Anatomy of DNA methylation signatures: Emerging insights and applications

Abstract

DNA methylation (DNAm) signatures are unique patterns of DNAm alterations defined for rare disorders caused by pathogenic variants in epigenetic regulatory genes. The potential of DNAm signatures (also known as "episignatures") is just beginning to emerge as there are >300 known epigenetic regulatory genes, ∼100 of which are linked to neurodevelopmental disorders. To date, approximately 50 signatures have been identified, which have proven unexpectedly successful as predictive tools for classifying variants of uncertain significance as pathogenic or benign. The molecular basis of these signatures is poorly understood. Furthermore, their relationships to primary disease pathophysiology have yet to be adequately investigated, despite clear demonstrations of potential connections. There are currently no published guidelines for signature development. As signatures are highly dependent on the samples and methods used to derive them, we propose a framework for consideration in signature development including sample size, statistical parameters, cell type of origin, and the value of detailed clinical and molecular information. We illustrate the relationship between signature output/efficacy and sample size by generating and testing 837 DNAm signatures of Kleefstra syndrome using downsampling analysis. Our findings highlight that no single DNAm signature encompasses all DNAm alterations present in a rare disorder, and that a substandard study design can generate a DNAm signature that misclassifies variants. Finally, we discuss the importance of further investigating DNAm signatures to inform disease pathophysiology and broaden their scope as a functional assay.

Figure 1

DNAm signatures generated for Kleefstra syndrome by iteratively resampling 2–9 affected individuals for the discovery cohort (n = 10 total) versus age- and sex-matched control subjects (n = 40) (A) Relationship between the number of affected individuals in the discovery group (x axis) and mean signature size (y axis), i.e., number of significant CpGs (FDR-corrected p value < 0.05, mean group difference > 10%). (B) Mean number of CpGs in each signature that overlap with the 10-sample signature CpGs (239 CpGs). (C) Sensitivity of each signature (FDR-corrected p value < 0.05), tested on the separate validation set of KS-affected individuals (n = 10; 609 signatures). (D) Sensitivity of the corresponding 609 signatures meeting an FDR-corrected p value < 0.05 (x axis) versus an FDR-corrected p value < 0.1 (y axis). With fewer than eight affected individuals in the discovery group, the FDR10 signatures were more sensitive than the corresponding FDR5 signatures. At eight or more individuals in the discovery group, the signatures performed equally well or the FDR5 signatures performed better.