Last July, the National Institutes of Health and the Food and Drug Administration announced that they would begin tripling the amount of annual funding for a new clinical proteomics program. Now it seems the money has begun to bear fruit. Last week, Emanuel Petricoin and Lance Liotta, co-directors of the program, and their colleagues published a paper in The Lancet describing a diagnostic test for ovarian cancer based on serum protein profiles.
“This is the first time that the concept of proteomic patterns [has been] used as a diagnostic marker,” said Liotta.
To create a set of markers, the researchers generated protein profiles from serum samples of 50 patients with ovarian cancer and 50 unaffected women, using a type of Ciphergen’s SELDI-chip that binds a subset of low molecular weight proteins. The researchers determined the masses of the proteins — but not their identity — using time-of-flight mass spectrometry. An iterative algorithm developed by Correlogic Systems, a Bethesda-based bioinformatics company, then produced a protein pattern containing five markers that discriminated cancerous from non-cancerous samples.
In a validation step, the scientists tested this pattern on an independent set of 116 masked samples and found that it correctly identified all cases of cancer and almost all unaffected samples, yielding a positive predictive value of 94 percent. CA125, the most widely used biomarker for ovarian cancer, even in combination with ultrasound, only had a positive predictive value of 30 percent, Petricoin said.
Long Road to the Clinic
This does not mean the new test will take over current ones anytime soon. “We are not proposing that this test will be used as any replacement,” said Petricoin. “If it holds true, we hope that this will be used as an adjunct to give the clinician more information.”
The next step on the way to the clinic will be a larger study involving about 1,000 women from across the country who have been monitored because they are at a heightened risk for developing ovarian cancer. The researchers are currently preparing this study and hope to complete it within several months.
Promising results provided, they will eventually seek out a company to develop the test and get it approved by the FDA, Liotta said. He would like to see it used in high-risk patients first, and “if it continues to hold up, we will at some point dream about extending it to the general population,” he said.
In contrast to gene expression profiles, which require cellular material — for example from biopsies — proteomics-based tests only need a few drops of blood. “The strike zone is perfect for proteomics because it’s just a bag of protein information,” said Liotta. However, while a gene expression-based test reveals exactly which genes differ between samples, the nature of the proteins in these “fingerprints” remains an enigma.
With or Without Ciphergen
Although the researchers used Ciphergen’s SELDI system to create the protein profiles, they said they are not wed to the technology. “Right now the Ciphergen platform is the most amenable, but there are other mass specs that could be made [to work] in the same high-throughput fashion” combined with other fractionation methods, Petricoin said. The self-organizing genetic algorithm, on the other hand, is crucial to the approach, Liotta added, because it “learns, gets smarter, and grows as more data is added” to the training set over time.
Meanwhile the researchers have been working on similar diagnostic tests for other types of cancer, including prostate, breast, pancreatic, colon, and lung cancer, some of which “look very promising,” Petricoin said. They are also preparing to study whether the pattern can predict relapse or response to therapy in ovarian cancer patients. A further goal is to find protein patterns that can predict drug toxicity or detect exposure to infectious agents.