Glioblastoma (GBM) exhibits profound intratumoral molecular heterogeneity that contributes to treatment resistance and poor survival. Specifically, each tumor comprises multiple molecularly-distinct subpopulations with different treatment sensitivities. This heterogeneity not only portends the pre-existence of resistant molecular subpopulations, but also the communications between neighboring subpopulations that further modulate tumorigenicity and resistance. In fact, a minority tumor subpopulation with EGFRvIII mutation has been shown to potentiate a majority subpopulation with wild-type EGFR to increase tumor growth, cell survival, and drug resistance. This type of cooperativity presents clear implications for improving GBM treatment. Yet compared to other tumor types, the interactions in GBM remain critically understudied. A significant barrier to studying the interactions between molecularly-distinct subpopulations is the challenge of tissue sampling in GBM. In particular, contrast-enhanced MRI (CE-MRI) routinely guides surgical biopsy and resection of the MRI enhancing core, but fails to address the diverse subpopulations of the surrounding non-enhancing parenchyma (so called ?brain around tumor? or BAT). These unresected residual subpopulations in BAT represent the main contributors to tumor recurrence, which can exhibit different therapeutic targets (and interactions) compared with enhancing biopsies. To address the limitations of tissue sampling, imaging techniques can help quantitatively characterize tumors in their entirety, including unresected BAT regions. Our group has used multi-parametric MRI and image-guided biopsies to develop and validate machine-learning (ML) models of intratumoral genomic heterogeneity, with particular focus on the BAT zone.
In Aim 1, will we collect and molecularly profile a large set of image-recorded stereotactic biopsies in primary GBM patients to quantify the diversity of molecularly-distinct subpopulations, as well as their phenotypic niches, throughout the BAT zone. We will assess local heterogeneity at the biopsy level and also co-localize regional patterns and rates of recurrence on serial MRI.
In Aim 2, we will use these biopsies and spatially matched MRI metrics to refine our existing ML predictive models. We will use these ML models to co- localize spatial patterns of molecularly-distinct subpopulations (and their phenotypic niches) to quantify their risk of regional recurrence.
In Aim 3, we will functionally validate the subpopulation interactions observed in Aims 1 and 2 using patient derived xenograft (PDX) models. We will also validate these interactions in human GBM using a subset of spatially matched biopsies from primary and recurrent tumors in the same patients. This proposal leverages our unique expertise in image-guided tissue analysis and MRI-based computational modeling to study the diversity of molecularly-distinct subpopulations and the evolving competitive landscapes in human GBM. This work will help risk stratify patients in future targeted clinical drug trials and should also facilitate new strategies (e.g., adaptive therapy) to exploit subpopulation co-dependency for therapeutic benefit.
We propose here a novel strategy that identifies the potential molecular underpinnings of treatment resistance within the zone where tumor almost always recurs. By developing image-based models that quantify the diversity of molecularly-distinct subpopulations (and their interactions) within individual tumors, the proposed work will improve image-based diagnosis and therapeutic strategies in the paradigm of individualized oncology.
