Transcriptome, differentially expressed genes and GO annotation
In this study, the Affymetrix GeneChip® Porcine Genome Array and Q-PCR were used to monitor the whole-genome expression profile of porcine MLN in response to S. Choleraesuis infection. Quite similar to the porcine MLN transcriptome seen in a study on the response to S. Typhimurium , expression of more than 16,000 transcripts were reproducibly detected and of these, 11,020 (68.7%) transcripts had a significant hit to human RefSeq. Thus, we believe the transcriptional response detected in our study represents a high proportion of the porcine genomic response to S. Choleraesuis infection within the MLN. A technical limitation of this work is that the Affymetrix chip cannot assay for transcript that are not represented by probesets. An analytical limitation is that we primarily analyzed only those genes with a relatively large fold change, due to the large number of genes showing differential expression. While a significant number of genes passed our filters, there are additional genes and possibly pathways yet to be analyzed in these data, and all data is provided in Additional files and to NCBI-GEO. Statistical analysis of differential expression revealed 1,853 genes changed their expression level within at least one of 10 possible pair wise comparisons during infection, and about 63% of these differentially expressed genes were annotated using BLAST analysis. Limitation of availability of full-length porcine cDNA and the fact that many human/mouse genes do not have functional annotation, the non-annotated genes are currently not very helpful in furthering our understanding of gene functions and pathways responding to infection. However, the discovery of these non-annotated genes as part of the transcriptional response to bacterial infection is novel, and, as better annotation becomes available, will contribute to an improved understanding of the pathways of host response to bacteria infection across host species.
Compared to the response in porcine MLN during the S. Typhimurium infection, where only 97 genes were induced significantly at 48 hpi , a strong host transcriptional response at 48 hpi with S. Choleraesuis was observed, as 954 genes were differentially expressed at 48 hpi compared to non-infected pigs (Fig 1). This is consistent with the clinical manifestations that the rectal temperature of infected animals peaked at 48 hpi. As expected, we found many genes with large gene expression changes, such as cytokines, chemokines and heat shock proteins; these generally overlap with sets of genes that have been implicated in host response to infection by others [20, 21], indicating our data can be integrated with similar results in other species. However, we also identified a number of genes that had not previously been shown to be involved in host response to bacterial infection, such as STEAP4, PSTPIP2, LIPG, HK3, STXBP1, ITPR1, CA3, KRT17 and PLN. None of these genes responded significantly to S. Typhimurium during infection . Thus we predict that these induced genes are specific to the response to the S. Choleraesuis serovar, as compared to S. Typhimurium. Even though the biological function of these genes during infection remains unclear, these data more fully describe the transcriptional response to S. Choleraesuis and suggest additional functional roles for these genes. Furthermore, the differentially expressed genes with known immune functions and those with unknown functions add to the list of candidate genes to investigate for associations between immune related traits and DNA-level variation; polymorphisms at these candidate genes might result in valuable markers for enhancing disease resistance, pig health, and food safety through molecular breeding methods. Even though we are focusing on the early response (8, 24 and 48 hpi) during the infection, the genes expressed differentially at day 21, include CCL5, CCR5, TGFB3 and AIF1, and provide additional gene profiling data during the chronic infection stage.
Co-expression of genes of known function with poorly characterized or novel genes may provide a simple means to gain leads as to the functions of genes for which information is not currently available. In our study, hierarchical cluster analysis was performed on 1,853 genes that were differentially expressed during infection, and we find the heat map was driven by the 954 differentially expressed genes at 48 hpi. A similar heat map analysis of the transcriptional response to S. Typhimurium shows that the major peak of 151 responding genes occurred at 24 hpi during infection , thus we propose that porcine MLN had a different response peak time for S. Typhimurium and S. Choleraesuis infection. This is consistent with the results from a previous study using SSH technology to investigate expression at 24 hpi . As we observed in porcine MLN response to S. Typhimurium , some ribosome- and/or translation-related genes were repressed at 8 hpi and 24 hpi. This effect is also similar to the response to LPS in skeletal muscle of neonatal and adult pigs [22, 23] and to the response to endotoxin in human blood leukocytes , where a large number of genes involved in translation were repressed. Thus, we speculate that an early pathogenic effect of multiple serovars of Salmonella on the host is suppression of translation.
Three gene groups (Fig. 2, group C, D and E) in the induced cluster exhibited different expression patterns but shared a common feature, a peak expression response at 48 hpi. As large numbers of NFκB target genes were found in groups C and D, we discuss their expression features in the context of NFκB signaling pathways below.
A predominant Th1 immune response was observed during infection, as 6 of 11 investigated Th1 associated genes, including TNF, IFNG, and some IFNG-signaling responsive genes (SOCS1, STAT1, WARS and IRF1), were strongly up-regulated at 24 hpi and/or 48 hpi (Fig. 4B). This is consistent with the results in porcine lung during S. Choleraesuis infection  and in porcine MLN in both S. Typhimurium and S. Choleraesuis infection [8, 12]. However, IL12A and IL12B, which are thought to also favor Th1 cell development, did not change their expression level during infection. Their low expression level in response to S. Choleraesuis has also been detected by QPCR analysis of infected porcine MLN by our group . A low expression level of IL12 has been also observed in the pig in response to porcine reproductive and respiratory syndrome virus  and in pig MLN response to S. Typhimurium infection  and Toxoplasma gondii  infection. We speculate that lack of IL12 induction during S. Choleraesuis infection might stifle IFNG induction and negatively affect host defense against Salmonella.
Our microarray data shows that apoptosis pathway related genes displayed a strong induction at 24 hpi and reached a peak response at 48 hpi in response to S. Choleraesuis infection. Multiple genes within gene families involved in apoptosis pathways were induced significantly, such as genes in the caspase family (CASP1, CASP3 and CASP4) and in the transglutaminase family (TGM1, TGM2 and TGM3). The caspases are a family of cysteine proteases important not only in starting and executing apoptosis, but also in processing and maturation of the inflammatory cytokines IL-1b and IL-18 . The CASP1 gene is known to be important in inflammation due to its role in maturation of IL-1b and IL-18. Even though it is unclear whether CASP1 plays a direct role in apoptosis, overexpression of CASP1 has been shown to cause apoptosis in a variety of cell lines . A recent study revealed that IPAF is an activator of CASP1 and IL-1b in Salmonella infected macrophages ; unfortunately, we did not find any oligonucleotide set representing the IPAF gene on the Affymetrix microarray. Like CASP1, CASP4 has been shown to process pro-IL-18 and IL-1F7b, albeit inefficiently, and to cleave CASP3 into its active form. One important characteristic of CASP4 is its robust induction by INFG and the NFκB complex [30, 31].
Genes from the transglutaminase family can be significantly induced in porcine lung during S. Choleraesuis infection  and in porcine MLN during S. Typhimurium infection . In this study, TGM1, TGM2 and TGM3 showed exactly the same expression pattern: activation was initiated at 24 hpi and peaked at 48 h post S. Choleraesuis infection in porcine MLN. The TGM3 gene showed a very strong induction at 48 hpi, with the fold change about 600 compared to non-infected animals by Q-PCR. There is evidence that TGM2 is a NFκB dependent gene , and researchers have shown that increased TGM2 activity can trigger NFκB activation through an unusual IκB polymerization reaction rather than IκB degradation through IKK signaling . Although transglutaminase genes have been demonstrated to elevate their expression level during inflammation and to play a physiological role in mediating defense against injury or infection in various cell types , their role in apoptosis is not yet clear.
Another important cellular immune response to infection that occurs in the MLN is antigen processing and presentation, as phagocytic cells in the MLN communicate with T cells for further immune activation and initiation of adaptive immune responses. Three of four antigen processing related genes (Fig. 4B) exhibited an increased expression level at 48 hpi, which is consistent with the gene expression patterns that were observed in porcine lung during S. Choleraesuis infection by . Interestingly, two markers involved in the antigen presentation activation pathway, CD80 and CD86, did not show significant expression changes at 24 hpi and 48 hpi. These data tempt us to speculate that the host DC-mediated antigen presentation pathway was altered early after infection. Antigen presentation by murine DC cells can be inhibited by S. Typhimurium [34, 35], and we observed that both CD80 and CD86 were down-regulated at 8 hpi in S. Typhimurium infected MLN . Thus, we predict that lack of a strong DC-mediated antigen presentation might be a mechanism that permits S. Choleraesuis to escape the porcine GALT and cause a systemic infection.
One characteristic of host MLN transcriptional response to S. Choleraesuis is that several groups of genes with annotations for calcium binding activities changed their RNA expression levels significantly during infection. Calcium binding proteins are important molecules in the transduction of calcium signaling, which evoke various cellular processes, such as cell migration, cell differentiation, cell death and cell growth . The S100 gene family is the largest group of calcium binding proteins; two members of this gene family, S100A9 and S100A12, were up-regulated significantly with a large fold change at 48 hpi relative to non-infected animals in our study. Induction of S100A9 has been also observed in porcine Peyer's patch in response to S. Choleraesuis infection . Overexpression of S100A9 and S100A12 at the site of inflammation has been well described in humans, and S100A8/S100A9 and S100A12 are used as clinical laboratory markers for inflammation . The exact biological function of these genes remains to be defined in greater detail, although there is evidence that they have anti-microbial properties , and are involved in induction of apoptosis . Other genes with GO annotation for calcium binding activity, including annexins (ANXA1, ANXA5 and ANXA8) and transglutaminases (TGM1, TGM2 and TGM3), showed differential expression at 24 and/or 48 hpi compared to non-infected animals. Calcium is an important second messenger in cells and changes in calcium pathways can evoke various cellular processes . Our report is the first to describe the strong induction of multiple calcium binding protein genes in MLN of S. Choleraesuis infected pigs and adds new information to host transcriptional response to gram-negative bacteria infection in pig and other species.
NFκB signaling pathway
NFκB is a latent transcription factor held in the cytoplasm through binding by its inhibitors IκB. During stimulation or infection, IκB is phosphorylated and degraded, allowing NFκB translocating into the nucleus to induce expression of many genes . The NFκB target genes control a variety of cellular processes. Both microarray and QPCR data analysis revealed that many known NFκB target genes were significantly up-regulated from 24 hpi to 48 hpi, indicating a strong NFκB response during acute S. Choleraesuis infection. Both the magnitude and timing of this response was different from the response to S. Typhimurium, where suppression of the NFκB pathway from 24 to 48 hpi was observed . Thus understanding the transcriptional profiles of NFκB target genes is an important step to further understand its regulatory function in bacterial infection. Different activation times for NFκB target genes during stimulation has been intensively studied [39, 40]. Analysis of LPS-stimulated mouse macrophages demonstrated that NFκB binding occurs in two distinct waves due to different rates of NFκB recruitment . Tian et al. (2005) identified "Early", "Middle" and "Late" expression profiles of NFκB target genes during the TNF stimulation in epithelial cells, since these genes have peak response at 1 h, 3 h and 6 h, respectively. Three mechanisms for these differential expression patterns of NFκB target genes were suggested; a) Early and Late gene promoters are bound by NFκB complexes containing different subunits; b) there are different environments in which the NFκB binding sites are located between the Early and Late gene promoters or c) genes in the Late group undergo an additional rate-limiting step necessary for promoter activation .
In our study, two groups of NFκB target genes were identified, an ''Early'' group and a ''Late'' group. To determine the biological functions of genes in the Early versus the Late groups, GO annotations were assigned. GO terms for cytokine activity or chemokine activity were enriched in the Early genes (Fisher's exact test p = 0.053), while the Late group was predominantly composed of genes with signal transduction and cell metabolism annotations. This is consistent with the functional categorization of NFκB dependent genes in ''early'' and ''late'' groups of . The rapid induction of cytokines and chemokines during infection clearly may be important for recruiting immune cells to infection sites.
Both microarray and Q-PCR data suggested that NFKBIA (which encodes IκBα) increased its RNA level slightly at 48 hpi. There is overwhelming evidence that demonstrates that expression of this inhibitory gene is activated by NFκB in a negative feedback loop, which provides an effective mechanism for controlling NFκB activity . It has also been shown that NFκB DNA binding activity is at its peak when NFKBIA is newly synthesized . Thus, as we observe up-regulation of NFKBIA and strong activation of other NFκB target genes at 24 hpi, we speculate that NFκB activity was initiated at 24 hpi and transcriptional responses to NFκB peaked at 48 hpi. This contrasts to the response of NFκB target genes to S. Typhimurium infection, where NFKBIA was not up-regulated significantly at 48 hpi, and many genes were suppressed from 24 hpi to 48 hpi .
Identification of potential porcine NFκB targets
Using TFM-Explorer and Clover, it was possible to identify promoter regions containing NFκB motifs in human genes orthologous to the porcine genes which had Early (within 24 hpi) and Late (by 48 hpi) response to Salmonella infection. We have confidence that our analysis has identified strong candidate NFκB target genes, as the cutoff for identifying NFκB motifs within promoters was stringent, given that the non-zero possibility that groups of randomly selected genes used as the background in the TFM Explorer analysis will contain some NFκB targets. Some of the known targets were missed by our TFM-Explorer analysis, but the input sequence window used was only 2,000 total bases. Thus for these known targets, there might be some binding sites outside of the input sequence range. However, we were able to identify 95%, 65% and 83% of the known targets in the E83, L319 and A544 groups, respectively, indicating that our sensitivity was adequate to find the large majority of known target genes.
Results for DNA regulatory motif analysis of human orthologs for the stimulated gene groups provided evidence for 51 putative NFκB targets in the Early group and 145 in the Late group. These putative targets were also analyzed by GO annotation to further develop our understanding of the regulatory response to Salmonella infection (data not shown), as performed on the known NFκB target genes (Figure 5), These putative target genes, including SOCS1, SCARB2, CEBPD, CXCL16, IL1RAP, and CASP7, are involved in multiple cellular processes, including cell adhesion, regulation of transcription, immune response, and receptor activity. While there is no additional experimental evidence that these genes are direct NFκB targets, many of them have been previously reported to be involved in the response to bacterial infection [8, 42, 43]. Further, the motif-finding software TFM-Explorer was able to predict nearly all of these promoters as containing NFκB motifs, even when known NFκB target promoters were removed from the test set. This indicates that these putative NFκB target genes have important regulatory regions in common with bona fide NFκB targets, and that these common motifs can be found independently in the known motif data. Thus, although further experimental work is needed to confirm that these genes are true NFκB targets, our analysis detecting NFκB regulatory motifs in these genes further supports the hypothesis that they constitute an important part of the early anti-bacterial infection response.
To supplement our PubMed literature search that found none of these genes were known NFκB direct targets, Pathway Studio (a PubMed text-mining software) analysis was run on the complete sets of Early and Late genes. Pathway Studio identified seven Early group genes having literature evidence of direct binding of NFκB to their promoter region (CEBPD, IL1B, IL6, IL8, FOS, EGR1, PTGS2), while in the late group only two were identified as having literature evidence of direct binding (TNF, IFNG). In addition, in the Early group, ten other genes were identified as having a connection with NFκB in the literature, usually sharing a common expression pattern (CSF3, CYCS, OAS1, CEBPB, CELP, IER3, SOD2, CCL2, HLA-A, CXCL2). For the Late group, an additional 11 were also found to have a connection with NFκB (LYZ, SP6, IL10, TLR2, BCL2, APP, HXB, CCR5, ANXA5, IL15, CDKN1A). TFM-Explorer and Clover found motifs at the promoters of: a) all genes with direct binding evidence; b) all but one of the NFκB-connected genes (CYCS) identified in the Early group (94%); and c) all but two (CCR5, ANXA5) of the 13 (77%) in the Late group. Of the genes that were predicted to contain an NFκB binding site by both programs and had evidence from Pathway Studio as having a direct binding relationship with NFκB, only CEBPD and EGR1 were not previously on the "known" list of targets at available websites (see Methods). However, due to more recent literature evidence provided by the Pathway Studio software, they should be considered "known" NFκB targets. Thus, this literature analysis indicates that the large majority of genes predicted to be NFκB targets (Fig 7), based on co-expression and motif data, have not yet been recognized as direct NFκB target genes. As this evidence is based on human promoter sequences, these genes may be regulated by NFκB during cellular responses to Salmonella in other species as well, including human.