Complexity of the cellular organization of the tumor microenvironment as an ecosystem remains to be fully appreciated. Here, for a comprehensive investigation of tumor ecosystems across a wide variety of cancer types, we performed integrative transcriptome analyses of 4.4 million single cells from 978 tumor and 474 normal samples in combination with 9,510 TCGA and 1,339 checkpoint inhibitor-treated bulk tumors. Our analysis enabled us to define 28 different epithelial cell states, some of which had prognostic effects in cancers of relevant origin. Malignant fibroblast signatures defined according to the organ of origin demonstrated prognostic significance across diverse cancer types and revealed FN1, BGN, THBS2, and CTHRC1 as common cancer-associated fibroblast genes. Novel associations were revealed between the AKR1C1+ inflammatory fibroblast and myeloid-derived PRR-induced activation states and between the CXCL10+ fibroblast and squamous/LAMP3+ DC/SPP1+ macrophage states. We discovered tumor-specific rewiring of the tertiary lymphoid structure (TLS) network, involving previously unappreciated DC1, and pDC.. Along with other TLS component states, the tumor-associated germinal center B cell state identified from adjacent normal tissues was able to predict responses to checkpoint immunotherapy. Distinct groups of pan-cancer ecosystems were identified and characterized along the axis of immunotherapy responses. Our systematic, high-resolution dissection of tumor ecosystems provides a deeper understanding of inter- and intra-tumoral heterogeneity.
|
