Objective: Abnormal DNA methylation can regulate carcinogenesis in lung adenocarcinoma (LUAD), while transcription factors (TFs) mediate methylation in a site-specific manner to affect downstream transcriptional regulation and tumor progression. Therefore, this study aimed to explore the TF-methylation-gene regulatory relationships that influence LUAD prognosis. Methods: Differential analyses of methylation sites and genes were generated by integrating transcriptome and methylome profiles from public databases. Through target gene identification, motif enrichment in the promoter region, and TF prediction, TF-methylation and methylation-gene relation pairs were obtained. Then, the prognostic TF-methylation-gene network was constructed using univariate Cox regression analysis. Prognostic models were constructed based on the key regulatory axes. Finally, Kaplan-Meier curves were created to evaluate the model efficacy and the relationship between candidate genes and prognosis. Results: A total of 1878 differential expressed genes and 1233 differential methylation sites were screened between LUAD and normal samples. Then 10 TFs were predicted to bind 144 enriched motifs. After integrating TF-methylation and methylation-gene relations, a prognostic TF-methylation-gene network containing 4 TFs, 111 methylation sites, and 177 genes was constructed. In this network, ERG-cg27071152-MTURN and FOXM1-cg19212949-PTPR regulatory axes were selected to construct the prognostic models, which showed robust abilities in predicting 1-, 3-, and 5-year survival probabilities. Finally, ERG and MTURN were downregulated in LUAD samples, whereas FOXM1 and PTPR were upregulated. Their expression levels were related to LUAD prognosis. Conclusion: ERG-cg27071152-MTURN and FOXM1-cg19212949-PTPR regulatory axes were proposed as potential biomarkers for predicting the prognosis of LUAD.

Background: Lung adenocarcinoma (LUAD) has a complex tumor heterogeneity. This study aimed to identify LUAD subtypes and build a reliable prognostic signature based on the activity changes of the hallmark and immunologic gene sets. Methods: Changes in the activities of the hallmark and immunologic gene sets were analyzed based on The Cancer Genome Atlas (TCGA)-LUAD dataset, followed by identification of prognosis-related differential gene sets (DGSs) and their related LUAD subtypes. Survival analysis, correlation with clinical characteristics, and immune microenvironment assessment for subtypes were performed. Moreover, the DGSs among different subtypes were identified, followed by the construction and evaluation of a prognostic risk score model and nomogram. The tumor mutation burden (TMB) of different risk groups wascompared. Results: Two LUAD subtypes were identified based on the activity changes of the hallmark and immunologic gene sets. Cluster 2 had worse prognosis, more advanced tumorand clinical stages, and higher immune infiltration than cluster 1. Moreover, a prognostic risk score signaturewas established using two LUAD subtype-related DGSs, which could stratify patients at different risk levels. A shorter survival time and higher TMB levels were observed in the high-risk patients. The established nomogram accurately predicted the survival outcomes. Conclusions: Our findings revealed that our constructed prognostic signature could accurately predict the survival outcomes and immune microenvironment of patients with LUAD, which was helpful in predicting the prognosis and guiding personalized therapeutic strategies for LUAD.