How to inform a QSP model with gene expression data – exemplified with a model on intestinal epithelial integrity and RIPK2 inhibition
Dr. Raphaela Geßele, Dr. Ramona Schmid, Dr. Sudha Visvanathan
Boehringer Ingelheim
Objectives
Quantitative system pharmacology (QSP) models are widely used in drug discovery and development to understand the complex interactions between drugs and biological systems. A QSP model is typically composed of compartments which represent organs involved in the disease and species accounting for cells and extra cellular signaling proteins, e.g. of the immune system, in a mathematical and programmatical manner. This structure does not naturally support genetic interactions as genetic pathways are not represented directly. However, gene expression data becomes available more vastly. And more than that, gene data offers great value for biological accuracy to learn about mechanistic details.
In this study, we demonstrate how network inference can leverage gene expression data to inform a QSP model, by showcasing an application to RIPK2 inhibition in an intestinal epithelial barrier model in the context of Inflammatory Bowel Diseases (IBD). We thus exemplify how to implement the inhibition of intra cellular proteins (genetic pathways) in a QSP model.
Methods
We first constructed an ODE based QSP model (Matlab R2021a) for intestinal epithelial barrier integrity in IBD composed of an intestinal lumen, an epithelial layer, a lamina propria, and a blood compartment. Species in the model are, for example, different types of epithelial cells, defensins produced by them, different cell types of innate and adaptive immune cells, and microbes entering from the lumen. Pathogen entry from the lumen through the epithelial barrier depends on the concentration of defensins in the mucus, which is part of the epithelial layer, and the percentage of healthy versus pro-inflammatory epithelial cells, and holes in the epithelial barrier. Furthermore, cytokines released by macrophages in the lamina propria degrade pathogens. To monitor the inflammation status and intactness of the epithelial layer, we implemented fecal biomarkers and a measure of epithelial intactness, which is referred to as “leakiness”. Model equations follow mass action law and model parameters were based on literature [1-5] and in house data. Remaining parameters were sampled from plausible ranges, which were educated guesses based on related parameters.
To implement RIPK2 inhibition we analyzed gene expression data from intestinal tissue asking the question which genes corresponding to pro-inflammatory cytokines are regulated strongest downstream of RIPK2. To achieve this, we used the network inference algorithm GENIE3 (GEne Network Inference with Ensemble trees [6]) on in house gene expression data from tissue biopsies of IBD patients. After ranking all genes regulated by RIPK2 we further narrowed down the targets to a list of 20 genes associated to inflammation related cytokines. Ranking for the strongest five cytokine targets of RIPK2 using GENIE3, we implemented a dimensionless parameter “kRIPK2” in the QSP model, which allows for simulation of overexpression of RIPK2 and thus the respective cytokines. As RIPK2 can be expressed to different extents in macrophages and epithelial cells, we implemented kRIPK2 as a factor in front of pro-inflammatory cytokine production by macrophages. Which one of the modeled cytokines had the factor in front was decided based on the results of the network inference algorithm.
Results
Our results showed that network inference on gene-expression data widen the methods to include targets in a QSP model in a data-driven manner. We made use of gene expression data from IBD patients to implement RIPK2 inhibition and simulate different disease and treatment scenarios. Parameter sweeps of “kRIPK2" showed the impact of RIPK2 inhibition by resulting in significantly different outcomes for the implemented biomarkers. Simulations with high values of “kRIPK2” resulted in damage of the epithelial barrier and higher levels of fecal biomarkers. These high "kRIPK2" values can be seen as corresponding to patients with chronic inflammation in the intestine (as e.g. in IBD), whereas low values can be matched to either healthy or RIPK2-inhibition-treated patients.
Conclusion
Network inference is a novel tool to check consistency and enhance knowledge of genetic pathways using gene expression data. We explored this approach in order to implement an intracellular target, namely RIPK2, in a QSP model focusing on extracellular processes in various organs (cell and protein level). The process can be seen as a prototype of how to inform QSP models with gene expression data.
[1] Rogers, K. V., Martin, S. W., Bhattacharya, I., Singh, R. S. P. & Nayak, S. A Dynamic Quantitative Systems Pharmacology Model of Inflammatory Bowel Disease: Part 1 – Model Framework. Clin Transl Sci 14, 239–248 (2021).
[2] Balbas-Martinez, V., Asín-Prieto, E., Guillen, Z. P. P. & Trocóniz, I. F. A Quantitative Systems Pharmacology model for the key Interleukins involved in Crohn’s Disease. J Pharmacol Exp Ther 372, jpet.119.260539 (2019).
[3] Roda, G. et al. Intestinal epithelial cells in inflammatory bowel diseases. World J Gastroentero 16, 4264–4271 (2010).
[4] Günther, C., Neumann, H., Neurath, M. F. & Becker, C. Apoptosis, necrosis and necroptosis: cell death regulation in the intestinal epithelium. Gut 62, 1062 (2013).
[5] Teshima, C. W., Dieleman, L. A. & Meddings, J. B. Abnormal intestinal permeability in Crohn’s disease pathogenesis. Ann Ny Acad Sci 1258, 159–165 (2012).
[6] Huynh-Thu, V. A., Irrthum, A., Wehenkel, L. & Geurts, P. Inferring Regulatory Networks from Expression Data Using Tree-Based Methods. Plos One 5, e12776 (2010).