3.4. Find Master Regulators in Networks
When we followed the upstream activation pathways of the TFs potentially involved in the (co-)regulation of the liver cluster 3 genes, we found TAB1 as one potential master regulator of the up-regulated genes (Figure 6). Mapping expression values from the whole liver experiment showed no highly up- or down-regulated genes for the involved proteins of the identified pathway.
TAB1 is a protein that binds to and regulates the activity of the mitogen-activated protein kinase MAP3K7, also known as TGF-β-activated kinase 1 (TAK1). This kinase mediates TGF-β and TNF-α signals and, via some phosphorylation events, activates the NF-κB pathway and the MAPK pathways, the latter targeting transcription factor (TF) AP-1 and related TFs. This way, TAK has been shown to play a dual role as both a tumor-promoting and suppressing agent, depending on the cellular context [30,31,32]. In liver, the role of TAK1 as tumor suppressor has been demonstrated [32]. Based on these findings, TAK1 has been discussed as a potential target for cancer treatment [33,34].
Figure 6  Master regulator TAB1 was identified for the cluster of up-regulated genes in the liver. The master regulator is shown at the top-most position of the schematic overview (pink rectangle), connecting molecules up to 10 steps upstream (green rectangles) starting from the identified transcription factors (blue rectangles). Known complexes are highlighted by the dark-green hexagonal frames. The diagram is a result of the workflow shown in Figure 5. See Supplementary Figure SF2 for high-resolution version.   In addition, caspase 6 was found to be a common master regulator of up- and down-regulated genes in liver (Figure 7). This gene encodes a cysteine-aspartic acid protease (caspase). Caspases are activated by proteolytic processing cascades [35,36]. Their sequential activation is essential for cell apoptosis [37]. However, caspase 6 seems to be an exception in that its activation does not necessarily depend on other caspases and, thus, its role in apoptosis might be a different one compared to the other caspase family members, subject to further proofs [38].
Figure 7  The potential master regulator caspase 6 (pink rectangle) was identified for the cluster of both up-and down- regulated genes in naphthalene-treated mouse liver. The master regulator is shown at the top-most position of the schematic overview (pink rectangle), connecting molecules up to 10 steps upstream (green rectangles) starting from the identified transcription factors (blue rectangles). The diagram is a result of the workflow shown in Figure 5. See Supplementary Figure SF3 for high-resolution version.   The upstream strategy applied to the up-regulated genes of lung cluster 4 revealed PTK6 (protein tyrosine kinase 6) as one of the top-most six upstream regulators (Figure 8). Expression mapping showed that many of the identified potential TFs are either up- or down-regulated (blue and red border lines in Figure 7). Down-regulated expression of protein tyrosine kinase 6 (PTK6) is correlated with poor survival in esophageal squamous cell carcinoma [39]. A previous study showed over-expression of PTK6 in non-small-cell lung cancer (NSCLC) and evaluated its pathological and prognostic significance [40]. The results confirmed that NSCLC patients with overexpressed PTK6 had a poor survival prognosis, rendering PTK6 inhibitors candidate drugs for treating this kind of cancer [40].
Figure 8  Identified master regulator PTK6 is shown at the top-most position of the schematic overview (pink rectangle), connecting molecules up to 10 steps upstream (green rectangles) starting from the identified transcription factors (blue rectangles). Strong red border lines indicate up-regulated genes and blue border lines down-regulated genes. The diagram is a result of the workflow shown in Figure 5, mapped with expression values. See Supplementary Figure SF4 for high-resolution version.   Usp22 was found to be a common master regulator for up- and down-regulated genes in mouse lung. Usp22 encodes ubiquitin carboxyl-terminal hydrolase 22 (Figure 9). As a component of the histone acetylation (HAT) complex SAGA, Usp22 removes the ubiquitin residues from histones H2A and H2B, which leads to a transcriptional (co-)activation [41,42,43]. Human USP22 is known to play a role in different types of cancer [44,45,46]. In particular, it has been demonstrated that overexpression of USP22 is associated with non-small-cell lung cancer (NSCLC) and causes a poor survival prediction [44].
Figure 9  The potential master regulator Usp22 is shown at the top-most position of the schematic overview (pink rectangle), connecting molecules up to 10 steps upstream (green rectangles) starting from the identified transcription factors (blue rectangles). Strong red border lines indicate up-regulated genes and blue border lines represent down-regulated genes. The diagram is a result of the workflow shown in Figure 5, mapped with expression values. See Supplementary Figure SF5 for high-resolution version.   Altogether, we noticed that the suggested master regulators for both tissues are involved in promoting tumor progression and/or apoptosis. Those found in the liver seem to be of a more general function, whereas those identified from the lung dataset have the potential to specifically trigger the development of lung tumors (NSCLC). Thus far, we have not been able to directly compare the results of our analysis with what other tools aiming at upstream analyses would result in, such as IPA [47]. It is our aim to model a mechanistically plausible upstream pathway, for which the most crucial first step is the identification of all relevant TF-target gene relations. For this, we apply a de novo rather than a knowledge-based strategy. Our approach stresses the importance of regulation through TF combinations and secures the required flexibility for the analysis of new cellular systems, e.g., tumors that have not yet been studied and in which the existing TF repertoire has usually been redirected to govern a significantly different genetic program, e.g., as described in [48]. Optimally, each newly studied cellular system would be experimentally characterized for genomic locations of all ~1600 TFs (in case of mammals), as was done exemplarily for one TF (BCL6) in a previous study [49], which is not yet feasible. We therefore feel that our approach represents a good and realistic compromise between reliable knowledge-based pathway reengineering and flexible de novo analysis of regulatory genome regions.