NEW YORK (360Dx) – Over the last year, proteomics firm Somalogic has shifted focus from using its aptamer-based SomaScan platform for fee-for-service research work towards developing it as a clinical platform.
The company has entered collaborations with parties including DeCode Genetics and the Leeds Centre for Personalized Medicine and Health to test its platform in large patient cohorts with the aim of building clinical assays to assess questions including a patient's risk of having a cardiovascular event or developing diabetes.
While protein biomarker development typically follows a pattern in which researchers measure large numbers of proteins to identify potential markers and then winnow those down to small panels of diagnostically useful analytes to which they build rigorously validated targeted assays for clinical use, Somalogic plans to run the full SomaScan panel, which currently consists of 5,000 protein measurements, on every patient, with clinicians then looking at the measurements from those 5,000 that are relevant to the clinical question at hand.
This approach has raised questions among some as to how the company will establish that all 5,000 assays in the SomaScan panel have the level of analytical performance typically expected of a clinical test.
The SomaScan platform is "an exciting technology for proteomic discovery work," said Manuel Mayr, professor of cardiovascular proteomics at King's College and author of the commentary. However, he added, "not all of these [Somamers] may be ready for prime time in terms of clinical use."
Last year, Mayr and his King's College colleague Abhishek Joshi published a commentary in Circulation raising questions about whether the assays in the SomaScan panel had been sufficiently validated for clinical use. Mayr is an inventor on patents on biomarkers linked to cardiovascular disease and diabetes that could potentially compete with Somalogic's efforts in this space.
"The way aptamers target proteins is not so well understood for all of the 5,000 proteins that are claimed in the [SomaScan] assay," Mayr said. "Some will be detected with high specificity, but others will likely have cross-reactivity to other isoforms or will display non-specific binding."
He noted that he has not used the SomaScan platform in his own research but cautioned that "the transition from a research platform to as clinical assay is a big step to make."
"My worry is that we are measuring 5,000 proteins [with the SomaScan platform], and that is very exciting, but some of those 5,000 proteins may not be measured as accurately as is required for a clinical assay," he said. "Clinical validation of one protein requires a significant amount of resources, and doing it for 5,000 proteins is an even bigger challenge, especially since we don't understand aptamer-protein binding yet as well as we do antibody-protein binding, and matrix effects must be excluded."
Mayr added that development of a good clinical protein assay would involve validation work using internal standards to all proteins being measured and the ability to measure not just relative changes in expression but absolute concentrations.
"What matters for a clinician is what is the normal value and what is the abnormal value," he said. "If I can't give a reliable absolute concentration for a [protein] and am giving just a relative fold change, then this is clinically not very useful."
He gave the example of the biomarker troponin, which is used to diagnose heart attacks.
"There we basically have an assay where you determine what is the normal value in the population and based on this 99th percentile you can then determine what is an abnormal value, and you can make decisions about myocardial infarction because you know what the reference concentration is," Mayr said. "So, a simple approach would be to take a current clinical assay for troponin and compare it to the Somalogic measurement of troponin. What is the correlation? If that correlation is weak then it is not a suitable assay for clinical use in its present form."
In their commentary, the King's College scientists pointed to work done using SomaScan as part of the UK's Interval study in which researchers used the platform to quantify plasma protein levels of 3,301 apparently healthy, genotyped subjects. In that study, 14 percent of the Somamers showed non-specific binding, either to a protein other than the target protein (7 percent) or to a protein isoform (7 percent).
Other studies have similarly pointed to the potential challenges of adapting the SomaScan platform to clinical work. For instance, in a study published last year in Diagnostics, a team led by researchers at the University of Tennessee Health Science Center compared SomaScan measurements with ELISA measurements for three proteins linked to renal disease, FGF23, FGFR1, and FGFR4. They found that while the correlation between the assays was "good to near-perfect" for FGF23 and FGFR4, the assays for FGFR1 were poorly correlated.
Earlier work from Somalogic and Quest Diagnostics also demonstrates the challenges of developing a clinical-grade assay. In a 2013 paper published in PLOS One, the companies detailed an effort to develop a Somamer for measuring serum levels of the EGFR extracellular domain (ECD) in cancer patients. They found, however, that while the assay showed good linearity and high specificity for EGFR ECD, measurements were affected by levels of cetuximab and panitumumab in patients being treated with those drugs, likely due to the blocking by these drugs of the Somamer binding site on EGFR ECD. ELISA measurements were unaffected by the presence of these drugs.
While the study authors suggested that the Somamer assay might nonetheless prove useful for assessing patient drug response or guiding dosing, the work provided an example of the challenges of clinical assay development.
Alan Williams, Somalogic's chief development and operations officer, said that the company's approach to building clinical assays differed in some respects from traditional methods.
While Mayr stressed the importance of measuring absolute concentrations of target proteins, Williams said that Somalogic is "generally treating the [SomaScan] platform as a semi-quantitative platform, as opposed to treating it as an analytical platform like you might do with an analyzer. And what we're doing is using that semi-quantitative platform with the 5,000 analytes to directly build models against clinical end points."
Sheri Wilcox, Somalogic's senior director, discovery services, noted that the company is working its way through the SomaScan assays to thoroughly characterize the Somamer-protein binding and identify likely off-targets.
"We are working our way through as much as we can to evaluate what is the likelihood of binding to alternate proteins in a biological sample, particularly for closely related proteins," she said.
Wilcox suggested that the discrepancy in the FGFR1 assay results observed in the Diagnostics study could be due to differences in the FGFR1 isoform used in development of the Somamer versus development of the ELISA. She said that according to UniProt, FGFR1 has 21 different isoforms due to alternative splicing.
"We don’t yet have the data to know exactly which of the isoforms our reagent can bind, but the unusually large number of isoforms is likely contributing to the discrepancy between our measurement and that particular ELISA," she said.
Wilcox said that when Somalogic looked broadly at the question of ELISA-Somamer correlation it had found that ELISAs and Somamer-based assays correlated with one another as frequently as do different ELISAs to the same target.
Ultimately, Williams said, the company's strategy is based on making measurements in large cohorts and observing reproducible signals that they can train against specific endpoints.
"In a number of cases we're using retrospectively collected samples where there are hard outcomes, and we are able to essentially identify signatures on the platform that are associated with those outcomes," he said. "We're pretty confident in what the platform is measuring, which is the availability of the specific epitope that [Somamer] binds to.
"The challenge is that the availability of that epitope can be impacted by all sorts of things," he said. "It can be impacted by the fact that there's a complex formation that either exposes or hides the epitope. There could be post-translational modifications that actually change the epitope such that the [Somamer] no longer binds."
Somalogic believes that so long as it is able to train its signatures in large enough and diverse enough cohorts, "it really doesn't matter which of those things is impacting the epitope availability," Williams said. "What we are confident in is that the availability of that epitope is actually correlated with disease."
As Mayr's comments indicate, this is a somewhat different way of thinking about clinical measurements, and, Williams said, "it definitely creates an educational challenge," for the company in terms of bringing clinicians aboard.
"It hasn't necessarily created an adoption challenge, although we're super early with regard to a commercialization of the clinical strategy," he said. "We have yet to really get feedback on how the [healthcare] providers as well as the patients kind of respond to that level of information."