The lists of differentially expressed genes were analyzed for biological functionalities using the DAVID (The Database for Annotation, Visualization and Integrated Discovery) software 343536, version 6.6. The gene lists were analyzed against the gene list of the human genome since the annotation of the human genome, especially with physiological data, is more advanced than the chicken genome. Therefore, all genes were converted to human Entrez identifiers. The false discovery rate, multiple testing correction for statistical significance 37, and the fold enrichment analyses were manually included for all analyses. The tissue-specific profiles and functional annotations and clusterings of the gene lists were investigated.

Due to the platform differences used in the individual studies the number of genes available in all studies was reduced as expressed in Figure 2. The Figure shows that 7,643 genes were common to all microarray platforms and thus available for meta-analysis over all studies. Different platforms may use different probes for the same genes, and the probes may differ in hybridization characteristics. However, since differential expression of genes was measured within a platform for each of the experiments this will not affect the meta-analysis.

The results of the meta-analysis DE-group were expressed in several lists of genes with regulated expression in more than one or all experiments (Table 1 – the Table with the gene lists is Additional file 1). The Table shows that approximately 3,000 genes are differently expressed, irrespective of age of infection or Salmonella serovar of infection. Differential expression was between control and infected animals of the same age. Furthermore, most differently expressed genes were the same in all experiments irrespective of sampling time point post-infection (Figure 3). The Figure shows that the large majority of the genes are shared by the (e) and the (t) categories.

Number of differentially expressed genes in the groups Differentially Expressed (DE) genes and Integration Driven Discoveries (IDD), the latter of which is specific for genes only found in the meta-analysis. Categories depend on the experimental differences: (e) experiment, (t) time post infection, (et) both (e) and (t), (e-t) experiment specific, and (t-e) time specific.

The IDD group (i.e. the list of genes found only differently expressed in the meta-analysis) genes were predominantly in the (t) category (Figure 4). A few genes were differently expressed in the meta-analysis only for the (e) category.

The gene lists were analyzed for biological functional groups using the DAVID software. First the expression profile of differentially expressed genes of the DE-group was compared with normal tissue-specific expression profiles from the same time points and the same tissue type (Table 2). Significant results were only obtained for the DE group, (e), (t), and (et) categories. The Table indicates that the sampled intestinal tissue showed expression profiles related to a number of different tissue types. The most significant tissue expression profile is epithelium. This cell type is abundantly present in intestinal tissue. Several other cell types and tissues also showed similarities for tissue-specific expression profiles, some of them not relevant in intestine (data not shown).

To investigate cell types and tissues related to the differentially expressed gene profiles the DAVID software compared the lists of differentially expressed genes with normal physiologic expression of tissue-specific gene profiles. N is the number of differently expressed genes found in a tissue, FE is Fold Enrichment, Benjamini is the P-value after correction for multiple testing, and FDR is False Discovery rate.

Clustering of the functional annotations of the gene lists of the differential expression (DE) group, i.e. the Number of Differentially Expressed genes (i.e. in the experiments and in the meta-analysis), and categories: (e): experiment - i.e. 4 studies, (t): time after infection – 14 different time point post-infection, and (et): overlap between the (e) and (t) groups. The t-values were deduced from the other groups because the number of data was too large for the DAVID software to analyze directly. 1: Number of the cluster within group.

The results of the differently expressed gene lists were then analyzed for functional biological mechanisms. The results are shown in Additional file 2. The results showed lists of biological functions for the (e) and the (t) categories of the DE group. Moreover, the (et) category showed that both lists were largely similar, and the top of the lists were even identical. The top of the lists indicated that phosphorylation of proteins, acetylation in the cytoplasm and lumen of other cellular components, and ATP consuming processes were important biological mechanisms during chicken host reaction to Salmonella infection. The meta-analysis showed additional significant results for both the experiment (e-t) and the time (t-e) categories of the phosphoprotein biological function during chicken host reaction to Salmonella infection. Similarly, in the IDD group the time (t-e) category further indicated additional significant results especially for the phosphoprotein biological function.

Finally, a cluster analysis was performed for the lists of biological functional annotations. The DE-group (e) and (et) categories showed over 600 clusters. Due to the fact that the list of the (t) category in the DE group was longer than 3000 entries, clustering was technically not possible for the DAVID software. Since the lists of functional annotations of the (et) and the (t) categories of the DE group were very similar the (t) category clusters were deduced from the (e) and (et) categories clusters (see below).

Table 3 shows the results for clusters with enrichment scores larger than 2. An enrichment score indicate whether the number of genes in a cluster is equal to the expected number of genes (due to the number of genes of that physiological group in the genome and on the microarray) or higher or lower than expected. A high enrichment score thus indicates that the physiological trait of the cluster may be significant to the trait. The clusters were ordered by enrichment scores. The range of enrichment scores was from over 10 to almost zero. Enrichment scores less than 2 were omitted, leaving between 35 and 40 clusters in the DE-group (e) and (et) categories, respectively. For completeness, the Additional file 3 includes the biological functions per cluster in detail, i.e. larger than 1. The content of each cluster is a group of biological functions taken together from various databases centered on a specific theme. For example (see Table 3), cluster 1 groups biological functions together related to the lumen of cell organelles, especially relating to the nucleus. Cluster 2 groups nucleotide binding and phosphorylation functions (especially serine and threonine phosphorylation), while cluster 3 groups mitochondrial membrane functionalities. The Table contains both cell / tissue morphological clusters and (macro) molecular biogenesis functional clusters. The functions of several protein domains were also relevantly clustered.

Taken together functional clustering analysis showed that these functional annotations can be grouped together in higher order biological morphological structures and biological processes. The clusters can be divided in 21 clusters describing metabolic processes – of which three were related to energy metabolism, eight clusters describing (cell) morphological features, three clusters specifically pointing to protein domains involved, one cluster related to apoptosis and one cluster is a collection of processes, making it difficult to recognize a central theme. Apart from apoptosis, these clusters describe normal cellular physiological processes taking also place in non-infected animals, e.g. during growth and development of the tissues and organs. Nevertheless, these processes also participate in the host reaction to infection with Salmonella.

Apart from small differences in the order of clusters the (e) and (et) categories of the DE group differ only in a few clusters from each other. A specific tyrosine phosphorylation was found in the (et) category but not in the (e) category of the DE group while the (e) category showed a protein domain WD cluster and a cell movement cluster, both not found in the (et) category. Finally, it should be noted that in none of the other categories (DE and IDD groups) a significant cluster with enrichment score of at least 2 could be found.

A meta-analysis is performed using the original raw data from a number of individual experiments. Since the experiments may have different objectives and use different technologies, the experiments or data may not be directly comparable. In our study we compared data from four studies using: (1) genetically different chicken lines differing in Salmonella susceptibility), (2) different Salmonella serovars, (3) different sampling time points, (4) different sampled tissues, (5) different microarray types, and (6) different ages of bacterial challenge. Intuitively, it would be expected that these differences would affect the meta-analysis: (1) Genetically different lines of chicken, differing in Salmonella susceptibility, were expected to differ in reaction mechanism and/or reaction severity. (2) A pathogen specific host reaction was expected to different Salmonella serovars. (3) Sampling at different times post infection was suggested to show different temporal expression patterns related to the stage of infection. (4) Expression patterns are also expected to differ between different tissues or cell types. (5) Finally, different microarrays contained different sets of genes, so results from one study were expected to be missing from another study and vice versa. (6) The age of challenge of the birds would be expected to produce very different responses due to the poorly developed immune system of day old chicks compared to 3 week old birds. Despite of all these differences our meta-analysis indicated that the chicken lines react to Salmonella infection through comparable mechanisms irrespective of Salmonella serovar and tissue type, and therefore it may be concluded that we identified common mechanisms of the host response to the bacterial challenge. However, due to the different experimental ages of the animals used in the diverse studies, this conclusion may be hampered by the developmental differences of tissues and organs in the animals. It can be expected that at least part of the mechanisms found may relate to this. This could have been investigated only if control samples of all experimental ages in the individual datasets would have been available. But often these control samples are only available for the last experimental sampling age. Further experiments are needed to elucidate this point.

Although it is not certain, it can be expected that the results would have been more comprehensive if all experiments were performed under standard procedures. Similarly, the functional annotation analysis to elucidate potential biological mechanisms of the functional reaction of chicken to Salmonella infection would have been more robust.

The first indication that the chicken host reaction to Salmonella infection was similar between the diverse experiments was obtained from the similarities in the gene lists for the differently conducted experiment (e) and time (t) categories, i.e. the (et) category. One unexpected finding was that the expression profiles related to several different cell types. Intestinal tissue is composed of many different cell types that could be indicated by a mixture of expression profiles. Furthermore, localized infection will change tissue expression profiles, which will be exacerbated by the influx of immune cells, which will further change the overall expression profile. However, the results indicated similarities to the expression profiles of several cell types including many unrelated tissues like liver and brain. The epithelium cell type of the intestine was the highest ranking tissue in all three analysis groups. Also platelet and muscle tissue, and perhaps colon expression profiles were recognizable – these cell types are also included in the intestinal tissue. Other cell types may also be on the list for several reasons. One reason may be that a cell type has a high turnover rate like epithelial cells in intestine tissue. In these cell types the general mechanism for cell division will be activated and therefore all these cell types appear on the list. Especially in developing young-age animals this may be expected. Finally, cell types and tissues may have been included in the list because we used the human physiological information instead of chicken physiological information for the DAVID software to create the list, e.g. lymph tissue may be inserted for that reason (although the chicken intestine contains a limited number of Peyer’s patches as lymphoid tissues 38).

The functional annotation is the result of the analysis of the DAVID software using the same gene lists to analyze several different databases containing biological function information. Due to the similarities within the gene lists the DE group (e), (t), and (et) categories showed similar functional annotations. Furthermore, the top category functional annotation “phosphoprotein” was also found in the differently expressed genes unique for both the experiments and the time (t) array after infection, and in the time-related genes found specifically in the meta-analysis (IDD-group). These results suggest two mechanisms: (1) the analysis is robust and indicated the same biological functionalities for all experiments despite the experimental differences, and (2) the meta-analysis adds new genes and data to the already existing data, but does indicate new biological mechanisms for the reaction of chicken hosts to the Salmonella infection. On the other hand, protein phosphorylation is an important regulatory mechanism for protein function in normal tissue and changed phosphoprotein content of the cell may have important physiological consequences for cellular metabolism (see below).

The clustering of biological functional annotations showed only in the lower part of the list differences between the DE (e) and DE (t) category. While these differences themselves were statistically significant, the place on the list may suggest that the differences in the reaction of chicken to Salmonella are small. Alternatively, these differences point towards differences in the expression profiles related to time point after infection. However, due to the structure of the dataset these differences may also relate to deviations in the general chicken reaction mechanism caused by different chicken breed/lines, different tissues or different Salmonella serovars (jejunum vs. caecum; S. Enteritidis vs. S. Typhimurium).

While most clusters of differently expressed genes were similar in both the experiment (e) and the time (t) categories, some interesting differences were obtained. It should be noted that clusters found in one category but not in the other may be the result of real missing clusters or clusters failing to reach the enrichment score limit in one of the two categories. Three clusters were found in the differently expressed genes group experiment (e) category, but not in the time (t) category suggesting that these genes were not, or less regulated in time after infection and may be constitutively active during the chicken host reaction to Salmonella infection: (E1) Tyrosine phosphorylation, (E2) Protein domain WD (repeat), and (E3) cell movement. Two clusters were found in the differently expressed genes group time (t) category but not in the experiment (e) category, suggesting that these genes were especially regulated at different moments in time after infection of the chicken: (T1) ubiquitin / proteasome mediated proteolysis, and (T2) Intracellular vesicles.