PROJECT SUMMARY/ABSTRACT The tumor microenvironment (TME) is comprised of diverse immune and stromal elements – each with context-dependent phenotypic states and distinct functions – that interact with cancer cells to form unique cellular communities. In recent years, major advances have been made in understanding the cross-talk between tumor and TME cell populations in shaping metastasis, and in leveraging it for therapies. However, a pan-cancer characterization of single-cell communities within the TME, both in primary and metastatic tumor deposits, is currently lacking. Moreover, circulating cell-free nucleic acids in peripheral blood plasma have emerged as promising biomarkers for noninvasive detection of cancer, and for issue-of-origin mapping. However, no liquid biopsy assays have been developed to monitor the cell states and cellular communities of the TME. I hypothesize that large-scale profiling of TME communities could present new therapeutic opportunities to transform cancer treatment. To study TME communities at scale, I recently developed EcoTyper, a new machine learning framework for delineating cell states and multicellular communities, termed ecotypes, from bulk tumor expression data. Using EcoTyper, I constructed the first global atlas of transcriptionally-defined cell states and ecotypes in >6,000 primary bulk tumor samples from 16 types of carcinoma and >1,000 diffuse large B cell lymphomas. Although these atlases are major milestones toward understanding the TME, they do not achieve single-cell resolution. While efforts to construct pan-cancer single- cell atlases have been described, they do not identify multicellular communities, nor do they provide automated methods to discover new cell states or interrogate them in new data. I propose that large-scale ecotype profiling (1) can be performed at single-cell resolution via dedicated improvements to the EcoTyper platform, (2) can delineate the determinants of progression to metastatic disease, (3) and can be used to noninvasively monitor clinically relevant heterogeneity in the TME from liquid biopsies. In the K99 phase, I will significantly improve upon EcoTyper by extending it to identify cell states and ecotypes from the joint analysis of large collections of single-cell RNA sequencing (scRNA-seq) data. I will also define a global single-cell atlas of cell states that extends our previously published pan-carcinoma atlas; and will derive a global atlas of ecotypes across multiple metastatic sites, including liver, brain and lymph nodes, by analyzing thousands of metastatic carcinomas. In the R00 phase, my group will develop bioinformatics tools for resolving epigenomic signatures of ecotypes, including methods that leverage single-cell and bulk methylation data to define methylation signatures of TME ecotypes, and will leverage them to test whether tumor ecotypes can be reliably detected from circulating nucleic acid molecules.