(A-30) Stochastic Modeling of T Cell Killing in Melanoma

Introduction:: Cancer is a major health concern worldwide, and the adaptive immune system plays a crucial role in suppressing cancer progression. To reinvigorate the immune response against tumors, immunotherapy has emerged as a promising approach. Therapeutic design and tumor environment dynamics can be inferred from mechanistic modeling [1]. While ordinary differential equation (ODE)-based models have been widely used, they are limited in capturing cell type heterogeneities and the spatio-temporal dynamics of individual cells [2]. Agent-based modeling (ABM) is an alternative approach that can overcome such shortcomings and accomplish multi-scale coupling to inform individual cell fate from molecular interactions [3]. However, the inclusion of finer mechanistic details and spatial resolution brings high computational burdens. To speed up ABM simulations, the model solution can be sampled by the Gillespie algorithm with Poisson leaping [4]. This paper develops a stochastic mechanistic model framework that employs agent-based modeling and Gillespie algorithm to reveal the role of T cell killing in driving tumor regression.

Materials and Methods:: Our model includes four cell types: tumor cells, intermediate T cells, activated cytotoxic T cells, and antigen-presenting cells (APCs). Each cell type has its own source, proliferation, and degradation rates. Cell-cell interactions are modeled by two-step competitive binding of surface proteins: after primary activation, the probability of secondary stimulation or inhibition determines cell fates. Specifically, APCs can activate intermediate T cells, and activated T cells can kill tumor cells or get inhibited via PD1-PDL1 binding. To faithfully capture the details of these physiological processes, ABM simulations are performed in NetLogo with a spatial setup of immobilized tumor cells and free-floating immune cells. Meanwhile, the Gillespie algorithm is implemented in Python for sensitivity analysis and parameterization based on current literature and a temporal single-cell RNA-seq murine melanoma dataset [5]. The performance of the Poisson leaping-approximated Gillespie algorithm is then compared to and validated by ABM simulations.

Results, Conclusions, and Discussions:: The stochastic model developed can capture tumor progression dynamics with cell type heterogeneity and randomness. In comparison to ABM, the Gillespie algorithm with Poisson leaping significantly speeds up simulations by more than 100-fold, while preserving trajectory sample mean and variance. Immune-sensitive and immune-unresponsive states are identified from parameter optimization on the melanoma dataset. In the immune-sensitive state, the intermediate T cell population increases drastically, as a source for activated T cell expansion, and the tumor cell population follows an exponential decay. Tumor regression is the most sensitive to MHC and TCR concentrations on active T cells and tumor cells, respectively, while the source rate of APC drives the peaking of activated T cells. In the immune-unresponsive state, active T cells are sustained at an abundant level yet fail to interact with an exponentially growing tumor cell population. The active T cell population is maintained by the decrease in naïve T cells recruited from lymph nodes to cancer sites and the increase in APCs presenting tumor antigens. The tumor population is no longer sensitive to immune-related parameters and, instead, depends on its natural proliferation rate. This model framework can be extended to other types of cancers and provides insights for designing effective immunotherapies.


Acknowledgements (Optional): : I would like to express my sincere gratitude to my mentor Dr. Feilim MacGabhann for his patience, guidance, and support of this project. I would like to thank the authors and those who greatly inspired me in the project: Wanyan, for her immense knowledge of immunology; Yunxiao, for his expertise in Gillespie algorithm and high-performance computing; Dr. Subas Acharya, for his insightful suggestions in stochastic processes; Dr. Donald Geman, for his kind affirmation in my project ambition; Ethan Levy, for his enthusiasm in formulating the project motivation. Last, I would like to thank my family for their unconditional love for supporting me in pursuing science.


References (Optional): : [1] Darvin, Pramod, et al. "Immune checkpoint inhibitors: recent progress and potential biomarkers." Experimental & molecular medicine 50.12 (2018): 1-11.

[2] Figueredo, Grazziela P., et al. "Comparing stochastic differential equations and agent-based modelling and simulation for early-stage cancer." PloS one 9.4 (2014): e95150.

[3] Gong, Chang, et al. "A computational multiscale agent-based model for simulating spatio-temporal tumour immune response to PD1 and PDL1 inhibition." Journal of the Royal Society Interface 14.134 (2017): 20170320.

[4] Gillespie, Daniel T. "Exact stochastic simulation of coupled chemical reactions." The journal of physical chemistry 81.25 (1977): 2340-2361.

[5] Davidson, Sarah, et al. "Single-cell RNA sequencing reveals a dynamic stromal niche that supports tumor growth." Cell reports 31.7 (2020): 107628.