Mapping the single cell state basis of metastasis in space and time

点击翻译按钮获取中文摘要

We propose to leverage recent advances in machine learning and systems biology to enable high dimensional molecular assessment of the dynamic cell state transitions driving metastasis. We hypothesize that the interaction between a cancer cell's intrinsic reactivation of developmental programs with its spatiotemporal context determines its metastatic potential. We will exploit developmental changes in the mammary epithelium to define their cell state basis and map the aberrant reuse of these transcriptional programs in metastatic disease. Both normal mammary epithelium and breast tumors undergo dramatic changes in differentiation and tissue architecture, and loss of differentiation correlates with poor patient outcomes. We developed 3D culture assays that recapitulate epithelial morphogenesis and cancer growth, invasion, and metastatic colony formation. The key concepts arising are that: (1) a conserved process of dedifferentiation and loss of polarity accompanies both normal and neoplastic morphogenesis and (2) the cancer cells in luminal and basal breast cancer recapitulate basal epithelial and hybrid EMT programs. Recent advances in single cell sequencing, spatial transcriptomics, and machine learning enable transcriptome-wide resolution of these states in tissue, quantitative comparison of normal and cancerous cell states, and the identification of targetable cell state regulators. Aim 1: Map cell states in space and time during development, tumor formation, and metastasis. We will generate scRNA-seq data from normal glands, ductal carcinoma in situ, and invasive tumors collected at different ages and also longitudinally in 3D culture. We will use our CoGAPS algorithm to infer cell states and their temporal progression. We will then use our patternMarker2 statistic to identify cell state makers for MERSCOPE analysis in tissue. We will map these states in normal glands, primary tumors, and metastases isolated from genetically engineered mouse models (GEMM) and patient derived xenografts (PDX). Aim 2: Model the dynamics of differentiation state during development and cancer progression. To define the effect of cell state on metastatic progression, we will construct an ecosystem-style multinomial diversity model. We will initialize the model with literature-based parameter values to predict the interactions between cell type and cell state. We will then extend the model to use the weights assigned by CoGAPS to each cell, thereby linking gene regulatory programs to the cell state changes driving metastasis. Aim 3: Validate candidate regulators of metastatic cell state transitions in 3D culture and in vivo. To isolate the genes regulating metastasis, we will use our transfer learning algorithm, projectR, to score each cancer cell for its relative utilization of scRNA-seq-defined molecular programs. We will then use our projectionDriver statistic to identify differentially expressed (DE) genes at sites of cancer invasion, relative to the tumor interior. DE genes will be tested genetically in 3D culture assays modeling invasion and colony formation and then in orthotopic and tail vein metastasis assays in vivo.