Multiscale spatiotemporal modeling of acute primary viral infection and immune response in epithelial tissues

Simulations of tissue-specific effects of primary acute viral infections like COVID-19 are essential for understanding differences in disease outcomes and optimizing therapeutic interventions. We present a multiscale model and simulation of an epithelial tissue infected by a virus, a simplified cellular immune response and viral and immune-induced tissue damage. The model exhibits basic patterns of infection dynamics: widespread infection, slowed infection, recurrence, containment and clearance. Inhibition of viral internalization and faster immune-cell recruitment promote containment of infection. Fast viral internalization and slower immune response lead to uncontrolled spread of infection. Because antiviral drugs can have side effects at high doses and show reduced clinical effectiveness when given later during the course of infection, we studied the effects on infection progression of both treatment potency (which combines drug effectiveness and dosage) and time-of-first treatment after infection. Simulation of a drug which reduces the replication rate of viral RNA shows that even a low potency therapy greatly decreases the total tissue damage and virus burden when given near the beginning of infection. However, even a high potency therapy rapidly loses effectiveness when given later near the time of peak viral load in the untreated case. Many combinations of dosage and treatment time lead to stochastic outcomes, with some simulation replicas showing clearance or control of the virus (treatment success), while others show rapid infection of all epithelial cells in the simulated tissue subregion (treatment failure). This switch between a regime of consistent treatment success and failure occurs as the time of treatment increases. However, stochastic variations in viral spread mean that high potency treatments at late times are occasionally effective. The model is open-source and modular, allowing rapid development and extension of its components by groups working in parallel. We're extending the model through already calibrated ODE models to have more biological meaningful behaviors. ODE models can be calibrated in a straightforward manner, however they don't contain information about space which is meaningful. As ODE based models are not spatial they need to be spatialized in some manner for our use, we've developed a method to generate spatial models from ODE models and have the spatial model recover the ODE predicted population behaviors. While the spatialized model shows differences, we recuperate the overall ODE model behavior, and we are exploring how spatiality itself causes those differences. With this work we will bring forth more ways in which ODE models can be useful, e.g., having their overall behavior inform and predict how the COVID infection spreads through the lungs.