In papers published simultaneously in last week’s Nature, MDS Proteomics and Cellzome publicly described for the first time their approaches for applying proteomics to understanding cellular function, and by extension, to drug discovery.
While many scientists may be familiar with the two companies’ general strategies, the disclosure is significant because it represents the first peer-reviewed presentation of their technology platform, and a demonstration in the yeast Saccharomyces cerevisiae of what may in the future be possible to learn from the human proteome.
“The paper illustrates that this isn’t just hypothetical,” said Mike Moran, MDS Proteomics’ chief scientific officer and a co-author of the MDS paper. “We built a platform to study human proteins and human protein interactions and just decided to do a test run on yeast, partly [because] we’d be more likely to share that information publicly.”
Pilot project or not, the two companies’ research using yeast shows that their methods have reached a certain level of maturity. Both groups claimed to have used “bait” proteins to pull out interacting proteins representing about 25 percent of the yeast proteome, perhaps the largest analysis of protein complexes completed to date.
The similarities extend to the two groups’ methods. Both rejected yeast two-hybrid systems in favor of tagging techniques to pull out complexes of proteins from cell lysates, and combined their tagging approaches with mass spectrometry to identify the components of each complex. Abandoning the yeast two-hybrid approach was justified, the MDS and Cellzome researchers argued, because their methods identified a much larger fraction of the yeast protein-protein interactions described in the literature.
In fact, the MDS researchers stated that on average the interactions they identified were three times more likely to match those in the literature than those in published yeast two-hybrid studies. Researchers at Cellzome, for their part, claimed their interactions match 56 percent of the interactions described in the yeast protein database (YPD), compared with an overlap of only 10 percent when considering yeast two-hybrid reports.
“When we compare our data to binary interaction data like yeast two-hybrid [data] the overlap is very poor,” said Anne-Claude Gavin, a Cellzome scientist and the first author of the Cellzome paper. “For us a protein complex is more than a simple sum of interactions. You [might need] cooperative binding to have an interaction between two proteins [because] you would need a third partner there.”
Little Differences Count
Upon closer inspection, however, the MDS and Cellzome groups do not exactly see eye-to-eye. For one thing, the two groups used different methods to pull out the protein complexes from the yeast cell lysates. Whereas Cellzome opted for a two-step purification method using a TAP (tandem affinity purification) tag, the MDS scientists pulled out protein complexes using the Flag epitope tag in a one-step immuno-affinity purification.
The difference is significant, Gavin said, because MDS’ one-step method requires that the tagged bait protein be overexpressed in the yeast cells, a condition that may allow the tagged protein to migrate away from its typical cell compartment. “If it’s a nuclear protein and you overexpress it, it can also become cytoplasmic if you really saturate the system,” she said, leading to observed protein interactions that do not actually occur in vivo.
MDS scientists countered that their one-step method is superior because it involves fewer dilution steps, which would tend to lower the technique’s sensitivity. Having just one step, Moran said, makes the MDS method more sensitive to “transient-type” interactions, while the Cellzome approach “may be better for generating data on more stable interactions.”
In addition, the two groups chose competing mass spectrometry techniques for identifying the constituents of the protein complexes. MDS scientists, using Finnigan LCQ Deca ion traps, applied tandem mass spectrometry to identify the amino acid sequence of every digested peptide from the complexes. Cellzome researchers, on the other hand, relied on ABI Voyager DE-STR MALDI-TOF mass spectrometers to perform peptide-mass-fingerprinting on the yeast peptide fragments.
“We abandoned MALDI because it’s really only useful when you’ve got a fully annotated genome as in yeast. When you have really complex mixtures of proteins and an unannotated database as in the human context, a MALDI platform just won’t work.”
Gavin agreed that studying human samples requires tandem mass spectrometry, and said Cellzome has already developed capabilities in LC/MS/MS mass spectrometry. Using MALDI in the yeast study was just a shortcut, she said.
Myriad Lies in Waiting
Although MDS and Cellzome have grabbed the spotlight for the moment, they aren’t the only companies attempting to trace the interactions in the human proteome. Myriad Proteomics, a joint venture between Myriad Genetics, Hitachi, and Oracle, has also promised to “map” the human proteome within a few years time. In fact, last week Myriad Genetics’ CEO Peter Meldrum said the company’s scientists will move into their new proteomics laboratory by the end of January.
Myriad, for it’s part, hasn’t yet abandoned the yeast two-hybrid approach, preferring instead to combine the much-maligned method with the affinity purification techniques championed by Cellzome and MDS. From Myriad’s perspective, two approaches are better than one, but making the jump to human systems is still the ultimate challenge:
“[The MDS and Cellzome scientists] have validated the use of mass spectrometry in protein-protein interaction mapping at a systems scale,” said Jay Boniface, senior director of protein science and proteomics at Myriad. “Such a detailed study in humans will be far more than an order of magnitude more difficult.”