Construction of a prognostic model for lung adenocarcinoma based on bioinformatics analysis of metabolic genes

Background: Long-term observations and studies have found that the occurrence and development of lung adenocarcinoma (LUAD) is associated with certain metabolic changes and that metabolic disorders are directly related to carcinogenic gene mutations. We attempted to establish a prognostic model for LUAD based on the expression profiles of metabolic genes. Method: We analyzed the gene expression profiles of patients with LUAD obtained from The Cancer Genome Atlas (TCGA). Univariate Cox regression was used to assess the correlation between each metabolic gene and survival. The survival-related metabolic genes were fit into the least absolute shrinkage and selection operator (LASSO) to establish a prognostic model for LUAD. After 100,000 times of calculations and model construction, we successfully established a prognostic model consisting of 16 genes that can classify patients with LUAD into high-risk and low-risk groups. Further, the protein-protein interaction (PPI) network was built to determine the hub gene from16 metabolic genes. Finally, the top one hub gene was validated by real-time reverse transcription quantitative polymerase chain reaction (RT-qPCR) and immunohistochemistry in our 50 paired LUAD and adjacent tissues, and the prognostic performance of 16 metabolic genes was validated in GEO LUAD cohorts. Results: Univariate Cox regression analysis and LASSO regression analysis results showed that the prognostic model established based on 16 metabolic genes could differentiate patients with LUAD with significantly different overall survival (OS) and that the prognosis of the high-risk group was worse than that of the low-risk group. In addition, the model can independently predict the OS of patients in both the training cohort and the validation cohort (training cohort: HR =2.44, 95% CI: 1.58-3.74, P<0.05; validation cohort: HR = 2.15, 95% CI: 2.52-2.70, P<0.05). The decision curve analysis further showed that the combination use of the prognostic model and clinical features could better predict the survival of patients and benefit patients. In addition, Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analyses revealed several basic signaling pathways and biological processes of metabolic genes in LUAD. Combined with the clinical features and metabolic gene characteristics of patients with LUAD, we also constructed a survival nomogram with a C-index of 0.701 to predict the survival probability of patients. The calibration curve confirmed that the nomogram predications were consistent with the actual observation results. The top one hub gene was TYMS, which was determined by PPI. TYMS levels in LUAD were detected by RT-qPCR and the expression of TYMS was significantly up-regulated in the LUAD tissue of all 50 pairs (t=11.079, P<0.0001). Simultaneously, the correct of the prognostic model was validated, based on the data in GSE37745. Conclusions: We constructed and validated a new prognostic model based on metabolic genes. This model could provide guidance for the personalized treatment of patients and improve the accuracy of individualized prognoses for patients with LUAD.