Using a two-stage discovery strategy guided by cancer genetics data, ETH Zurich researchers have identified four protein biomarkers linked to prostate cancer.
The study, which was detailed in an article published this month in the online version of the Proceedings of the National Academy of Sciences, points to a potential way forward for protein biomarker development, Ruedi Aebersold, professor at the Swiss Federal Institute of Technology's Institute of Molecular Systems Biology, told ProteoMonitor, suggesting that the discovery pipeline must evolve from broad shotgun-style serum proteome comparisons to more targeted approaches guided by the underlying biology of the diseases under investigation.
The research also serves as a demonstration of a biomarker discovery platform under development at biotech firm ProteoMedix – an ETH Zurich spin-off co-founded in 2010 by Aebersold and several of his colleagues to commercialize protein biomarker diagnostics based on the targeted strategy described in the PNAS paper. Since launching last year the company has raised $133,000 in venture funding and is currently collaborating with researchers at the University of Applied Sciences Northwestern Switzerland on a clinical prototype of a prostate cancer diagnostic based on the work.
"The basic observation is that there has been an enormous amount of resources expended to compare serum samples from healthy and disease-affected individuals, and that these efforts have largely not been successful," Aebersold said. "It's not that we don't find differences, but that we find too many differences to make sense of the data, and we find differences in many proteins that are not likely to be relevant to the disease."
What's needed, therefore, are strategies for narrowing the pool of candidate proteins prior to mass spec analysis. To this end, the ETH Zurich researchers relied on cancer genetics data to guide their search, generating mice with prostatic PTEN mutations – which are involved in roughly 80 percent of prostate cancers – and then using label-free mass spec on a Thermo Fisher Scientific LTQ-FT instrument to examine protein expression differences in prostate tissue taken from wild-type and PTEN-knockout animals.
That left them with several hundred candidate biomarkers that they then analyzed according to PTEN dependency, prostate specificity, and detectability in serum, further narrowing the list down to 126 proteins. They then searched for these proteins in sera taken from 143 prostate cancer patients and controls, identifying 39 proteins that could be consistently detected and quantified.
Using prostate tissue taken from the same set of cases and controls, the researchers also analyzed the patients' PTEN expression levels, combining these two datasets to determine which of the 39 candidate proteins best predicted PTEN status, ultimately arriving at a four-protein signature composed of HYOU1, asporin, cathepsin D, and olfactorymedin-4 that, when combined with prostate specific antigen, diagnosed prostate cancer with sensitivity of 85 percent and specificity of 79 percent compared to sensitivity of 78 percent and specificity of 63 percent for PSA alone.
According to the PNAS paper, the panel could potentially be used as an early detection test accompanying PSA testing, "reducing false-positive outcomes and therefore avoiding anxiety and biopsies in men who have an elevated PSA but do not harbor cancer.”
The study, Aebersold noted, looked only at N-linked glycoproteins – a limitation imposed by the sensitivity constraints of MRM mass spec. In order for that technique to quantitate proteins in the nanogram-per-mL range – the concentration at which most clinically useful biomarkers exist in serum – peptide enrichment steps are needed, and the enrichment technique selected by the ETH Zurich team – a hydrazine chemistry-based approach originally developed in Aebersold's lab – works only with glycoproteins.
That enrichment decision was "strictly pragmatic," Aebersold said, "because it allows us to get to the required concentration range," although he noted that given that glycoproteins comprise roughly 80 percent of the protein biomarkers currently in clinical use, the choice "is probably not a senseless compromise."
Beyond the limitations imposed by the glyco-enrichment, in general targeted strategies like that used in the PNAS study run the risk of missing potential protein markers not predicted by cancer genome data, Aebersold admitted. However, he said, "I think there is no alternative."
"If we continue to forever to compare serum samples, I don't think it will be successful," he said. "I think it is a very sound assumption that proteins changing in tissue are changing for a reason, and this reason is the [genetic] lesions incurred by the [tumor] cells."
More narrowly focused discovery platforms could also play an important role in moving proteomics into the clinic, he suggested. While there are hopes that automation of assays like SISCAPA could make mass spec a more clinically viable technique (see story this issue), ELISAs will likely continue to dominate the verification and validation arenas for the foreseeable future, Aebersold said. And given that generating a new ELISA can cost as much as $1 million, better strategies are needed for narrowing down candidate panels.
"If you come out of a screen with 50 proteins of interest, it's simply not a tractable task to generate an ELISA for each one," he said. "What we maintain here is that by going through the mass spectrometry to scale up from a few cases to a few hundred, a number of proteins will fall out, and we will have a somewhat solidified marker signature consisting of a few proteins. And then the justification to spend the investment to generate an ELISA is stronger."
The ETH Zurich researchers are now applying their process to several other cancers, including ovarian and pancreatic cancer. While the prostate cancer work involved hunting for candidate proteins in mouse models, the team is now exploring bioinformatic approaches wherein, Aebersold said, "we get [cancer] genomic sequences and from there try to computationally infer which signaling networks are perturbed and based on these networks generate hypotheses about which cell and tissue glycoproteins are affected."
Other scientists are using cancer genetics data to guide their biomarker research, as well. Perhaps most notably, the second phase of the National Cancer Institute's Clinical Proteomic Technologies for Cancer initiative – a five-year, roughly $100 million effort announced in June – calls for researchers to look for potential protein biomarkers in samples characterized by the NCI's Cancer Genome Atlas project, similarly using that genetic data to guide the discovery process (PM 07/09/2010).
Aebersold said, however, that he doesn't think proteomics as a field has fully embraced the notion of targeted discovered yet.
"There are relatively few studies that go in that direction," he said. "I think most proteomics researchers try to use more powerful mass spec instruments to go deeper into the serum proteome. I don't think that's the problem. I think the problem is that by doing all the serum comparisons in an unbiased manner we simply end up with too many [protein expression] changes we cannot associate with the disease."
Have topics you'd like to see covered in ProteoMonitor? Contact the editor at abonislawski [at] genomeweb [.] com.