March 3, 2026 | The quest to marry different DNA methylation technologies for the purpose of early cancer detection has been ongoing for several decades now. Among the recent breakthroughs is an “epigenetic instability index” (EII) measuring the randomness of methylation rather than just consistent, fixed changes.
Those “stochastically” altered cancer signals are detectable in blood and could potentially be employed with minimal residual disease (MRD) techniques and triaging of patients coming in for routine cancer screenings and, longer term, for purposes of multi-cancer early detection (MCED), according to Hariharan Easwaran, Ph.D., associate professor of oncology at the Johns Hopkins University School of Medicine.
The novel EII metric could provide a more robust and universal biomarker for early cancer detection than currently available methods, suggests the findings of a proof-of-concept study that was published recently in Clinical Cancer Research (DOI: 10.1158/1078-0432.CCR-25-3384). For early-stage lung and breast cancers, it did an exceptional job distinguishing patients from healthy individuals.
In separate ongoing studies, the researchers have found that the approach has other applications in oncology, including prostate cancer screening from biopsy samples, reports Sara-Jayne Thursby, Ph.D., a postdoctoral researcher in Easwaran’s lab.
Publicly available cancer DNA methylation datasets from 2,084 samples were analyzed to identify a panel of 269 specific genomic regions that capture most DNA methylation variability across multiple cancer types, she says. Those regions could now be used to design biomarker panels.
“We prioritized stochastic methylation alterations over the magnitude or direction of the change,” says Thursby. Methylation changes also occur as people age, but in more gradual ways. With cancer, changes in methylation tend to be both extreme and chaotic.
This is the latest innovative cancer detection technique to come out of Johns Hopkins, a historic leader in the field, says Easwaran, including mutation-based and, more recently, fragmentomics-based approaches. It is also where methylation panels and the study of DNA methylation in cancer were pioneered back in the 1980s, which were foundational to the development and eventual market approval of Cologuard for the detection of colorectal cancer.
Easwaran’s work on DNA methylation dates to 2009 and development of the EII metric is the latest chapter whereby he and his Johns Hopkins colleagues are seeking to expand on the knowledge that the process is highly heterogeneous and acts as a dynamic, unstable signal that drives cancer evolution “very early on,” he says. In these studies, Easwaran worked closely with his long-time Hopkins collaborator Tom Pisanic, Ph.D. Other key collaborators include Zhicheng Jin, Jacob Blum, Andrei Gurau, Michaël Noë, Robert B. Scharpf, Victor E. Velculescu, Leslie Cope, Malcolm Brock, and Stephen Baylin.
It has been known for decades that prolific epigenetic stochasticity begins at the earliest stages of cancer development, based on studies in cancer tissues sometimes termed epigenetic polymorphism or epigenetic entropy, Easwaran continues. But it took some time for epigenetic instabilities to be integrated into cell-free DNA (cfDNA) detection technologies, and the Easwaran lab was an early advocate of the approach.
Most initial studies employing epigenetic markers involved genes known to be cancer-methylated, says Easwaran. With the advent of The Cancer Genome Atlas, it was possible to look globally at all genes associated with specific cancers as well as the smaller universal set of pan-cancer markers. This has revolutionized the discovery of various biomarker panels.
The EII study was originally presented as an abstract at the 2024 AACR meeting, and between then and publication in the peer-reviewed journal, Easwaran and Thursby compared their EII approaches with traditional techniques relying on absolute methylation values. This involved retrieving and streamlining the codes for comparing EII with robust methods that model the population-level DNA methylation differences in cancer samples. Much of the work presented involved retrieving and analyzing data that were publicly available for the analyses, Easwaran emphasizes.
Key findings are that EII-based metrics could, with remarkable accuracy, detect both early-stage lung cancer (81% sensitivity/95% specificity) and breast cancer (68% sensitivity/95% specificity), says Thursby. The approach also showed promise in detecting signals from colon, brain, pancreatic, and prostate cancers.
It is a universal cancer biomarker because it captures a fundamental hallmark of cancer—the breakdown of normal epigenetic regulation which in turn leads to variability in gene expression that is independent of the tissue of origin, she says. cfDNA sequencing datasets were not readily available for four of the six cancer types in the validation exercise.
In early work, “We looked at overall methylation variability in cancer patients and healthy patient populations in general,” says Easwaran. The methylation landscape, when analyzing normal tissues, looks alike person to person while tumors from any two cancer patients are markedly dissimilar.
The reason goes back to their epigenetic instability, Easwaran continues. Cancers undergo rapid early changes in methylation that vary from cell to cell, as these changes do not occur in a coordinated manner. Even if broadly the same genomic regions are affected, actual sites undergoing methylation changes vary. The general distribution of those methylation patterns across regions is likely one reason why the EII method performs so well, he adds.
Thursby says that she experimented with different types of machine learning techniques, to learn which would do the best job of “bringing out the value of our [epigenetic instability] measure,” before settling on the random forest method. It was a perfect fit for the noisy methylation data they were dealing with. The computational task was to partner stochastic changes with a variability metric to quantify the instability of an epigenome, a metric that doesn’t have a defined numerical value for a normal sample.
“This applies even to regular methylation,” adds Easwaran, due to age-related changes. Some of the same methylation changes are seen with both cancer and aging, and the relationship is not fully understood, complicating detection efforts. “Epigenetic biomarkers also face the challenge that stochastic methylation errors and epigenetic drift can blur the distinction between age‑related and tumor‑associated changes … whereas the kind of pronounced epigenetic instability seen in cancers is generally less typical of normal aging.”
Random forest, which is an ensemble-based machine learning algorithm, potentially ruled out all the noise to better distinguish tumor and normal patterns, learning that was then applied to other datasets, notes Thursby. “I wanted a machine learning algorithm that would be complementary to the metric.”
The study’s authors foresee the EII metric one day being the basis of a single MCED test, along with the tissue-of-origin information in DNA methylation data, says Easwaran, which represents work now underway. It could also be used in primary tissues for triage in “borderline cases” where oncologists are questioning negative results of mammograms or prostate-specific antigen tests. In the latter instance, he adds, epigenetic instability would be measured in tissue extracted via needle biopsies.
Existing DNA methylation tests are already being used for all these purposes, says Easwaran, but “information about instability can pick up the very early signals.” Integration of the two diagnostic modalities could also be “very easily done.”
Immediate next steps include identifying the single best methylation profiling technology, standardizing it, and then using it across vast and diverse datasets. The contenders include the enzyme-based methylation sequencing of New England Biolabs, the 6-base sequencing method of Biomodal, and the 5-base sequencing assay of Illumina.
The research team further plans to do targeted sequencing of the 269 newly identified regions in the genome where epigenetic instability is concentrated, says Easwaran. Deep sequencing can potentially capture rare, tumor-derived fragments in blood or tissue.
Hopkins collaborators have clinical trials already underway for various MRD tests, and the EII metric could perhaps be a beneficial addition. But much remains to be done over the short term, says Thursby, including analytical validation of the novel biometric and determining its value.
Epigenetics remains a frontier for exploring ways to leverage it for cancer detection, and Easwaran is therefore interested in pursuing collaborative opportunities. One of the biggest barriers to progress has been the shortage of publicly available datasets and streamlined codes for doing the work, he says in making his case for open science.