Skip to main content
Premium Trial:

Request an Annual Quote

Whole-Genome Liquid Biopsy Method Discerns Cancer-Associated Mutational Signatures


NEW YORK – Results of a new study led by researchers at Memorial Sloan Kettering Cancer Center have demonstrated that low-coverage whole-genome sequencing (WGS) of circulating cell-free DNA (cfDNA) can identify mutational signatures associated with the development of cancer and potentially other diseases. Investigators hope to study whether the method could aid early-detection efforts or unlock new opportunities to assess risk and intervene earlier in the disease process.

The researchers published their method, called "Pointy," last week in Nature Communications, and showed proof of concept for using the approach to differentiate cancer patients from healthy controls with high sensitivity and specificity.

Current methods for blood-based cancer early detection incorporate things like point mutations, copy number alterations, and epigenetic signals like methylation and DNA fragmentation patterns. But mutational signatures could also play a role.

"These [signatures] can tell you a lot about either the endogenous or exogenous process that is forcing selective pressure," said Luis Diaz, the report's senior author and a medical oncologist at MSKCC. "For example, tobacco use is well represented in mutational signatures, as is UV light exposure, as are other carcinogens, things like mismatch repair, or the simple fact of aging."

Diaz added that his team wanted to use liquid biopsy "to see if we can get a snapshot of what a person has been exposed to during their lifetime … [or] a snapshot of the tumor bulk that's evolving in the case of cancer."

These broad patterns of somatic change, not limited to specific known point mutations, have been discerned using tissue sequencing in prior studies, but making this work with liquid biopsy posed a challenge for Diaz's team.

"Anyone can take whole-genome sequencing data and look at it, but the problem is that a lot of that is errors or artifacts," he said.

To overcome this issue, the study's first author Jonathan Wan tailored a method combining signature extraction and machine learning with a variety of tools to mitigate technical and biological noise.

The group then compared data from 215 cancer patients and 227 healthy individuals whose cfDNA samples were sequenced at between 0.3 and 1.5X coverage, using the Pointy approach to look for both pathological and physiological mutation signatures.

In an initial cohort of colorectal cancer patients, the team saw signatures indicative of the aging process and of microsatellite instability, which is known to be a significant driver in these tumors. Both categories of signature correlated strongly with both circulating tumor DNA fraction and tumor mutational burden.

Using machine learning, they trained a classifier to discern cases as either cancers or controls based on the presence and contribution of specific signatures and then expanded to a multi-cancer cohort to test the method across multiple tumor types. These included 37 non-small cell lung cancers, 48 breast cancers, 27 colorectal cancers, 27 gastric cancers, 26 ovarian cancers, 34 pancreatic cancers, and 206 non-cancer controls.

Across the cohort, the proportion of patients with at least one cancer-associated signature detected ranged from 85 percent in NSCLC to 38 percent in pancreatic cancer. By stage, the rate of detection of at least one signature ranged from 70 percent in stage I disease to 75 percent in stage IV.

Overall, the authors calculated an area under the receiver operating curve of 0.96 for detection of cancer versus healthy samples across all tumor types. Notably, detection rates were similarly high in all tumor stages.

Diaz and his coauthors wrote in the study that this first stab at assessing mutational signatures in cfDNA had notable limitations. In the future, the group is hoping to see if sequencing more deeply, and using matched germline analysis to filter the data, can allow for even more accurate characterization.

According to Diaz, the researchers were most excited about applying the tool to early cancer detection, but a number of other exciting possibilities have popped up including the opportunity to study a larger group of healthy individuals to hopefully discover more about the evolution of cancer risk and the process of oncogenesis.

"Every project has a moment where you kind of see beyond the horizon," Diaz said. "At first we were looking at this as another feature set to help detect cancer early … [but] that's not what gets me excited. I came from another angle, that this is going to allow us to see risk."

This could include establishing signature cutoffs that indicate significant versus mild risk. "I might be able to test someone's blood and say, listen, you've been exposed to enough tobacco, smoke, or radiation, or UV light that you're at a higher risk for developing the cancer," Diaz said.

Alternatively, researchers could begin to understand more about mechanisms of resilience against accumulation mutations. "You could have smoked your whole life, and for some reason you don't have a mutation signature for tobacco. Why is that? That means there's something interesting about the DNA repair mechanism in that person, so it kind of changes the playing field," he added.

Another exciting question is whether the same might be possible in other disease contexts. For example, Wan, the study's first author is currently working to apply the same approach in brain diseases.

For early detection, meanwhile, Diaz and his coauthors wrote that future work could also seek to combine mutational signatures with other parameters like DNA fragmentation and methylation into a larger machine learning ensemble.

"DNA is the most stable. Epigenetics is stable but less stable. Transcription and expression are harder to capture, and protein, we haven't found a really good way to do that yet," Diaz said. "I think that obviously if you can collapse all these together, you get the best picture possible. But we're still kind of in that messy middle ground right now."

As far as commercial prospects go, the multi-cancer cohort used in the team's study came from Johns Hopkins spinout Delfi Diagnostics, which is advancing a fragmentomics-based cancer early detection platform.

Delfi hasn't made any claim on the Pointy methodology, but since the company's approach already involves low-pass WGS, it would theoretically be easy to add mutational signature analysis, Diaz said.