Context-Dependent Modulators of Cardiac Fibroblast Phenotype

Fibrosis, the accumulation of excess extracellular matrix, is a characteristic of many pathologies. But it is currently poorly understood, partially because it was long thought to be simply a reaction to injury or a side effect of healing, rather than a treatable disease process. However, fibrosis is increasingly a focus of research as anti-fibrotic therapy has the potential to improve the prognosis for patients with fibrosis. In most organs fibrosis is associated with increased mortality because it disrupts normal organ function, and it can be progressive and permanent. In the heart, fibrosis is particularly devastating because increased extracellular matrix stiffens cardiac tissue and the mechanical properties of the heart are crucial to efficient pumping. Even post-myocardial infarction, when fibrosis is necessary to prevent rupture of the ventricular wall, the character and extent of matrix production is an important determinant of cardiac health. This is particularly true since cardiomyocytes do not repopulate the infarct scar, leaving the matrix and the fibroblasts who maintain it the important role of preserving pump function.
Another reason the mechanisms of fibrosis remain unknown is that it is a complex, heterogeneous response to a variety of injuries. In order for fibrosis to develop, a multitude of signals from the extracellular matrix, inflammatory cells, injured parenchymal cells, and endothelial cells drive matrix protein production or degradation. Often the relative amount of different matrix proteins, cell types, and chemical stimuli are specific to the organ and the initiating injury, meaning each fibrosis type is unique. Furthermore, fibroblasts, thought to be the main cell type involved in fibrosis development, are highly plastic. That is, they are capable of adopting a variety of phenotypes in response to different signaling contexts, and that results in a variety of effects on the extracellular matrix. Fully investigating the complex signaling milieu and consequent cell decision-making is a difficult, almost intractable, experimental challenge. What drives fibrosis is exactly the sort of question that is suited to a systems biology approach. With computational modeling it is easier to interrogate fibrosis as a system to determine which cell type, which signaling pathway, or which protein is the major modulator of fibrosis development in a given setting. Ultimately, a systems biology investigation of fibrosis could identify potential therapeutic approaches for prevention or reversal of fibrosis through targeting fibroblast signaling.
This study represents a first step in that direction by focusing on cardiac fibroblast signaling and decision-making. We used the literature on cardiac fibroblast and general fibroblast signaling interactions to compile a manually curated network of fibroblast signaling pathways. This network is effectively a review of the current understanding of fibroblast signaling. This network was used in a logic-based ODE model that can predict changes in relative activity of network members given an extracellular signaling context (both mechanical and chemical).
We applied this model to the question: how does signaling context determine which node or pathway can modulate collagen expression? We first investigated how pathways are organized into modules and how these modules are affected by which stimulus is applied to the model. This led us to the prediction that mechanical signaling depends on the TGFβ pathway to induce αSMA production, and this was validated in rat cardiac fibroblasts. The model was also used to predict how dynamic signaling in the post-MI setting affects collagen I expression in fibroblasts. We predicted that IL1 signaling is the dominant pathway determining fibroblast phenotype in the early (0.5 day) post-MI signaling context, and that TGFβ is the dominant pathway at the 7 day time-point. We found that ROS is a context-dependent regulator of collagen I mRNA expression in both dynamic and steady-state simulations. We also hypothesize, based on model predictions, that nodes such as PKC or IL1RI are pro-fibrotic modulators of fibroblast signaling in the post-MI dynamic signaling context. This use of a model to investigate wound healing process is an important advance, as the dynamics of acute wound healing are difficult to study. Ultimately, the model predicted that up-regulation of a node is more likely to be pro-fibrotic than anti-fibrotic. This leads to the general hypothesis that cardiac fibroblasts are primed to increase collagen production in response to a variety of stimuli.
As drug discovery is an important use of computational modeling, we developed a pipeline for in silico drug screening using the fibroblast signaling model. With this pipeline we were able to predict drugs that can improve (triflusal) or worsen (arsenic trioxide or anti-thymocyte globulin) cardiac remodeling post-MI through regulation of collagen I expression by fibroblasts.
In this study we developed a signaling network and computational model that together provide a framework for understanding how fibroblasts decide to adopt a certain phenotype. I outline in this dissertation a few important applications for this model: linking topological structure to function, screening for potential pro- or anti- fibrotic activity of drugs, and determining the context-dependence of collagen modulators. But this network and model are flexible enough to be applied to other questions. Gene expression data could be used to inform model parameters and make predictions about how organ-specific or patient-specific fibroblasts respond to different signaling contexts. This model could be incorporated into a multi-scale model of tissue-level fibrosis, or stochastic variation could be added to the model to predict how populations of fibroblasts lead to changes in matrix composition. Finally, pathways could be added onto this network either through further manual curation or through network expansion using inference techniques and high-throughput expression or proteomics data. The capabilities of this model highlight how it can be useful in organizing current knowledge to make hypotheses about fibroblast behavior.