NEW YORK – PrognomiQ, a spinoff of proteomics technology firm Seer, hopes to launch its first test, a blood-based lung cancer screening assay, within a year after completing a large case-control study that it published in a recent preprint.
The firm split off from its parent company in 2020 to allow Seer to maintain its focus as a platform provider and avoid the possibility of competition with its own diagnostic customers.
The Seer platform is intended to boost the depth of coverage of proteomics experiments while still allowing for relatively high throughput and uses nanoparticle-based enrichment of proteins that researchers can then identify and quantify using technologies like mass spectrometry.
"Our mission as a company was to apply this technology in combination with other tools that are available for biomarker discovery to come up with what we call a true, unbiased, multiomics discovery set for improved liquid biopsy," said PrognomiQ CEO Philip Ma.
Although the company has collected and is analyzing samples across several cancer types, it decided to focus its first commercial efforts on lung cancer because it is the most frequently fatal cancer in the US and has a more established clinical roadmap.
Like colorectal cancer, which other firms like Guardant Health and Exact Sciences are targeting with blood-based screening, lung cancer offers a more straightforward path to physician adoption and reimbursement because there are established protocols for what to do with a positive screening test.
"With colorectal cancer, if you get a positive test, you go and get a colonoscopy. In the case of lung cancer, if we have a positive test, you should get a follow-on imaging study," Ma said.
Hilary Robbins, a researcher with the International Agency for Research on Cancer who has also investigated proteomic signatures for lung cancer detection but is not involved with PrognomiQ, said in an email that she couldn't speak directly to the company's data, but there are some common pitfalls in studies of this kind that the investigators should be mindful of as they move forward.
"For clinical translation of early cancer detection biomarkers, the road is long. Stepwise, rigorous research is needed to quantify the added utility of biomarkers beyond other tools and assess whether they provide sufficient added benefit to warrant implementation," she wrote.
Robbins highlighted as crucial steps independent validation and controlling for other risk factors to confirm an assay isn't providing redundant information.
She also raised the concern that control groups that don't appropriately match the cases can contribute to inflated performance that later drops when a test is studied in its intended use population.
PrognomiQ took steps to address both of these issues in its study, dubbed MOSAIC. Investigators analyzed samples from a total of 2,513 subjects split among cases, general controls, and a second group of controls with potentially confounding lung conditions.
"The reason it's important for this population to include those comorbidities is that when you smoke, you cause a lot of damage to your lungs and to your body and you are more likely to get chronic obstructive pulmonary disease, emphysema, pulmonary fibrosis, all of which are not cancer," Ma said. "Because we ultimately want to go after a high-risk smoker population, we wanted to include in our controls these comorbidities, which is an important differentiator between our approach and how others have done this type of study."
Using the Seer platform for protein discovery, as well as RNA-seq, metabolomics, and targeted immunoassays, the PrognomiQ researchers detected a large swath of molecules — 6,109 peptides, 40,171 mRNA transcripts, 9,368 intronic regions, 241 metabolites, and four targeted proteins that were differentially abundant between the lung cancer and non-cancer cohorts. They then winnowed these, using machine learning, into a set of 682 biomarkers that could best discriminate lung cancer from controls.
To make sure the assay content wasn't just replicating the predictive nature of clinical features like age or smoking status, the investigators compared their performance with a classifier based only on clinical factors, which it outperformed significantly. Further analysis testing the multiomic classifier's association with specific clinical features further reinforced that the panel was specifically predictive of cancer status.
Tested in an independent validation set of 398 study subjects that were held out from the initial training, the test demonstrated 89 percent sensitivity across cancer stages, 80 percent sensitivity for stage I tumors, 88 percent sensitivity for stage II cancers, and 99 percent sensitivity for stages III and IV at 89 percent specificity.
Next steps for the company include further refinement of the classifier to try to weed out unnecessary content while maintaining high sensitivity. "Some of these 682 (markers) are hugely important, but there is a really, really long tail, where molecules might be statistically significant, but we don't know if they are ultimately going to be clinically significant," Ma said.
The plan will then be to evaluate that final assay in a prospective intent-to-treat population, which the company has already begun enrolling, with a goal to reach 15,000 subjects, something Robbins stressed is crucial to determining whether a test actually leads to earlier detection.
A study of diagnostic or post-diagnostic samples simply can’t confirm that, she wrote in her email.
PrognomiQ hopes the data from that trial will support a bid for approval of its test by the US Food and Drug Administration. In the meantime, the company intends to launch the test as an LDT, potentially within the next year, which will also allow it to collect real-world data.
"We are very clear about the purposes of the lab-developed test … because the way we're thinking about the test in the US at least is that it's going to likely require a Class III designation, which means that the level of evidence will need to be very high," Ma said. "Unless we get that, reimbursement is probably going to be slow so we're focusing primarily on using our LDT to generate clinical data to support the FDA filing and ultimately for reimbursement. We're not thinking about the LDT as a huge revenue generator."
In the longer term, the firm is working on applying the same approach to other disease areas, including pancreatic cancer, colon cancer, and breast cancer. "We know those areas are competitive, too, so we'll be hopeful as we look at the performance and as we look at the public strategic opportunity, but down the line, it could be a multi-cancer application," Ma said.