NEW YORK – Investigators at Oxford University have developed a multiomic blood test for the detection of multiple cancers, delving into a crowding pool of methods aiming to bring liquid biopsy into the cancer screening market.
Named TriOx, the team's test analyzes multiple features of DNA in the blood to identify fragments that originated in a cancerous tumor. In a study published in January in Nature Communications, researchers demonstrated that the assay could detect tumors, including early-stage ones, across six cancer types, with more than 90 percent sensitivity.
The methodology behind TriOx, which includes sequencing of DNA methylation patterns, shares some characteristics with other platforms being advanced by companies like Grail or Guardant Health and is based on a technique called TAPS (TET-assisted pyridine borane sequencing), previously developed by the Oxford group and currently owned by Exact Sciences through its acquisition of Base Genomics, an Oxford spinout.
Unlike the bisulfite sequencing commonly used for methylation analysis, which alters the underlying DNA structure, TAPS preserves DNA sequence data, allowing for analysis of methylation, gene copy number, and mutations in tandem.
Anna Schuh, one of the study's lead authors and a professor of molecular diagnostics at Oxford, said that TAPS offers a "gentle" approach, which involves enzymatic oxidization of methylated cytosines into a form called 5-carboxylcytosine. That compound is then reduced into another chemical form known as dihydrouracil, which is read as a thymine during sequencing based on its similarity. These thymine reads indicate the presence of a methylated cytosine when compared to the reference genome. Unmethylated cytosines are preserved, and thus DNA copy number and mutations can still be detected.
In contrast, bisulfite sequencing changes 95 percent of unmethylated cytosine to thymine, altering the genetic code so significantly that analysis of other DNA features requires a second, untreated sample.
In developing the TriOx test, the investigators experimented with various approaches to optimize their analysis of these three targets. For example, to distinguish cancer-associated DNA alterations from potentially confounding germline molecules, the group adopted a deep whole-genome sequencing approach (at least 80X) with each plasma sample paired with a matched germline sample to weed out somatic changes in normal, noncancerous cells.
For methylation, the team settled on an approach focusing on fragments rather than CPG loci and limited their calling of cancer to instances of increased methylation versus decreased methylation compared to baseline. Overall, the authors described their approach as a "strict de-noising strategy."
In the study, the investigators collected blood samples from patients with and without cancer who had been referred by their general practitioner for follow-up of suspicious symptoms alongside a control group of asymptomatic individuals without cancer, with 61 cancers and 30 non-cancer subjects in total. The authors stressed that their focus on samples from patients who were referred to surgery with curative intent allowed them to enrich for earlier-stage cancers.
Overall, the team determined that its assay could distinguish between people who had cancer and those who did not with about 95 percent sensitivity and 90 percent specificity.
More specifically, TriOx correctly identified as negative eight out of nine non-cancer controls and as positive 56 of 59 cancer cases. The assay also correctly identified the cancer origin in 25 out of 36 colorectal, four out of eight esophageal, three of four ovarian, and two of six pancreatic cancers. It could not identify the origin of any of the five renal cancers in the cohort, and the authors hypothesized that this may be because among the 377 hypermethylated regions used in its analysis, only four were specific to renal cancer.
Additional in silico validation demonstrated that the method could maintain a strong discrimination — an area under the receiver operating curve of 86 percent — at ctDNA fractions as low as 0.7 percent.
Finally, the authors also experimented with using TriOx to track tumor burden via ctDNA in colorectal cancer patients who had their tumors surgically removed, essentially a molecular residual disease application, which is another area of swelling commercial interest for liquid biopsy technologies.
The research team is now working to validate the test across more cancer types and in larger patient groups. Although a prospective study in a large cohort would be required to confirm clinical utility, the authors performed some modeling to hypothesize impact.
Using the cancer prevalence of 6.7 percent seen in Oxford's prior prospective study of Grail's Galleri test, which similarly focused on patients being referred due to suspicion of cancer, and assuming the test sensitivity and specificity seen with TriOx, the negative predictive value would be expected to be about 99 percent. "A negative test result would reduce the risk of cancer down from 6.7 percent to 1.2 percent," the authors wrote.
Assuming a primary care pre-referral population with a pretest probability of having cancer of around 3 percent, investigators calculated that TriOx's negative predictive value would be 99.5 percent with the posttest probability of cancer following a negative test decreasing from 3 percent down to 0.5 percent.
Exact Sciences hasn't shared data on the development of its own commercial multi-cancer screening assay, currently dubbed Cancerguard, since late 2022, when it presented a poster on research featuring a combined methylation and protein assay reflective of the technology it brought in with its acquisition of another company, Thrive Earlier Detection.
The firm hasn't said to what extent, or if, TAPS has been incorporated, but the new data on TriOx suggest that the two methods could be comparable in terms of optimizing early-stage sensitivity and high specificity.
The firm did report last year on outcomes data from the DETECT-A study, a prospective interventional trial of Thrive's previous version of Cancerguard, CancerSeek. Nine cancer types were diagnosed in 26 participants by that test. With at least four years of follow-up following their enrollment in the study, 13 of the 26 diagnosed patients were alive and cancer-free. Seven of these had cancers without recommended standard-of-care screening modalities. All eight treated stage I or II participants and 12 of 14 surgically treated participants were alive and cancer-free after four years.
If similar data can be gleaned from other ongoing prospective studies of MCED tests, it may begin to allay lingering skepticism regarding the ultimate value of such approaches to the healthcare ecosystem, the Oxford authors wrote.
Moving forward, Schuh said that she and her colleagues are currently focused on optimizing TriOx for cancer detection in terms of detection performance and cost and on extending its operational range to long-read whole-genome sequencing technologies.
"Eventually, we would like to assess its performance in a real-life scenario using a prospective cohort of up to 10,000 patients [in the UK's National Health System], ideally within the next five years," she added.
Schuh is also cofounder and chief medical officer of another spinout company, SerenOx, which has licensed TriOx and is currently seeking Series A investors, with a strategy focused on building a clinical niche for the test in non-US markets, including resource-poor regions.
"It is important to note that these days cost challenges apply to all healthcare settings. Cancer is a problem for all, rich and poor," said Schuh.
"But early cancer detection is even more important to the poor as it saves dramatically on the price paid to pharma and the healthcare providers for accessing treatment of late-stage cancer," she added.
She and her team are currently studying how reducing the depth of sequencing affects detection performance, and how long-read sequencing technologies and the use of multiple genomic data modalities can help alleviate any losses in performance that may result from a decrease in sequence depth.
For potential commercialization, SerenOx would hope to achieve a price target closer to $100, compared to the $1,000 price tag of Grail's Galleri test.