Tumor immune microenvironment remodeling predicts response to checkpoint inhibitor therapy.

Abstract

Immune checkpoint inhibitors (ICIs) have transformed cancer therapy, yet the basis of variable patient responses remains unclear. We assemble a longitudinal single-cell RNA sequencing atlas of 441 samples from 241 patients across ten cancer entities to map treatment-associated remodeling of the tumor immune microenvironment (TIME). With a hierarchical reference-guided deep-phenotyping framework, we define 77 immune and stromal subtypes and resolve four conserved TIME subtypes. Approximately 40% of tumors shift between states during therapy, and the transition is more predictive of outcome than the baseline state. Favorable transitions toward inflamed or B cell-enriched subtype track with improved response and survival, while persistence in or shifts toward myeloid dominance indicate resistance. We derive a 59-gene signature that predicted response and survival for 1,383 baseline tumors across 19 independent cohorts. These findings establish immunotype transitions as a central determinant of ICI efficacy, offering new avenues for response prediction and rational immunotherapy design.