CancerSEEK: Blood Test That Detects Eight Common Cancers

Alexander M. Castellino, PhD

January 18, 2018

A new blood test, dubbed CancerSEEK, can identify eight common cancers ― ovary, liver, esophagus, pancreas, stomach, colorectal, lung, and breast cancer ― say researchers from the Johns Hopkins Kimmel Cancer Center in Baltimore, Maryland.

The new test measures circulating tumor DNA (ctDNA) from 16 genes as well as eight protein biomarkers, and then uses machine-based learning to analyze the data.

"The use of a combination of selected biomarkers for early detection has the potential to change the way we screen for cancer, and it is based on the same rationale for using combinations of drugs to treat cancers," said Nickolas Papadopoulos, PhD, senior author and professor of oncology and pathology, in a statement.

In the future, the blood test could be used to detect cancers "early enough that they could be cured by surgery alone, but even cancers that are not curable by surgery alone will respond better to systemic therapies when there is less advanced disease," said one of the corresponding authors, Anne Marie Lennon, MD, PhD, professor of medicine, surgery, and radiology and clinical director of gastroenterology. "Earlier detection provides many ways to improve outcomes for patients," she added.

The new test is described in a report published online January 18 in Science.

It is still early days for the test. "To actually establish the clinical utility of CancerSEEK and to demonstrate that it can save lives, prospective studies of all incident cancer types in a large population will be required," the researchers conclude.

Nevertheless, they believe that this study "lays the conceptual and practical foundation for a single, multi-analyte blood test for cancers of many types."

The researchers also estimate that the cost of this single test will be less than $500. "This is very much an estimate of what we think it will likely cost if it was implemented as a widely used screening test," Dr Lennon told Medscape Medical News.

If this does turn out to be the case, this blood test would be considerably less expensive and more convenient that current screening tests, such as colonoscopy and mammography.

Novel Approach: Looking at Both Mutations and Proteins

"A novelty of our classification method is that it combines the possibility of observing various DNA mutations together with the levels of several proteins in order to make the final call," said co-corresponding author Cristian Tomasetti, PhD, associate professor of oncology and biostatistics, who developed the algorithm, in a release.

From initially looking at several hundred genes and 41 protein biomarkers that might detect cancers prior to metastasis, the Hopkins researchers settled on interrogating segments of 16 genes and analyzing levels of eight protein biomarkers.

The eight proteins are cancer antigen 125, carcinoembryonic antigen, cancer antigen 19-9, prolactin, hepatocyte growth factor, osteopontin, myeloperoxidase, and tissue inhibitor of metalloproteinases 1.


The first component of CancerSEEK determines mutations in ctDNA obtained from the blood sample. With purified DNA from plasma, the researchers amplified the sample with multiplex–polymerase chain reaction using a robust 61-amplicon panel that had the ability to detect rare mutations expected in plasma ctDNA.

The 61 primer pairs were designed to amplify 66 to 80 base-pair segments of the DNA in regions of interest from 16 genes. The amplified products were uniquely labeled with a DNA barcode and were matched to reference sequences present in the Catalog of Somatic Mutations in Cancer dataset or ones that were predicted to be inactivating mutations in tumor suppressor cells.

"Keeping the mutational panel small is essential to minimize false positive results to keep such tests affordable," said Joshua Cohen, an MD-PhD student at the Johns Hopkins University School of Medicine, who is the article's first author.

The second component of CancerSEEK is determining levels of eight cancer protein biomarkers. According to the researchers, the reason for incorporating this component into the mix is because "early-stage tumors do not release detectable amounts of ctDNA, even when the most sensitive techniques are used to identify them."

The eight proteins selected could be reproducibly evaluated through a single immunoassay platform.

Details of the Study

For the study, the researchers recruited 1005 patients with stage I to III cancers of the ovary, liver, esophagus, pancreas, stomach, colorectum, lung, or breast who had not yet undergone surgery. The median age of the patients at diagnosis was 64 years. The most common stage at presentation was stage II (49%); 20% of patients had stage I disease; and 31% of patients had stage III disease.

The study also included 812 healthy individuals (median age, 55 years) who had no history of cancer. These individuals served as a control population.

Patients were excluded if they had undergone prior neoadjuvant therapy, were determined to have stage IV disease after resection, or if their blood had been drawn while anesthesia was being administered.

Peripheral blood was collected from the cancer patients before they underwent surgical resection.

The CancerSEEK blood test was used to detect the presence of mutations in 2001 genomic positions of the 16 genes. Luminex bead-based immunoassays were performed using panels that determined the levels of 39 proteins, including the eight protein biomarkers.

A patient's test result was classified as positive if the frequency of a mutation in one of the 16 genes or the levels in one of the eight proteins, or their combination, was significantly elevated with respect to the control population.

The algorithm then used machine-learning tools and statistical analyses to determine the median sensitivity and specificity through 10 iterations of 10-fold cross-validations.

"One of the most important attributes of a screening test is the ability to detect cancers at an early stage," the researchers note.

The median sensitivity of CancerSEEK was 73% for stage II cancers, 78% for stage III cancers, and a low 43% for stage I cancers.

For stage I cancers, the highest sensitivity was for liver cancer (100%), and the lowest was for esophageal cancer (20%).

The median overall sensitivity — the ability to find cancer — was 70% for the eight common cancers, which account for more than 60% of cancer deaths in the United States. The sensitivities ranged from 98% for ovarian cancer to a low 33% for breast cancer.

For the cancers for which there are currently no screening tests available — ovary, liver, stomach, pancreas, and esophagus cancers — the sensitivities ranged from 69% to 98%.

The specificity of CancerSEEK was >99%, with a positive score found in only 7 of the 812 healthy individuals who acted as the control population.

Dr Lennon told Medscape Medical News that for the seven healthy individuals who tested positive, those results were presumed to be false positives. "However, it is feasible that they could have an undiagnosed cancer and were asymptomatic. Unfortunately, we don't have any follow-up on them," she said.

Determining Tissue of Origin

To determine the tissue of origin in patients with a positive CancerSEEK test, the researchers devised an algorithm that took into account the ctDNA, the level of any of the eight protein biomarkers as well as the 31 other proteins determined by the panels in the immunoassay, and the sex of the patient. For this analysis, esophageal and gastric cancer patients were grouped together, because endoscopy was the optimum follow-up in both cases.

Using this algorithm for 626 cancer patients who scored positive on CancerSEEK, and without using any other information, the researchers were able to localize the origin to two anatomic locations in a median of 83% of patients and to a single organ in a median of 63%.

"Given that driver gene mutations are usually not tissue-specific, the vast majority of the localization information was derived from protein markers," the researchers write. The accuracy was highest for colorectal cancer (84%) and lowest for lung cancer (39%).

"Such multi-analyte tests would not be meant to replace other non-blood based screening tests, such as those for breast or colorectal cancers, but to provide additional information that could help identify those patients most likely to harbor a malignancy," the researchers note.

Study Limitations

A key limitation of the study was that patients with known cancers were enrolled.

"Though none of our patients had clinically evident metastatic disease at the time of study entry, most individuals in a true screening setting would have less advanced disease and the sensitivity of detection is likely to be less than reported here," the researchers acknowledge.

Experts not involved in the research agree. In an accompanying Science news article, proteomics researcher Lance Liotta, from George Mason University in Manassas, Virginia, points out that cancer-related proteins used by the test reflect tissue damage and can also appear in people with inflammatory diseases such as arthritis. "That means the 1% false positive rate will likely be higher in less healthy populations," he notes.

Another cancer researcher quoted in the same article, Alberto Bardelli, from the University of Turin in Italy, also homes in on this point. The test picked up 43% of cancers in stage I patients, so is it likely to work in a general population who shed less DNA? "We are not there yet," he said.

"Our vision is that the test would be used in asymptomatic individuals to screen for cancer, similar to what we're currently doing with mammography or colonoscopy," Dr Lennon told Medscape Medical News.

The Johns Hopkins team has already started to use the test for this. Working in collaboration with the Geisinger Health System in Pennsylvania, they have already begun to use CancerSEEK on blood samples from female volunteers aged between 65 and 75 years who have never had cancer. This 5-year study of up to 50,000 women, estimated to cost $50 million, is being funded by a private philanthropic group, The Marcus Foundation.

However, cancer experts feel the test is not yet ready for the clinic.

In an interview with Medscape Medical News, Daniel Hayes, MD, Stuart B. Padnos Professor of Breast Cancer Research, University of Michigan Comprehensive Cancer Center, described the study as "a good first step."

This test "has potential to be an assay for screening cancers in the future," he commented, but "generating data and showing clinical utility for cancer screening generally is a huge undertaking with many pitfalls, especially since both high sensitivities and specificities are important."

"In screening tests, specificity is especially important — specificity for cancer versus non-cancer and specificity for one tissue versus another. The study suggests that the test can do both," he said.  

Some caveats, also acknowledged by the study authors, need to be highlighted, Dr Hayes said. "This is a hand-picked cancer population and even then the sensitivity is not as high as one would like. The control is a group of healthy adults. We know that the general population who is a target for screening is not healthy with respect to benign conditions. Inflammatory and benign conditions can at the least make serum proteins go up and down, and we really don't know about circulating cell free DNA. How will the test fare in the general population?"

Another limitation, also acknowledged by the investigators, is the use of a single dataset, which was also used for an internal validation. We have to view this concern with a healthy skepticism, since what worked in one dataset may or may not work in another validation population. The test needs to be confirmed in a different dataset, at the least.

Even though this is not a randomly selected dataset, it is not a fatal error in regard to generating preliminary data, Dr Hayes said.

"However, that being said, this test cannot be applied as yet to the general population. The investigators have work to do. Even if they get validation of their results in a second existing data/specimen set, they probably need to do a prospective study of walk-in-the-door population of individuals, and importantly generate data that finding these cancers improves survival," he commented.  

Other reactions to the news, from cancer experts in the United Kingdom, can be found on the Science Media Centre website. They include the two opinions reproduced below.  

Nicholas Turner, PhD, professor of molecular oncology at the Institute of Cancer Research, London, and consultant medical oncologist at The Royal Marsden NHS Foundation Trust, said: "This paper is a step along the way to a possible blood test to screen for cancer, and the data presented is convincing from a technical perspective on the blood test. Potentially we can use a blood test like this to screen for undiagnosed cancers in asymptomatic people, and identify cancers at a point where they are more curable.

"However, the study does not support use of the blood test outside research studies, as it has not shown yet whether the blood test has the characteristics required for population screening, nor whether the blood test will improve outcome. There was a 1% 'false positive' rate with the blood test. Although it is unknown if these were really false positives, perhaps those people had undiagnosed cancer, a false positive rate of 1% could be quite a concern for population screening. There could be a lot of people who are told they have cancer, who may not have it. We can only learn about how much this is, or is not, a problem through larger studies. Much larger studies will also be needed to address whether using these blood tests can help improve survival rates.

"We need larger population studies to assess blood tests like this, and it is important that we wait for the results of these studies before people consider having the tests."

Paul Pharoah, PhD, professor of cancer epidemiology at the University of Cambridge, said: "There has been a lot of interest in using so called liquid biopsies as a relatively non-invasive way to detect cancer. The DNA from cancer cells is mutated and this mutated DNA can get into the blood stream (circulating tumour DNA, ctDNA) and be detected by sequencing the DNA in blood. There have been many studies demonstrating that this can detect cancers at the time of diagnosis or can be used to detect cancer relapses. However, there is a question about whether or not such a test will be able to detect cancer before it becomes symptomatic (i.e. at an early stage) and therefore be useful as a cancer screening test.

"The use of protein biomarkers for detecting cancer has been around for many years. While there are many that have proven useful in the diagnosis of symptomatic cancer, none have been shown to be useful for detecting cancer in people without symptoms (i.e as a screening test), with perhaps the exception of PSA for detecting prostate cancer.

"This group has developed a test that has combined ctDNA analysis with an analysis of protein biomarkers as a method to detect cancers. The authors used this new, combination test on blood samples from just over 1,000 clinically diagnosed patients with one of eight cancers — ovary, liver, stomach, pancreas, oesophagus, colorectal, lung, or breast  — and found that the cancer was successfully identified in about 70% of cases with false positive tests in 1% of individuals without cancer (specificity of 99%).

"This is a reasonable sensitivity and specificity, but there are some major caveats. Firstly, 80% of the cancer evaluated were stage 2 or stage 3 cancers (fairly advanced). Demonstrating that a test can detect advanced cancers does not mean that the test will be useful in detecting early stage symptomatic cancer, much less pre-symptomatic cancer. The sensitivity for the stage 1 cancers in the study was only 40%.

"Second, the study is somewhat unclear about how the relevant cut-offs for the test were identified in order to determine a positive and negative test. It seems that the results from some of the cases and controls evaluated in this study were used for that purpose. If so, the results are bound to overestimate the effectiveness of the test and an independent validation data set would be needed to demonstrate the real test performance.

"Third, the contribution of the protein biomarker component of the test to the sensitivity and specificity is not described. Given that these protein biomarkers have not, in general, proved useful for early detection, if their contribution to the performance of this test is substantial, it would question the likely utility of the test in clinical practice.

"Before seeing the results of this test, I regarded liquid biopsy as a potentially useful approach for the early detection of cancer. Having seen these results I still regard liquid biopsy as a potentially useful approach for the early detection of cancer. In other words, I do not think that this new test has really moved the field of early detection very far forward — it remains a promising, but yet to be proven technology."

The original article contains a full listing of the investigators' relevant financial relationships.

Science. Published online January 18, 2018. Abstract


Comments on Medscape are moderated and should be professional in tone and on topic. You must declare any conflicts of interest related to your comments and responses. Please see our Commenting Guide for further information. We reserve the right to remove posts at our sole discretion.