Using microarray analysis, we were able to generate a temporal gene expression profile for 12,257 genes concurrently. Analysis of a relatively large number of genes reduces investigator bias and allows the generation of a gene expression model. Selection of the genes printed on the microarray was independent of our expectations about the effects of Salmonella infection or the role of SipA, SopA, SopB, SopD or SopE2. We expected that this method would yield a dataset that would include genes with a known role in Salmonella infection as well as identification of expression changes in genes whose role in host response was unexpected. We identified alterations in 38 genes whose functions are as yet unknown. Undoubtedly, the function of these transcripts will be better understood as our understanding of the bovine genome continues to improve and may enhance our understanding of host response to Salmonella infection. The microarray data presented here were validated by confirmation of a portion of the altered genes using real time PCR as well as consistency of our results with previously published data. The limitations of this approach are that only changes in expression of genes spotted on the array are measured and that the intestinal tissue used is comprised of multiple cell types, such that it is not possible to differentiate gene expression from one cell type, such as PMNs, from another type, such as epithelial cells; however, we have employed laser-capture microdissection in other studies to evaluate specific gene expression in specific intestinal cell types. Additionally, the data from this study are limited to gene transcription and do not describe alterations in protein expression or state of activation or phosphorylation that may affect activity in the pathways discussed. Studies to address these issues are currently underway in our labs. Ideally, unbiased evaluation of gene expression would include expression data for all genes but at the time that this study was performed, deep-sequencing techniques were cost-prohibitive. Upcoming studies are in progress to utilize these methods and to further develop our understanding of bovine host response to Salmonella.
Limitations of our study included the use of four animals instead of a larger number of animals. Previous studies in our laboratory indicated that four animals would be sufficient to detect significant differences between wild type Salmonella
and the sipA
. Additionally, we wished to limit the number of animals sacrificed for the study and there were financial limitations on the number of animals and number of microarrays that we could perform. Recognizing that low number of replicates is a common problem in microarray studies, techniques to compensate for low replicates were employed. Specifically, an intensity dependent smoothing variance function was used in the Bayesian inference significance tests 
. This method borrows strength across the large number of genes measured across all time points. The genes were ranked by intensity levels and a new standard deviation computed by finding the mean of the variances of the next 50 higher genes and the next 50 lower genes surrounding the gene of interest. This produced a Bayesian estimator of the standard deviation that is found to be a reliable approach for statistical testing compensating for the low replicates.
The primary result of this work was the establishment of a temporal gene expression profile and development of an in silico
model for host response to both wild type S. Typhimurium and a ΔsipA sopABDE2
mutant. SipA, SopA, SopB, SopD, SopE2 were previously demonstrated to have a combined role in the generation of fluid, induction of cytokines and C-X-C chemokines, and PMN recruitment in bovine ligated ileal loops 
. The major difference between WT and MUT infections was the presence of a biphasic response with increased activity at one hour post infection in the intestinal loops infected with WT. Using dynamic Bayesian network modeling (DBN) in conjunction with data from KEGG, the Gene Ontology and GenBank, patterns of pathway activation were established for both WT and MUT infection. The differences in gene activity seen in the WT but not MUT were studied in a subset of pathways selected from categories of cell mobility, cell communication, cell growth and death, immune system, membrane transport, and signal transduction. These pathways could be linked through the interactions of two Salmonella
effectors, SopB and SopE2, and the products of three bovine genes, RAC1
, and AKT2
. Although this subset of interactions does not explain the entire cascade of interactions observed in the data presented herein, they do serve to validate the model and the DBN approach Our original hypothesis and intent was to generate a profile of host gene expression associated with inflammation and diarrhea using two Salmonella
strains, wild type and a mutant that generates reduced fluid secretion and inflammatory responses during intestinal infection. Although we have generated gene expression profiles for these two strains, we have not correlated activation or repression of these pathways with the production of fluid in the bovine intestine as a measure of diarrhea. In vitro
studies employing gene silencing are underway to determine the roles of the 49 pathways identified as differing between the WT and MUT conditions.
In addition to the effects on pathway regulation and mechanistic gene identification, this study identified effects on individual genes in response to infection. We identified increased expression of CXCL9
, and CXCL11
. This is consistent with previous work in both rhesus macaques and mice which demonstrated that IL17, IL12, and CXCL10 are increased in response to wild type S.
Typhimurium and that this response is ablated in animals with diminished numbers of CD4+ T cells in the intestinal mucosa 
. Identification of IFNγ as a mechanistic gene in both WT and MUT loops supports a role for T cells and natural killer cells (both of which produce this powerful cytokine) which is independent of SPI-1 secreted effectors. Increased expression of CXCL9
, and CXCL11
was observed in WT and MUT loops (). It has been suggested from work in murine cell lines that TLR4 signaling induces expression of Type I Interferons (interferon-α and interferon-β) 
. Increased expression of CXCL9, CXCL10, and CXCL11 suggests either that IFN-γ induces the expression of these chemokines or that Type I interferons are increased during the first 12 hours of infection in the bovine intestine. Support for a role of Type I interferons includes studies in mice, which demonstrate that IFNαβ are increased by Salmonella
. Additional indirect evidence includes the increased expression of genes such as MX2
that is induced during viral infections by IFN-αβ 
. IFN-αβ is critical for the function of natural killer cell activity 
. Further study is warranted to define the source of expression of these chemokines and their role in recruitment of T cells and natural killer cells to the intestine during Salmonella
In conclusion, this report catalogued the temporal pattern of intestinal gene expression in response to wild type S. Typhimurium in a natural Salmonella host, the calf, in response to S. Typhimurium infection. We have identified a temporal gene expression pattern comparing wild type S. Typhimurium and a strain attenuated for production of diarrhea and inflammation, and we have catalogued the differences observed due to the combined effects of SipA and SopABDE2 on host gene expression. Pathway analysis and the construction of a system level genetic interaction model of the host response have improved our ability to identify important pathways, genetic mechanisms, and novel genes previously not associated with S. Typhimurium infection. Further, since the genetic model defined by a dynamic Bayesian network can be trained and used for modeling the effects of silencing selected genes and inferring the possible host response, it will be an invaluable tool for: 1) selecting potential intervening gene targets; 2) guiding future experiments for validating host response models; and 3) identifying dynamic gene expression patterns for early diagnostics.