Simulating the time series of a selected gene expression profile in an agent-based tumor model

Abstract

To elucidate the role of environmental conditions in molecular-level dynamics and to study their impact on macroscopic brain tumor growth patterns, the expression of the genes Tenascin C and PCNA in a 2D agent-based model for the migratory trait is calibrated using experimental data from the literature, while the expression of these genes for the proliferative trait is obtained as the model output. Numerical results confirm that the gene expression of Tenascin C is indeed consistently higher in the migratory glioma cell phenotype and show that the expression of PCNA is consistently higher among proliferating tumor cells. Intriguingly, the time series of the tumor cells’ gene expression exhibit a sudden change in behavior during the invasion of the tumor into a nutrient-abundant region, showing a robust positive correlation between the expression of Tenascin C and the tumor’s diameter, yet a strong negative correlation between the expression of PCNA and the diameter. These molecular-level dynamics correspond to the emergence of a structural asymmetry in the form of a bulging tumor rim in the nutrient-abundant region. The simulated time series thus supports the critical role of the migratory cell phenotype during both the tumor system’s overall macroscopic expansion and the evolvement of regional growth patterns, particularly in the later stages. Furthermore, detrended fluctuation analysis (DFA) suggests that for prediction purposes, the simulated gene expression profiles of Tenascin C and PCNA that were determined separately for the migrating and proliferating phenotypes exhibit lesser predictability than those of the phenotypic mixture combining all viable tumor cells typically found in clinical biopsies. Finally, partitioning the tumor into distinct geographic regions of interest (ROI) reveals that the gene expression profile of tumor cells in the quadrant close to the nutrient-abundant region is representative for the entire tumor whereas the expression profile of tumor cells in the geographically opposite ROI is not. Potential implications of these modeling results for experimental and clinical cancer research are discussed.