In-depth characterization of the DNA methylation aging landscape links epigenetic noise and genotype-specific epigenetic aging to inflammaging

According to recent estimates, over 50% of the human DNA methylome (DNAm) changes with age. These age-associated changes display a significant diversity of patterns that have not yet undergone any comprehensive characterization. Based on a priori-knowledge we here guide a deep-learning based convolutional neural network to classify over 800,000 CpGs in blood into 10 distinct age-related DNAm patterns, which we extensively characterize in relation to chromatin state, immune-cell composition, inflammaging, epigenetic clocks, genotype and clonal hematopoiesis. One pattern defined by methylation quantitative trait loci (mQTLs) where the age-association is genotype-specific, uncovers a number of potential DNAm mediators of inflammatory diseases, which are distinct to those of existing causal clocks. Another pattern, representing age-dependent noise, is associated with inflammaging and all-cause mortality, and can be dissected into a component driven by variations in the memory CD8 + T-cell fraction, and another which is cell-type independent. In summary, by exposing and fully characterizing the diversity of age-related DNAm patterns we here improve our mechanistic understanding of these patterns.