DNA Methylation Signatures in Blood Distinguish Patients at Risk of Post-operative Atrial Fibrillation Following Cardiopulmonary Bypass

Post-operative atrial fibrillation (AF) is the most common complication for cardiac surgery, with an incidence of 27-40% and is associated with increased 30-day mortality. Several models exist to predict post-op AF using clinical data, however none have been widely implemented, and all have room for improvement (average area under ROC curve is ~0.70). We tested whether DNA methylation, assayed from pre-operative blood draws, could add predictive value. We enrolled a cohort of 55 adult patients undergoing cardiopulmonary bypass, with a 36% incidence of post-op AF. Reduced representative bisulfite sequencing provided single-base resolution for an average of 3 million CpGs with 10x sequencing coverage per sample. We first tested whether these CpGs were correlated with AF. Of the 809,569 CpGs that passed thresholds in all samples, 16,774 were significantly correlated with AF (corrected p-value <=0.01 using methylKit in R). We used the R package Glmnet to perform logistic regression on the 16,774 CpGs in 40 of the 55 patients using 5-fold cross validation to build a model and selecting the 381 CpGs which were included in 95% of the 5,000 models. Our assumption being that these 381 CpGs were predictive no matter the subset of patients selected in the randomly generated groups employed during cross validation. Using these 381 CpGs as inputs we built a final model that ultimately incorporated 51 CpGs. None of the 51 CpGs were significantly correlated with age (corrected p-value <=0.01), ruling out age as a confounding factor. We tested our model on the remaining 15 patients. Our model correctly predicted 5 of the 5 AF cases, and 9 of the 10 AF-free cases. Validation of our model in a larger cohort is ongoing, and future directions will incorporate our epigenomic signature with readily available clinical risk factors to further optimize the model.