A lingering bottleneck in getting a biomarker to the clinic and into widespread use is the lack of good methods to verify candidate markers from discovery experiments. A new computational approach developed by Vincent Fusaro, a postdoc at the Harvard Center for Biomedical Informatics, will allow researchers to better design mass spec assays for biomarker verification. New methods are sorely needed: only five new protein biomarkers have been approved by FDA in the last five years for measurement in plasma or serum.
"The main limitation of all these discovery 'omics experiments is that the 'ome that we're probing has very high dimensionality, and the number of samples that we're able to analyze is relatively small," says Steven Carr, who leads the proteomics platform at the Broad and co-supervised Fusaro on this project. "The result is that you cannot take information directly out of a discovery 'omics experiment and say you've found biomarkers. All you have identified are candidates, and you must do additional experiments in order to verify those markers before you can start calling them biomarkers."
Antibody-based approaches are still used, but assays like ELISA are far from perfect at this crucial game. "The content of the catalog of antibodies that you'd like to have is incomplete. So there is a need to bridge that gap between discovery and the clinic," he adds.
The core technology that has emerged for verifying candidate biomarkers is stable isotope dilution-multiple reaction monitoring-mass spec, where peptides representing the protein of interest are detected and quantified against the signal response of stable isotope-labeled control versions of those peptides. MRM-based assays begin with the selection of a subset of -peptides, referred to as signature peptides, which are used as "quantitative surrogates for each candidate protein," Carr says.
Typically, coming up with a handful of good signature peptides is done by parsing experimental data or databases, like the Global Proteomic Machine or the Peptide Atlas. The main criteria for selecting signature peptides are high response, or high ion current signal intensity, and uniqueness in the genome, says Carr. "A good signature peptide will allow you to configure an MRM assay with sensitivity sufficient to assay the candidate protein down to the high-picogram to low-nanogram-per-milliliter level, in the context of plasma," he says. "A poorer-quality signature peptide will not have as good sensitivity, linearity, or dynamic range."
Often, however, experimental data has missed detecting the majority of peptides in a sample, or database information is incomplete. "Having a means of predicting, from sequence alone, which peptides for any given protein are likely to work well for targeted mass spec assay development is a crucial missing link in the biomarker development pipeline," Carr says. This is where the enhanced signature peptide predictor comes in.
The ESP predictor is a classification algorithm that evaluates 550 physicochemical properties for a peptide and predicts the likelihood that the-peptide will generate a high response in the mass spec. Carr and colleagues trained the ESP predictor to distinguish between high-responding peptides and low-responding peptides. "The ESP predictor reports a probability of high response for each peptide," he says. "We typically rank the peptides from high to low probability and select the top five peptides to develop a targeted mass spec-based assay." When they compared using the ESP predictor to using database information, the ESP predictor was more successful in predicting signature peptides.
"The ESP predictor is not just for clinical proteomics, but is useful for any case where a targeted, quantitative assay for a protein is desired — a general need in many biological applications of proteomics," Carr says.