Innate immunity is the first line of defence offered by host cells to infections. Macrophage cells involved in innate immunity are stimulated by lipopolysaccharide (LPS), found on bacterial cell surface, to express a complex array of gene products. Persistent LPS stimulation makes a macrophage tolerant to LPS with down regulation of inflammatory genes (\"pro-inflammatory\") while continually expressing genes to fight the bacterial infection (\"antibacterial\"). Interactions of transcription factors (TF) at their cognate TF binding sites (TFBS) on the expressed genes are important in transcriptional regulatory networks that control these pro-inflammatory and antibacterial expression paradigms involved in LPS stimulation.
LPS, a major component of gram-negative bacterial cell surface, is a potent stimulator of macrophages. LPS acts via the TLR4 receptor to trigger downstream signalling and expression of pro-inflammatory and antibacterial genes . This induction needs to be under tight control since dysregulated inflammation can cause a number of pathological disorders such as septic shock, autoimmunity, atherosclerosis and cancer . Various mechanisms of negative regulation of TLR-induced gene expression have been proposed to dampen uncontrollable inflammation  and these collectively lead to the phenomenon of \"LPS tolerance\"  wherein there is decreased expression of pro-inflammatory genes when there is prolonged LPS administration. Foster et al  have characterized the gene expression profiles of macrophages differentially treated with LPS to classify the genes into various phenotypic states including a tolerant state obtained by an initial LPS treatment. Their analysis of the genes expressed in the tolerant phenotype categorized the genes as belonging to \"tolerizable\" or \"non-tolerizable\" sets depending on no induction or further induction respectively during a second LPS treatment compared to the first one. Although LPS tolerance could prevent pathological inflammatory conditions in chronic bacterial infections, there is a strong need for a persistent antibacterial response to keep the infections under control. The set of genes that exhibit the tolerizable phenotype can be considered \"pro-inflammatory\" while those belonging to the non-tolerizable phenotype as \"anti-bacterial\".
In order to correlate gene expression with transcriptional regulation, we set out to identify characteristic TF-TFBS interactions unique to the two classes of genes. We approached this by employing three independent TFBS motif detection tools with different algorithmic paradigms and to arrive at a list of predicted TFBS common to the three tools viz., MEME, MotifModeler and PASTAA (Middle part of Figure 1). MEME is an expectation-maximization tool that fits a two-component finite mixture model to the input sequences for motif prediction . MotifModeler uses a model selection approach that best fits a set of motifs to gene expression values (both up and down regulated) in co-ordinately expressed genes . PASTAA detects TFBS based on the prediction of binding affinities of a TF to promoters and their association with tissue specific expression of corresponding genes . To reduce false positives, we considered only the top 70% predicted TFBS from each tool. MEME, MotifModeler and PASTAA individually identified 17, 301 and 350 motifs respectively from the pro-inflammatory genes while four motifs were found to be common to all three tools (Figure 4A). Figure 4B lists the TFBS common to all tools found in the pro-inflammatory gene set and the corresponding profile TFs from TRANSFAC . We compared scores from the three TFBS prediction tools for the four motifs identified in our pro-inflammatory data set, and similarly predicted from a random set of 228 genes. As shown in Table 1, the scores of MotifModeler and PASTAA from the pro-inflammatory gene set are significantly higher than the scores from the random gene data indicating genuine enrichment of these motifs in our genes with a potential for transcriptional control of pro-inflammatory specific gene expression. MEME did not predict any of these four motifs in the random gene set. The two motifs predicted from PASTAA and MotifModeler when a similar analysis as described above was performed on the random gene set showed reciprocal higher (PASTAA) or relatively similar (MotifModeler) scores for the random genes compared to the pro-inflammatory genes (Table 2) stressing the specificity of the four motifs (Figure 4B, and Tables 1 &2) for regulating pro-inflammatory gene expression.
Gene UIDs and corresponding expression values of a co-regulated set of genes were given as input and this software works by taking a set of random motifs of fixed size and mapping them onto putative regulatory regions of genes of interest. A linear model was established by considering the expression values and the efficacy of selected motifs. In this model each motif was evaluated based on its contribution to transcriptional regulation. This was iterated many times to calculate a cumulative transcription contribution score (TCS) that was used for motif selection. Least square method was used to dictate inhibitory and stimulatory effects of the predicted motifs. For the analysis of random gene data, we used random expression values, generated by a PERL script, within the range of the maximum and minimum expression values of test genes. 153554b96e