Identification of a methylomics-associated nomogram for predicting overall survival of stage I-II lung adenocarcinoma

Abstract

The aim of this paper was to identify DNA methylation based biomarkers for predicting overall survival (OS) of stage I-II lung adenocarcinoma (LUAD) patients. Methylation profile data of patients with stage I-II LUAD from The Cancer Genome Atlas (TCGA) database was used to determine methylation sites-based hallmark for stage I-II LUAD patients' OS. The patients were separated into training and validation datasets by using median risk score as cutoff. Univariate Cox, least absolute shrinkage and selection operator (LASSO) and multivariate Cox analyses were employed to develop a DNA methylation signature for OS of patients with stage I-II LUAD. As a result, an 11-DNA methylation signature was determined to be critically associated with the OS of patients with stage I-II LUAD. Analysis of receiver operating characteristics (ROC) suggested a high prognostic effectiveness of the 11-DNA methylation signature in patients with stage I-II LUAD (AUC at 1, 3, 5 years in training set were (0.849, 0.879, 0.831, respectively), validation set (0.742, 0.807, 0.904, respectively), entire TCGA dataset (0.747, 0.818, 0.870, respectively). Kaplan-Meier survival analyses exhibited that survival was significantly longer in the low-risk cohort compared to the high-risk cohort in the training dataset (P = 7e - 07), in the validation dataset (P = 1e - 08), and in the all-cohort dataset (P = 6e - 14). In addition, a nomogram was developed based on molecular factor (methylation risk score) as well as clinical factors (age and cancer status) (AUC at 1, 3, 5 years entire TCGA dataset were 0.770, 0.849, 0.979, respectively). The result verified that our methylomics-associated nomogram had a strong robustness for predicting stage I-II LUAD patients' OS. Furthermore, the nomogram combined clinical and molecular factors to determine an individualized probability of recurrence for patients with stage I-II LUAD, which stood for a major advance in the field of personalized medicine for pulmonary oncology. Collectively, we successfully identified a DNA methylation biomarker and a DNA methylation-based nomogram to predict the OS of patients with stage I-II LUAD.

Figure 1

Workflow of model generation and subject enrolment.

Figure 2

Candidate methylation sites selection on the basis of the LASSO Cox regression model. (A) LASSO Cox regression with L1 regularization. Tenfold cross-validation for tuning parameter selection in the LASSO model via minimum criteria (the 1-SE criteria). (B) The LASSO Cox regression model was employed to determine the most robust hallmarks. LASSO coefficient profiles of the 84 methylation sites. A coefficient profile plot was created against log (lambda) sequence. Vertical line was used at the value selected by using tenfold cross-validation, where optimal lambda resulted in 25 non-zero coefficients.

Figure 3

Boxplots of methylation β values against risk group in the entire TCGA dataset. “High Risk” and “Low Risk” represent the high-risk and low-risk samples, respectively. The median risk score was applied as a cutoff. Vertical coordinates represent the β-value of 11-DNA methylation sites respectively. Mann–Whitney U test was employed to assess the diferences between the high-risk score and low-risk score groups.

Figure 4

Kaplan–Meier and ROC analysis of patients with stage I–II LUAD in in training, validation and whole datasets. (A, C, E) Kaplan–Meier analysis for stage I–II LUAD patients between the low-risk and high-risk. The x-axis is follow-up time, the y-axis is OS. The log-rank test indicated the higher risk scores were significantly correlated with worse OS (P < 0.05). (B, D, F) 1-, 3-, 5-year ROC curves of the 11-DNA methylation signature. Blue line, green line and red line represent 1-, 3-, 5-year ROC curves respectively.

Figure 5

Methylation risk score analysis of 393 stage I–II LUAD cases in the entire TCGA dataset. (A) Methylation risk score distribution against the rank of risk score. The red triangle represented the high-risk samples, the green ball represented the low-risk samples. (B) Survival status of stage I–II LUAD patients against the rank of risk score. The green ball referred to alive samples, the red ball referred to the dead samples. (C) Heatmap showed 11 methylation sites profiles in low- and high-risk groups. Each row of the heat map represented a profile of a methylation site.

Figure 6

Exploration of the 11 DNA methylation signature-based biological pathways. (A) Heatmap of top 20 enriched pathways associated with high risk group. Each row of the heat map represented a pathway and each column represented a stage I–II LUAD sample. (B) Correlation graph between risk scores and top 20 pathways. Each red ball represents a pathway, and each transverse line represents one sample. The score characteristic of each sample was shown in the graph.

Figure 7

In order to quantify the risk assessment and survival probability for individual stage I–II LUAD patients, a nomogram was developed in the entire TCGA dataset according to the 11 DNA methylation signature-based risk score, age and cancer status.

Figure 8

Analysis of the 11-DNA methylation biomarker-related nomogram in the entire TCGA dataset. (A) The horizontal axis stood for clinical factors, the vertical axis stood for the percentage of importance. (B) 1-, 3-, 5-year ROC curves for the 11-DNA methylation biomarker-related nomogram. (C–E) referred to the 1-, 3-, 5-year nomogram calibration curves, respectively. (F) The DCA for the nomogram. The net benefit was plotted versus the threshold probability. The red line represented the nomogram. The blue line represented the treat-all and the green line represented the treat-none.

Figure 9

ROC curves illustrated the effectiveness of the methylation-related nomogram and a few known biomarkers in predicting the prognosis of stage I–II LUAD patients. The AUC in Rotunno et al. is 0.810, in Chen et al. (0.656), in Kuo et al. (0.802), in Wu et al. (0.940), in Sun et al. (0.765), in Zhao et al. (0.760), in Sun et al. (0.786).