Determinants of combination GM-CSF immunotherapy and oncolytic virotherapy success identified through in silico treatment personalization

Abstract

Oncolytic virotherapies, including the modified herpes simplex virus talimogene laherparepvec (T-VEC), have shown great promise as potent instigators of anti-tumour immune effects. The OPTiM trial, in particular, demonstrated the superior anti-cancer effects of T-VEC as compared to systemic immunotherapy treatment using exogenous administration of granulocyte-macrophage colony-stimulating factor (GM-CSF). Theoretically, a combined approach leveraging exogenous cytokine immunotherapy and oncolytic virotherapy would elicit an even greater immune response and improve patient outcomes. However, regimen scheduling of combination immunostimulation and T-VEC therapy has yet to be established. Here, we calibrate a computational biology model of sensitive and resistant tumour cells and immune interactions for implementation into an in silico clinical trial to test and individualize combination immuno- and virotherapy. By personalizing and optimizing combination oncolytic virotherapy and immunostimulatory therapy, we show improved simulated patient outcomes for individuals with late-stage melanoma. More crucially, through evaluation of individualized regimens, we identified determinants of combination GM-CSF and T-VEC therapy that can be translated into clinically-actionable dosing strategies without further personalization. Our results serve as a proof-of-concept for interdisciplinary approaches to determining combination therapy, and suggest promising avenues of investigation towards tailored combination immunotherapy/oncolytic virotherapy.

Fig 1. Pictorial representation of the tumour growth model.

Quiescent cells activate to begin division by transiting into the G₁ phase of the cell-cycle. Cells exit G₁ to enter the active phase and complete division. Most susceptible cells in the active phase re-enter quiescence after mitosis, however certain dividing cells may mutate into an immuno-resistant lineage (red dotted arrow). Immune interactions are driven by phagocytes who come into contact with quiescent and G₁ phase susceptible cells (dashed yellow lines). Tumour-immune interactions increase pro-inflammatory cytokine concentrations to recruit additional phagocytes to the tumour site (blue dotted line). Cells and cytokine are denoted by circles, processes by squares, and rates by arrows.

Fig 2. Summary of in silico clinical trial platform.

a) Individual in silico patient parameter values are sampled from a normal distribution of values based on an average parameterization. The model is then simulated for each individual and predictions are tested for physiological relevance. If realistic, the virtual individual is cloned n times and each clone is assigned to n separate cohorts. b) Each cohort undergoes a different treatment protocol, from which summary statistics are collated and compared between other cohorts. Cohorts may also undergo therapy optimization (see Methods) from which an empirical distribution for the probability of administering a given dose is inferred. c) The probabilistic treatment protocol is based on probability distributions inferred from the individualization based on procedures in a and b.

Fig 3. Treatment with oncolytic virus provides improved outcomes over immunotherapy in virtual clinical trial.

A) Kaplan-Meier curves for patients in the immunotherapy and virotherapy arms of the virtual trial; B) The relative survival benefit for identical virtual patients. The ratio of survival time on T-VEC against survival time on GM-CSF for identical virtual patients (line of best fit, slope = 0.0035) establishes a causal relationship between treatment type and survival time, indicating that oncolytic virus therapy provided slightly larger survival gains in those with longer doubling times when compared to GM-CSF.

Fig 4. Optimal personalized dose scheduling for each of the 300 virtual patients.

Dose size presented as a multiple of the standard dose with immunotherapy in shades of purple, and virotherapy in shades of green. The nth horizontal row corresponds to the nth virtual patient, while the m-th vertical column corresponds to the dose administered on day m.

Fig 5. In silico clinical trial predicts improved outcomes for both probabilistic dosing strategies and maintenance therapy versus standard combination therapy.

Kaplan-Meier curves for Arm 1: patients receiving Standard Combination Therapy (dotted turquoise line), Arm 2: Maintenance Treatment (solid light blue line), Arm 3: Probabilistic dosing regimen determined through the in silico clinical trial (dashed dark blue line).