Capturing DNA methylation heterogeneity and examining its relationship to somatic mutations in Alzheimer’s disease

Background One of the barriers to the identification of molecular targets in Alzheimer’s disease (AD) is its biological heterogeneity. However, computational approaches used to quantify heterogeneity in cohorts can be improved to produce more nuanced output. We develop novel and refine existing statistical methods to capture molecular heterogeneity and examine it on multiple ‐omic levels in the context of AD. Method As AD is associated with aging, we consider key age‐associated molecular changes: somatic mutations and a novel DNAme biomarker of aging, Stochastic Epigenetic Mutations (SEM). SEM are outliers in DNAme values at genomic sites when compared with the rest of cohort and can quantitatively capture DNAme heterogeneity. However, it was unclear whether the existing SEM detection methods were optimal for capturing biologically meaningful signals. We test statistical approaches to optimize these methods using the strength of association between SEM burden and chronological age as our metric of choice in brain and whole blood. Furthermore, we test the reliability of these methods in technical replicates. Hypothesizing that SEM coming from different types of DNAme probes (e.g., normally unmethylated or methylated) may contain information on distinct processes, we develop machine learning methods to subcategorize SEM and test their associations with AD pathology. Finally, we assess the relationship between somatic mutations, SEM, and AD pathology in multi‐region brain datasets generated by our lab from the ROSMAP cohorts. Result We refined the existing approaches for SEM detection using the skewness adjusted Interquartile Range to correct for the non‐symmetrical DNAme distributions. Additionally, we subcategorized SEM and found that only certain SEM types are linked with the hallmarks of AD; we also observe similar associations between AD pathology and somatic mutations burden. Finally, we saw that these associations are present only in certain brain regions that are known to be affected by AD. Conclusion We show that DNAme contains important heterogeneous signals associated with AD and somatic mutations. We will perform mediation and enrichment analyses to elucidate the crosstalk of the two age‐associated molecular changes in the context of AD and hope that this will bring us one step closer to identifying novel therapeutic targets for this debilitating disease.