Epigenetic Instability and Liquid Biopsy: Rethinking Early Cancer Detection

Johns Hopkins investigators describe a liquid-biopsy early-detection framework that emphasizes epigenetic instability—defined as increased variability in DNA methylation—rather than reliance on absolute, fixed methylation markers, using the Epigenetic Instability Index (EII) as the central signal.

EII is presented as a cfDNA methylation–based, variance-driven metric designed to quantify within-sample heterogeneity of methylation levels across predefined CpG-island regions. Rather than asking whether specific loci are methylated to a particular degree, the assay asks how variable methylation is within selected regions of circulating DNA. The authors argue that conventional absolute methylation signatures can become overfit to derivation cohorts and may lose performance when applied to broader, more heterogeneous populations, whereas variability-based signals may be more robust across settings.

The development workflow clarifies where the work sits on the translational maturity curve. Using 2,084 TCGA tissue samples (179 normal and 1,905 tumor samples across five cancer types), the investigators identified 269 CpG-island regions (B-269) that captured a substantial fraction of cancer-associated methylation variability. These regions were then locked and applied to independent cfDNA datasets for breast cancer and lung adenocarcinoma, with machine-learning classifiers trained and evaluated using an 80:20 split and cross-validation. While this design supports technical feasibility and signal consistency across datasets, it does not substitute for prospective validation in intended-use, population-level screening cohorts.

Performance is reported at a fixed operating point of 95% specificity, with sensitivity evaluated as the dependent variable. At this threshold, sensitivity for stage IA lung adenocarcinoma is reported at approximately 81%. Anchoring specificity in this manner defines the false-positive rate by design and frames sensitivity as the proportion of true cancers detected under that constraint. Within the authors’ framework, the stage IA lung result is presented as an early indication that a variability-based methylation metric can distinguish early cancer from healthy controls under stringent specificity requirements, rather than as definitive screening performance.

Using the same 95% specificity anchor, early-stage breast cancer sensitivity is reported at approximately 68%. Additional analyses in TCGA tissue data show elevated epigenetic instability signals in colon adenocarcinoma, glioblastoma, and pancreatic adenocarcinoma, but these findings are exploratory and derived from modeling rather than cfDNA screening cohorts. Accordingly, the authors characterize these multi-tumor observations as promising early signals, not as evidence of validated, cross-cancer screening readiness.