The core insight is that some disease mechanisms never show up cleanly in DNA sequence. Pathogenic changes can hide in non-coding regulatory elements, repetitive regions, complex structural rearrangements, imprinted loci or genomic stretches that are difficult to align with short sequencing reads. Others alter splicing, allele-specific expression, chromatin state, RNA stability or translation in a tissue-specific or developmental-stage-specific manner. Mosaic variation may fall below the detection threshold of standard assays or be absent from the tissue that was sampled. In these situations, the sequence result alone is insufficient to establish causality, and clinicians need a way to interrogate the molecular consequences of variation rather than the variation itself.

An episignature is best understood as a molecular phenotype rather than a primary genetic lesion. It does not usually identify the causal variant directly; instead, it reflects the downstream epigenomic consequences of disrupted gene regulation. When a pathogenic variant affects a gene involved in chromatin regulation, it can alter the genome-wide regulatory state and generate a reproducible methylation pattern detectable in peripheral blood. Sequencing identifies candidate DNA variants, whereas episignature testing evaluates whether those variants, or an unresolved disease mechanism, have produced a methylation profile consistent with a recognised disorder. The distinction matters because a methylation episignature provides functional evidence of disease mechanism, not direct variant detection. The term is used generically in the review and should not be conflated with EpiSign, the trademarked clinical assay developed by the Western University and London Health Sciences Centre group.

The biology behind these signatures follows a writer-eraser-reader framework. Writers install regulatory marks on DNA, histones and RNA, erasers remove them, and readers interpret them by recruiting downstream molecular machinery. Pathogenic variants can strike any component. A disorder may result from failure to deposit an activating mark, inability to remove a repressive mark, inappropriate recognition of methylated DNA, disrupted chromatin remodelling, or altered interpretation of modified RNA. Among these systems, DNA methylation is the most mature and widely implemented clinical biomarker. Variants in DNMT3A cause Tatton-Brown-Rahman syndrome with a recognisable methylation profile; TET3-related neurodevelopmental disorder illustrates disruption of cytosine oxidation and demethylation pathways; and MECP2-related Rett syndrome demonstrates how altered reading of methylated DNA can cause severe disease.

Histone modifications add further clinically relevant layers. H3K4 methylation, associated with active promoters and enhancers, is deposited by KMT2-family methyltransferases and removed by demethylases such as KDM1A and the KDM5 family; variants in KMT2D and KDM6A cause the two forms of Kabuki syndrome, while KDM5C variants cause X-linked intellectual disability with a characteristic methylation pattern. H3K9 methylation maintains heterochromatin, and its disruption underlies EHMT1-related Kleefstra syndrome. H3K27 methylation by the Polycomb complex regulates developmental gene control, and H3K36 methylation is particularly important because it links histone marks to DNA methylation itself: NSD1 and SETD2 deposit distinct H3K36 marks that are recognised by PWWP domains in DNMT3A and DNMT3B, and pathogenic variants in these genes produce the broad methylation changes seen in Sotos syndrome and Luscan-Lumish syndrome. Histone acetylation follows the same paradigm, with EP300, CREBBP, KAT6A and KAT6B as key writers, and Rubinstein-Taybi syndrome and KAT6B-related disorders illustrating the clinical consequences.

Strikingly, methylation profiling can even distinguish mechanistically distinct disorders arising from the same gene. In KAT6B-associated conditions, truncating variants in different regions of exon 18 produce separate episignatures: those causing Genitopatellar syndrome, predicted to escape nonsense-mediated decay, generate a profile consistent with a probable gain-of-function mechanism, while those causing Say-Barber-Biesecker-Young-Simpson syndrome, predicted to cause loss of function, exhibit a different signature. A similar principle applies in Menke-Hennekam syndrome, where genome-wide methylation profiling has identified distinct domain-specific episignatures associated with variants affecting the ZZ, TAZ2 and ID4 regions of CREBBP and EP300, demonstrating that these signatures can be mechanism-specific rather than gene-specific.

The scope of episignature analysis is also expanding beyond classical neurodevelopmental chromatinopathies. A recent study identified a distinct DNA methylation episignature in Diamond-Blackfan anemia syndrome, a genetically heterogeneous ribosomopathy and inherited bone-marrow-failure disorder, showing that reproducible methylation phenotypes can arise from disease mechanisms outside canonical epigenetic regulatory pathways. Reproducible methylation patterns can even reflect environmental exposures: a peripheral-blood episignature has now been defined for clinically confirmed fetal alcohol syndrome, with potential utility as a supportive molecular biomarker, though such profiles should complement rather than replace established clinical diagnostic criteria.

Generating the data is a technical exercise in itself. Clinical methylation profiling usually begins with genomic DNA from peripheral blood, which undergoes bisulfite conversion so that unmethylated cytosines are read as thymine while methylated cytosines remain protected, though conventional arrays cannot distinguish 5-methylcytosine from 5-hydroxymethylcytosine. The converted DNA is hybridised to probes targeting CpG sites across the genome, and fluorescent signals yield beta values, which approximate methylation proportions, and M-values, a log-ratio transform preferred for statistical modelling. Quality control and normalisation pipelines account for age, sex, batch effects and blood cell composition, ensuring that observed patterns reflect disease-associated biology rather than technical noise.

Interpreting the data is a pattern-recognition problem in high dimensions: each sample carries methylation values across hundreds of thousands of CpG sites, while rare-disease cohorts may contain only a handful of confirmed cases. Differential methylation analysis identifies candidate loci, with covariate adjustment and false-discovery-rate correction to control artefacts. Informative CpGs are then used for unsupervised visualisation through hierarchical clustering heatmaps, multidimensional scaling, principal component analysis and UMAP, which show whether a patient clusters with known cases. Supervised classifiers, typically support vector machines, perform the diagnostic prediction step; in the EpiSign framework this produces an MVP score, a classifier-derived confidence measure of similarity to a reference episignature that is explicitly not a p-value, a disease severity metric or an ACMG/AMP pathogenicity class. Validation statistics, including sensitivity, specificity and predictive values, determine how much weight a result deserves, and independent validation is essential because small discovery cohorts are vulnerable to overfitting.

The clinical utility is strongest in three settings: reclassifying a variant of uncertain significance in a gene with a validated episignature, resolving phenotypically suggestive but sequence-unsolved cases, and discriminating between clinically overlapping syndromes in the same regulatory pathway. A concordant methylation profile in a patient with a VUS in KMT2D, NSD1, DNMT3A or KDM5C can provide functional evidence supporting variant reclassification. But caution is essential. A negative result may reflect tissue specificity, age-related methylation effects, low-level mosaicism, a hypomorphic variant or a mechanism not represented in the training data, and should never be interpreted automatically as evidence of benignity. Most genetic disorders still lack validated episignatures, peripheral blood may not capture methylation changes confined to neural tissue, and reference datasets must become more ancestrally and geographically representative to ensure equitable diagnostic performance.

The future points toward integration. Long-read sequencing platforms, particularly nanopore sequencing, can simultaneously detect sequence variants, structural variants, repeat expansions and DNA methylation from the same molecule without bisulfite conversion, and phasing can establish whether variants and methylation changes occur on the same allele. Machine-learning approaches have already transformed a blood-derived Down syndrome episignature into cell-type-agnostic classifiers that performed across six tissue types, offering a route to prenatal applications. Multi-omic integration with RNA sequencing, chromatin accessibility assays such as ATAC-seq, CUT&Tag histone profiling and proteomics promises a more complete picture of disease mechanism, while epitranscriptomic marks such as RNA m6A represent an emerging but still investigational layer. For the foreseeable future, DNA methylation episignatures will remain the most clinically mature epigenomic application, but the broader trajectory is clear: rare-disease diagnostics is moving from variant detection toward integrated interpretation of genotype, molecular function and phenotype, blurring the line between genomic and epigenomic medicine.

Subject of Research: Clinical use of DNA methylation episignatures for diagnosing rare neurodevelopmental and chromatin disorders

Article Title: Clinical Epigenomics in Rare Diseases: Interpreting DNA Methylation Episignatures

Article References: Goel, H., Goh, S., & Stuart, L. (2026). Clinical Epigenomics in Rare Diseases: Interpreting DNA Methylation Episignatures. Molecular Genetics & Genomic Medicine, 14(10), Article e70319. https://doi.org/10.1002/mgg3.70319

Image Credits: AI Generated

DOI: 10.1002/mgg3.70319

Keywords: DNA methylation, episignatures, rare disease, chromatinopathies, neurodevelopmental disorders, genetic diagnostics, variants of uncertain significance, epigenomics, long-read sequencing, machine learning, Kabuki syndrome, multi-omics