NEW YORK – Investigators are making progress advancing a genome-wide methylation profiling method designed to detect cancer-associated epigenetic signals without the need for ultra-deep sequencing or detection of specific mutations.
Two publications piloting different potential clinical applications appeared in Nature Medicine this week. In one, a team led by Dana Farber investigators Toni Choueiri and Matthew Freedman used the method, called cfMeDIP-seq, to try to distinguish samples from patients with kidney cancer, including early-stage disease, from normal controls, using both blood and urine.
According to the authors, the methylation sequencing approach was nearly 100 percent accurate in their small case-control cohort when used on blood samples. Although detection dropped in urine, they believe that technical improvements can improve the method's performance in this even less invasive sample type.
In a second study, researchers led by a group at Toronto's Princess Margaret Cancer Centre used the same sequencing method to glean methylation signatures that both detect intracranial tumors and can help discriminate among different cell-of-origin subtypes that can be difficult to distinguish using standard-of-care imaging.
CfMeDIP-seq (cell-free methylated DNA immunoprecipitation and high-throughput sequencing) was developed by a team led by Daniel De Carvalho — also a co-author on both new studies this week.
Epigenetic methods have increasingly come to dominate among genomic approaches to non-invasive early cancer detection test development. Many techniques being advanced use bisulfite conversion techniques to analyze methylation patterns across the genome. Whole-genome screens can be used first to generate specific signatures that differentiate cancers from non-cancers.
Unlike these, cfMeDIP-seq uses immunoprecipitation to enrich a sample for CpG-rich DNA fragments, allowing genome-wide methylation profiling without the need to predefine a specific signature. By pulling out the fragments most likely to harbor informative methylated sites, it avoids wasting sequencing reads unnecessarily in areas of the genome unlikely to be epigenetically informative.
De Carvalho discussed his team's ongoing work to optimize and validate the technique during the American Association for Cancer Research's annual meeting this week, where he said that one big improvement has been the development of a set of 52 synthetic spike-in fragments that are added to a sample before immunoprecipitation and sequencing.
These allow for an added ability to adjust for both biological and technical bias, and for absolute quantification of circulating tumor DNA.
According to De Carvalho, the new application of cfMeDIP-seq to brain cancers, which he and colleagues reported in Nature Medicine this week, is a logical one. Other research has already established that DNA methylation in tumor tissue can help oncologists differentiate types of brain tumors that might be otherwise obscure.
Because obtaining tissue samples from the brain is difficult and dangerous, a blood-based method would be ideal, but because so little tumor DNA makes it into the blood from the brain, a liquid biopsy technique would have to be extremely sensitive.
In their study, De Carvalho and his coauthors analyzed samples from a mixed cohort of patients with several different types of intracranial tumors.
The authors reported that cfMeDIP-seq, with its genome-wide scope, had the sensitivity to distinguish different brain tumors from one another — for example, differentiating gliomas from extracranial cancers that had metastasized to the brain, or identifying different primary brain tumors that might otherwise be indistinguishable using standard-of-care imaging.
Authors argued that having a blood-based method for diagnosis and subtyping would remove the need for risky invasive biopsy for tumors that don't require surgery for treatment or which present too high a risk.
Even for patients that would benefit from, and are slated to receive surgery, having pre-knowledge about a tumor's biology could help improve surgery planning and safety, the authors argued, citing the example of hemangiopericytomas that "are almost indistinguishable on MRI from meningiomas, but pose a risk for excessive blood loss if the surgeon is not prepared."
In the Dana-Farber team's kidney cancer study, meanwhile, researchers used cfMeDIP-seq to profile samples from 99 patients with early and advanced kidney cancer, 15 patients with stage IV urothelial bladder cancer, and 28 healthy, cancer-free control subjects.
According to the authors, kidney cancer, more specifically renal cell carcinoma, is associated with significant mortality in the US and has no clinically validated non-invasive detection biomarkers other than imaging used in screening for those with known hereditary syndromes.
"An accurate, highly sensitive and specific noninvasive test, alone or in combination with imaging, could transform clinical management by enabling early detection … and reducing unnecessary kidney biopsies and nephrectomies," the team wrote.
Matthew Freedman, MD, a medical oncologist at Dana-Farber and co-senior author of the report, said in an interview this week that early cancer detection methods have exploded over the last few years, but that the approach developed by De Carvalho and colleagues offers what he and his colleagues view as a particularly promising take.
"I've always thought that our biggest impact on mortality in cancer was going to come through early detection," Freedman said. "So, we had been thinking about getting into this space, but because there is so much going on, we weren't sure where to invest our time and resources. But when I saw [the early data on cfMeDIP-seq], I was convinced."
When he and his colleagues used cfMeDIP-seq on blood serum samples from their kidney cancer cohort, they saw "near-perfect" classification of patients across all stages, Freedman and his coauthors wrote in the study this week.
They calculated an area under the receiver operating curve of 0.99 for methylation-based discrimination of kidney cancers versus normal controls and 0.98 for distinguishing renal cancers from urothelial bladder cancer cases.
In urine, performance was a bit lower, with an AUC of 0.858 for distinguishing kidney cancer cases from controls, but the group argued that because the data were generated using a protocol that had been optimized from plasma, they should be able to up that performance via technical and computational optimization. This could include "size selection of cfDNA to enrich for tumor-derived DNA and utilizing tumor methylation data to inform cfDNA methylation analysis." The team has applied for a patent specific to the urine-based application.
According to Freedman, studies to better optimize urine performance are underway, as well as work to make sure that the approach won't be confounded by other disease states. "For this [paper], we had people without kidney disease that we know had no evidence of kidney cancer in a very, very stringent way. But we also want to look at inflammatory kidney diseases, and we want to make sure that this test is as specific as possible," he said.
"We're also working on doing things in a prospective manner … where blood and plasma were obtained prior to the clinical diagnosis, Freedman added. "We're working on really all of those fronts now — now that we know that that test works and that it works well."
During the AACR meeting, De Carvalho said that he and his team have also applied cfMeDIP-seq to longitudinal cancer monitoring and minimal residual disease detection in early-stage patients. The team hasn't yet published this work, but they have seen success in both minimal residual disease detection and disease recurrence surveillance in a study of localized head and neck cancer patients, he said during his presentation.
Using both their methylation method, and mutation profiling via CAPP-seq, the group found that while mutation-based sensitivity for MRD dropped off at low ctDNA fractions, cfMeDIP-seq was able to maintain a signal.
By developing a cutoff point for methylation-based MRD, the group could discriminate patients with better and worse overall survival.
Importantly, De Carvalho said, cfMeDIP-seq offers a tumor-naive option for MRD assessment, unlike many of the approaches already being commercialized in the space, which require up-front tumor tissue sequencing to define a patient-specific panel of mutations that is then tracked in blood.
Beyond this, the group has also begun to explore their epigenetic approach for discriminating localized versus advanced disease and predicting survival in prostate cancer.
De Carvalho said the team also is working with a Canada-wide consortium called CHARM — Circulating tumor DNA in Hereditary and High-risk Malignancies, in which it is testing samples from individuals with known inherited cancer risk genes, comparing cfMeDIP-seq with mutation-based cancer detection to see if developing tumors can be detected (and treated) early in this population.
So far, the consortium has banked 810 samples from 641 patients, he said.