NEW YORK – A new low-cost method for profiling genome-wide methylation in cell-free DNA (cfDNA) accurately detected the presence and tissue of origin of early cancer in multiple cancer contexts, in a study published today by a team led by University of California, Los Angeles researchers.
The method, called cfMethyl-Seq and published in Nature Communications, consists broadly of creating a sequencing library of only CpG sites, preventing enzymatic digestion of those cfDNA fragments, and applying a computational method for extracting cancer-specific cfDNA methylation features. Respectively, these three components of cfMethyl-Seq reduce the cost of library preparation, increase the cfDNA signal, and identify the cancer's tissue of origin.
CpG sites are methylation-amenable regions of DNA where a cytosine nucleotide is followed by a guanine. When these so-called CpG islands occur in regulatory regions of DNA such as promoters, methylation at those sites silences the linked gene. While this is crucial to development and to tissue-specific biological processes, it often occurs aberrantly in cancer.
Library preparation in cfMethyl-Seq involves blocking both ends of all cfDNA fragments in a sample in order to prevent enzymatic digestion by the MspI enzyme, which degrades all fragments lacking at least two CCGG sites. Remaining fragments are ligated to adapters containing duplex unique molecular identifiers to avoid the occurrence of multiple fragments with the same start and end sites — a known by-product of enzymatic digestion.
Narrowing the sequencing space to CpG islands reduces the overall cost of sequencing, as these regions account for only approximately 3 percent of the human genome. In the current study, the UCLA team found their sequencing costs reduced by approximately a factor of 12.
The researchers applied this technique to 217 patients with colon, liver, lung, or stomach cancer, and 191 individuals without cancer but which included patients with other diseases.
Tests of cfMethyl-Seq in this population revealed an overall sensitivity of 80.7 percent and a specificity of 97.9 percent, with sensitivities for individual cancer types ranging between 75.9 percent and 92.3 percent. The method also accurately predicted tissue of origin in 89.1 percent of cases.
A key finding of the study was that different types of methylation appeared to correlate with distinct cancer features.
Specifically, the UCLA team identified distinct methylation patterns between cancer and non-cancer tissue and found that cancer-specific hypermethylation provided the strongest signal in early detection, while cancer-specific hypomethylation proved most useful in determining tissue of origin.
The team compared cfMethyl-Seq to reduced representation bisulfite sequencing (RRBS), another technique used to enrich CpG-rich regions. In contrast to cfMethyl-Seq, RRBS can only enrich CpG-rich regions from intact genomic regions of DNA, rather than from cfDNA.
"RRBS is not applicable to cell-free DNA because it's based on cutting and then site selection," said Jasmine Zhou, a professor of pathology and laboratory medicine at UCLA and the study's senior author.
RRBS also lacks the fragment blocking step found in cfMethyl-Seq, resulting in a less efficient library preparation.
Although cfMethyl-Seq can, in principle, get similar results to RRBS and other methylome sequencing techniques such as cfMeDIP, Ben Berman, a professor of computational biology specializing in DNA methylation at the Hebrew University of Jerusalem, cautioned that the new method may not be ideal for investigating all cancer methylation questions.
"One drawback of this technique compared to a capture-based or whole-genome approach," he said, "is that you lose the fragmentation information — fragment length, fragment end motifs, et cetera. It has been demonstrated that methylation and fragmentomic information can provide complementary cancer information."
Billy Lau, a researcher at Stanford University who specializes in DNA methylation sequencing, called cfMethyl-Seq a "nice method" that had been validated with a decent cohort size.
"While a lot of other players have moved onto hybridization capture, using a restriction enzyme is a low-cost method of reducing the search space for finding relevant methylation biomarkers," he said. "There's a significant cost to developing and using targeted panels, so this is a nice study that shows you don't need to be too precise about what exactly you're targeting to get a good signal."
In addition to her UCLA position, Zhou is the founder and CEO of Early Diagnostics, a company spun out in 2017 from UCLA that hopes to eventually offer cfMethyl-Seq as a lab-developed test. For that, she said, the company is seeking funding, both to conduct larger studies and to establish a CLIA laboratory for their eventual LDT offering.
At the moment, cfMethyl-Seq is available through EarlyDx for research purposes only, as part of the company's CancerRadar, a service that couples cfMethyl-Seq to computing algorithms for early cancer detection.
EarlyDx recently won a $2 million Phase 2 Small Business Innovation Research grant from the National Institutes of Health, and, said Zhou, "we're trying to go through a Series A financing [round] try to conduct a new trial. That's our major goal right now."
EarlyDx also currently offers a machine learning and cloud-based bioinformatics platform called EarlyDx Cloud, which supports researchers in analyzing cfDNA-based liquid biopsy results for cancer detection, treatment monitoring, and therapy selection.