With electron transfer-dissociation fragmentation becoming a more standard part of proteomics researchers' mass-spec workflows, the Association of Biomolecular Resource Facilities’ Proteome Informatics Research Group this year evaluated the community's ability to identify proteins based on ETD spectra.
The results of the study, presented at ABRF’s 2011 conference held this week in San Antonio suggest that, while the field has become more adept at working with ETD data, there remains room for improvement. This goes both for individual researchers' skill levels and the EDT options provided by mass-spec instrument and software vendors.
As opposed to the more commonly used collision induced dissociation, or CID, fragmentation, ETD relies on free-radical-driven chemistry to fragment peptide ions for mass-spec analysis.
CID remains the dominant fragmentation method in proteomics research, but recently ETD has grown in popularity due to its ability to retain many post-translational modifications and provide even fragmentation along the entire length of large peptides.
Because ETD generates different fragmentation patterns than CID, however, proteomics researchers have faced a learning curve as they familiarize themselves with the peptide fragments generated by this technique.
The ABRF study provided 35 research groups with ETD fragmentation data generated in yeast lysate on a Thermo Scientific LTQ Orbitrap XL LC-MS/MS, and had the participants identify as many peptides from the set as they could using a 1-percent false-discovery rate.
The results were mixed. More than 2,400 spectra IDs were agreed upon by over half the participants, but there was less consensus on many others. The most prolific team identified over 4,000 spectra and over 2,000 unique peptides, while the least prolific identified just over 500 spectra and around 400 peptides.
On the whole, researchers were "a little too confident about how reliable their results were," said Lennart Martens, iPRG chair and a researcher at Ghent University. This led them to make calls of which they weren't entirely certain and increased the average false-discovery rates from the specified 1 percent into the 2 percent to 3 percent range, he noted.
In making their IDs, the researchers used more than two dozen different protein-identification and database programs, including commercial products like Bioinformatics Solutions' PEAKS software and Agilent's Spectrum Mill; some free open-source products like ProteoWizard; and a number of in-house solutions. It turned out that not one of the tools performed noticeably better than any of the others, Martens said.
Sorting the results by the search engine, it was clear that "it's not so much the search engine as it's the beast behind the search engine," he said. "That seems to make the most difference."
However, that isn't to say that there are no software or instrument improvements to be made with regard to ETD. While the technique works well with particular kinds of peptides – particularly those with high charge densities – it typically does a poorer job fragmenting peptides with low charge densities.
ETD, therefore, is often used in concert with CID, with mass spectrometers deciding based on preset parameters which method to use for fragmentation given the properties of the peptide passing through the system at that moment.
Developed by University of Wisconsin, Madison, researcher Joshua Coon in 2008, this CID-ETD decision-tree technique has only recently become available as a feature on commercial mass specs. And, as Andreas Huhmer, proteomics marketing director at Thermo Fisher Scientific, told ProteoMonitor, "constant improvements are being made" to the logic trees underlying the method.
In a talk he gave after the presentation of the iPRG study results, University of California, San Francisco researcher Robert Chalkley showed mass-spec data from the same yeast sample set that suggested ways Thermo Fisher might further improve their ETD decision trees.
Chalkley ran the sample using five different acquisition methods: On a Thermo Scientific LTQ Orbitrap XL using only ETD; on the XL using its CID-ETD decision tree to switch between methods; on a Thermo Scientific LTQ Orbitrap Velos using only ETD; on the Velos using its CID-ETD decision tree; and on the Velos using its HCD, or higher-energy collisional dissociation, -ETD decision tree.
The Velos using the CID-ETD decision trees was the most effective, acquiring 18,900 spectra from which it identified 4,404 peptides. In comparison, the Velos using HCD-ETD acquired 12,170 spectra and identified 3,124 peptides; the XL using CID-ETD acquired 10,215 and identified 3,933; the Velos using just ETD acquired 9,691 and identified 3,100; and the XL with ETD alone acquired 8,032 and identified 2,511.
In the course of this work, Chalkley identified several points at which Thermo Fisher's CID-ETD decision tree might not be working optimally. In particular, for triply charged precursor ions with m/z between 650 and 900, the company's instruments default to CID. Acquiring such ions with EDT instead, however, Chalkley was able to more than double his spectrum IDs.
He also suggested that the vendor's decision tree should enable ETD fragmentation of doubly charged ions, which it currently doesn't allow. Although ETD has typically worked poorly with doubly charged precursors, Chalkley said that the technique can be effective with such ions if supplemental activation — a process by which extra energy is imparted to aid the EDT fragmentation — is used.
Thermo Fisher researchers have seen Chalkley's presentation, Huhmer said, adding that "we have to go back and look at the data and convince ourselves that there's an improvement. It's a fantastic dataset, so we'll look into it."
He noted that one reason for the discrepancy could be that, while much of Chalkley's work was done on an Orbitrap Velos, the company's CID-EDT decision tree was developed largely on an Orbitrap XL EDT.
"This might be an opportunity to look at the decision tree in the Velos mass spectrometers, because you get an improvement in fragmentation with the Velos," he said. "So it might be an opportunity to improve the decision tree, as well."
There's room for improvement in terms of search engines, as well, Chalkley said. In particular, in the case of doubly charged precursors many search engines use "a poor model to predict how [these] peptides fragment in ETD data," he told ProteoMonitor. "They don't consider some of the fragment ion types that are observed."
He also said that the advantages of EDT for phosphoproteomics — touted by a variety of instrument vendors — have been overblown. While the technique has revolutionized the study of PTMs like O-GlcNac and O-glycosylation, phosphorylated peptides are generally low-charge due to the negative charge of the phosphate, which decreases the efficiency of ETD fractionation.
"Overall the benefits of ETD over CID on phosphopeptides are marginal," Chalkley said, calling efforts to push the technique for that purpose a "sales pitch."
Have topics you'd like to see covered in ProteoMonitor? Contact the editor at abonislawski [at] genomeweb [.] com.