Treatment failure in cancer is often driven by the emergence and expansion of treatment-resistant cell populations. Evolution-based adaptive therapies aim to exploit competitive interactions between drug-sensitive and drug-resistant cells to extend disease control. In solid tumors, however, competition is shaped not only by cell-intrinsic fitness differences but also by local physical interactions arising from tissue architecture and cell packing. The authors argue that successful physics-informed adaptive therapy requires integrating tissue mechanics with experimentally grounded models of tumor evolution to link single-cell interactions to population-level outcomes.
To investigate how physical interactions influence therapy response, the study used genetically tailored, therapy-mimicking experiments in spatially structured 2D melanoma colonies. These wet-lab experiments were combined with mechanistic, particle-based tissue simulations that served as digital twins. The authors describe a real-to-sim-to-real workflow in which observations from experiments inform simulations, and simulation-derived insights guide further experimental tests. This integrative approach was used to probe how mechanical interactions between individual cells give rise to macroscopic treatment responses.
Across the experimental and simulation systems, the investigators observed that changes in local mechanical crowding induced by treatment alter how resistant cells are spatially confined or permitted to expand. Specifically, treatment-induced modulation of cellular crowding initiated a cascade of events: altered crowding led to changes in the spatial organization of sensitive and resistant cells, enabling resistant clones to expand and to spatially escape competitive constraints, which in turn precipitated therapy failure. These results highlight that mechanical crowding and local cell–cell physical interactions can be primary determinants of whether resistance remains confined or becomes invasive at the population level.
Recognizing that the physical mechanisms observed at the cellular level have implications for treatment scheduling and strategy, the authors abstracted the underlying dynamics into an aggregated mathematical model. This abstraction linked mechanistic drivers—crowding, confinement, and escape—to observables that could be monitored at the population scale. By mapping the physical mechanisms to simpler model parameters, the authors created a bridge between detailed mechanobiology and higher-level models suitable for therapy optimization.
Building on the abstracted mathematical model, the team applied a reinforcement-learning framework to explore therapy optimization in the presence of mechanics-driven resistance dynamics. From this computational optimization they derived a distilled, rule-based treatment-decision heuristic intended to capture the key element(s) of the optimized policies in a more interpretable form. The heuristic was designed to be experimentally testable and to embody the mechanistic insight that controlling or anticipating changes in crowding and escape can influence long-term treatment control.
The authors report that the distilled treatment-decision heuristic was tested experimentally in their 2D melanoma system. This real-world test served to connect simulation-driven policy design back to the biological system and to evaluate whether the mechanistically derived heuristic can alter treatment outcomes in practice. The experiments, simulations, and mathematical abstraction together supported the conclusion that monitoring and responding to mechanics-related population observables can inform adaptive therapy decisions.
The combined experimental and computational results identify mechanical crowding and escape as central, governing principles in resistance-mediated therapy failure for solid tumors in this 2D melanoma model. The study provides a conceptual and methodological basis for integrating tissue mechanics into evolution-informed, adaptive treatment strategies. By demonstrating a real-to-sim-to-real pipeline—from single-cell mechanical interactions to reinforcement-learning optimization and an experimentally tested heuristic—the work suggests that mechanistically informed decision rules may improve adaptive therapy performance in spatially structured tumors.
This summary is based on the abstract and front-matter of the preprint. Specific experimental parameters, quantitative outcomes, detailed algorithmic formulations, and supplementary results were not provided in the abstract and are not reported here. The article is a preprint and has not undergone peer review; consult the full manuscript and supplementary material for complete methods and data.