B. Price1, G. An1, C. Cockrell1 1University Of Vermont College Of Medicine / Fletcher Allen Health Care, Surgery, Burlington, VT, USA
Introduction: The natural history of volumetric muscle loss (VML), such as wounds of the type suffered from significant blast-injury, is healing via scar formation. The functional recovery of patients with VML would be greatly enhanced if the VML could be replaced with functional muscle instead of scar. We have performed simulation experiments with an agent-based model of VML that have suggested a crucial tipping point in terms of scar formation versus muscle regrowth that is heavily influenced by the differentiation of fibroblasts to myofibroblasts. A review of the literature suggested an important role for α-smooth muscle actin (α-SMA) in the transformation of fibroblasts into myofibroblasts, and this prompted an in silico examination of the effect of inhibition of α-SMA accumulation on the ability to increase the muscle composition of a VML.
Methods: We have developed a mechanism-based agent-based model (ABM) to simulate the temporal evolution of a volumetric muscle loss (VML). Cells represented in the ABM include satellite stem cells, myoblasts, fibroblasts, and immune cell subtypes. Machine-learning techniques are used to develop a set of model parameterizations which encompass the biological heterogeneity seen clinically. Simulation experiments were performed that used Deep Reinforcement Learning (DRL) to incorporate the effects of inhibiting α-SMA accumulation along with other immunomodulatory controls to steer the system to maximize muscle in the healed wound.
Results: The DRL algorithm discovered treatment policies which increased the amount of functional muscle regenerated from 2% (in the untreated case) to 80% (when AI-controlled). The discovered therapeutic policy was complex, but upon examination, two primary objectives could be inferred: 1) control of inflammation such that muscle regeneration is not impaired, and 2) sufficient interruption of the differentiation of fibroblasts into myofibroblasts by inhibition of α-SMA accumulation, thus interrupting the conversion of granulation tissue to fibrotic scar.
Conclusion: In this work, we provide an in silico proof-of-concept study that demonstrates the importance of appropriate suppression of myofibroblast transformation in order to increase final wound muscle composition, and the need for a multi-modal intervention policy to accomplish this goal. These findings are a further demonstration of the utility of an integrated workflow that uses high-fidelity mechanism-based simulation with advanced machine learning and artificial intelligence to discover potential control strategies for complex pathophysiological processes.