Skip to main content
Premium Trial:

Request an Annual Quote

UNC Pancreatic Cancer Test Aims to Classify Subtypes to Improve Therapy Selection

Premium

NEW YORK – A group of researchers from the University of North Carolina at Chapel Hill have validated an array-based assay for subtyping pancreatic cancer that they believe may help determine which treatments are most effective in certain patients.

The Purity Independent Subtyping of Tumors, or PurIST, test assigns a subtype based on a 16-gene expression signature measured using the array-based NanoString nCounter Elements XT technology and a proprietary algorithm. Its analytic validation, which looked at the test's performance in 74 formalin-fixed, paraffin-embedded biopsy or resection specimens, was published last month in the Journal of Molecular Diagnostics.

Margaret Gulley, a director of molecular pathology at UNC and one of the test's developers, said that the test has been under development for a number of years and was originally developed using RNA sequencing, but the need for a more rapid result led the researchers to seek other, faster methods of testing. There are "a lot of limitations" in pancreatic cancer testing, including the fact that getting large pieces of tissue is particularly difficult because of the pancreas's location within the abdomen, so a clinically useful test must be able to use "tiny bits of tissue," she said.

When developing the test, the researchers could have used a variety of different methods, but the NanoString XT technology was the fastest that allowed the researchers to look at multiple analytes simultaneously, Gulley said. NanoString was acquired by Bruker in May.

The research team also sought to develop an assay that would be robust even if the tissue sample was not handled the same way at every step of the pre-analytic process because RNA is "not a stable molecule." "We needed an assay that would work and be robust despite … the variables," she said, such as how quickly a patient's tissue was formalin-fixed.

The test is not just measuring RNA levels but measuring them as a proportion to other RNA levels, she added.

Naim Rashid, a biostatistician at UNC who helped develop the PurIST algorithm, said that the researchers used a machine-learning approach to narrow down which 16 genes were most related to the different pancreatic cancer subtypes. Once those genes were selected, the team constructed eight gene pairs, which "forms the basis of the test," he said. For each gene pair, the model asks whether the first gene in a pair's expression is greater than the second gene's expression, and those measurements are then plugged into a prediction model, which provides a score that indicates the probability of the cancer being the basal or classical subtype. If the probability is greater than ½, the patient is predicted to have the basal subtype, and if the score is less than ½, the patient is predicted to have the classical subtype.

When training the prediction model that the test is based on, the team used expression data and survival outcome data from UNC as well as publicly available expression and survival data from prior studies, Rashid noted. They also utilized survival and treatment response data available in certain datasets to validate the clinical relevance of the predictions.

The gene pair approach also helps avoid some issues with gene expression variations in different settings, Rashid said. "Rather than asking whether the absolute level of expression of a gene is correlated with subtype, we just ask within a patient whether the relative expression, the ordering of expression of two genes, is correlated with subtype." This is a "slightly easier problem to deal with, because now you're just looking at a relative comparison within a sample as opposed to looking at raw values and having to worry about adjusting those raw values for technical effects between patients and across studies," he added.

Past retrospective data has shown that patients with the basal subtype typically have worse survival outcomes compared to classical patients, "so there's some prognostic relevance to the way in which we classify pancreatic cancer using this system," Rashid said. In addition, historical data has also shown that patients with the basal subtype tend to have low or no response to FOLFIRONOX, a common first-line therapy used for pancreatic cancer, Rashid said.

In preliminary data published by other research groups, a patient's subtype has been shown to have an impact on the effectiveness of certain treatments, Gulley also noted.

According to the Journal of Molecular Diagnostics paper, the array-based version of the test showed 97 percent concordance with the comparator method, which was full transcriptome RNA sequencing. That method, however, is costly and takes a significantly longer time to return a result. RNA sequencing can take about a week, while the NanoString array has a turnaround time of about three days. Gulley noted that since the JMD paper was written, the researchers have managed to lower the turnaround time to two days.

Using an array is also "more straightforward" and less labor-intensive than RNA sequencing, Gulley said.

Jen Jen Yeh, a professor of surgery and pharmacology at UNC and another developer of the test, said that UNC is opening a clinical trial for metastatic pancreatic cancer that will use the PurIST test in the fall. While pancreatic cancer subtypes have been shown to potentially be useful in retrospective data, the "use of subtypes in clinical practice is still evolving and new,” Yeh noted.

"In order to understand whether subtyping itself is actually of utility for patients … [it] needs to be tested in the setting of a clinical trial, so we know whether or not it's actually helpful," she said.

In Rashid's view, the test could hopefully be used after a patient has received a pancreatic cancer diagnosis and undergone a tissue biopsy to provide the predictive subtype of the cancer, a general assessment of their prognosis, and potential options for first-line therapies.

Yeh noted that the test is currently being used for research and clinical trials, and because UNC is an academic institution, it is "not in a position to make technologies widely available commercially." As a result, commercialization of the test's array-based version would likely come from a partnership with an outside partner.

However, the PurIST algorithm has already been commercialized in one form, she said. The algorithm was licensed to GeneCentric in 2018 and then sublicensed by Tempus AI, which has commercialized its version of the assay as a laboratory-developed test. Tempus' test uses exome capture and sequencing instead of an array and has been analytically and clinically validated in accordance with CLIA and the College of American Pathologists guidelines, Tempus VP of Molecular Pathology Nirali Patel said. Tempus has published an abstract and a preprint describing those results, and an additional clinical validation study is under peer-review for publication.

Tempus' test has also not been run in a clinical trial, but Patel noted that the firm "performed a robust analytical comparison between RNA-seq and PCR and found no systematic differences in the quantification of gene expression for select genes in PurIST between the two methods."

The test is ordered clinically alongside Tempus' xR RNA sequencing panel, with results delivered the same day as xR. Results from the xR test are returned within 10 days of specimen receipt, a spokesperson for the company noted.

Yeh emphasized the importance of a "head-to-head comparison" between current and future subtyping tests performed on different platforms, such as Tempus PurIST versus UNC PurIST, on the same patients in a clinical trial setting to ensure that the same result is being returned each time and see how it influences treatment planning.