Gene-interaction perturbation network analysis identifies distinct subtypes and actionable targets for aggressive thyroid cancer: an integrative multi-omics and machine learning approach for precise medicine

High-risk thyroid cancer such as aggressive papillary thyroid cancer (PTC) and anaplastic thyroid cancer (ATC) poses significant clinical challenges due to their tumor heterogeneity and limited therapeutic strategies. We constructed a genetic interaction (GI) perturbation network based on functional interactions from the Reactome database and identified distinct molecular subtypes of thyroid cancer with unique biological characteristics. Furthermore, a GI-based machine learning-assisted prognostic model was established for predicting progression of thyroid cancer patients, and RRM1 was identified as an essential gene that is significantly correlated with progression of thyroid cancer. Drug repositioning and molecular docking analyses demonstrated that gemcitabine targeting RRM1 can serve as a promising therapeutic strategy for ATC. Subsequent in vitro and in vivo experiments validated the therapeutic potential of gemcitabine in targeting RRM1, with significant efficacy in inducing ATC cell apoptosis and inhibiting tumor growth in a xenograft mouse model. Using single-cell RNA-sequencing (scRNA-seq) analysis and multiplex immunohistochemistry (mIHC), we revealed a unique interaction between RRM1-positive epithelial cells and a specific subset of cancer-associated fibroblasts (CAF) in ATC, highlighting the importance of the tumor microenvironment in driving aggressiveness. In summary, our findings highlight the critical role of dynamic gene-interaction network in thyroid cancer progression and offer a promising therapeutic strategy by targeting RRM1 for patients with high-risk thyroid cancer.