MOSAIC: Imaging Human Tissue State Dynamics In Vivo

SUMMARY: PROJECT 2: IMAGING THE DYNAMIC TISSUE STATE IN PATIENTS IN VIVO With a dismal median survival of 16 months, glioblastoma (GBM) is the most common malignant primary brain tumor within adult patients. Response to the standard-of-care (SOC) is widely variable across patients. Identifying optimal targeted treatments traditionally relies on tissue sampling to identify patient-relevant targets.

Yet, tissue sampling has many severe limitations and costs (time, money, and facility access), and ultimately provides only limited scope both spatially and temporally thus always leaving behind residual tumor cells that have not been sampled. Multi-parametric magnetic resonance imaging (MRI) measures an array of

complementary physiologic biomarkers that correspond with diverse tumor phenotypes (e.g., proliferation, inflammation, angiogenesis), and it serves as the clinical mainstay for monitoring therapeutic response and disease progression. As tumor cell signaling may be mediated through interactions (i.e.,“cross-talk”) with

surrounding non-tumoral cells in the regional microenvironment, there is a critical need to define the degree to which this cross-talk influences local tissue state, phenotypic expression, and disease progression. Understanding these associations should help refine the clinical interpretations of imaging

phenotypes to improve guidelines for non-invasive diagnosis and disease monitoring. There is an urgent need for image-based radiomics tools that can 1) predict which patients will respond to a given treatment and 2) can observe/track that response over time. Overall Hypothesis: Tissue states, represented as combinations of cellular constituents and phenotypes, can be

resolved on clinical imaging to a level sufficient to identify transitions in these states with and without treatments in individual patients in vivo. Our two aims in this project investigate this hypothesis in two separate settings, Aim 1) Standard of Care, Aim 2) Immunotherapy. In these aims, we will characterize the landscape of phenotypic states, build image-based

models to predict tissue state from images, investigate how predicted tumor states correspond with outcomes, quantify dynamics of states from pre- to post-therapy, and finally build mechanistic models to understand the critical driving differences in the flow of cells in local phenotype state space leading to the overall tumor state.