Ultra-resolution Deconvolution of Spatial Transcriptomics Unveils Spatiotemporal Cellular Dynamics in Complex Microenvironments

Abstract

AbstractThe current challenge in analyzing sequencing-based spatial transcriptomics (ST) is to precisely resolve the spatial distribution of cell subpopulations with low abundance and detect context-dependent variation in cell states within complex tissue microenvironments. Here, we introduce an ultra-resolution ST deconvolution algorithm (UCASpatial) that improves the resolution of mapping cell subpopulations to spatial locations by leveraging the contribution of genes indicative of cell identity through entropy-based weighting. Using both in silico and real ST datasets, we demonstrate that UCASpatial improves the robustness and accuracy in identifying low-abundant cell subpopulations and distinguishing transcriptionally heterogeneous cell subpopulations. Applying UCASpatial to murine wound healing, we recapitulate known spatiotemporal dynamics of multiple cell subpopulations involved in wound healing and reveal the spatial segregation of distinct macrophage subpopulations embedded within different cellular communities at the wound center. In human colorectal cancer (CRC), we link genomic alterations in individual cancer clones to multi-cellular characteristics of the tumor immune microenvironment (TIME) and reveal the co-evolution of tumor cells and TIME within the same tumor. We show that the copy number gain on chromosome 20q (chr20q-gain) in tumor cells orchestrates a T cell-excluded TIME, indicative of resistance to immunotherapy in CRC, and is associated with tumor-intrinsic endogenous retrovirus silencing and impaired type I interferon response. Our findings present UCASpatial as a versatile tool for deciphering ultra-resolution cellular landscapes in ST and exploring intercellular mechanisms in complex and dynamic microenvironments.