Loop of N2-polarized neutrophils and exhausted CD8 + T cells induces immunotherapy resistance in NSCLC.

While neutrophils represent a prominent myeloid component in non-small cell lung cancer (NSCLC), the specific immunosuppressive functions of N2-polarized neutrophils and their mechanistic interactions with CD8⁺ T cells remain incompletely characterized. Furthermore, the development of clinically applicable models for prognostic stratification and immunotherapy response prediction, grounded in these molecular interactions, represents a critical unmet need. We integrated large-scale single-cell RNA sequencing datasets to delineate the tumor immune microenvironment, performing pathway enrichment and cell-cell communication analyses. Key molecular features derived from these interactions were employed to construct a deep neural network model. This model was trained and validated on bulk RNA sequencing cohorts to predict immunotherapy response. Additionally, we developed the N2_Neu-CD8⁺ Tex Loop Score (NTLS) for prognostic assessment and evaluated its pan-cancer applicability. Our analysis revealed a previously uncharacterized positive feedback loop between N2 neutrophils and exhausted CD8⁺ T cells (Tex). Neutrophil-derived ICAM1 engages with ITGAL/ITGB2 on CD8⁺ T cells, suppressing their NF-κB signaling and reinforcing the exhausted phenotype. In a feed-forward manner, Tex-derived CCL5 signals via CCR1 on N2 neutrophils, activating their NF-κB pathway and further upregulating ICAM1 expression. This ICAM1-Integrin and CCL5-CCR1 axis creates a self-sustaining immunosuppressive circuit. A deep learning model, built upon genes central to this loop, accurately predicted immunotherapy outcomes in NSCLC and melanoma. The derived NTLS score proved effective for prognostic stratification and was validated across multiple independent cohorts and cancer types. This study defines a pathogenic positive feedback loop, driven by ICAM1-Integrin and CCL5-CCR1 interactions, through which N2 neutrophils and Tex cells cooperatively establish an immunosuppressive niche that drives immunotherapy resistance. The computational models we developed, based on this molecular circuitry, offer robust tools for patient stratification and hold significant translational promise.