Background: Lung cancer causes more deaths worldwide than any other cancer, but timely interventions targeting high-risk individuals may reduce mortality. Lung cancer screening programmes apply validated risk prediction models to identify high-risk individuals who would likely benefit from lung scans. However, only a very small proportion of these scans detect cancer. Recent studies have identified biomarkers in peripheral blood that can improve the accuracy of lung cancer risk estimates within the subsequent 2-3 years. DNA methylation is a potentially complementary source of biomarkers as it is strongly associated with long-term exposure to lung cancer risk factors such as smoking and air pollution.

Objectives: We perform a preliminary analysis of how the capacity of DNA methylation to predict lung cancer risk compares to PLCOm2012, a questionnaire-based lung cancer prediction tool that forms the basis of most screening protocols.

Methods: This study measured DNA methylation (DNAm) in blood samples collected from 213 lung cancer cases up to two years before diagnosis and 131 control participants from the European Investigation into Cancer and Nutrition (EPIC). DNAm was measured using Twist Targeted Methylation Sequencing to 3900 genomic regions known to be associated with lung cancer risk, lung cancer risk factors, aging, blood cell types, protein abundance and ancestry. DNAm associations with lung cancer risk were adjusted for age, number of sequencing reads, and DNAm-based blood cell count estimates and compared to associations additionally adjusted for PLCOm2012.

Results: Of the 118,986 DNAm sites covered by at least 10 sequencing reads in all samples, 406 sites were differentially methylated between lung cancer cases and controls (p < 4.2 x 10-7). After additionally adjusting for PLCOm2012, the number of associations was reduced to 22, all near or coinciding with sites known to be strongly associated with smoking history at genes GFI1, F2RL3 and AHRR.  Performance of top DNAm sites for differentiating between cases and controls was comparable to that of PLCOm2012 (e.g. AUC = 0.77 for the cg05575921 site in the AHRR gene compared to 0.74 for PLCOm2012). Performance was only slightly attenuated (AUC = 0.76) among individuals deemed high-risk by PLCOm2012 (i.e. PLCm2012 > 1.5%).  

Conclusions/implications: These preliminary findings suggest that DNAm is comparable to PLCOm2012 and may capture additional lung cancer risk information even among individuals identified as high-risk by PLCOm2012. We note that, in the next few weeks, we will expand this preliminary analysis to include data for ~1900 additional study participants. Described analyses will be repeated and further developed, including development and evaluation of a novel DNAm risk model. These new findings will be presented at the conference.