Logic-based mechanistic machine learning on high-content images reveals how drugs differentially regulate cardiac fibroblasts.

Fibroblasts are essential regulators of extracellular matrix deposition following cardiac injury. These cells exhibit highly plastic responses in phenotype during fibrosis in response to environmental stimuli. Here, we test whether and how candidate anti-fibrotic drugs differentially regulate measures of cardiac fibroblast phenotype, which may help identify treatments for cardiac fibrosis. We conducted a high-content microscopy screen of human cardiac fibroblasts treated with 13 clinically relevant drugs in the context of TGFß and/or IL-1ß, measuring phenotype across 137 single-cell features. We used the phenotypic data from our high-content imaging to train a logic-based mechanistic machine learning model (LogiMML) for fibroblast signaling. The model predicted how pirfenidone and Src inhibitor WH-4-023 reduce actin filament assembly and actin-myosin stress fiber formation, respectively. Validating the LogiMML model prediction that PI3K partially mediates the effects of Src inhibition, we found that PI3K inhibition reduces actin-myosin stress fiber formation and procollagen I production in human cardiac fibroblasts. In this study, we establish a modeling approach combining the strengths of logic-based network models and regularized regression models. We apply this approach to predict mechanisms that mediate the differential effects of drugs on fibroblasts, revealing Src inhibition acting via PI3K as a potential therapy for cardiac fibrosis.

Network Analysis Reveals a Distinct Axis of Macrophage Activation in Response to Conflicting Inflammatory Cues.

Macrophages are subject to a wide range of cytokine and pathogen signals in vivo, which contribute to differential activation and modulation of inflammation. Understanding the response to multiple, often-conflicting cues that macrophages experience requires a network perspective. In this study, we integrate data from literature curation and mRNA expression profiles obtained from wild type C57/BL6J mice macrophages to develop a large-scale computational model of the macrophage signaling network. In response to stimulation across all pairs of nine cytokine inputs, the model predicted activation along the classic M1-M2 polarization axis but also a second axis of macrophage activation that distinguishes unstimulated macrophages from a mixed phenotype induced by conflicting cues. Along this second axis, combinations of conflicting stimuli, IL-4 with LPS, IFN-γ, IFN-ß, or TNF-α, produced mutual inhibition of several signaling pathways, e.g., NF-κB and STAT6, but also mutual activation of the PI3K signaling module. In response to combined IFN-γ and IL-4, the model predicted genes whose expression was mutually inhibited, e.g., iNOS or Nos2 and Arg1, or mutually enhanced, e.g., Il4rα and Socs1, validated by independent experimental data. Knockdown simulations further predicted network mechanisms underlying functional cross-talk, such as mutual STAT3/STAT6-mediated enhancement of Il4rα expression. In summary, the computational model predicts that network cross-talk mediates a broadened spectrum of macrophage activation in response to mixed pro- and anti-inflammatory cytokine cues, making it useful for modeling in vivo scenarios.

Logic-based mechanistic machine learning on high-content images reveals how drugs differentially regulate cardiac fibroblasts.

Fibroblasts are essential regulators of extracellular matrix deposition following cardiac injury. These cells exhibit highly plastic responses in phenotype during fibrosis in response to environmental stimuli. Here, we test whether and how candidate anti-fibrotic drugs differentially regulate measures of cardiac fibroblast phenotype, which may help identify treatments for cardiac fibrosis. We conducted a high content microscopy screen of human cardiac fibroblasts treated with 13 clinically relevant drugs in the context of TGFß and/or IL-1ß, measuring phenotype across 137 single-cell features. We used the phenotypic data from our high content imaging to train a logic-based mechanistic machine learning model (LogiMML) for fibroblast signaling. The model predicted how pirfenidone and Src inhibitor WH-4-023 reduce actin filament assembly and actin-myosin stress fiber formation, respectively. Validating the LogiMML model prediction that PI3K partially mediates the effects of Src inhibition, we found that PI3K inhibition reduces actin-myosin stress fiber formation and procollagen I production in human cardiac fibroblasts. In this study, we establish a modeling approach combining the strengths of logic-based network models and regularized regression models, apply this approach to predict mechanisms that mediate the differential effects of drugs on fibroblasts, revealing Src inhibition acting via PI3K as a potential therapy for cardiac fibrosis.

Network model integrated with multi-omic data predicts MBNL1 signals that drive myofibroblast activation.

RNA-binding protein muscleblind-like1 (MBNL1) was recently identified as a central regulator of cardiac wound healing and myofibroblast activation. To identify putative MBNL1 targets, we integrated multiple genome-wide screens with a fibroblast network model. We expanded the model to include putative MBNL1-target interactions and recapitulated published experimental results to validate new signaling modules. We prioritized 14 MBNL1 targets and developed novel fibroblast signaling modules for p38 MAPK, Hippo, Runx1, and Sox9 pathways. We experimentally validated MBNL1 regulation of p38 expression in mouse cardiac fibroblasts. Using the expanded fibroblast model, we predicted a hierarchy of MBNL1 regulated pathways with strong influence on αSMA expression. This study lays a foundation to explore the network mechanisms of MBNL1 signaling central to fibrosis.

Network model-based screen for FDA-approved drugs affecting cardiac fibrosis.

Cardiac fibrosis is a significant component of pathological heart remodeling, yet it is not directly targeted by existing drugs. Systems pharmacology approaches have the potential to provide mechanistic frameworks with which to predict and understand how drugs modulate biological systems. Here, we combine network modeling of the fibroblast signaling network with 36 unique drug-target interactions from DrugBank to predict drugs that modulate fibroblast phenotype and fibrosis. Galunisertib was predicted to decrease collagen and α-SMA expression, which we validated in human cardiac fibroblasts. In vivo fibrosis data from the literature validated predictions for 10 drugs. Further, the model was used to identify network mechanisms by which these drugs work. Arsenic trioxide was predicted to induce fibrosis by AP1-driven TGFß expression and MMP2-driven TGFß activation. Entresto (valsartan/sacubitril) was predicted to suppress fibrosis by valsartan suppression of ERK signaling and sacubitril enhancement of PKG activity, both of which decreased Smad3 activity. Overall, this study provides a framework for integrating drug-target mechanisms with logic-based network models, which can drive further studies both in cardiac fibrosis and other conditions.

Computational model predicts paracrine and intracellular drivers of fibroblast phenotype after myocardial infarction.

The fibroblast is a key mediator of wound healing in the heart and other organs, yet how it integrates multiple time-dependent paracrine signals to control extracellular matrix synthesis has been difficult to study in vivo. Here, we extended a computational model to simulate the dynamics of fibroblast signaling and fibrosis after myocardial infarction (MI) in response to time-dependent data for nine paracrine stimuli. This computational model was validated against dynamic collagen expression and collagen area fraction data from post-infarction rat hearts. The model predicted that while many features of the fibroblast phenotype at inflammatory or maturation phases of healing could be recapitulated by single static paracrine stimuli (interleukin-1 and angiotensin-II, respectively), mimicking the reparative phase required paired stimuli (e.g. TGFß and endothelin-1). Virtual overexpression screens simulated with either static cytokine pairs or post-MI paracrine dynamic predicted phase-specific regulators of collagen expression. Several regulators increased (Smad3) or decreased (Smad7, protein kinase G) collagen expression specifically in the reparative phase. NADPH oxidase (NOX) overexpression sustained collagen expression from reparative to maturation phases, driven by TGFß and endothelin positive feedback loops. Interleukin-1 overexpression had mixed effects, both enhancing collagen via the TGFß positive feedback loop and suppressing collagen via NFκB and BAMBI (BMP and activin membrane-bound inhibitor) incoherent feed-forward loops. These model-based predictions reveal network mechanisms by which the dynamics of paracrine stimuli and interacting signaling pathways drive the progression of fibroblast phenotypes and fibrosis after myocardial infarction.

Multiscale Coupling of an Agent-Based Model of Tissue Fibrosis and a Logic-Based Model of Intracellular Signaling.

Wound healing and fibrosis following myocardial infarction (MI) is a dynamic process involving many cell types, extracellular matrix (ECM), and inflammatory cues. As both incidence and survival rates for MI increase, management of post-MI recovery and associated complications are an increasingly important focus. Complexity of the wound healing process and the need for improved therapeutics necessitate a better understanding of the biochemical cues that drive fibrosis. To study the progression of cardiac fibrosis across spatial and temporal scales, we developed a novel hybrid multiscale model that couples a logic-based differential equation (LDE) model of the fibroblast intracellular signaling network with an agent-based model (ABM) of multi-cellular tissue remodeling. The ABM computes information about cytokine and growth factor levels in the environment including TGFß, TNFα, IL-1ß, and IL-6, which are passed as inputs to the LDE model. The LDE model then computes the network signaling state of individual cardiac fibroblasts within the ABM. Based on the current network state, fibroblasts make decisions regarding cytokine secretion and deposition and degradation of collagen. Simulated fibroblasts respond dynamically to rapidly changing extracellular environments and contribute to spatial heterogeneity in model predicted fibrosis, which is governed by many parameters including cell density, cell migration speeds, and cytokine levels. Verification tests confirmed that predictions of the coupled model and network model alone were consistent in response to constant cytokine inputs and furthermore, a subset of coupled model predictions were validated with in vitro experiments with human cardiac fibroblasts. This multiscale framework for cardiac fibrosis will allow for systematic screening of the effects of molecular perturbations in fibroblast signaling on tissue-scale extracellular matrix composition and organization.