부산대학교 도서관

Abstract Assessing the hypoxic status within the tumor microenvironment (TME) is crucial for its significant clinical relevance in evaluating drug resistance and tailoring individualized strategies. In this study, we proposed a robust pan-cancer hypoxic quantification method utilizing multiple public databases, diverse bioinformatics, and statistical methods. All tumor samples were classified into four subtypes: non-hypoxic/TMEhigh (C1), hypoxic/TMEhigh (C2), non-hypoxic/TMElow (C3), and hypoxic/TMElow (C4). We systematically analyzed multi-omics data and single-cell RNA-sequencing (scRNA-seq) data to reveal distinct immune landscape patterns and genomic characteristics among the four subtypes across pan-cancer. Furthermore, we employed multiple machine learning approaches to construct a hypoxic-TME model to enhance the predictive accuracy of immunotherapy response. Additionally, drug repositioning was implemented for cancer patients predicted as non-responders to immunotherapy. A pan-cancer analysis identified PDK1 as a hub gene linking tumor hypoxia, glycolysis, and immunotherapy resistance. In vivo experimental validation further confirmed that targeting PDK1 could improve the response to immunotherapy. Overall, our study may offer valuable insights for integrating hypoxic-TME classification into tumor staging and providing personalized strategies for cancer patients.