Systematic dissection of tumor-normal ecosystems at single-cell resolution for identification of immunotherapy targets

Abstract

Understanding the complex tumor microenvironment (TME) poses significant challenges in cancer treatment. To comprehensively investigate the tumor-normal ecosystems, I conducted an integrative analysis of 4.9 million single cells from 1,070 tumor and 493 normal samples with pan-cancer 137 spatial transcriptomics, and 1,261 immunotherapy-treated bulk tumors. Using a deep neural network-based algorithm, I identified optimal dual antigen-targets for chimeric antigen receptor therapy in solid tumors. Numerous cell states within the tumor-normal ecosystems were defined and gene signatures related to TME hallmark in different cell types and organs were highlighted. The distinctions between inflammatory fibroblasts marked by AKR1C1 or WNT5A in terms of their interactions and spatial relationships were characterized. Co-occurrence analysis revealed a group of interferon-enriched community states, including components of tertiary lymphoid structure (TLS), which exhibited distinct rewiring patterns in tumor, adjacent normal, and healthy normal tissues. The prognostic power of the interferon-enriched community states was confirmed in immunotherapy-treated patients (n=1,261) across 4 cancer types. Spatial transcriptomes discriminated TLS-enriched from non-enriched cell types among immunotherapy-favorable components and my TLS signature well-captured TLS-enriched cell types across various cancer types. This systematic and detailed examination of tumor-normal ecosystems deepens the understanding of both inter- and intra-tumoral heterogeneity.