Overcoming the Intratumor Heterogeneity By Multi-Scale Spatial Characterizations of Tumor Microenvironment for Integrative Colorectal Cancer Classification

Description

Colorectal cancer (CRC) is the third most common cancer and the second-leading cause of cancer deaths worldwide. The heterogeneous nature of this disease hampers the selection of patients that can benefit most from adjuvant therapy and makes it difficult to develop targeted agents. In 2015, together with other teams in the colorectal cancer subtyping consortium (CRCSC), we contributed to the identification of four consensus molecular subtypes (CMSs) with distinct molecular properties and clinical characteristics. In recent years, a wide array of studies has been published that described specific prognostic, and more importantly, predictive properties of the four CMSs, providing aggregating evidence of its clinical utility. However, since CMS was developed based on bulk-tumor transcriptomic profiles, the subtyping system captures mainly intertumor heterogeneity. Therefore, CMS has been challenged by the fact that the confounding factor of the tumor-infiltrating stroma may lead to the potentially unstable classification of cell lines and patient-derived xenografts (PDXs). Till now, how to integrate intratumor and intertumor heterogeneity for more robust disease classification remains a challenging task. Histopathological images derived from tissue slides contain rich information about cell morphologies and tissue architecture, providing a cost-efficient and easily accessible assay to dissect the tumor microenvironment. In our pilot study, we provided compelling data to show that the CMS4-mesenchymal subtype of CRC remains heterogeneous and can be further classified according to spatial diversity index (SDI), which quantifies the ecological diversity of various tissue types within the tumor and tumor-proximal regions. In the proposed project, our 1st objective is to perform more comprehensive, multi-scale “tissue-omics” profiling of spatial characteristics from H&E stained histology slides and identify prognostic features for more systematic dissection of intratumor heterogeneity. Integrating the obtained tissue-omics profiles and gene expression data, our 2nd objective is to employ machine learning methods to identify biologically coherent and clinically relevant subtypes. To gain insights into the subtype-specific molecular mechanisms, our 3rd objective is to perform multi-omic and clinical characterizations, as well as regulatory network analysis. To support the project, we have curated a large data cohort from 19 public datasets and two in-house generated datasets, involving over 4600 patient samples, cell lines, and PDXs in total. To achieve these primary objectives, we will integrate image processing, bioinformatic analyses, machine learning, and artificial intelligence, as well as necessary experimental validation fully supported by our close collaborators at the Academic Medical Center Amsterdam and Sun Yat-sen University.