Comparison and combination of blood DNA methylation at smoking-associated genes and at lung cancer-related genes in prediction of lung cancer mortality.

Epigenome-wide association studies have established methylation patterns related to smoking, the major risk factor of lung cancer (LC), which are distinct from methylation profiles disclosed in LC patients. This study simultaneously investigated associations of smoking-associated and LC-related methylation markers with LC mortality. DNA methylation was determined by HM450K assay in baseline blood samples of 1,565 older adults in a population-based case-cohort study. The associations of 151 smoking-associated CpGs (smoCpGs) and 3,806 LC-related CpGs (caCpGs) with LC mortality were assessed by weighted Cox regression models, controlling for potential confounders. Multi-loci methylation scores were separately constructed based on smoCpGs and caCpGs. During a median follow-up of 13.8 years, 60 participants who had a first diagnosis of LC died from LC. The average time between sample collection and LC diagnosis was 5.8 years. Hypomethylation at 77 smoCpGs and 121 caCpGs, and hypermethylation at 4 smoCpGs and 66 caCpGs were associated with LC mortality. The associations were much stronger for smoCpGs than for caCpGs. Hazard ratios (95% CI) were 7.82 (2.91-21.00) and 2.27 (0.75-6.85), respectively, for participants in highest quartile of Score I (based on 81 smoCpGs) and Score II (based on 187 caCpGs), compared with participants in the corresponding lower three quartiles. Score I outperformed Score II, with an optimism-corrected C-index of 0.87 vs. 0.77. In conclusion, although methylation changes of both smoking-associated and LC-related genes are associated with LC mortality, only smoking-associated methylation markers predict LC mortality with high accuracy, and may thus serve as promising candidates to identify high risk populations for LC screening.