Tumors are complex ecosystems consisting of malignant, immune, and stromal elements whose dynamic interactions drive patient survival and response to therapy. A comprehensive understanding of the diversity of cellular states within the tumor microenvironment (TME), and their patterns of co-occurrence, could provide new diagnostic tools for improved disease management and novel targets for therapeutic intervention. To address this challenge, we developed EcoTyper, a novel machine learning framework for large-scale identification of TME cell states and their co-association patterns from bulk, single-cell, and spatially resolved tumor expression data. EcoTyper starts by “purifying” cell type-specific gene expression profiles of epithelial cells, immune, and stromal cell types from bulk tissue transcriptomes using CIBERSORTx (Newman et al., Nat Biotechnol 2019). It then identifies transcriptional states for each cell type, validates them in scRNA-seq data, and uncovers co-occurrence patterns between cell states in order to define tumor cellular ecosystems. Applied to 6,475 tumor and adjacent normal samples from solid tumor types profiled by The Cancer Genome Atlas (TCGA), EcoTyper identified robust transcriptional states across 12 major cell types, including epithelial, fibroblast, endothelial, and 9 immune subsets. These states included both known and novel cellular phenotypes, nearly all of which could be validated in a compendium of scRNA-seq tumor atlases spanning ~140,000 cells. Most cell states were specific to neoplastic tissue, ubiquitous across tumor types, and significantly associated with overall survival, both in TCGA and in 9,062 held-out tumor specimens (Gentles/Newman et al., Nat Medicine 2015). We found that specific cell states co-occur in distinct cellular communities with characteristic patterns of ligand-receptor interactions, genomic features, clinical outcomes, and spatial organization. One such ecosystem defined a normal-like state that was strongly enriched in non-malignant samples. Others delineated novel pro- and anti-tumor inflammatory environments involving specific fibroblast, endothelial, and immune cell transcriptional programs. In summary, large-scale deconvolution of cell type-specific transcriptomes across thousands of solid tumors revealed a comprehensive atlas of TME cell states and cellular ecosystems. Our results provide a high-resolution portrait of cellular heterogeneity in the TME across multiple solid tumor types, with implications for novel diagnostics and immunotherapeutic targets.

References:

1. Newman, A.M., et al., Determining cell type abundance and expression from bulk tissues with digital cytometry. Nat Biotechnol, 2019. 37(7): p. 773-782.

2. Gentles, A.J., et al., The prognostic landscape of genes and infiltrating immune cells across human cancers. Nature medicine, 2015. 21(8): p. 938.

Citation Format: Bogdan A. Luca, Chloé B. Steen, Armon Azizi, Magdalena Matusiak, Joanna Przybyl, Nastaran Neishaboori, Almudena Espín Pérez, Maximilian Diehn, Ash A. Alizadeh, Matt van de Rijn, Andrew J. Gentles, Aaron M. Newman. Atlas of clinically-distinct cell states and cellular ecosystems across human solid tumors [abstract]. In: Proceedings of the Annual Meeting of the American Association for Cancer Research 2020; 2020 Apr 27-28 and Jun 22-24. Philadelphia (PA): AACR; Cancer Res 2020;80(16 Suppl):Abstract nr 3443.