Subtype-resolved tumor-intrinsic programs in immune-cold breast cancer: mechanistic interactions, evidence gaps, and translational priorities

Most breast cancers respond poorly to immune checkpoint blockade (ICB), and hormone receptor-positive (HR +) tumors, which constitute most diagnoses, are particularly refractory. This review synthesizes evidence on tumor-intrinsic determinants of the immune-cold phenotype in breast cancer under an operational framework designated ICE, comprising three functionally defined domains: Invisibility (failure of antigen generation, presentation, and innate sensing), Coercion (active exclusion and post-infiltration suppression of effector cells), and Exhaustion-by-Metabolism (metabolic attrition of infiltrating lymphocytes). ICE is not proposed as a newly discovered taxonomy of immune evasion; its added value lies in integrating breast-cancer subtype resolution, evidence provenance, systems-level interactions, negative clinical evidence, and mechanism–biomarker–intervention matching within a single operational framework. Four features define this framework. The domains are overlapping rather than sequential, and an explicit assignment rule places multifunctional regulators such as MYC, hypoxia-inducible factor 1α, and lactate in a primary domain while cataloguing secondary effects. Tumor–immune interaction is treated as bidirectional, positioning tumor-intrinsic programs as both causes and consequences of immune pressure. Major mechanistic and translational claims are assigned claim-specific evidence tiers. Resistance to immune-mediated killing, including death receptor loss, is classified as cross-dimensional cytoprotection rather than a recognition defect. Applying this discipline exposes an asymmetric evidence base: antigen presentation defects and metabolic suppression rest on substantial preclinical data with limited clinical validation, several widely cited mechanisms derive from non-breast models, and the translational record includes cross-cancer constraints—failure of IDO1 inhibition in a melanoma phase III trial and minimal STING-agonist monotherapy activity in early mixed-solid-tumor studies—together with breast-cancer-specific negative findings from IMpassion131 and the unfavorable benefit–risk profile of abemaciclib plus pembrolizumab. Cross-dimensional systems-level structures, including the hypoxia–adenylosuccinate lyase–fumarate–STING axis and the tumor–macrophage arginine–polyamine relay, are identified as the most instructive examples of domain interaction, and seven falsifiable predictions are advanced, each specifying a measurable biomarker and a decisive experiment.