Just as important, though, is that Luminex can do that not just with proteins like cytokines—one of the platform’s original target markets—but can query samples at the genetic and gene expression levels as well. For cytochrome P450, for example, “we have some larger assays that look at a lot of different alleles and then narrow down the key markers that are important for predicting response to different drugs like Warfarin,” Dr. Dunbar said. These kinds of studies allow the small contributions of individual or clusters of biomarkers to be teased out from otherwise impenetrable background noise.
More than 8,000 systems have been placed over the last 12 years, with over 50 Luminex assays having been approved for in vitro diagnosis by the FDA—including one for HLA matching that has “kind of become the gold standard,” Dr. Dunbar noted. Such a track record reduces the risk associated with regulatory clearance.
Due to duration of a study, kit shelf life, and other factors, it is not unusual for samples for quantification of a biomarker study to be run on different lots of immunoassay kits in support of both preclinical and clinical studies. Yet for a variety of reasons—going up and down the supply chain of critical reagents that go into them—“there is quite a bit of variability in terms of the quality of these commercial kits,” laments Afshin Safavi, Ph.D., senior vp of BioAgilytix Labs.
Even home-brewed assays rely heavily on the critical reagents, he pointed out. “We don’t call it a kit, but in essence we are developing a kit in our shop internally.”
Like it or not, the researcher needs to shoulder some of the burden of making sure the kits perform the same way lot-to-lot, year-to-year—and if not, to come up with systems to bridge the data that are generated by different lots.
To do this, Dr. Safavi recommended at a minimum running a series of quality controls and samples in both old and new lots. For larger studies “our practice is actually to repeat the lot-bridging process over three consecutive days, and that way you generate a larger number of data points that are more statistically significant,” he said. “And then we come up with a correction factor…to normalize that data to the previous lot.”
With the trend to multiplex assays, the bridging process becomes more critical than ever because the panel of proteins included in different biomarker kits comprising a multiplex assay do not always vary in the same way as each other from lot to lot, Dr. Safavi observed.
When applying a correction factor, it is important also to know what kind of tolerance there is for differences, and this can depend upon the intended use of the kit, the disease area being supported, as well as the changes you may predict, he says. If a 500% change is expected due to drug treatment, 40% variability is not going to have much effect on the decisions likely to be made based on the test, but if a 20% change is expected due to drug treatment even a 5% change may affect the quality of the data generated and impact the decision process.
Dr. Safavi was quick to emphasize that his comments applied only to research kits used in biomarker studies. “When it comes to pharmacokinetics or immunogenicity assays, they have their own sets of processes for qualifying and bridging reagents and assays and lots.”
A Case in Point
Tests developed in a regulated environment generally adhere to much tighter tolerances, with far more involved in their validation. When Roche developed the cobas 4800 BRAF V600 mutation test as a companion diagnostic to the metastatic melanoma therapeutic Zelboraf, for example, “no less than 25 analytic performance verification studies were required by the PMA [FDA premarket approval] process as well as for European and other regulatory agencies,” recalled Walter Koch, Ph.D., vp and head of global research for Roche Molecular Diagnostics. “Multiple labs, multiple operators, multiple reagent lots over different days, with over 1,400 samples, all showed that the test was highly reproducible.”
Development of the diagnostic and the drug proceeded together apace, taking a mere five years from the IND to their approval last year—“those are pretty much record times, I think,” Dr. Koch said. “It shows the power of patient selection if you’ve got a good hypothesis of the biomarker, a good drug target, and a good drug that hits that target.”
The assay, based on extracting DNA from FFPE samples for PCR, had its challenges to overcome, he pointed out. Formalin reacts with nucleic acids and causes them to degrade (“in the worst case only as much as 10% of the DNA is amplifiable”). Melanin in these highly pigmented cells is an inhibitor of PCR (“we had to devise sample-preparation strategies that eliminated the melanin from carrying into the PCR reactions”). And tumor tissue is heterogeneous and may contain different percentages of tumor content (and “not necessarily all the tumor cells have mutant copies—you can have mixed populations”).
After all that, Roche had a conundrum. The FDA required it to compare the test results with Sanger sequencing, which can miss mutations found in less than about 25% of cells. The cobas test, on the other hand, was able to detect down to about 5% of cells carrying the mutated BRAF gene. Dr. Koch explained: “Our clinical trials had shown that those patients benefited from the drug. Yet the gold standard said, ‘Oh, you’ve got a false negative or a false positive’.” The issue was ultimately resolved by validating the procedure against 454 next-gen sequencing. “That was a challenge, because the agency had not looked at such data before.”
Dr. Koch believes that targeted therapies and companion diagnostics are becoming inextricably linked. He urges coordinating and aligning their development from the beginning, and making sure that the diagnostic is in place and ready to go for the start of pivotal trials.