see Additional File 1) provided their cell counts did not differ by more than 10.000 per ml. Otherwise tissue from the udder quarter with the lowest cell count was used. Thus four samples of set iii were pooled from one mock inoculated and one untreated udder quarter. The fifth sample (sample ID n24h-4) was derived from tissue of a single untreated quarter since the noninfected udder quarters of this cow differed in somatic cells counts.

The impact of the E. coli infection on gene expression was estimated by hierarchical clustering of the transcriptome profiles (Figure 3). The heatmap obtained with the transcriptome profiles 6 h after inoculation showed a uniform coloration with no separation of healthy controls and infected samples (Figure 3A). This indicated that the gene expression of the mammary tissue was only marginally affected in the early stage of infection. The same comparison 24 h after infection with E. coli resulted in a sharp separation of the infected and the control group (Figure 3B) and a distinct blue-red coloration indicated substantial transcriptome changes. At the same time point, the effect of E. coli on transcriptome changes in neighboring noninfected udder quarters was less pronounced (Figure 3C). Compared with Figure 3B, the length of the vertical branch between the two principal groups is considerable lower, indicating a less pronounced difference of the expression profiles.

Nevertheless two almost perfectly separated groups were obtained: three of four noninfected samples pooled from one mock inoculated and one untreated udder quarter (n24h-1, -2, -3) clustered together with the sample derived from a single untreated quarter (n24-4), while the fourth pooled sample (n24h-5) was placed into the well-separated control group.

DEG (see Additional File 2) were identified by comparison of transcriptome profiles of infected and neighboring quarters with the profiles of the control animals. The highest number of DEG was found 24 h post infection in infected quarters (2154 at 0.01 fdr), followed by 476 (0.01 fdr) in their neighboring noninfected quarters and only 13 (0.1 fdr) in infected quarters at 6 h after inoculation. Despite the noisy data at 6 h after infection 6 of the DEG were also detected as differentially expressed at 24 h in infected quarters and 2 in neighboring quarters. At 24 h post infection, 294 DEG of the infected udder quarters were found among the DEG of the neighboring noninfected quarters.

This overlap indicates either the existence of a systemic response to a local infection with E. coli or a local effect after application of 2 ml PBS in mock inoculated and infected udder quarters. The first reaction should take place in all udder quarters of infected cows, the second however should be absent in their untreated ones.

Local and systemic responses to E. coli revealed by cluster analysis

In order to look for local and systemic responses to E. coli infection, all DEG detected in infected quarters (2154) and noninfected neighboring quarters (476) were subjected to self organizing tree algorithm (SOTA) [18] clustering (Figure 4).

The calculated clusters revealed distinct reactions of the mammary gland in response to inoculation with E. coli. The first occurred exclusively in infected quarters (clusters 1 and 6) and comprised the local restricted response of the infected tissue. The second reaction took place both in infected and neighboring noninfected udder quarters (clusters 2 and 3) and pointed to a systemic response of the mammary gland to inoculation with E. coli in 2 ml PBS. Another type of reaction was recorded in the clusters 4 and 5. These 78 DGE were found induced specifically in the neighboring quarters, whereas in the infected quarters they were repressed (cluster 4) or unchanged (cluster 5). Their transcription was apparently regulated in a location specific manner. Nevertheless the different transcript levels in infected and neighboring noninfected udder quarters indicated that systemic signals are involved in their regulation.

The extent of apparently systemic gene regulation displayed in clusters 2 und 3 is much greater than the overlap of DEG from infected and neighboring quarters (294 genes). Roughly half of the 2335 genes fall into clusters that showed expression changes in infected as well as in neighboring quarters. This discrepancy is caused by the fact that DEG in infected quarters generally showed higher changes in expression than DEG in neighboring quarters. Due to the fold-change cut-off, genes with low fold-change are not assigned to the DEG but are nevertheless considered by SOTA clustering. Conversely, some genes differentially expressed in infected as well as in neighboring noninfected quarters were clustered by SOTA with the genes expressed locally in infected quarters (Figure 4, cluster 1), most likely because of their much higher induction in infected tissue than in tissue of neighboring quarters (e.g. S100A9, S100A12, SAA3, SDS, CP). For interpretation, those DEG, however, were treated as systemically reacting since the differential gene expression was regarded superior to clustering.

Based on annotations of the DEG more detailed analyses were performed. The genes of each cluster were analyzed in search of regulatory elements within their promoter region. Furthermore we identified biological processes involved in the systemic or local response of the mammary gland to the pathogen, which were represented by significantly more genes than expected by random sampling.

Characterization of the local response 24 h after infection

Analysis of the enrichment of functional annotations revealed that the locally up-regulated genes of cluster 1 (Figure 4) have a distinct profile of functions that is not found among the members of the other clusters and that is dominated very prominently by ontology terms related to immune response and inflammation. In addition ontology terms like chemokine/cytokine signaling and diapedesis were among the most prominent. This correlated with our clinical findings of a sharp increase in leukocytes solely in the milk of infected udder quarters. Many genes of this cluster were associated with classical inflammatory events (Figure 5) and TFBS enrichment analysis (Table 1) predicted the classical inflammatory transcription factors NFκB and STAT1 as regulatory elements. Major genes related to fever induction, IL6 and IL1B, were still up-regulated (16-fold and 4-fold respectively) in infected udder quarters 24 h after infection although the inner body temperature was already declining and approached basal levels at this time point. Furthermore, the majority of observed acute phase genes (e.g. HP, PTX3, FTL) was assigned to cluster 1. As evidenced in Figure 5, almost all DEG of the ontology response to infection were found locally restricted to the infected quarters and only a few were observed also in neighboring quarters. The highly significant enrichment of biologically meaningful terms (e.g. adjusted p-value 1e-38 for 'immunity and defense') is also an important indication for the validity of the presented microarray data.

Besides the response to infection TFBS enrichment analysis gave hints towards other processes. Among the genes of cluster 1, ELF5 was predicted as involved transcription factor. ELF5 is related to development of mammary epithelium [19] and milk production [20].

Characterization of the apparent systemic response 24 h after infection

The transcriptional response in the quarters neighboring the infected ones could in part be characterized as an immune response. Components of the antigen processing and presenting machinery as well as some cytokines are up-regulated. Of note are CXCL2 and CXCL14, whose products show direct antimicrobial properties against E. coli[21] and could probably function as a barrier against invasion of E. coli into the neighboring udder quarters. DEG of other important immune-related agents, as for example the CC-type of chemotactic cytokines, were restricted to the infected udder quarter. A number of genes, like GPX3, GLRX2, GSTM1, GSS, MT1A, MT2A and SOD2, indicate the presence of oxidative stress. Differential gene expression related to xenobiotic stress is evident by up-regulation of ATF5, which is responsible for the induction of the detoxifying cytochrome CYP2B6 [22], the latter being up-regulated only in the neighboring quarters (cluster 2).

Finally, there are signs of building up a first line of defense, like the up-regulation of genes with antimicrobial functions (S100A8/9/12, GNLY, CXCL2, CXCL14) or acute phase genes (SAA3) involved in recognition of pathogens (LBP) or in the complement cascade (C6, BF, C4BPA, IF).

NFIL3 as a candidate for signal transduction

Among the predicted transcription factors for the DEG, NFIL3 was the most interesting candidate. Its binding motif was enriched in the clusters of both systemically up- and down-regulated DEG (clusters 2 and 3), in line with its function as activator/repressor of inflammatory and apoptotic genes. Furthermore, NFIL3 was found in infected quarters as 3.5-fold up-regulated DEG 24 h after infection. Interestingly, up-regulation of NFIL3 expression was found as soon as 6 h after infection. Five of the DEG at 6 h belong to the apparently systemically regulated genes at 24 h and only 3 of them were found local in infected udder quarters. Interestingly at 6 h the DEG also comprised a couple of predicted NFIL3 target genes, and TFBS for NFIL3 were significantly enriched. The putative target genes regulated by NFIL3 suggest a systemic up-regulation of genes involved in cell proliferation (e.g. KLF6, FGF2) as well as apoptosis and a down-regulation of genes related to cell adhesion and extracellular matrix.

Analyses of mRNA expression by quantitative real-time RT-PCR

In order to investigate the influence of the inoculation solution (2 ml of physiological saline), the expression of the selected 14 genes was determined in all sampled udder quarters of infected animals and the values of untreated udder quarters and mock inoculated udder quarters were compared. As shown in Additional File 4, the treatment of the neighboring udder quarters with saline did not provoke a significant difference in expression of these 14 genes.

The results of qPCR allowed an estimation of animal specific variation of gene expression. The ΔCP values summarized in Additional File 5 clearly demonstrate that the transcript levels of some genes like HP or MT2A differed up to one order of magnitude in infected udder quarters 24 h hours after inoculation with E. coli 1303, whereas other genes like PIR or LBP were found to be uniformly expressed among the different animals.

Animals

This study included 15 healthy German Holstein Frisian heifers in mid lactation (3 to 6 months post partum). The trials were conducted at the Clinic for Ruminants, LMU Munich (Oberschleißheim, Germany) with the approval of the ethics committee of the regional government of Upper Bavaria, Germany (No. 55.2-1-54-2531-108-05). Only animals without previous diagnosis of clinical or subclinical mastitis and a reported somatic cell count <50,000/ml were included in the study. Quarter milk samples were collected and tested weekly before the trial to ensure that they contained <50,000 somatic cells/ml and were free of mastitis pathogens. Two different infection models were used in which the animals were inoculated in one quarter with E. coli and killed after 6 h (n = 5) or 24 h (n = 5), respectively. Five heifers served as controls, received no treatment and were killed after 24 h.

Somatic cell count (SCC) and bacteriological examination

Three weeks before the trial the following parameters were documented twice daily for each animal: milk yield, somatic cell count (SCC, California Mastitis Test, CMT) and rectal body temperature. Quarter milk samples were collected weekly for determination of the SCC and bacteriological examination. The SCC was determined with a Fossomatic 5000^® (FOSS Electric, Hillerod, Denmark) cell counter. To check for the absence of bacteria, 15 μl quarter milk samples were plated on Columbia sheep blood agar, Gassner agar and Edwards agar (Oxoid, Wesel, Germany). Samples were incubated for 48 h at 37°C.

Inoculum dose

Bovine mastitis isolate E. coli 1303 belongs to the major E. coli phylogenetic group A (E. coli collection of reference strains, ECOR-A). According to a multiplex PCR-based screening for virulence-associated genes of pathogenic E. coli, this strain does not represent an extraintestinal or intestinal pathogenic E. coli isolate. Only the genes coding for type 1 fimbriae, F17 fimbriae, antigen 43, the ferric citrate siderophore system and the EAST1 toxin could be detected. Bacteria were kept cryo-conserved (Mikrobank-System Cryobank™, Mast Diagnostika, Reinfeld, Germany) for subsequent infections. Bacteria were plated on Columbia sheep blood agar and incubated (37°C) for 24 h. A few colonies were transferred to a tube of brain-heart infusion broth (Oxoid, Wesel, Germany) and incubated for 6 h (37°C); then a 100-μl sample was transferred to a tube of trypticase soy broth (9.9 ml, Oxoid, Wesel, Germany). Serial dilutions were made after 18 h to prepare the desired inoculum dose of 500 cfu/2 ml 0.9% sterile, pyrogen-free saline. The inoculum dose was plated for control and ranged from 421-613 cfu/2 ml.

Experimental pathogen inoculation, sampling of milk, blood, and tissue

To achieve comparable hormonal conditions, cows were synchronized by two injections of (+)-cloprostenol (Dalmazin^®, Selectavet, Weyarn, Germany) at 12-day intervals. Animals were challenged three days after the second injection during estrus.20 IU oxytocin (Veyx-Pharma, Schwarzenborn, Germany) were applied intravenously and the udder was entirely milked out. Teats were cleaned and disinfected with 70% ethanol and 500 cfu E. coli strain 1303 were administered intracisternally in one quarter through the teat canal. Two milliliters of 0.9% sterile pyrogen-free saline without bacteria were inoculated into another quarter as placebo. The control group was treated the same way but without inoculation of an udder quarter.

The udder secretions were sampled in 12 h intervals before regular milking. Bacteriological examination was performed from foremilk samples, the SCC was determined in 10 ml whole milk after each udder quarter had been milked separately with a quarter milker (WestfaliaSurge, Bönen, Germany).

Blood samples (10 ml) were taken from the jugular vein aseptically using EDTA-vacutainers (Becton Dickinson, Heidelberg, Germany) at 0, 3, 6, 12 and 24 h after start of the trial. Finally, cows were killed 6 h or 24 h after trial start with a penetrating captive bolt gun followed by exsanguination. Leukocyte counts in blood samples were carried out with a Sysmex pocH100i (Sysmex, Norderstedt, Germany). Tissue samples were collected aseptically from slaughtered cows within 10 min after killing. A piece of tissue (5 × 5 × 5 cm) was removed from a deeper location of the udder quarter, 7 cm dorsal of the milk cistern, of which a smaller tissue piece (1 × 1 × 0.5 cm) was transferred to a tube containing 5 ml RNAlater^® (Applied Biosystems Incorporation, Foster City, USA) and served as original material for all further analyses.

Microarray hybridization

Total RNA was isolated by disrupting small pieces of approx. 500 mg of bovine mammary gland in 5 ml of TRIzol reagent (Invitrogen, Carlsbad, USA) with a tissue homogenizer (Heidolph, Schwabach, Germany). Homogenate was processed following the TRIzol manufacturer's protocol to yield highly pure total RNA. Purity (A260/A280 >1.9, A260/A230 >2.2) and integrity of RNA (28S:18S rRNA ratio ~1.5:1) was assessed by spectrophotometry (Nanodrop ND-100, NanoDrop Technologies LLC, Wilmington, DE, USA) and agarose gel electrophoresis.

Labeled cRNA probes for array hybridization of the 24 h samples (infected udder quarters, neighboring quarters and controls) were prepared with an Ambion Kit (MessageAmp™ II-Biotin Enhanced Single Round aRNA Amplification Kit). The probes for analyses of the 6 h time point (infected udder quarters and controls) were prepared with an Affymetrix kit (GeneChip^® Expression 3' Amplification One-Cycle Target Labeling Kit). RNA samples from noninfected udder quarters of the same animal were pooled when they had comparable somatic cells counts (detailed description of samples: see Additional File 1). The controls were performed as replicates to ensure comparability of the sets.

10 μg of the resulting fragmented cRNA were hybridized overnight to GeneChip^® Bovine Genome Arrays (Affymetrix, Santa Clara, CA, USA) processed in an Affymetrix Fluidics Station 450 and scanned with the Affymetrix 3000 7G scanner. Hybridizations were performed in three sets of 10 arrays each: (set 1) 5 infected quarters 24 h after infection and 5 controls; (set 2) 5 samples from neighboring quarters 24 h after infection (see Additional File 1) and 5 controls; (set 3) 5 challenged quarters 6 h after infection and 5 samples from untreated controls. The statistic software R and packages from the bioconductor [38] microarray suite (affy, affyPLM) were used for processing the CEL files and to identify possible outliers due to technical or handling artifacts. The microarray data has been deposited in the Gene Expression Omnibus database (accession GSE15025).

Statistical analysis and bioinformatics

Microarray raw data were summarized and normalized with the RMA method [39] and filtered for a detection criterion (MAS5 'present'-calls determined by Bioconductor's affy-package in R). DEG were identified by the packages LIMMA [40] and SAM [41]. LIMMA was performed for the analysis of the sample sets 24 h after infection, excluding profiles of animal 5 (samples i24h-5 and n24h-5) as a biological outlier and control animal 3 (sample c₁24h-3 and c₂24h-3) as a technical outlier. Due to its higher sensitivity in case of noisy data SAM was used for analysis of the 6 h time point. Cut-off criteria were a higher than 2-fold change and a false discovery rate of <1% for LIMMA and <10% for SAM analysis. The expression values of set 1 and set 2 for 2335 DEG were subjected to cluster analysis using the SOTA module of MeV4.2 [42] with the Pearson correlation as clustering metric and default values except for Winning Cell Migration Weight (0.006). The number of clusters was set to six, based on figure-of-merit calculations. The members of the resulting clusters were subjected to enrichment analyses with different online tools e.g. pantherdb [43] and CoPub [44] for GO terms and pathways and oPOSSUM [45] for TFBS. These analyses were based on human orthologs of the DEG as described by Hintermair [46] and the human genome or the orthologs present on the microarray as the background. Enriched functional annotations and pathways are represented as a heatmap of the Bonferroni-adjusted p-values with a cut-off of 0.05. The human ortholog gene symbols of differentially expressed bovine genes are listed in Additional File 2.

Real-Time RT-PCR

Quantitative real-time PCR reactions were performed with the same RNA samples as used for microarray hybridization using the LightCycler DNA Master SYBR Green I protocol [47]. The following primers were designed to amplify specific fragments referring to selected differentially expressed genes of each cluster (length of the PCR products in square brackets): cluster 1: HP (haptoglobin; for 5'-ACCTGGTATGCGGCCGGGA; rev 5'-TCCGAACCCAGTCCAGAATGGAGG [100 bp]), SAA3 (serum amyloid A3; for 5'-ATGACGCTGCCCGAAGGGGA; rev 5'-TGTCAGGCAGGCCAGCAGGT [210 bp]), PTX3 (pentraxin 3; for 5'-AAAGGGAGACTGGAGAAGGC; rev 5'-TGGCCAAAATGAAATTAACCA [173 bp]), cluster 2: S100A8 (S100 calcium binding protein A8; for 5'-AAAAAGGGAATTACCACGCC; rev 5'-ATCACCAGCACGAGGAACTC [163 bp]), MT2A (metallothionein 2A; for 5'-TCTCCGGACCCCAGCCTCCA; rev 5'-GCAGGAGGCGCACTTGCAATC [119 bp]), LBP (lipopolysaccharide binding protein; for 5'-CCTTGCCCTCAAACTCTCAG; rev 5'-CTTCCCCCTCCCTCTGTTAC [114 bp]), cluster 3: AQP3 (aquaporin 3; for 5'-TCGGTGGAGTTGGGTGGGGG; rev 5'-AGCCCCCTGAATAGAAGAAAGGGC [158 bp]), ALOX15 (arachidonate 15-lipoxygenase; for 5'-GGGTGGGATTCACCACGTGTCC; rev 5'-ACTGAGGCCAGATACCTCCAAC [170 bp]), cluster 4: ACAS2 (acyl-CoA synthetase short-chain family member 2; for 5'-GGCATGCACTTGCCCCGAGA; rev 5'-ACAGGCACTGCCATCCGGT [122 bp]), ADHFE1 (alcohol dehydrogenase iron containing 1; for 5'-AGGGCAGCCTGGACAAACGC; rev 5'-GATCCATGGAGCAAGGGCAGTC [102 bp]), cluster 5: PIR (pirin; for 5'-TTGGACCTGATGATGCACAGCAAA; rev 5'-ACGTGGACACTGTCACCTTCTCCC [76 bp]), THRSP (thyroid hormone responsive; for 5'-AAGTCAAAAGCATCTGGCATGT; rev 5'-ACTCAGAGTTGAGGACTCGGCTT [134 bp]), cluster 6: HPGD (hydroxyprostaglandin dehydrogenase; for 5'-TCTGTCTTCCACTGTAATGCTCAAAGC; rev 5'-AGTGAAGGCCACCAGTACAAAGACT [80 bp]), LPL (lipoprotein lipase; for 5'-TGACTTGTTGTTGGCATCCCCC; rev 5'-AAGTCAGAGTTCCCAGGGCCA [129 bp]) and SF3A1 (splicing factor 3a, subunit 1; for 5'-ACAAGGGTCCAGTGTCCATC; rev 5'-AGACCAGCACCTGTCCATTC [84 bp])) as housekeeping gene [48]. All amplified PCR fragments were sequenced (3100-Avant Genetic Analyzer; Applied Biosystems) to verify the resulting PCR product. In addition, the specific melting point (MP) of the amplified product carried out within the LightCycler standard PCR protocol confirmed product identity [49]. In each PCR reaction cDNA according to 50 ng total RNA was amplified in a total reaction volume of 20 μl (5 μM primer forward and reverse each, 1x LightCycler DNA Master SYBR Green I; Roche) using the LightCycler480 II instrument (Roche). The annealing temperature was 60°C for all PCRs.

Data analysis of real-time RT-PCR

The cycle number required to achieve a definite SYBR Green fluorescence signal (= crossing point; CP) was calculated by the second derivative maximum method (LightCycler software, version 1.5.0). The CP is correlated inversely with the logarithm of the initial template concentration. The CP of the housekeeping gene SF3A1 showed no significant statistical difference in all analyzed samples. Therefore, it was used to normalize the CP for the target genes (ΔCP). Differences between values obtained for udder quarters of the control group and infected or neighboring quarters were stated by the ΔΔCP as well as the 'mean fold changes qPCR' [50]. The normal distribution was tested by the Kolmogorow-Smirnov method, followed by a Student's t-test to confirm significant differences between groups.