Towards a Multiscale Modeling of Cardiac Allograft Vasculopathy: An Agent-Based Computational Approach

Cardiac Allograft Vasculopathy (CAV) is one of the major manifestations of chronic rejection following heart transplantation (HTx). It is a progressive form of coronary artery disease (CAD), characterized by concentric and diffuse intimal thickening, leading to significant lumen narrowing and graft failure. CAV arises from the interplay of i) immunological processes, such as infiltration of lymphocytes (LYMs) and macrophages (MPs), and ii) non-immunological factors like disturbed hemodynamics. Today, its underlying mechanisms remain poorly understood. Current diagnostic methods rely on invasive procedures and remain limited in their ability to detect the disease at an early stage. Thus, effective treatments are scarce, with re-transplantation remaining the only viable long-term solution, although challenging to access after a first rejection. While in vivo models have been the golden standard for CAV research, they are constrained by ethical issues, restricted access to key time points, and challenges in acquiring multiscale measurements. In this context, in silico models offer a complementary tool to overcome the limitations of in vivo studies. In this work, it is hypothesized that a multiscale computational model of CAV will simulate the disease with high fidelity and temporal resolution, providing mechanistic insights into the interplay between its main drivers. Accordingly, this work aims to develop a 2D agent-based model (ABM) to replicate the CAV pathogenesis over a 4-week follow-up. Disease progression was modeled via i) immune cell infiltration, ii) cell wall migration driven by chemoattractant molecules released by inflammatory cells, and iii) inward remodeling. Specifically, starting from 3D reconstructions of murine Left Coronary Artery (LCA) geometries (n=6), 2D cross-sections were extracted from each artery and used as spatial domains of the ABM simulations. The arterial wall was initialized with agents identifying LYMs, MPs, smooth muscle cells (SMCs) and extracellular matrix (ECM). Probabilistic rules were assigned to the different agents, and their activities (cell proliferation/migration and ECM synthesis/degradation) were governed via Monte Carlo method to replicate biological variability. Finally, hemodynamic and immunological inputs captured the multiscale features of the vasculopathy. To account for both pathology triggers, the ABM was coupled with two external modules. A computational fluid dynamics model computed wall shear stress distributions as hemodynamic input, while a LYM transport model estimated local LYM concentrations as immunological input. Model calibration was performed using literature data to reproduce both the observed degree of stenosis and known trends in immune cell infiltration. A batch of 3000 simulations was used to train a Random Forest Regressor and optimize the parameters driving the ABM’s probabilistic equations. Once calibrated, the ABM successfully captured key features of CAV onset and evolution, including progressive immune cell accumulation, concentric intimal thickening, and lumen narrowing over the simulated follow-up period. This work demonstrates the potential of multiscale in silico modeling to investigate CAV evolution. The ABM’s flexibility allows for the dynamic integration of heterogeneous biological inputs and scales, making it a powerful tool to simulate disease progression and investigate the interplay between CAV drivers. Future work will focus on validating the model with in vivo data and identifying new biomarkers.