A differentially-methylated-region signature predicts the recurrence risk for patients with early stage lung adenocarcinoma

Abstract

Predicting prognosis in lung cancer patients is important in establishing future treatment and monitoring plans. Lung adenocarcinoma (LUAD) is the most common and aggressive type of lung cancer with dismal prognosis and prognostic stratification would help to guide treatment. Aberrant DNA methylation in tumors occurs earlier than clinical variations, and keeps accumulating as cancer progresses. Preliminary studies have given us some clues that DNA methylation might serve as a promising biomarker for prognosis prediction. Herein, we aimed to study the potential utility of DNA methylation pattern in predicting the recurrence risk of early stage resectable LUAD and to develop a risk-modeling signature based on differentially methylated regions (DMRs). This study consisted of three cohorts of 244 patients with stage I-IIIA LUAD, including marker discovery cohort (n = 39), prognostic model training cohort (n = 117) and validation cohort (n = 80). 468 DMRs between LUAD tumor and adjacent tissues were screened out in the marker discovery cohort (adjusted P < 0.05), and a prognostic signature was developed based on 15 DMRs significantly related to disease-free survival in early stage LUAD patients. The DMR signature showed commendable performance in predicting the recurrence risk of LUAD patients both in model training cohort (P < 0.001; HR = 4.32, 95% CI = 2.39-7.80) and model validation cohort (P = 0.009; HR = 9.08, 95% CI = 1.20-68.80), which might be of great utility both for understanding the molecular basis of LUAD relapse, providing risk stratification of patients, and establishing future monitoring plans.

Keywords: DNA methylation; lung adenocarcinoma; prognosis; recurrence.

Figure 1

Work flow of the study. Abbreviations: LUAD: lung adenocarcinoma; DMR: differentially methylated region; LASSO: least absolute shrinkage and selection operator.

Figure 2

Identification of DMRs in the patients with early stage LUAD. (A) Difference analysis yielded a total of 468 DMRs, including 438 hypermethylated regions and 30 hypomethylated regions in the marker discovery cohort (LUAD: n = 39; normal: n = 39). (B) Principal component analysis between LUAD and normal lung samples based on the methylation value of 468 DMRs. (C) Heatmap of the 468 DMRs in the marker discovery cohort. (D) Annotation information of the 468 DMRs, including location and the function of corresponding genes.

Figure 3

Construction of a 15-DMR prognostic signature in the model training cohort. (A) LASSO penalized Cox regression analysis yielded 15 DMRs for the recurrence risk predicting model construction. (B) Hazard ratio with 95% confidence interval (CI) and P-value of the 15 DMRs in the univariable Cox regression analysis for disease-free survival (DFS). (C) Heatmap of the methylation value of the 15 DMRs in low-risk (n = 45) and high-risk (n = 72) groups. (D) Kaplan-Meier (KM) curves for DFS in the model training cohort. (E) Time-dependent receiver operating characteristic (ROC) curves in the model training cohort. (F) Comparison of the risk score in the different clinical subgroups including age, sex, smoking history, and TNM stage.

Figure 4

Performance of the DMR signature in the model validation cohort. (A) Heat map of the methylation value of the 15 DMRs in low-risk (n = 27) and high-risk (n = 58) groups. (B) KM curves for DFS in the model validation cohort. (C) Time-dependent ROC curves in the model validation cohort. (D) Comparison of the risk score in the different clinical subgroups including age, sex, smoking history, and TNM stage.