干扰素预激免疫图谱预测免疫检查点抑制剂治疗期间的免疫相关不良事件

Immune checkpoint inhibitors (ICIs) have transformed cancer therapy but frequently induce immune-related adverse events (irAEs), which can disrupt treatment and worsen outcomes. The mechanisms predisposing certain individuals to irAEs remain unclear, and reliable strategies for early prediction and prevention are urgently needed.

We analyzed pre-treatment peripheral blood mononuclear cell (PBMC) transcriptomic data from 88 patients receiving ICIs, including 22 who subsequently developed irAEs. Immune infiltration signatures were used to build machine learning models with SHAP-based interpretability. Immune-related differentially expressed genes were then incorporated into the Chemical-Induced Gene Signature (CIGS) framework to predict candidate reversal compounds. Selected compounds were further evaluated in Jurkat T cells to experimentally validate their effects on interferon- signaling and underlying mechanisms.

Baseline immune infiltration patterns showed strong predictive value for subsequent irAE development, with the Random Forest model achieving the best performance (AUC = 0.97). SHAP analysis revealed that activated T-cell and NK-cell signatures were dominant predictors, indicating a pre-existing immune-primed state in irAE-prone individuals. Single-cell analysis identified two irAE-enriched myeloid clusters with lung-associated inflammatory features, suggesting baseline myeloid priming. T cells and NK cells from irAE samples exhibited marked upregulation of interferon-stimulated genes and strong enrichment of type I and type II interferon pathways. Perturbation-based screening identified multiple compounds capable of reversing these interferon-amplified signatures, and in vitro experiments demonstrated that alpinetin and momelotinib suppress interferon- signaling through distinct STAT1-and JAK-STAT-dependent mechanisms.

irAEs may arise from the convergence of pre-existing myeloid inflammation and interferon-driven lymphocyte activation before therapy. Our study provides a predictive framework for identifying high-risk patients and highlights mechanistically grounded compounds for potential irAE mitigation.