A Data-Enhanced Agent-Based Model for Simulating 3D Cancer Spheroid Growth: Integrating Metabolism and Mechanics

TL;DR

The study addresses how tumor metabolism and mechanics jointly drive 3D cancer spheroid growth. It introduces a data-enhanced agent-based model that couples an ATP-driven metabolic network with a mechanical interaction framework, calibrated using Bayesian inference and a Gaussian-process surrogate trained on microfluidic spheroid data. The results show the model can reproduce experimental growth trajectories, reveal key parameters governing spheroid size, and predict how ECM density modulates spheroid architecture, with epithelial-like phenotypes dampening ECM effects. This integrative approach bridges in vitro observations and in silico predictions, offering a versatile tool for exploring microenvironmental influences on tumor development.

Abstract

Cancer research has shifted from a purely gene-centric view to a more holistic understanding that recognizes the critical role of the tumour microenvironment, where mechanics and metabolism are key drivers of disease progression. However, the intricate interplay between these multifactorial mechanisms remains poorly understood. To address this gap, we present an agent-based computational model (ABM) that integrates tumour metabolism and mechanics to study 3D cancer spheroid growth. Our approach unifies the metabolism and mechanical aspects of tumour development within an integral model for cancer spheroid formation and growth. In addition to that, we performed a computational calibration of the parameters and tested the model versatility to reproduce different cellular behaviours. Our model reproduced qualitatively and quantitatively the experimental results of spheroid growth obtained in the lab and also allowed to discern different dynamics that cancer cells can present under the same conditions, providing insight into the potential factors contributing to the variability in the size of spheroids. Furthermore, it also showed its adaptability to reproduce diferent cell lines and behaviours by tuning its parameters. This study highlights the significant potential and versatility of integrative modelling approaches in the field of cancer research, not only as a tool to complement in vitro studies, but also as independent tools to derive conclusions from the physical reality.

Figures (16)

• Figure 1: a) Geometry and design of the microfluidic devices described in the present work, with a central chamber containing the gel with the 3D cells embedded and two side channels to feed the nutrients. plou_individual_2018 b) Size comparison between a microfluidic chip and a 1 euro coin. c) A scheme of the 2D area projection of the spheroid measured in silico and in vitro.
• Figure 2: a) Scheme of the forces that act in the cell as part of the mechanic model. b) Graphic representation of the attraction and repulsion potential between two cells where the x-axis represents r = $x_j$ - $x_i$, being $x_i$ and $x_j$ the position of the particle $i$ and $j$ . $R$ an $R_A$ are the characteristic distances of the two potentials. These two distances define the range of action of both potential functions.
• Figure 3: Scheme of the different states the cell can have in the model. $ATP_{prolif}$ level establish over which level of ATP the cells proliferate and migrate. $ATP_{death}$ establish below which level cells die. If the cell's ATP levels are between these two levels, cells are quiescent and they do not migrate nor proliferate. Depending on the ATP level of the cells, different forces play a rol.
• Figure 4: Schematic representation of the coupling of the mechanical and metabolical models.
• Figure 5: General scheme of the calibration process. First, a Gaussian process is trained using outputs from multiple simulations of the computational model with different parameter combinations (green boxes). This surrogate model is then employed to perform a sensitivity analysis, identifying influential and non-influential parameters (yellow boxes). Finally, Bayesian calibration is carried out using only the GP with the experimental data available to obtain the calibrated parameter set (red boxes).