Uppsala University scientists have published a proof-of-principle study in the current edition of PLoS One demonstrating the use of next-generation sequencing as a read-out platform for protein biomarker research.
Called ProteinSeq, the technique uses NGS to quantitate proteins detected via proximity ligation assays, and could offer a more precise, sensitive, and multiplexible alternative to traditional sandwich immunoassays, said Ulf Landegren, an Uppsala professor and leader of the study.
Landegren is also the founder, a board member, and the second largest shareholder of Olink Biosciences, which has commercialized the PLA method for protein biomarker detection and has rights to any technology developed in his lab. Currently, the company uses real-time PCR as the read-out for PLA, but is "very interested in the possibility of using sequencing as a read-out," Landegren told ProteoMonitor.
PLA uses pairs of antibodies attached to unique DNA sequences to detect a protein of interest. When the antibodies bind their target, the attached DNA strands are brought into proximity and ligate, forming a new DNA amplicon that can then be quantified.
Using NGS instead of real-time PCR for that quantification step could enhance the precision of PLAs, Landegren said. NGS provides a digital measurement, he noted. "So now we're counting events instead of measuring levels of fluorescence or the amplification cycles, so I believe this is a strong approach to get very high-precision measurements."
It could also enable easier multiplexing and pooling of samples, allowing researchers to tag different patient samples with different DNA sequences, pool them, and read them out in parallel.
Optimization work will be required for ProteinSeq to fulfill its promise, though, Landegren said. While the method has potential for high precision, it demonstrated an average CV of 22 percent across the set of 35 proteins the Uppsala researchers measured in the PLoS One study. This was lower than the 29 percent average CV obtained using qPCR as the read-out platform, but well above the 10 percent that Landegren said would likely be needed for a clinical assay.
"We're not very proud of our CV in this study," he said. "This is our first proof of principle, so it will all have to be optimized further."
As Larry Gold, CEO of aptamers-based biomarker discovery firm Somalogic told ProteoMonitor in a May interview discussing the potential of NGS for protein detection, the method runs into difficulty when confronted with samples with high dynamic range (PM 5/6/2011).
In order to achieve acceptable CVs, NGS platforms will likely need to measure at least 1,000 copies of the target analyte, he said. In samples with low dynamic range, that doesn't pose a significant problem.
In broad biomarker-discovery screens measuring hundreds of different proteins in blood, however, the large dynamic range means that measuring 1,000 copies of low-abundance proteins will require sequencing billions of copies of DNA linked to higher-abundance proteins.
"It's not obvious that for big plexes [NGS] has an advantage over [DNA] hybridization" techniques like those used by SomaLogic in its biomarker discovery assays, Gold said at the time.
In the ProteinSeq study, Landegren's team tried to address this issue by decreasing the concentration of PLA probes targeting high-abundance proteins, which, he said, lowered the reporting rate in a predictable manner.
"It's a way of reducing sensitivity [to high-abundance proteins]," he said. "TNF-α, for instance, is present [in plasma] at 10-10-fold the concentration of albumin, so you don't want to keep sequencing the tag for abundant proteins billions of times before you find the rare ones, so it's helpful to find that we were able to reduce the sensitivity for abundant proteins."
By varying the concentrations of the PLA probes, "we're able to detect only a fraction of the [high-abundance proteins] but still proportionally to their numbers," Landegren said. "So once we know roughly what concentration the target proteins are going to be at, then we can adjust conditions so that all proteins are going to report at more or less the same rate."
He added that for any assay, "you need to make the [PLA] conjugates and test them on the sample to get a feeling for how abundant the [different] proteins are. And once you know that, then you can reduce the sensitivity by adjusting the concentration of the reagents."
Using this approach, Landegren said he "is fairly confident" the Uppsala scientists will be able to bring CVs for the method down to a level acceptable for clinical assays. He suggested that the falling cost of sequencing will also ease precision issues as it will allow researchers to sequence more molecules.
"At this point that's a relatively expensive way to solve this [precision] problem, but I think that is going to go away," he said.
Other researchers, most notably those at Belgian biomarker firm Pronota, are also looking into NGS as a read-out platform for protein detection.
While Pronota has typically used mass spec for its protein biomarker work, the company has also been investigating using aptamers as protein-affinity probes. These short, single-stranded oligonucleotides act as capture agents for proteins of interest and, after the capture step, can be read by NGS to determine the number of proteins present in a sample.
Last June the company announced that it completed a proof-of-concept study for a protein biomarker diagnostic platform using next-generation sequencing (PM 6/4/2010). In December, a team including then-CEO Koen Kas and several other Pronota researchers published a paper in Analytical Chemistry describing the use of an Illumina Genome Analyzer II sequencer to measure serum levels of endogenous immunoglobulin E captured by aptamers.
Kas stepped down from his position as CEO in August, telling ProteoMonitor at the time that he was leaving to pursue work on a new venture that might also touch on the intersection of NGS and proteomics. He declined to elaborate, however, saying that it was too early to comment further.
For the PLoS One study, Landegren's team used an Illumina Genome Analyzer IIx for the sequencing. His lab also has a benchtop Ion Torrent machine that Life Technologies awarded him in February to support the ProteinSeq work as part of its European Ion Torrent Personal Genome Machine Sequencer Grants Program. Life Tech has licensed the PLA technology from Olink and offers it as part of Applied Biosystems' TaqMan product line.
In the study, the researchers used the ProteinSeq method to analyze sets of 35 proteins in 5 µL of plasma, comparing results obtained on the platform to those obtained using standard sandwich ELISAs. Among the 35 proteins, ProteinSeq showed a lower limit of detection for 27 of them, while ELISAs were more sensitive for eight of the proteins. ProteinSeq had a median dynamic range of five orders of magnitude compared to three orders for the ELISAs.
The increased sensitivity is not due to the use of NGS but rather from the multiple antibodies used for protein detection via PLA, Landegren said. As marketed by Olink, PLA uses paired antibodies in a homogeneous phase to detect proteins. For the ProteinSeq work the researchers used a solid phase assay consisting of capture antibodies immobilized on paramagnetic microparticles plus the paired DNA-linked antibodies, meaning that to be detected a protein needed to be recognized by three different antibodies, as compared to the two antibodies required by traditional PLA and sandwich immunoassays. In related work, Landegren's lab has also recently expanded the PLA technique to require recognition by five antibodies simultaneously (PM 5/13/2011).
Key to further improving the method's sensitivity will be optimizing the antibodies used, Landegren said.
The researchers used polyclonal antibodies from R&D Systems for the study, "and we found that some 80 to 90 percent of those antibodies gave us pretty good results." However, he said that for some of the ProteinSeq assays the detection was worse than in sandwich assays by a factor of around eight, and for others ProteinSeq was better by a factor of 1,000.
"So I think there is still a great difference between the different antibodies, and if you really want to push this and build extremely sensitive assays, then we will have to work much harder to find the optimal reagents."
In addition to comparing the ProteinSeq method to conventional sandwich assays, the researchers used it to examine differential protein expression in the plasma of 63 cardiovascular disease patients and 19 matched healthy controls, identifying cystatin-B, P-selectin, and kallikrein-6 as potential CVD biomarkers.
Landegren said he initially plans to use the technology to do large-scale analyses of biobanked samples in hopes of identifying protein biomarkers for cancer, cardiovascular disease, and neurodegenerative diseases. Currently, his lab has developed around 45 assays for the platform, and, Landegren said, he believes they will be able to multiplex well over 100 assays.
He said that he expects Olink to be the commercial outlet for the technology as it develops, noting that it is "very compatible" with the company's current PLA offerings.
Regarding potential biomarkers identified through his lab's work on the method, Landegren said that while "the business model of Olink is not biomarker discovery, obviously there would be tremendous upside for the company if we were able to identify particularly promising biomarkers."
He added, though, that no decisions had been made about Olink's potential role in commercialization of any biomarker discoveries.
Have topics you'd like to see covered in ProteoMonitor? Contact the editor at abonislawski [at] genomeweb [.] com.