NEW YORK (GenomeWeb) – Recent studies have continued to explore how genomic technologies can detect cancer in blood samples without identifying specific mutations — by measuring things like the size of DNA fragments, their epigenetic modifications, and other features.
Results from new research suggest that such methods could either serve as an adjunct to tumor DNA mutation assays, making cancer detection easier and cheaper, or that they might be sensitive enough on their own to serve as standalone assays.
In one study, published last month in Science Translational Medicine, a European team led by investigators at Cancer Research UK and the University of Cambridge described a method they developed and tested that can distinguish the presence of cancer based on the size of DNA fragments in blood, as well as enrich for circulating tumor DNA for mutation analysis.
Inspired by the discovery over the last several years that fragments of circulating DNA from fetuses can be picked out from the blood of a pregnant woman based on shorter length, the group used whole-genome sequencing to survey the sizes of ctDNA fragments in 344 plasma samples from 200 cancer patients to find a threshold for tumor versus normal DNA.
Cell-free DNA fragmentation "is not random [but] is linked to the way DNA is released from the cells in circulation," Florent Mouliere, the study's first author — now a professor at the VU University Medical Center Amsterdam — explained in an email.
According to him and his coauthors, the study uncovered two things. The first is that pre-selecting fragments between 90 and 150 base pairs long can improve the detection of circulating tumor DNA in cancer patients by concentrating the DNA fragments that are most likely from a tumor and excluding the larger background of non-tumor material.
In the study, the team found that the number of cancer mutations they could detect increased by an average of 53 percent when they used size selection.
"We are working to integrate size selection into various analysis methods to improve sensitivity to detect ctDNA," Nitzan Rosenfeld, the study's senior author added in an email. This needs to be done in the right setting though, he said. "For example, if one is looking for particular mutations that are present at very few copies, any manipulation that results in loss of material, such as size selection, might also reduce sensitivity."
"We are therefore initially implementing this in conjunction with wider-scale sequencing approaches. For example … in analysis by whole exome sequencing when ctDNA levels are intermediate — too low to detect mutations without size selection, but high enough that mutations can be detected above background even if there is loss of material," he explained.
Beyond helping to improve mutation detection, the CRUK team also showed that genome-wide fragmentation can be used to detect the presence of cancer on its own: using machine learning to create a fragmentation-based classifier to segregate cancer and healthy individuals.
In the study, the researchers used logistic regression and random forest analysis, creating various models and defined size ranges to classify plasma samples as being derived either from healthy individuals or from patients with cancer.
One model performed especially well, correctly classifying samples derived from patients with cancer in between 65 and 94 percent of cases, depending on the levels of circulating tumor DNA present.
In another study, described last week in Nature Communications, investigators from the University of Queensland, Australia reported an approach they developed to characterize a genome-wide methylation landscape they call a "methylscape," which they hope can serve as a universal cancer biomarker.
The team collected evidence that DNA polymeric behaviour is strongly affected by this methylscape patterning, which translates to a difference in DNA solvation and gold affinity between cancerous and normal genomes. In their study, the investigators exploited these biological differences to create simple electrochemical one-step assays, which, with an analysis time under 10 minutes and minimal sample preparation, might offer a new way to detect tumors with low cost and effort.
So far, the group has tested the technology on about 200 samples across different types of human cancer alongside healthy controls, seeing accuracy as high as 90 percent. These have not been enriched for early-stage cancers, however.
"It's not perfect yet, but it's a promising start and will only get better with time," the authors wrote in a statement.
In an email, senior author Matt Trau, a professor of bioengineering at the University of Queensland, wrote that he and his colleagues have filed for IP protection on the electrochemical assay approach. "We are looking for collaborators who could share appropriate samples," he added, which will let the group begin to test their ability to pick up cases of cancer in their earliest stages.
The CRUK team that published its DNA fragment size approach has also not yet tested its method on early-stage samples, Rosenfeld wrote.
Moving forward, he said the group is planning to combine what they described with other, hopefully complementary methods "such as detection of point mutations or other biomarker[s] … so that a combined test can compensate for the limitations of each individual technique."
An open question as all of these techniques advance is whether they are truly assaying unique signals or picking up different readouts of what are the same epigenetic or structural features. DNA fragmentation, for example, is understood to be influenced by epigenetic factors.
In a commentary appearing alongside the CRUK paper in STM last month, Ellen Heitzer and Michael Speicher from the Medical University of Graz, Austria echoed this, highlighting the need for a better understanding of how various epigenetic mechanisms influence the difference in fragment size that is seen across DNA from different cellular origins.
Meanwhile, the research groups are not the only ones exploring genome-wide methylation or DNA fragmentation size.
Dennis Lo from the Chinese University of Hong Kong, for example, published a study in PNAS in October on a method to detect cancer by analyzing the end coordinates of circulating DNA fragments. In an email, he wrote that the CRUK report in STM is "along the same line" of what he and his team previously described, though it offers new evidence of the utility of fragment size analysis in a large sample cohort of multiple types of cancer.
"The essence [of all of this] is that circulating fetal … [or] tumor DNA molecules … exhibit a size profile which is slightly shorter than the background plasma DNA. Hence one can use size discrimination … to try to enhance the signal detection," he wrote.
Lo added that in addition to the recent PNAS study, he and his team have used the combination of DNA fragment sizing and counting in work to develop a test for nasopharyngeal carcinoma that relies on the detection of Epstein-Barr virus DNA, which, like circulating tumor DNA, presents as smaller fragments. That test is now being commercialized with US firm Grail.
Investigators from the University of Washington have also reported that they can determine the origin of circulating DNA based on an inference of nucleosome positioning.
Although not specifically looking at the size of molecules, that team's approach, which is being commercialized through a spinout, Bellwether Bio, also relies on an analysis of the fragmentation patterns of DNA molecules in blood.
Mouliere suggested that regardless of the specific target, strategies that don't rely on mutation detection in general clearly offer advantages. "Previous ctDNA-based approaches [have focused] on detecting mutations or copy number alterations, which is well adapted to a market where the main demand for ctDNA testing is for precision medicine," Mouliere wrote in his email.
"But the vast majority of ctDNA released by cancer cells is non-mutated," he added, so considering the difficulty of detecting rare mutations in early-stage cancer, tapping into this much more plentiful resource makes sense.