[Therapeutic Targets by Pathology AI and Spatial Transcriptomics in Breast Cancer]

Recent advances in pathology foundation models have markedly improved the accuracy and generalizability of histopathological image analysis in breast cancer. However, the mechanisms of resistance to CDK4/6 inhibitors in hormone receptor-positive, HER2-negative advanced breast cancer remain incompletely understood. This article outlines a strategy to identify therapeutic targets by integrating pathology AI with spatial transcriptomics. We developed AI-directed spatial transcriptomics (AID-ST), a framework that compares gene expression profiles between drug-sensitive and drug-resistant regions identified by pathology AI. In a preliminary analysis of clinical breast cancer specimens, this approach suggested that KRAS pathway activation is a major driver of resistance, accompanied in part by Polycomb dysregulation, RB loss, PI3K pathway alteration, and acquisition of stem-like features. Additional spatial analyses of paired pre- and post-treatment specimens supported these findings and further suggested a role for the tumor microenvironment, including EMT- and IL6/JAK/STAT3-related changes, in promoting resistant phenotypes. These results indicate that integrating pathology AI with spatial transcriptomics may enable systematic classification of resistance subtypes and prioritization of actionable therapeutic targets in breast cancer.