NEW YORK – Researchers at Grifols subsidiary Alkahest have published results from a study comparing the performance of multiple plasma proteomic workflows, including a number of the most commonly used commercial technologies.
Detailed in a preprint published this month in BioRxiv, the findings add to other recent evaluations of plasma proteomics technologies. Jochen Schwenk, a professor of translational proteomics at KTH Royal Institute of Technology and chair of the Human Proteome Organization's Plasma Proteome Project, suggested that the Alkahest study is particularly valuable as it is one of the first such efforts conducted by a group without connections to any of the platforms being evaluated.
Schwenk, who was not involved in the study, said that he "really liked" that it was "done from a user perspective." He noted that many previous evaluations have been done by researchers looking to "maybe benchmark their own method or benchmark [a method] they closely relate to against something else."
The study looked at four affinity-based platforms — Standard BioTools' SomaScan 7K and SomaScan 11K and Olink's Explore 3072 and Explore HT — and three mass spectrometry-based workflows — Seer's Proteograph XT and Biognosys' TrueDiscovery and SureQuant offerings. The Alkahest team ran plasma samples from 78 individuals on each of the platforms, assessing the technical performance of each as well as cross-platform correlation and the ability of each platform to identify markers linked to biological factors including age, race, sex, smoking status, and body mass index.
Among the study's takeaways is that mass spectrometry-based methods like Seer's Proteograph and Biognosys' TrueDiscovery provide a depth of coverage that is competitive with affinity-based platforms. Over the last several years, mass spec-based technologies for plasma proteomics have advanced dramatically as, led by Seer, a new wave of enrichment techniques have come to market. The Alkahest study offers the latest demonstration of this trend.
Across the 78 samples, Seer's Proteograph XT measured a total of 5,943 proteins, with roughly 3,000 of those proteins measured in all samples. Biognosys' TrueDiscovery platform measured 3,575 proteins across the 78 samples with around 2,000 of those proteins measured in all samples.
The Proteograph XT platform uses nanoparticles to enrich plasma proteins prior to mass spec analysis, boosting depth of coverage, while the TrueDiscovery platform improves depth of coverage by depleting samples of the highest abundance proteins prior to running them on mass spec. Depletion is an older and, generally speaking, less effective approach, as is evidenced by the substantially larger number of proteins identified via the Proteograph XT platform. Biognosys last year launched its P2 Plasma Enrichment product, which uses nanoparticle enrichment and, according to Oliver Rinner, the company's director and CEO, offers substantially better depth of coverage than the TrueDiscovery approach.
"I think in the [Alkahest] study, they pushed the depletion approach as far as you can push it," Rinner said, noting that Biognosys has moved almost completely to its P2 Plasma Enrichment product for discovery proteomics experiments. He added that the company is seeing depth of coverage with that platform comparable to the deepest published studies using other nanoparticle platforms, and the firm plans to present data on the system at the American Society for Mass Spectrometry meeting in June.
Both the Proteograph XT and TrueDiscovery platforms showed relatively high levels of variance with the former having a median coefficient of variation of 26.4 percent and the latter a median CV of 29.8 percent. Rinner noted that high CVs are a known challenge for depletion-based workflows like TrueDiscovery. The Seer figure, on the other hand, is significantly higher than other published CVs for the platform. In the plasma proteomics technology evaluation published in January by the lab of University of Wisconsin-Madison researcher Josh Coon (and funded in part by Seer), Proteograph XT had a median CV of 10.4 percent. Other studies have also found the Proteograph XT platform to have a median CV in the 10 percent range.
In an email, Asim Siddiqui, SVP and scientific fellow at Seer, and Serafim Batzoglou, the company's chief data officer, said that the samples in the Alkahest study had been run by the company in its Seer Technology Access Center. They noted that sample complexity and pre-analytical variability could have factored into the higher CV figure.
Sara Ahadi, associate director of multi-omics and integrative analysis at Alkahest and senior author on the preprint, said that a variety of details could lead to different CV figures, including how many proteins a study looked at and whether or not it used any limit of detection cut-offs or otherwise filtered results prior to calculating CVs.
"The devil is in the details," she said. "There are many ways you can look at these analyses."
More generally, Siddiqui and Batzoglu questioned the value of the study's cross-platform CV comparisons as CVs for each technology were calculated using different sample sets and different numbers of samples, which they said made an "apples-to-apples" comparison difficult. The Alkahest researchers calculated median CVs of 5.3 percent for SomaScan 11K, 5.8 percent for SomaScan 7K, 8.3 percent for SureQuant, 11.4 percent for Olink Explore 3072, 26.4 percent for Proteograph XT, 26.8 percent for Olink Explore HT, and 29.8 percent for TrueDiscovery. When the researchers looked only at Olink Explore HT assays that returned results above the assay's limit of detection, the median CV for that platform fell to 12.4 percent.
The study also provides new data comparing the two leading affinity-based discovery proteomics platforms, SomaScan 11k and Olink Explore HT. On its face, the study appears to demonstrate the superiority of the SomaScan platform as it detected 9,645 proteins across the 78 samples with 96.2 percent of those proteins detected in all of those samples. Explore HT, meanwhile, detected a total of 5,416 proteins across the full sample set with only 35.9 percent detected in all 78 samples. The Alkahest team considered a protein detected by an affinity platform if it returned a result above the assay's limit of detection.
SomaScan also had lower CVs than did Explore, which has been a consistent finding across several studies.
Carl Raimond, president, proteomic sciences, at Thermo Fisher Scientific, which acquired Olink in July for $3.1 billion, raised questions, however, about whether the SomaScan assay is actually detecting the roughly 10,000 proteins it targets. He suggested that many of the proteins it targets would not be expected to be seen in healthy plasma.
"Is every protein in your body being expressed all the time and making its way into your blood?" he asked. "That's super questionable. I don't think most biologists would think that's the case."
Maik Pietzner, a bioinformatician at the MRC Epidemiology Unit at the University of Cambridge School of Clinical Medicine who uses both platforms, said that he is also interested in this question, noting that in his experience with the SomaScan platform, he and his colleagues have been able to link around half the proteins in the panel to different biological outcomes, indicating those measurements are providing useful information. For the other half, however, "we hardly detect anything … or else technical variables explain most of the variation," he said.
Pietzner said this does not necessarily indicate those SomaScan assays are not hitting their targets. Rather, it could simply mean the researchers have not run the assay in populations with diseases or conditions where those proteins play a role.
"It might, though, give you a hint that there is something there that we would benefit from resolving in terms of better understanding what we are looking at," he said.
Stephen Williams, chief medical officer at Standard Biotools, said that the company has generated orthogonal confirmation for 8,200 of the 11,083 assays in the SomaScan 11K panel that indicates they are hitting their stated target. That confirmation is based on data from experiments like mass spec analyses and cis-protein quantitative trait loci (pQTLs) identified in population proteogenomic studies, he said.
The Alkahest preprint is one of several recent publications finding that the Olink Explore HT consistently measures only around half of its 5,416 protein targets above its assays' limits of detection.
Raimond said that in large part this reflects the fact that the roughly 2,500 proteins added to the platform since the Explore 3072 are weighted towards intracellular proteins that would not necessarily be expected to be present in healthy plasma samples. He suggested that the point of the expanded panel is not to detect all of the proteins in every sample but rather to have coverage of a wider range of proteins that might show up in plasma in the case of a particular disease or biological condition.
"The fact that we are not lighting up everything in our panel in healthy plasma is, to some degree, by design," he said.
Nonetheless, based on the Alkahest study as well as the Coon lab's preprint, the Explore HT assay does appear to miss a number of plasma proteins detectable by other methods. Looking at the Alkahest study's mass spec-based data alone, Proteograph XT detected close to 3,000 more proteins across the 78 samples than did Explore HT, and the TrueDiscovery platform detected roughly 1,500 more. In the Coon lab's study, Proteograph XT identified roughly double the number of proteins as Explore HT.
Explore HT does still have certain advantages compared to those mass spec methods, including substantially higher throughput. KTH's Schwenk suggested that the value, or lack thereof, of Explore HT versus both the Explore 3072 and other plasma proteomics platforms will become more clear as larger datasets are produced on the system.
The Alkahest researchers also looked at the number of proteins measured by each platform that correlated with the biological factors of age, sex, race, hematocrit, total protein, smoking status, and BMI. Across these factors, SomaScan 11K and 7K detected the highest number of significant proteins with Proteograph XT next followed by Olink Explore HT.
Looking at the amount of biological variance captured by each platform, the researchers found that the TrueDiscovery, Explore 3072, Proteograph XT, and SomaScan 7K and 11K explained around 20 percent of variance, while Explore HT and TrueDiscovery explained around 14 percent.
Ahadi highlighted as notable the fact that the TrueDiscovery platform — a targeted mass spec assay that measured 551 proteins — captured a substantial amount of biological variance despite measuring only a relatively small panel of analytes.
"We're seeing that with an increased number of proteins we're not necessarily getting an increased number of markers," she said, suggesting that this is perhaps due to both to the fact that as panels expand they may bring in proteins that are less closely tied to the biology of interest and may include more difficult to measure proteins and assays with lower ratios of signal to noise.
Nonetheless, Ahadi said, there remains significant room for expanded proteomic discovery, particularly when taking into account the large number of different proteoforms and their roles to biological functions. Ideally, researchers will be able to explore this vast protein diversity while also measuring it robustly and specifically, she said.
"Every day papers are published with lists of markers, but do they really make it to a diagnostic for a disease? No." Ahadi said. "I think specificity in being able to identify the correct proteoform of a protein would help a lot in being able to get to an answer."