Questions
Cytokines, such as IL-6, are produced during inflammation and act on hepatocytes to initiate the acute phase response. This results in broad influence on the detoxification capacity of the liver via modulation of absorption, distribution, metabolism and excretion (ADME) gene expression. Several signaling pathways are known to be activated by IL-6, predominantly involving MAPK, PI3K, and JAK/STAT. Previous experiments suggest a larger role of MAPK and PI3K for the regulation of ADME gene expression than of JAK/STAT. ADME gene regulation by IL-6 is investigated further in this work.

Methods
Primary human hepatocytes from several liver donors were treated with IL-6 in combination with siRNAs, knocking out key kinases of the three major signaling pathways or chemical inhibitors of these kinases. The activities of key signaling proteins were measured by reverse-phase array and Western Blot analysis. Gene expression data were analysed using the Taqman® qPCR (Fluidigm) platform. In order to explain the regulation of ADME genes by IL-6, we first created a pre-knowledge model (PKN) based on information from literature. This model contains several signaling pathways (PI3K/Akt, JAK/STAT, and MAPK) that are connected to ADME gene expression. The signal transduction part is mainly an adaptation of an existing model (Ryll et al., Mol.BioSyst., 2011, 7, 3253- 3270), whereas the information for the gene-regulatory module was taken from several other sources. We calibrated the PKN with adapted optimization and compression routines of the software CNORfuzzy (Morris et al., PLoS Comp. Biol., 2011,7:3) with respect to the data, thus creating a compressed fuzzy logic model. The procedure was executed 100 times, which yielded a "mean" model. The weight of a transition in this model is the portion of optimized models, in which it is contained. We changed several transitions in the PKN model and repeated the optimizations afterwards in order to improve the fit between data and model predictions.

Results
The data show large variations between the different donors. However, after applying the CNORfuzzy methods to the data of a donor, the optimized and compressed model still contained the three main signaling pathways for all donors. For several ADME genes our model suggests a prominent regulatory role of one specific signaling pathway (often JAK/STAT), whereas for other genes regulation seems to be more complicated and cannot be clearly elucidated from the data.

Conclusions
Our logical model combines knowledge from literature and new data sets. It enables us to draw conclusions about the importance of specific influences for ADME gene regulation. One finding is that the model suggests a more prominent role of the JAK/STAT pathway than previously assumed.