Mathematical Modeling of Novel Cancer Immunotherapies

Immunotherapy has received great attention recently, as it has become a powerful tool in fighting certain types of cancer. Immunotherapeutic drugs strengthen the immune system's natural ability to identify and eradicate cancer cells. This work focuses on immune checkpoint inhibitor and oncolytic virus therapies. Immune checkpoint inhibitors act as blocking mechanisms against the binding partner proteins, enabling T-cell activation and stimulation of the immune response. Oncolytic virus therapy utilizes genetically engineered viruses that kill cancer cells upon lysing. To elucidate the interactions between a growing tumor and the employed drugs, mathematical modeling has proven instrumental. This dissertation introduces and analyzes three different ordinary differential equation models to investigate tumor immunotherapy dynamics.

The first model considers a monotherapy employing the immune checkpoint inhibitor anti-PD-1. The dynamics both with and without anti-PD-1 are studied, and mathematical analysis is performed in the case when no anti-PD-1 is administrated. Simulations are carried out to explore the effects of continuous treatment versus intermittent treatment. The outcome of the simulations does not demonstrate elimination of the tumor, suggesting the need for a combination type of treatment.

An extension of the aforementioned model is deployed to investigate the pairing of an immune checkpoint inhibitor anti-PD-L1 with an immunostimulant NHS-muIL12. Additionally, a generic drug-free model is developed to explore the dynamics of both exponential and logistic tumor growth functions. Experimental data are used for model fitting and parameter estimation in the monotherapy cases. The model is utilized to predict the outcome of combination therapy, and reveals a synergistic effect: Compared to the monotherapy case, only one-third of the dosage can successfully control the tumor in the combination case.

Finally, the treatment impact of oncolytic virus therapy in a previously developed and fit model is explored. To determine if one can trust the predictive abilities of the model, a practical identifiability analysis is performed. Particularly, the profile likelihood curves demonstrate practical unidentifiability, when all parameters are simultaneously fit. This observation poses concerns about the predictive abilities of the model. Further investigation showed that if half of the model parameters can be measured through biological experimentation, practical identifiability is achieved.