Multi-omics integrative analysis of stemness-associated pathological signatures to guide prognosis and therapeutic strategies in uveal melanoma.

BACKGROUND: Uveal melanoma (UVM) is the most common intraocular malignancy in adults and exhibits poor prognosis upon metastasis. Stemness, a hallmark of cancer aggressiveness, remains understudied in UVM, especially in relation to pathology-derived features and therapeutic implications. METHODS: We integrated single-cell RNA sequencing (scRNA-seq), bulk transcriptomic data, and histopathological whole-slide images (WSIs) to systematically investigate stemness-associated heterogeneity in UVM. High stem-like (HStem) malignant subpopulations were identified via CytoTRACE and further analyzed using hdWGCNA to determine signature genes. A machine learning-based pathological image-derived prognostic score (IPS) system was constructed using multiscale pathomic features and benchmarked across seven survival models. RESULTS: HStem cells exhibited elevated oncogenic and metabolic activity, strong fibroblast interactions, and enriched expression of stemness-related genes. Nine key pathomic image features were selected to construct the IPS model, which stratified patients by prognosis and predicted immunotherapy response. High-IPS tumors showed immune-cold phenotypes and were more sensitive to ABT-888, ATRA, AZ628, and temsirolimus. CONCLUSIONS: This study highlights a novel integrative framework combining pathomics and scRNA-seq to decode stemness-driven heterogeneity in UVM. The IPS model offers a non-invasive tool for risk stratification and therapeutic guidance, paving the way for precision oncology in rare ocular malignancies. Notably, the IPS was derived and internally validated within a single TCGA-UVM cohort, and its generalizability to other populations requires validation in independent multi-center cohorts.