September 22, 2026 | Epigenetic instability, a measure of the chaotic nature of methylation in specific regions of circulating cell-free DNA (cfDNA), has emerged as a promising new approach for early cancer detection. In a recent proof-of-principle study, the metric outperformed conventional absolute methylation signatures in detecting early-stage breast and lung cancers.
Details of the study were shared at the recent Next Generation Dx Summit by Hariharan Easwaran, Ph.D., an associate professor of oncology at the Sidney Kimmel Cancer Center at Johns Hopkins University, who also spoke with Diagnostics World News about the novel approach earlier this year. The focus is on a panel of 269 methylation markers in “CpG islands,” short regions of DNA with a high concentration of cytosine (C) and guanine (G) nucleotides linked by a phosphate bond.
Instability itself is suspected to be an early-stage process, says Easwaran, noting the sensitivity of the metric improved when amplified in any DNA methylation fragment. The testing was done using a lung dataset where samples had close to 500x coverage as well as a breast dataset with only 1x coverage—meaning, on average, every base pair in the sample's genome was read or scanned only a single time by the sequencing machine. This confirmed a robust signal-to-noise ratio even at ultra-low sequencing depths, which is where standard absolute metrics typically lose statistical power.
Researchers are now developing cost-effective, targeted methylation assays to capture and sequence the pan-cancer panel at high definition from cfDNA for multiple different tumor types, Easwaran reports. The leading commercial DNA methylation assay for multi-cancer early detection currently on the market is Grail’s Galleri test, whose sensitivity is higher in later stages and varies significantly depending on the cancer type.
Easwaran’s key interest is understanding how the cancer epigenome evolves in aging and how that impacts tumorigenesis. Epigenetic instability is a consequence of constant attacks from environmental and intercellular metabolic stressors, he says, but there’s a lot of heterogeneity to the pattern of that instability. Finding an early cancer signal in the chaos is thus a complicated quest.
Efforts to develop liquid biopsy tests for cancer detection have revolved largely around analytes such as circulating tumor DNA and circulating tumor cells and more recently lipids. Epigenetic markers, notably DNA methylation, have also been of great interest because they are ubiquitous across cancer types. Even with multiomics profiling, DNA methylation patterns remain the dominant signal, says Easwaran.
“DNA methylation is one of the best marks on the DNA for cancer detection ... [because] there are a lot of genome-wide changes in DNA methylation during cancer tumorigenesis,” he continues, the highest concentration of which are seen in CpG islands. In cancer cells more than normal cells, the C and G residues in these regions are more heavily methylated, while demethylation generally occurs with cancer in over 90% of the human genome, including gene bodies, intergenic regions, and repetitive elements. This epigenetic paradox makes the DNA methylation pattern of interest for early cancer detection.
In addition to dynamic changes in the genome, enormous changes in DNA methylation patterns occur during tumor evolution, Easwaran says. There is so much heterogeneity in DNA methylation, especially within the tumor itself, making it challenging to identify a single, consistent methylation signature that reliably serves as a diagnostic or prognostic biomarker across the entire tumor as well as patients. Many of the methylation changes are happening in the genes that control cancer biological pathways that play a role in tumor evolution and drug resistance mechanisms, he adds.
Measuring Variability
The epigenetic heterogeneity that happens early on in tumorigenesis, if better understood, could be useful for early cancer detection. It has long been observed that epigenetic changes accompany the development stage of adenomas, the primary precursors to colorectal cancer, he points to as an example. But the heterogeneity in that pattern makes it hard to pinpoint the transition to cancer, since this is also when the methylation profiles of cells get more diverse.
This has been shown in animal models where fibroblasts were forced to form tumors by introducing oncogenes, Easwaran says. Researchers used them to compare what happens during cancer development versus normal aging and senescence.
Different independent clones of tumors, in different stages of tumorigenesis development, were thereby generated. In senescent cells as well as normal healthy ones, DNA methylation patterns are highly programmatic “but the moment cells start to form cancers they have a lot of heterogeneity in methylation,” Easwaran shares. With cancer, epigenetic changes are more stochastic than what happens in normal homeostatic processes, including during aging.
It is classically known that cancer flips the normal DNA methylation landscape, driving widespread alterations at CpG sites, says Easwaran, returning to the epigenetic paradox. In cancer, CpG residues are often assumed to show uniform methylation changes, “but underlying this assumption lies a sea of tumor-derived fragments that have variable DNA methylation patterns,” which are missed by PCR- and sequencing-based assays focused on fixed, absolute changes in methylation levels.
This heterogeneity inevitably creates an “intermediate zone” of missed detection and false positives, where the signal becomes too weak for DNA methylation to be used as a marker, he says. The remedy pursued by Easwaran and his team was to try to take DNA methylation heterogeneity into account by measuring the variance across CpG residues and use that as a signal for cancer detection—and then compare its performance to the traditional approach based on DNA methylation averages. Similar approaches in the field have shown promising use of epigenetic instability, and a need for better understanding of these dynamics.
Tests such as Galleri belong to a class of epigenetic assays that distinguish cancer-derived DNA fragments from healthy ones by using probes to capture and sequence DNA methylation patterns at specific CpG sites across the genome, often targeting on the order of a million such sites to filter out background noise from normal, age-related methylation changes, says Easwaran. These assays are typically built on an underlying assumption that, within a given genomic region, CpG sites are either fully methylated or fully unmethylated across all DNA molecules in a sample.
An analysis of colorectal cancer datasets in the Cancer Genome Atlas paints a different picture. Most of the CpG methylation patterns seen in individual patient samples were highly variable, primarily falling in the intermediate 50% to 60% methylation range, rather than clustering near the fully methylated or fully unmethylated extremes. These intermediate methylation patterns are not completely accountable by tumor purity, Easwaran says.
This reflects a mixture of methylation states across individual DNA molecules within the same tumor sample—a level of heterogeneity that conventional assays, designed around fixed methylation thresholds, are likely to miss. In this context, the epigenetic measure of computed variance in methylation in CpG islands separated tumor samples from normal samples as well as the traditional average-methylation strategy measuring how much a single CpG site is methylated relative to a baseline average, says Easwaran.
The next step was to have machine learning use the novel metric to identify specific regions of the genome capturing most of this variability and turn that into a small set of panels suitable for “more diagnostically applicable” targeted methylation profiling assays, he continues. This is how Easwaran and his team arrived at the 269-marker panel that was further validated by sequencing cfDNA methylation data from independent liver and lung samples.
“For every read level, we actually measured the variance and then used that as a feature in our dataset and trained models to see what does better ... the variance or the classical methylation-based metrics,” says Easwaran. In breast as well as lung cancers, one notable finding was a uniform difference from lowest to highest variance between cancers and normals when the epigenetic-based instability metric was used across a range of methylated DNA fragments. With the methylation-based metrics, good differences were seen in general with advanced disease.
“We went ahead and made models using this read-level variance as a feature,” says Easwaran, comparing it with the classical approach of identifying differentially methylated regions through statistical modeling. Across cancer stages, the smaller set of 269 CpG island features consistently produced comparable or better classification performance than the differentially methylated region approach—including in lung cancer.
The framework would have potential for early cancer detection or more generally for “cases where you want to triage patients based on initial screening,” says Easwaran, noting that he and his co-inventors have applied for a patent on the technology for detecting cancer from cfDNA using novel epigenetic instability-based metrics. Detecting early epigenetic perturbation could eventually help bridge the domains of biological aging and cumulative, environmental exposure-driven epigenetic changes, potentially offering a more precise way to identify who should undergo low-dose CT scans, he notes, since the screening modality suffers from a high rate of false-positive results in low-incidence populations.