Mapping CAR T-Cell Design Space Using Agent-Based Models

Agent-based modeling in cancer biomedicine: applications and tools for calibration and validation

Computational models are not just appealing because they can simulate and predict the development of biological phenomena across multiple spatial and temporal scales, but also because they can integrate information from well-established in vitro and in vivo models and test new hypotheses in cancer biomedicine. Agent-based models and simulations are especially interesting candidates among computational modeling procedures in cancer research due to the capability to, for instance, recapitulate the dynamics of neoplasia and tumor - host interactions. Yet, the absence of methods to validate the consistency of the results across scales can hinder adoption by turning fine-tuned models into black boxes. This review compiles relevant literature that explores strategies to leverage high-fidelity simulations of multi-scale, or multi-level, cancer models with a focus on verification approached as simulation calibration. We consolidate our review with an outline of modern approaches for agent-based models' validation and provide an ambitious outlook toward rigorous and reliable calibration.

In silico study of heterogeneous tumour-derived organoid response to CAR T-cell therapy

Chimeric antigen receptor (CAR) T-cell therapy is a promising immunotherapy for treating cancers. This method consists in modifying the patients' T-cells to directly target antigen-presenting cancer cells. One of the barriers to the development of this type of therapies, is target antigen heterogeneity. It is thought that intratumour heterogeneity is one of the leading determinants of therapeutic resistance and treatment failure. While understanding antigen heterogeneity is important for effective therapeutics, a good therapy strategy could enhance the therapy efficiency. In this work we introduce an agent-based model (ABM), built upon a previous ABM, to rationalise the outcomes of different CAR T-cells therapies strategies over heterogeneous tumour-derived organoids. We found that one dose of CAR T-cell therapy should be expected to reduce the tumour size as well as its growth rate, however it may not be enough to completely eliminate it. Moreover, the amount of free CAR T-cells (i.e. CAR T-cells that did not kill any cancer cell) increases as we increase the dosage, and so does the risk of side effects. We tested different strategies to enhance smaller dosages, such as enhancing the CAR T-cells long-term persistence and multiple dosing. For both approaches an appropriate dosimetry strategy is necessary to produce "effective yet safe" therapeutic results. Moreover, an interesting emergent phenomenon results from the simulations, namely the formation of a shield-like structure of cells with low antigen expression. This shield turns out to protect cells with high antigen expression. Finally we tested a multi-antigen recognition therapy to overcome antigen escape and heterogeneity. Our studies suggest that larger dosages can completely eliminate the organoid, however the multi-antigen recognition increases the risk of side effects. Therefore, an appropriate small dosages dosimetry strategy is necessary to improve the outcomes. Based on our results, it is clear that a proper therapeutic strategy could enhance the therapies outcomes. In that direction, our computational approach provides a framework to model treatment combinations in different scenarios and to explore the characteristics of successful and unsuccessful treatments.

Reinforcement learning-guided control strategies for CAR T-cell activation and expansion

Reinforcement learning (RL), a subset of machine learning (ML), could optimize and control biomanufacturing processes, such as improved production of therapeutic cells. Here, the process of CAR T-cell activation by antigen-presenting beads and their subsequent expansion is formulated in silico. The simulation is used as an environment to train RL-agents to dynamically control the number of beads in culture to maximize the population of robust effector cells at the end of the culture. We make periodic decisions of incremental bead addition or complete removal. The simulation is designed to operate in OpenAI Gym, enabling testing of different environments, cell types, RL-agent algorithms, and state inputs to the RL-agent. RL-agent training is demonstrated with three different algorithms (PPO, A2C, and DQN), each sampling three different state input types (tabular, image, mixed); PPO-tabular performs best for this simulation environment. Using this approach, training of the RL-agent on different cell types is demonstrated, resulting in unique control strategies for each type. Sensitivity to input-noise (sensor performance), number of control step interventions, and advantages of pre-trained RL-agents are also evaluated. Therefore, we present an RL framework to maximize the population of robust effector cells in CAR T-cell therapy production.

Incorporating temporal information during feature engineering bolsters emulation of spatio-temporal emergence

MOTIVATION

Emergent biological dynamics derive from the evolution of lower-level spatial and temporal processes. A long-standing challenge for scientists and engineers is identifying simple low-level rules that give rise to complex higher-level dynamics. High-resolution biological data acquisition enables this identification and has evolved at a rapid pace for both experimental and computational approaches. Simultaneously harnessing the resolution and managing the expense of emerging technologies-e.g. live cell imaging, scRNAseq, agent-based models-requires a deeper understanding of how spatial and temporal axes impact biological systems. Effective emulation is a promising solution to manage the expense of increasingly complex high-resolution computational models. In this research, we focus on the emulation of a tumor microenvironment agent-based model to examine the relationship between spatial and temporal environment features, and emergent tumor properties.

RESULTS

Despite significant feature engineering, we find limited predictive capacity of tumor properties from initial system representations. However, incorporating temporal information derived from intermediate simulation states dramatically improves the predictive performance of machine learning models. We train a deep-learning emulator on intermediate simulation states and observe promising enhancements over emulators trained solely on initial conditions. Our results underscore the importance of incorporating temporal information in the evaluation of spatio-temporal emergent behavior. Nevertheless, the emulators exhibit inconsistent performance, suggesting that the underlying model characterizes unique cell populations dynamics that are not easily replaced.

AVAILABILITY AND IMPLEMENTATION

All source codes for the agent-based model, emulation, and analyses are publicly available at the corresponding DOIs: 10.5281/zenodo.10622155, 10.5281/zenodo.10611675, 10.5281/zenodo.10621244, respectively.

Design and mathematical analysis of activating transcriptional amplifiers that enable modular temporal control in synthetic juxtacrine circuits

The ability to control mammalian cells such that they self-organize or enact therapeutic effects as desired has incredible implications. Not only would it further our understanding of native processes such as development and the immune response, but it would also have powerful applications in medical fields such as regenerative medicine and immunotherapy. This control is typically obtained by synthetic circuits that use synthetic receptors, but control remains incomplete. The synthetic juxtacrine receptors (SJRs) are widely used as they are fully modular and enable spatial control, but they have limited gene expression amplification and temporal control. As these are integral facets to cell control, I therefore designed transcription factor based amplifiers that amplify gene expression and enable unidirectional temporal control by prolonging duration of target gene expression. Using a validated in silico framework for SJR signaling, I combined these amplifiers with SJRs and show that these SJR amplifier circuits can direct spatiotemporal patterning and improve the quality of self-organization. I then show that these circuits can improve chimeric antigen receptor (CAR) T cell tumor killing against various heterogenous antigen expression tumors. These amplifiers are flexible tools that improve control over SJR based circuits with both basic and therapeutic applications.

PASCAR: a multiscale framework to explore the design space of constitutive and inducible CAR T cells

CAR T cells are engineered to bind and destroy tumor cells by targeting overexpressed surface antigens. However, healthy cells expressing lower abundances of these antigens can also be lysed by CAR T cells. Various CAR T cell designs increase tumor cell elimination, whereas reducing damage to healthy cells. However, these efforts are costly and labor-intensive, constraining systematic exploration of potential hypotheses. We develop a protein abundance structured population dynamic model for CAR T cells (PASCAR), a framework that combines multiscale population dynamic models and multi-objective optimization approaches with data from cytometry and cytotoxicity assays to systematically explore the design space of constitutive and tunable CAR T cells. PASCAR can quantitatively describe in vitro and in vivo results for constitutive and inducible CAR T cells and can successfully predict experiments outside the training data. Our exploration of the CAR design space reveals that optimal CAR affinities in the intermediate range of dissociation constants effectively reduce healthy cell lysis, whereas maintaining high tumor cell-killing rates. Furthermore, our modeling offers guidance for optimizing CAR expressions in synthetic notch CAR T cells. PASCAR can be extended to other CAR immune cells.

A systems and computational biology perspective on advancing CAR therapy

In the recent decades, chimeric antigen receptor (CAR) therapy signaled a new revolutionary approach to cancer treatment. This method seeks to engineer immune cells expressing an artificially designed receptor, which would endue those cells with the ability to recognize and eliminate tumor cells. While some CAR therapies received FDA approval and others are subject to clinical trials, many aspects of their workings remain elusive. Techniques of systems and computational biology have been frequently employed to explain the operating principles of CAR therapy and suggest further design improvements. In this review, we sought to provide a comprehensive account of those efforts. Specifically, we discuss various computational models of CAR therapy ranging in scale from organismal to molecular. Then, we describe the molecular and functional properties of costimulatory domains frequently incorporated in CAR structure. Finally, we describe the signaling cascades by which those costimulatory domains elicit cellular response against the target. We hope that this comprehensive summary of computational and experimental studies will further motivate the use of systems approaches in advancing CAR therapy.

Investigating Two Modes of Cancer-Associated Antigen Heterogeneity in an Agent-Based Model of Chimeric Antigen Receptor T-Cell Therapy

Simple Summary

Chimeric antigen receptor (CAR) T-cell therapy has shown much promise in liquid tumors but often fails in solid tumors. This work uses a computational model to examine under what conditions this therapy might fail or be successful. The model includes interactions between cancer cells, CAR T-cells (treatment), and vascular cells (that feed and support tumor growth). From our results, we determined specific tumor conditions in which CAR T-cell therapy is predicted to fail and suggest a combination treatment that might improve the efficacy of the treatment.

Abstract

Chimeric antigen receptor (CAR) T-cell therapy has been successful in treating liquid tumors but has had limited success in solid tumors. This work examines unanswered questions regarding CAR T-cell therapy using computational modeling, such as, what percentage of the tumor must express cancer-associated antigens for treatment to be successful? The model includes cancer cell and vascular and CAR T-cell modules that interact with each other. We compare two different models of antigen expression on tumor cells, binary (in which cancer cells are either susceptible or are immune to CAR T-cell therapy) and gradated (where each cancer cell has a probability of being killed by a CAR T-cell). We vary the antigen expression levels within the tumor and determine how effective each treatment is for the two models. The simulations show that the gradated antigen model eliminates the tumor under more parameter values than the binary model. Under both models, shielding, in which the low/non-antigen-expressing cells protect high antigen-expressing cells, reduced the efficacy of CAR T-cell therapy. One prediction is that a combination of CAR T-cell therapies that targets the general population of cells as well as one that specifically targets cancer stem cells should increase its efficacy.