The MONSTAR-SCREEN-3 Spatial Atlas: Pan-cancer TME archetypes and clinically relevant tumor-immune ecosystems

Background: The SCRUM-Japan MONSTAR-SCREEN consortium is a nationwide molecular profiling initiative spanning all solid tumors except lung cancer and hematologic malignancies. The next-generation MONSTAR-SCREEN-3 (M3) project incorporates spatial transcriptomics to achieve a "quantum leap" by integrating novel modalities to decode the tumor microenvironment (TME) architecture governing therapeutic response. This study presents the foundational pan-cancer spatial atlas from the initial M3 dataset. Methods: MONSTAR-3 enrolls ~3,200 cases, and spatial analysis is being performed on eligible specimens. Among these, 331 FFPE tumors across 21 solid types were analyzed. Spatial profiling used the Xenium 5K platform within a standardized NCCE-GxD workflow (NCCE tissue preparation, GxD Xenium processing; ~10-week pipeline). Cell types were annotated using a consensus framework across four pipelines. Malignant epithelial cells were distinguished from benign epithelium through integrated spatial context, transcriptional features, and pathologist-guided review. We quantified malignant program heterogeneity and identified shared and distinct spatial niches linked to clinical variables. Results: We generated a high-confidence spatial atlas comprising ~20 million cells. Across 21 tumor types, TME was organized into conserved, pan-cancer spatial ecosystem archetypes that stratified prognosis and therapeutic response. Spatial heterogeneity in malignant programs correlated with advanced disease. We also defined 15 conserved pan-cancer cellular niches (metaclusters), each with distinct compositions and spatial architectures. Among these, two tumor-stromal interface archetypes emerged: MC9, an innate-immune-rich interface, and MC5, a lymphocyte-dominant interface. The MC9 niche was significantly enriched in the advanced-stage tumors, showed increasing prevalence with higher TNM stages, and its derived signatures were associated with poor prognosis in legacy cohorts (e.g., TCGA). Conversely, B-cell-dominant (MC10) and B-cell/CD4+ T-cell-rich (MC14) niches were associated with favorable outcomes. These lymphocyte-dominant structures were further resolved into simple aggregates, primary follicles, and secondary follicles indicative of TLS formation. Secondary follicle signatures correlated with improved prognosis and response in an independent immunotherapy-treated cohort. Conclusions: This study presents one of the largest pan-cancer spatial transcriptomic atlases and establishes a foundational component of the M3. We provide a robust framework for analyzing spatial transcriptomics and conserved, clinically actionable spatial ecosystems. This resource paves the way for spatially-driven patient stratification, next-generation biomarker development, and accelerated precision oncology within the SCRUM-Japan framework.