As a basis for characterizing transcriptomic signatures in LPS tolerance, we differentiated pro-inflammatory and antibacterial gene expression in LPS-stimulated macrophages. The genes in the Foster et al  microarray data were classified into pro-inflammatory and antibacterial sets using biologically relevant filters (Figures 1 and 2). Authors of that study had performed similar categorization to identify "tolerizable" and "non-tolerizable" classes corresponding to our pro-inflammatory and antibacterial classes respectively. Basically, Foster et al and we hypothesize that during tolerance although pro-inflammatory genes are down regulated, antibacterial genes need to be continually expressed. The two classes in their study were shown to have differences in chromatin modifications indicating transient silencing of some pro-inflammatory genes while priming of antibacterial gene expression . However, they did not attempt to identify any global TF-TFBS interactions that can attest to transcriptional signatures responsible for the distinction between the two classes and hence to LPS tolerance. We have identified and characterized such distinctive interactions employing pattern search algorithms combined with literature-based validation of the target genes in the microarray data.
Our filtering approach deducted only a small number (18) of antibacterial genes since the cut-off of even one-fold step-wise increase from the N to N+L to T+L stages along with the p-value threshold seems to be highly stringent and this may indicate a highly specific antibacterial phenotype for these genes. Due to the small size of this data set, subsequent analyses of motif search, and target identification and validation for this class of genes did not merit much attention. However, genes such as Lcn2  and Tirap  in this list (Table S1) have shown antimicrobial activity. Interestingly Lcn2 suppressed LPS-induced inflammatory cytokines in macrophages  indicating an additional anti-inflammatory role for this gene. It is likely that a number of antibacterial genes in our microarray data maintain the same level of expression at both the N+L and T+L stages to provide persistent bacterial protection and hence it would be worthwhile to analyze a larger set of these genes by filtering with a less stringent condition.
Based on the concept of coordinate gene expression being controlled by same or similar TFs binding to their cognate binding motifs, we identified respectively four and one TFBS motifs in the upstream regulatory regions of the pro-inflammatory and antibacterial genes (Figure 4). The essence of the specificity of these TF-TFBS interactions was provided by the intersection of the predictions from three tools of differing algorithmic approaches. Additionally, when compared to a random gene background the prediction scores for the four pro-inflammatory motifs were significantly higher in the test data set (Tables 1 &2). Our predictions identified interferon regulatory factors 1 and 8 (IRF-1 and IRF-8) as pro-inflammatory TFs (Figure 4, Table 1) and these proteins have been clearly implicated in macrophage-associated innate immunity [21, 22, 23].
We validated the TF-TFBS predictions in the pro-inflammatory class of genes by manually identifying the corresponding target genes implicated to be under control of the predicted TFs. This analysis produced a significant number of target genes and a small proportion of them (18 out of 141) showed the pro-inflammatory specific pattern of gene expression (Table 3, Figure 5). A similar gene expression pattern of the 18 genes identified in our analysis with data from the Mages et al  microarray analysis (Table S4) confirms the genuine pro-inflammatory phenotype of these transcriptional targets in macrophage tolerance induction, and indicates that the four predicted TFs most likely control the transcriptional regulation of these genes to establish the phenotype. A number of genes in this list have been experimentally associated with LPS-mediated macrophage activation including some important TFs such as IRF1, IRF2, Stat1, Stat3 and Nfkb1  that are essential members of presumptive transcriptional regulatory networks. Nfkb1 has been identified as a major player in the downstream signalling pathways of LPS-stimulated macrophages  with a crucial role in the transcriptional regulation of a number of target genes [9, 25]. More importantly, some of these target genes, such as Trim21, Ptgs2 and Nos2 (Figure 5) that show the sharpest drop in gene expression upon prolonged LPS treatment have been implicated in tolerance [26, 27, 28]. Nfkb1 acts as the controlling TF in LPS-induced expression of Ptgs2 and Nos2  while acting downstream of Trim21 . Here, since the microarray data represents a static view of gene expression, we cannot determine whether the Nfkb1 gene acts up or downstream of these genes as they all (Nfkb1, Trim21, Ptgs2 and Nos2) show the prototypical pro-inflammatory phenotype of down regulation (Figure 5) in induction of tolerance. A caveat to this classification may be the case of Nos2 that shows up in the pro-inflammatory category in our analysis. This gene has a direct role in killing intracellular pathogens [30, 31, 32] even though it shows down regulation in the tolerant state. A reason for the presence of such genes in our classification maybe due to the non-linearity of TF-TFBS interactions in that the dynamic modulation of transcriptional target gene expression does not always correlate with the corresponding TF binding to its site.
A limitation to this study is the lack of the tolerant (T) stage in the Foster et al  data set unlike the Mages et al  data. By employing the N+L stage instead of T (Figures 1, 2 and 5) we are likely to miss some genes (false negatives) that are tolerizable. However, the filtering approach that we employed is logically sound and did result in a significant number (228) of pro-inflammatory genes. Our motif prediction tools converged on 4 TF binding motifs that could co-ordinately regulate these genes (Figure 4). From a biological perspective, the 18 genes that are validated from the literature regards to being transcriptional targets of the four predicted profile TFs are most likely to be genuine candidates for establishing and/or maintaining the pro-inflammatory phenotype.
It is interesting to note that although the exhaustive work of Ramsey et al  looked at the dynamics of transcriptional programs in LPS-stimulated macrophages, they did not characterize the differential expression of the two categories of tolerant genes as belonging to the pro-inflammatory and antibacterial classes. We observed a number of common genes between our 228 pro-inflammatory set and their list of 1960 differentially expressed genes probably indicating a mixture of both classes of genes in their data. Litvak et al  have implicated one such gene, Cebpd, a TF, in a regulatory circuit discriminating between transient and persistent TLR4-stimulated signals. A search for the Cebpd binding motif in the regulatory regions of our two classes of genes (similar to their analysis) is likely to shed more light on the varying gene expression patterns in LPS tolerance induction.