ABSTRACT
Background: Hypoxia is a typical microenvironmental feature of most solid tumors, affecting a variety of physiological processes. We developed a hypoxia-related prognostic risk score (HPRS) model to reveal tumor microenvironment (TME) and predict prognosis of lung adenocarcinoma (LUAD). Methods: LUAD sample expression data were from The Cancer Genome Atlas (TCGA) and Gene Expression Omnibus (GEO) databases. Weighted gene co-expression network analysis (WGCNA) and least absolute shrinkage and selection operator (LASSO) Cox regression identified hypoxia-related genes (HRGs) to create HPRS. The prognostic value, genetic mutation and TME, and therapeutic response of distinct HPRS groups were analyzed. Univariate and multivariate Cox regression analysis identified independent factors associated with the prognosis of LUAD. A decision tree based on HPRS and clinicopathological variables was established using the classification system based on decision tree algorithm. A nomogram was constructed with important clinical features and HPRS by the RMS package. Results: A HPRS model with five HRGs was developed and verified in two separate cohorts of GEO. HPRS model divided patients with LUAD into two groups. High HPRS was related to high probability of genetic alterations. HPRS could predict the prognosis, TME, and sensitivity to immunotherapy/chemotherapy of LUAD. The decision tree defined four risk subgroups with significant OS differences. Nomogram with integrated HPRS and clinical features had acceptable accuracy in predicting LUAD prognosis. Conclusions: A HPRS model was developed to evaluate prognosis, genetic alterations, TME, and response to immunotherapy, which may provide theoretical reference for the study of molecular mechanism of hypoxia in LUAD.
ABSTRACT
BACKGROUND: The mechanism of cancer occurrence and development could be understood with multi-omics data analysis. Discovering genetic markers is highly necessary for predicting clinical outcome of lung adenocarcinoma (LUAD). METHODS: Clinical follow-up information, copy number variation (CNV) data, single nucleotide polymorphism (SNP), and RNA-Seq were acquired from The Cancer Genome Atlas (TCGA). To obtain robust biomarkers, prognostic-related genes, genes with SNP variation, and copy number differential genes in the training set were selected and further subjected to feature selection using random forests. Finally, a gene-based prediction model for LUAD was validated in validation datasets. RESULTS: The study filtered 2071 prognostic-related genes and 230 genomic variants, 1878 copy deletions, and 438 significant mutations. 218 candidate genes were screened through integrating genomic variation genes and prognosis-related genes. 7 characteristic genes (RHOV, CSMD3, FBN2, MAGEL2, SMIM4, BCKDHB, and GANC) were identified by random forest feature selection, and many genes were found to be tumor progression-related. A 7-gene signature constructed by Cox regression analysis was an independent prognostic factor for LUAD patients, and at the same time a risk factor in the test set, external validation set, and training set. Noticeably, the 5-year AUC of survival in the validation set and training set was all Ë 0.67. Similar results were obtained from multi-omics validation datasets. CONCLUSIONS: The study builds a novel 7-gene signature as a prognostic marker for the survival prediction of patients with LUAD. The current findings provided a set of new prognostic and diagnostic biomarkers and therapeutic targets.
