Predicting the probabilistic complex invasive map of GBM cells

Division Of Basic Sciences - Nci

Investigators

Abstract

Glioblastoma multiforme (GBM) is the most lethal adult primary brain cancer and remains incurable despite decades of research. Some of the challenges of dealing with the disease are due to the wide inter- and intra-tumoral heterogeneity. However, there is substantial evidence that GBM development, and mainly its spread, is not solely a cell intrinsic process driven by molecular perturbations, but that it also depends on the brain's topology and anatomy, including their diffusion properties. Several GBM invasion paths were well identified into the brain including the 1) extracellular matrix (ECM) of the brain parenchyma, 2) perivascular space, and 3) the white matter tracts, in addition to the 4) initial spread from the bulk of the tumor. In this program, we have been focused on two interconnected key possible mechanisms concerning the spatiotemporal evolution of the GBM cells. Could the different molecular states of a given GBM cell be sufficient stimulators to induce its decision to move with clear direction? or independent of a cell's molecular state, is the spread of the cancer primarily due to the physical properties of the brain geometry and anatomy? Or does reality lie somewhere between these two concepts? Each one of these possibilities would lead us to different treatments. A better understanding of the fundamental processes of the dynamics in spatial GBM spread, given molecular up to organ scale patient information, may pave the way toward a robust treatment approach that combats these survival mechanisms. The goal of these 2 projects is to address the tumor spread mechanisms by developing computational approaches to track motile cells (Project 1), and understand their phenotype transition (Project 2). To do this, we use information on the molecular systems (such as phenotype landscape), and the image signatures of the patient's brain and tumor, with or without specific treatments. Once we describe and quantify those systems, we can offer robust ways to predict spatiotemporal spread of those tumors given a treatment. Our work includes a wide range of computational approaches including on the single cell (Boolean network, complex dynamic network, nonlinear dynamical systems (ODEs)), cell population (ABM, PDEs FE) and organ scale (CNN, ML, fusion data, PDEs FE, image processing), as we explore and integrate broader measures of brain tumor behaviors, brain structure and function as sensors of tumor progression over space and time (single-cell and spatial tumor gene expression, MR time-series data, DTI, MRA, prefusion, clinical treatment and outcome etc.). We have developed tools that mix theoretical models and learning algorithms to overcome challenges in cancer research, based on complex patient datasets. It is essential to place such models, experimental or computational, in an anatomically realistic setting, which requires an interdisciplinary approach combining the fields of biology, biotechnology, mathematical modeling, scientific computation, and medical imaging. Work on this project concluded in FY21 following Dr. Lavi's departure.