Comprehensive multi-omics and single-cell analysis reveals a ferroptosis signature associated with tumor microenvironment remodeling and therapeutic decision-making across pan-cancer

Background: Ferroptosis, an iron-dependent form of regulated cell death, has been implicated in tumor biology and cancer immunotherapy, but its pan-cancer landscape and impact on the tumor microenvironment (TME) and treatment response remain incompletely defined. We aimed to establish the ferroptosis signature (Fp.Sig) as a quantitative ferroptosis index within the pan-cancer single-cell landscape, thereby guiding clinical patient stratification and therapeutic decision-making. Methods: We integrated profiles of 10 510 bulk samples across 33 The Cancer Genome Atlas cancer types, along with a large-scale single-cell analysis comprising 3 289 156 cells from 1559 samples across 19 cancer types, sourced from the Curated Cancer Cell Atlas and CancerSCEM. Furthermore, we incorporated CRISPR-Cas9 dependency screens targeting 17 387 genes in 378 cell lines, pharmacogenomic data covering 345 drugs in 738 cell lines, and 13 clinical trials involving 1091 patients treated with immune checkpoint inhibitors. The prognostic and predictive value of Fp.Sig was assessed through modeling and external validation in both comprehensive and prospective cohorts. Results: We developed Fp.Sig at the bulk multi-omics and single-cell resolution pan-cancer level by using 10 machine learning algorithms and 101 algorithm combinations. Fp.Sig stratified patients into molecular subgroups with distinct overall survival and retained independent prognostic significance after adjustment for age, sex, and stage. Higher Fp.Sig was associated with elevated genomic instability, oncogenic pathway activation, virus-related tumor states, and distinct telomere phenotypes. Increasing Fp.Sig correlated with greater lymphocyte infiltration but a more immunosuppressive TME enriched for regulatory T cells and myeloid-derived suppressor cells. Single-cell analysis revealed coordinated signatures in malignant and immune cells and identified VEGFA-NRP1 signaling between ferroptosis-high tumor cells and myeloid populations, implicated in TME remodeling. Fp.Sig was linked to oncogenic mutations, ferroptosis-related metabolic reprogramming, and differential sensitivity to targeted agents, including the TOP1 inhibitor SN-38. Lower Fp.Sig was associated with improved survival and higher response rates, and an Fp.Sig-based response model achieved an AUC of 0.78 in validation, with competitive or improved discrimination versus established predictors in both comprehensive cohorts and prospective cohorts, which demonstrates its potential to assist clinical decision-making. Conclusions: These data define a pan-cancer ferroptosis atlas and provide a ferroptosis score and web tool that may support risk stratification and therapeutic decision making (https://puh3.shinyapps.io/FpSig_predict/).