Catching cancer: could a new blood test rewrite oncology?

 Early identification of cancer is crucial in oncology, which means that any device capable of catching cancer cells sooner is a welcome development. Now, a team in China claims to have developed a blood test that can pick up cancer up to four years before symptoms appear. But can this game-changing technology live up to the hype, or is PanSeer too good to be true? Abi Millar finds out.


n July, a team of researchers in China piqued the attention of the oncology community worldwide. They had developed a blood test, PanSeer, which they claimed could detect five common types of cancer up to four years before symptoms emerged. Publishing their results in Nature, the researchers said the test could detect cancer in up to 95% of asymptomatic patients who later receive a diagnosis.

The test, developed by Singlera Genomics, uses DNA methylation techniques along with a machine learning algorithm to pick up on early signs of cancer. While this is far from being the only liquid biopsy in development, it is one of the first to detect cancer even before the patient shows symptoms.

“Over the long term, this could be something that is offered at a yearly check-up at the doctors’ office,” says Professor Ken Zhang of The University of California San Diego, who co-led the study. “It might become a first-line screening approach that can be deployed in the general population, starting with high-risk age groups or those with family history, before being expanded to the broader community.”

How the test works

The test works by screening particular regions of DNA found in blood plasma. Unlike many blood tests for cancer, which search for circulating tumour cells or circulating cell-free tumour DNA, PanSeer looks for methyl groups – a telltale sign of tumour DNA.

“DNA methylation is what we call epigenetic information,” says Zhang. “I use the analogy that, if you have a building, you can change the furniture in a room even though you don’t change its structure. In this analogy, the DNA is the building but the methylation is the furniture. When normal cells become cancer cells, they acquire new behaviour and the room acquires different furniture. So, if you know where to look, you can identify these changes.”

“PanSeer looks for methyl groups – a telltale sign of tumour DNA.”

These furniture re-fittings, so to speak, can happen at many different locations across the human genome. Here, the researchers selected 595 regions they knew to be particularly susceptible.

“We selected these regions based on prior knowledge that some of these regions might have abnormal patterns in one cancer type or another,” says Zhang.

“Our machine learning technique takes all the information we collected from these 595 regions, tries to integrate this information and come up with one single score between 0 and 1. This allows us to determine whether a person has cancer or not.”

The study design

The study used a longitudinal cohort – blood plasma samples collected in China between 2007 and 2014 as part of a wider research effort. The research in question, the Taizhou Longitudinal Study (TLS), was launched in 2007 by Zhang’s collaborator Professor Li Jin at Fudan University.

It involved over 120,000 healthy individuals aged between 25 and 90, and followed them for a decade via local health registries and repeated blood draws. Overall, the researchers collected and archived over 1.6 million samples.

In the PanSeer study, researchers located plasma samples from 191 asymptomatic patients who were diagnosed with stomach, oesophagus, colorectal, lung or liver cancer within four years of the blood draw. They also found samples from 414 patients who remained healthy throughout the time frame, and 223 samples (from biobanks) from patients who had already been diagnosed with one of the cancer types.

“[The algorithm] flagged cancer in 95% of the participants who had no symptoms, but went on to be diagnosed.”

Around half the samples were then fed into the algorithm, which learnt to identify the epigenetic differences between those who had cancer and those who did not. When the algorithm was applied to remaining samples, it was able to identify cancer with startling accuracy. It flagged cancer in 95% of the participants who had no symptoms, but went on to be diagnosed. It also identified 88% of the participants who already had cancer, and 96% of those who did not.

“Using this design, we try to prove that you can detect cancers years before people show up in the hospital,” says Zhang. “If we can do this over again, there’s a possibility we can save people’s lives or at least improve their life quality.”

As the paper emphasises, PanSeer doesn’t appear to be predicting patients who will later develop cancer. Instead, it works by identifying those who already have early stage cancer that existing methods can’t detect.

Limitations of the study

While the study has been broadly welcomed, it has a few limitations. The paper mentions concerns around sample contamination, and the sample size was relatively small. On top of that, the test can’t detect what type of cancer a person has.

“This DNA methylation pattern is a fundamental property of cancer cells, and the 595 regions we picked were common for all five cancer types,” says Zhang. “If we were to expand this we’d need to add additional locations of the genome where we have prior knowledge about tissue and organ specificity.”

“This DNA methylation pattern is a fundamental property of cancer cells.”

This isn’t beyond the realm of possibility. In previous studies, DNA methylation has been used to determine the type of cancer someone has. However, these studies have required a far larger panel of genomic regions, along with more than one blood vial from the patient. The PanSeer test has the advantage of being cheap and requiring only one draw of blood.

The researchers therefore envision using the test as a first-line screen. Any patient testing positive could then be followed up with a more expensive blood test or a CT scan.

“A full body CT scan would be able to easily identify the location of a lesion,” says Zhang. “That’s what you need anyway, because once you know a person has cancer the next obvious step would be a surgical procedure. You have to know the location or you won’t know where to cut.”

What comes next?

The next logical phase for the research would be a prospective cohort study, which follows up a large group of people in real time. While this kind of study is expensive, resource-intensive and lengthy, Singlera Genomics is in the process of raising funds for a clinical trial.

“The second direction we want to go in is to further improve this test – can we detect ten or 20 cancers and localise where the tumour DNA comes from?” says Zhang. “That means we need to expand the content of our assay to look at, say, 2,000 regions, and we might need to add the regions that can tell you whether it’s lung, liver, brain, etc.”

“Screening is very effective in managing cancer, because if you can detect cancer in the early stages.”

At present, it’s hard to say how long these efforts might take; it could be in the order of a decade or more. However, the rationale for conducting the research is clear – when cancer is detected at an earlier stage, we see a dramatic improvement in the survival rates.

“Screening is very effective in managing cancer, because if you can detect cancer in the early stages, then surgical procedures followed by chemotherapy will be effective,” says Zhang.

“For example, in breast cancer patients the five-year survival is about 90% just because we have mammograms and other screening approaches that allow us to detect cancers when they’re local. If we can do similarly early detection for other cancers then it’s a complete game-changer.”