Effect of deconvoluting single-cell transcriptomics on cellular programs regulated by cell-cell communication in colorectal cancer.

59 Background: Colorectal cancer (CRC) is the 3rd most common cancer. The consensus molecular subtypes among CRC tumors reflects the heterogeneity of composition and functional states of cells in the tumor microenvironments (TME), which in turn may underlie diverse behaviors of disease progression and responses to anticancer treatments. Cells within a TME communicate through intricate cell-cell communication (CCC) networks. The altered state of one cell can influence the states of other cells through ligand-receptor-mediated signal transduction. Gaining insight into the cellular states of cells and understanding their complex communications in a TME is vital for uncovering the heterogeneous mechanisms governing tumor growth, immune evasion, and therapy resistance. Methods: In this meta-analysis of eight single-cell cohorts encompassing 153 patients and 279 samples with more than 600,000 cells, we advance the understanding of CCC networks in CRC through a novel analytical framework. Employing hierarchical language modeling, we identify gene expression modules (GEMs) that mirror single-cell signaling states, crucial for deciphering the complexity of intercellular interactions. By applying causal discovery methods, we systematically uncover GEMs likely regulated by ligand-receptor signaling and cross-cell-type communication. We further validate the discovered CCC using spatial transcriptomic data by testing the spatial co-localization. Results: This analysis reveals nine cross-cell-type CCC programs, marked by highly correlated GEMs across various cell types, shedding light on the intricate CCC networks within the TME. The discovered CCC programs include malignant program, proliferation program, stromal interactions, stromal-myeloid interactions, innate immunity interactions, regulatory interactions, epithelial-lymphocyte interactions and exhaustion program. Each program is composed of GEMs from various cell types. Spatial transcriptomics further validate these findings by demonstrating the co-localization of GEMs within CCC programs in distinct spatial domains, emphasizing the spatial dynamics of tumor intercellular communication. We successfully construct the CCC subnetworks connected by ligand-receptor signaling. Our interactive website and analytical framework equip researchers with powerful tools to explore complex mechanisms, potentially uncovering novel drug targets and refining strategies for precision immunotherapies. Conclusions: This study provides an in-depth analysis of colorectal cancer by: 1) Cataloging GEMs that precisely depict the transcriptomic processes unique to individual cell clusters or shared among multiple cell clusters. 2) Presenting the CCC networks driven by ligand-receptor interactions within the TME supported by both single-cell RNA-seq data and spatial transcriptomic data.