SUMC reveals conserved and context-specific tumor microenvironment architectures across heterogeneous spatial datasets

Current spatial transcriptomics analyses face critical challenges in cross-sample integration due to batch effects, limiting the identification of conserved spatial patterns. Herein, we developed spatially unified meta-clustering (SUMC), which employs a robust three-stage clustering strategy to align spatial architectures across diverse datasets. When applied to a compendium of 114 non-small cell lung cancer (NSCLC) samples across 7 cohorts encompassing 341443 spatial spots, SUMC revealed conserved spatial patterns, including prognosis relevant epithelial patterns that stratify histological sub-types, as well as heterogeneous tertiary lymphoid structures (TLSs). We further uncovered a mature TLS sub-pattern enriched in core locations and associated with favorable prognosis, as well as different fibroblast niches that spatially co-localize with TLSs, whose molecular profiles are suggestive of distinct immunomodulatory functions. Furthermore, the validity of SUMC was confirmed through cross-platform and pan-cancer analyses, demonstrating its ability to decode fundamental spatial organization principles. In summary, SUMC provides a powerful tool for integrative spatial transcriptomics analysis and discovery of spatially resolved biomarkers.