Transforming blood-derived episignatures into cell-type-agnostic classifiers: A shortcut to prenatal episignatures.

Abstract

DNA methylation (DNAm) episignatures are stable disorder-specific epigenetic patterns that serve as valuable biomarkers for assessing variant pathogenicity and phenotypic outcomes in neurodevelopmental disorders (NDDs). However, episignatures derived from whole blood are inherently tissue- and cell-type specific, limiting their applicability in prenatal diagnostics. To explore the feasibility of developing episignatures capable of informing variant pathogenicity across tissues and developmental stages, we conducted a proof-of-concept study using Down syndrome, a common NDD caused by trisomy 21 (T21). We generated a blood-derived T21 episignature using a large cohort of 266 samples. Next, we used that episignature and publicly available DNAm data for 850 T21 and control samples across six different pre- and postnatal tissues to train machine-learning models, thereby enabling accurate prediction of T21 status across all tested tissues. Notably, our results show that models trained on postnatal blood-derived signatures as well as other tissues can generate cell-type-agnostic disease-specific patterns. This method also supported integrating well-characterized postnatal episignatures with a limited set of prenatal samples, which generated an episignature capable of accurately classifying prenatal samples. This approach forges a path for cross-tissue DNAm biomarker development and lays the groundwork for a workflow to rapidly integrate episignatures into prenatal diagnostics.