A team of investigators at the Johns Hopkins Kimmel Cancer Center has found a way to detect cancer early by measuring the randomness of DNA methylation rather than focusing on specific methylation changes. In their study, published in Clinical Cancer Research, the researchers found that by analyzing random variation in DNA methylation across selected CpG island regions in cell-free DNA, they could develop a quantitative measure called the Epigenetic Instability Index (EII) to provide early-stage cancer detection through a blood-based liquid biopsy.
Senior author Hariharan Easwaran, PhD, an associate professor of oncology at the Johns Hopkins University School of Medicine, noted that their approach to leverage methylation signals to detect cancer is markedly different from current methods. He said that during carcinogenesis, DNA methylation is often lost across large regions of the genome, while narrow regions rich in cytosine–guanine (CpG) dinucleotides (CpG islands) tend to acquire DNA methylation. These cancer‑associated methylation changes often arise in a mosaic, heterogeneous manner across cells and alleles, consistent with partially stochastic acquisition rather than a fully deterministic on/off switch at each locus, reflecting the underlying epigenetic instability in the maintenance of epigenetic information, which is particularly pronounced in cancers.
Because of this, the methylation changes in CpG-islands are considered good biomarkers. These epigenetic changes evolve in a very stochastic manner, with inter-individual and intra-tumor heterogeneity, and potentially indicate underlying instability in maintenance of DNA methylation at these regions.
“Typically, most studies select regions that show the largest cancer‑specific differences in mean DNA methylation (differentially methylated regions) compared with normal controls,” he said. “We hypothesized that, because methylation at these loci is heterogeneous and cohort‑dependent, both the magnitude of change and the specific regions identified can vary, limiting the robustness of such markers across cohorts. If methylation at these loci arises in a partly stochastic manner, then identifying regions based on their degree of epigenetic instability and developing cfDNA‑based metrics to quantify this instability could provide more reliable approaches for cancer detection.”
While some existing liquid biopsy methods can detect cancer by identifying absolute differences in methylation at individual CpG sites. According to the researchers, these approaches can perform well in narrowly defined populations but may not generalize across diverse patient groups. Prior research has shown that cancer evolution is marked by extensive epigenetic heterogeneity, with methylation changes arising in a stochastic manner rather than following uniform patterns.
The Johns Hopkins team focused on CpG islands, since these genomic regions typically have low levels of DNA methylation as well as low variability in healthy tissues. During carcinogenesis, these regions exhibit methylation unevenly across individual CpG sites. Because CpG islands are normally stable, shifts away from their normal state provide a measurable signal of epigenetic dysregulation associated with early cancer development.
To identify which regions would be best for identifying this variability, the researchers analyzed publicly available DNA methylation datasets from 2,084 cancer and normal samples. From this analysis, they identified 269 CpG island regions that captured most cancer-associated methylation variability across multiple cancer types. These regions formed the basis for developing the EII.
“We identified specific genomic regions that tend to be the most variable in DNA methylation marks during cancer,” said first author Sara-Jayne Thursby, PhD, a postdoctoral researcher in the lab of Easwaran. “In cell-free DNA in the blood, that variability shouldn’t be high, but if it is, it is indicative of a developing cancerous phenotype.”
Using cell-free DNA from blood samples, the team applied machine learning classifiers trained on the EII metrics derived from the 269 CpG island regions. The data showed that in lung adenocarcinoma, the EII distinguished stage IA cancer from normal samples with approximately 81% sensitivity at 95% specificity. For early-stage breast cancer, the approach achieved 68% sensitivity and the same specificity.
“Quantifying epigenetic instability is a novel, capable approach to distinguishing cancer from normal cases using cfDNA, performing better than standard approaches using absolute methylation changes,” the researchers contend. The data also showed signals consistent for the early detection of colon, brain, pancreatic, and prostate cancers.
The researchers developed and refined the EII by comparing its performance against conventional methylation-based measures in both tissue-derived datasets and cfDNA datasets. They further tested the approach across different DNA methylation profiling technologies, including whole-genome bisulfite sequencing and targeted bisulfite sequencing. According to the researchers, the EII does not rely on identifying specific biomarker sites, which may not appear consistently across patients, but instead captures a broader signal of tumor-associated variability.
Easwaran said that the EII could be used clinically in conjunction with other current testing methods. “Another application we are pursuing is implementing these approaches along with other cancer screening strategies, such as colonoscopy (colon cancers), PSA (prostate), where these techniques can provide complementary information,” he said.
“For example, patients with elevated prostate-specific antigen levels might use an EII-based blood test to help determine whether follow-up biopsy is warranted.”
The team is now expanding and refining the method and plans to validate the approach in larger, long-term clinical cohorts. They are also exploring whether tissue-specific EII models could help identify the tissue of origin of cancer signals. The researchers wrote that “general models can be created for detection of DNA fragments originating from any type of cancerous cells,” supporting further development of epigenetic instability metrics for multi-cancer early detection.
