Conserved principles of spatial biology define tumor heterogeneity and response to immunotherapy

Abstract

The complexity of tumor microenvironments (TMEs) poses a substantial challenge to understanding tumor heterogeneity and clinical outcomes. By studying an ensemble of 262 diverse solid tumors, we uncovered a conserved, hierarchical architecture of transcriptionally covarying regions we term Spatial Groups (SGs). SGs corresponded to discrete biological units as benchmarked against multiple spatial technologies, and their nested organization revealed context-dependent constraints within tumors. Using SGs for comparing tumors, we derived a pantumor classification where immune spatial heterogeneity was the dominant axis of variation. This classification stratified response to immune checkpoint blockade in an out-of-sample cohort of non-small cell lung cancer patients. Statistical approximation techniques defined a sparse set of protein markers capturing system-level properties of TME spatial biology, demonstrating a framework for distilling genome-wide information into clinically deployable diagnostics. Our findings position the architecture of SGs as a general model unifying TME structure with biological function and clinical translation.