
NEW YORK – A team led by researchers at the University of Wisconsin-Madison has published a comparison of six technologies for analyzing the plasma proteome.
The team's findings, detailed in a preprint published this month in BioRxiv, show substantial variation across the methods tested in terms of depth of coverage and reproducibility as well as trade-offs between cost, performance, and throughput.
Plasma is relatively easy to collect and potentially contains biomarkers reflecting the activity of a wide range of organs and bodily systems, making it an appealing sample type for proteomics research. Measuring the proteome in plasma is technically challenging, however, due to its high dynamic range and the fact that a few highly abundant proteins constitute the vast majority of a plasma sample's total protein mass.
Historically, this limited the depth of coverage researchers were able to achieve, especially using mass spectrometry-based techniques. Until recently, mass spec experiments in plasma typically topped out at around 500 proteins. Through extensive fractionation, it was possible to increase that number to several thousand proteins, but such experiments were extremely time-intensive and did not offer the throughput necessary for large-scale studies.
In recent years, however, a number of private companies including Seer, PreOmics, and Biognosys, as well as several academic labs have developed approaches for enriching the plasma proteome. Combined with improvements in mass spec technology, these methods allow researchers to measure thousands of proteins in plasma, with throughput of 20 samples or more per day.
At the same time, affinity-based platforms from Olink (now part of Thermo Fisher Scientific) and SomaLogic (now part of Standard BioTools) have continued to expand their content. Olink's Explore HT product assays more than 5,300 proteins in plasma, while Standard BioTool's SomaScan platform measures roughly 11,000 proteins.
"Over the last four, five years, we have seen tremendous advances in plasma proteomic capabilities," said Josh Coon, a professor of chemistry and biomolecular chemistry at the University of Wisconsin-Madison, the senior author on the preprint study.
Coon said that new enrichment techniques, advances in mass spec instrumentation, and the ongoing development of data-independent acquisition mass spec methods have combined to "really transform the depth and speed" of plasma proteomic experiments.
He noted that these developments have generated a swell of interest in plasma proteomics, but that "there hasn't been a good technical comparison across the methods," adding that this was the rationale for conducting the study.
Coon and his colleagues looked at six plasma proteomic approaches: analysis of unenriched, or neat, plasma; an acid depletion method developed by Boston Children's Hospital researchers Judith and Hanno Steen; particle-based enrichment methods from proteomics firms Seer and PreOmics; the Mag-Net particle-based enrichment method developed by the lab of University of Washington researcher Michael MacCoss; and Olink's antibody-based Explore HT platform. Aside from Olink, all the methods tested are mass spec-based.
Two notable plasma proteomics technologies that were not included in the study are Standard BioTools' SomaScan and Biognosys' P2 Plasma Enrichment.
Seer, Coon said, initiated the study, approaching him about conducting a comparison of several plasma proteomic technologies. The company covered the cost of the Seer, PreOmics, and Olink assays, as well as the plasma samples used. Coon's lab designed the study and shared the design with Seer prior to starting the experiments, which he said was necessary in order to get the firm's sign-off on the study's scope, given its role in funding the work. He added that his team told Seer it would publish its results regardless of the outcome, and that the company agreed.
Coon, who is a member of Seer's scientific advisory board and consults for Olink's parent company Thermo Fisher Scientific, said both Seer and PreOmics were sent drafts of the final manuscript and given opportunities to respond.
The researchers ran five sets of experiments to evaluate proteomic depth, limit of detection (LOD) and quantification (LOQ), reproducibility, linearity, and tolerance to lipid interference across the six different methods, conducting a total of 618 mass spec experiments and 93 Olink Explore HT experiments.
They found that of the six technologies, Seer's Proteograph XT delivered the highest number of protein identifications, with an average of 4,501 proteins detected across five technical replicates. Olink's Explore HT delivered an average of 2,236 identifications, while PreOmics' EnrichPlus delivered 2,187, Mag-Net 1,182, acid depletion 809, and neat plasma analysis 584.
To evaluate LOD and LOQ, the researchers spiked varying amounts of human plasma into a matrix of chicken plasma and determined how many human proteins each approach could detect and quantify at each dilution level. They found that Seer proved the most sensitive, detecting 1,966 proteins at a dilution level of less than 1 percent human to chicken plasma. PreOmics detected 493 proteins at that level, while Olink detected 339, Mag-Net 333, neat analysis 310, and acid depletion 193. For LOQ at that dilution level, Seer measured 1,808 proteins, PreOmics 441, Mag-Net 272, neat analysis 200, acid dilution 145, and Olink 25.
The authors noted that the chicken plasma matrix might have presented unique challenges to Olink's Explore HT due to the potential for cross-reactivity of its antibodies with chicken proteins.
Looking at the quantitative reproducibility, neat analysis had the lowest median coefficient of variation at 8.7 percent, while Seer's median CV was 10.4 percent, Mag-Net's was 12.6 percent, Olink's 13.9 percent, PreOmics' 25.2 percent, and acid depletion's 26.6 percent.
All five mass spec-based techniques showed solid linearity (R2 above 0.92), which the researchers measured by spiking in C-reactive protein (CRP). They did not measure Olink's linearity as Explore HT does not include an assay for CRP.
The six methods were largely unaffected by variations in plasma lipid levels.
Coon and his colleagues also looked at the ability of the different approaches to identify disease-related differences in proteins, analyzing 40 plasma samples from stage IV non-small cell lung cancer (NSCLC) patients and healthy controls. All of the methods found significant differences between the NSCLC patients and controls, with PreOmics' EnrichPlus and acid depletion identifying the largest number of differentially regulated proteins.
Most approaches identified significantly more proteins in the patient samples than in the plasma sample (sourced from BioIVT) used in the initial set of experiments. Seer identified 7,019 proteins across the patient samples, with 6,604 found in at least 50 percent of all samples; PreOmics identified 5,157 and 4,616 proteins, respectively; Mag-Net identified 4,559 and 3,392; Olink 4,470 and 2,199; acid depletion 2,186 and 1,728; and neat analysis 1,221 and 1,006.
Coon said that in his lab's experience, the BioIVT plasma "tends to yield a lower number of protein IDs for any method as compared to patient samples."
He added that in the cancer samples, "we tend to see evidence of tissue leakage and other proteins that often do not show up in pooled, healthy individuals."
Coon said one takeaway from the study is that Seer by and large outperformed its mass spec-based competition. This finding is largely in line with past data published on the various technologies, where Seer typically provided the deepest coverage.
He noted, however, that Seer is more costly than the other approaches (with the exception of Olink, which is priced comparably), calling that the product's "main downside."
Cost has been a concern for some potential Seer customers, but Omid Farokhzad, the company's president and CEO, said he believes its product's higher price is justified by its performance. He added that the company offers volume-based discounts off its list price as well as different pricing for industry and academic customers.
Seer President and CFO David Horn added that the company recently completed a detailed pricing study. "We've put a ton of time and work and effort into pricing, and I think where we are is very competitive," he said.
UW's MacCoss, who was not involved in the study, suggested that the comparison would have benefited from more detailed cost information. Assays were scored on cost by putting them in one of three tiers ($, $$, or $$$), with neat analysis, acid depletion, and Mag-Net receiving a $, PreOmics receiving a $$, and Seer and Olink receiving a $$$. MacCoss said Mag-Net runs around $9 per sample in reagent costs. Seer has quoted researchers costs per sample of around $600. Olink Explore HT prices vary, but as a point of reference, Vanderbilt University Medical Center's High-Throughput Biomarker Core charges between $318 and $1,242 per sample, depending on the type of researcher and sample volume.
One other area in which the study found Seer to lag is throughput. The Proteograph XT can run 12 samples per day compared to 24 samples per day for the other four mass spec-based approaches and 172 per day for Olink.
MacCoss questioned whether the Mag-Net experiments detailed in the study were using as much peptide load as the Seer experiments. He pointed to the total ion current figures provided in the paper for each of the mass spec methods, noting that they should be roughly equal if the same amount of peptide is being introduced into the mass spec in each method. Instead, the Seer and neat plasma experiments had total ion current figures roughly double the size of those observed for the three other mass spec approaches.
Garwin Pichler and Nils Kulak, founders and managing directors of PreOmics, likewise noted this difference in total ion current and suggested that this could have impacted the results.
Coon agreed that there was variation in the total ion current signal across the experiments, with Seer and neat plasma at the high end and Mag-Net, PreOmics, and acid depletion at the low end.
Reasons for this difference could be different chemical properties of the proteins enriched for by each technology and different ionization efficiencies of the resultant peptides, he suggested. The concentration of peptides in each sample, which was used to determine the amount of sample to inject into the mass spec system, was established using either the Thermo Scientific Quantitative Fluorometric Peptide Assay (for the Seer samples) or the Thermo Scientific Pierce Quantitative Colorimetric Peptide Assay.
In any case, Coon noted, the researchers "followed the protocols precisely for all the methods" they investigated.
Pichler and Kulak also highlighted the CV figure the study arrived at for PreOmics' EnrichPlus kits, noting that at 25.2 percent, it was significantly higher than the 7 percent to 15 percent they said internal and customer studies have typically shown. They added that the kit used in the study was "an early prototype version of EnrichPlus," and that the final product "has undergone substantial improvements in digestion efficiency, alkylation, and robustness."
Coon also highlighted as notable the relatively low number of proteins identified by the Olink Explore HT platform. While the platform is advertised as measuring 5,300 proteins in plasma, it averaged fewer than half that number in the BioIVT plasma samples and measured only 2,199 proteins across 50 percent or more of the samples from the NSCLC set.
"Having never used [Olink] before, I had this view that, well, Olink will have the best depth because they get about 5,400 proteins," Coon said. "But it turns out that they don't get them all, and they get about half of them in our study."
MacCoss similarly suggested this result would surprise many proteomics researchers, particularly those primarily focused on mass spectrometry.
"In mass spectrometry, when you develop a targeted assay, you expect that assay is detectable in kind of a standard, healthy individual," he said. "The fact that in a healthy individual pooled plasma sample only half the proteins were detected [by Olink] was news to me, and I can imagine that it might be news to other people, as well."
The study counted protein IDs only for Explore HT assays that returned values above their LOD. A separate study, published in July as a MedRxiv preprint, similarly found that the Explore HT returned values above the LOD for only around half its roughly 5,300 targets. That work was funded in part by Olink competitor SomaLogic. Josef Coresh, professor of medicine at NYU Langone Health and a former member of SomaLogic's scientific advisory board, is the senior author on that study.
Jochen Schwenk, a professor of translational proteomics at KTH Royal Institute of Technology and chair of the Human Proteome Organization's Plasma Proteome Project, said that researchers' approaches to dealing with Olink values below the LOD varies.
"Some remove the proteins if more than 50 percent of the samples are below LOD, some set values below LOD to the stated LOD, and some even keep the values generated below LOD, depending on how much uncertainty they want to accept," he said, noting that follow-up experiments can help identify the best approach for a given experiment. "Unreliable features disappear quite soon from the list of candidates."
Schwenk, who was not involved in the Coon lab's study, said that results from large population studies using Explore HT will provide a clearer view on the assay's performance. This month, for example, the UK Biobank said it plans to use the Explore HT to analyze 600,000 blood samples.
Karsten Suhre, professor of biophysics and physiology at Weill Cornell Medical College, similarly said that researchers take varying approaches to dealing with Olink results below the LOD. He suggested that in the case of the Coon paper, discarding values below the LOD made for a fairer comparison across the platforms.
In an email, Michael Gonzales, VP of global marketing at Olink, said that "the number of detectable proteins of an assay platform depends on the natural abundance of the target proteins, sample type, assay specificity, collection methods, and composition of the assay library," adding that "one would not expect all proteins coded in the genome to typically be present at all times, let alone all find their way into plasma."
Gonzales said that Olink recommends using all data, including measurements below the LOD, for applications including "exploratory investigations and statistical analyses." However, he said, for technical evaluations, such as those conducted in the Coon study, "only data above LOD should be used, since measuring reproducibility of noise is not an accurate reflection of performance."