While protein-protein interaction networks have been studied extensively in yeast cells via high-throughput experiments, few corresponding studies with mammalian cells have ever been done.
But a team of US and Canadian scientists have developed a method that they say significantly simplifies the process for analyzing protein interactions in humans, opening a window on the unknown functions of proteins and paving the way for the possible development of drug therapies for disease.
Their work appears in the July 20 edition of Molecular Cell.
The article is part of a broad project to map protein interactions that regulate cell growth, differentiation, and disease progression. Central to the project, called the Human Proteotheque Initiative, is a “discovery engine” that generates maps of protein interactions. The work outlined in Molecular Cell article is the first generation of the platform.
According to Benoit Coulombe, the lead author of the study, identifying protein-protein interactions via high-throughput experimental methods in yeast has become relatively routine, “so there are many proteomic studies that have come out in the past few years in yeast.”
Coulombe is the director of the Gene Transcription and Proteomics Laboratory, and the Proteomics Discovery Platform at the Institut de Recherches Cliniques de Montréal.
It’s a different story with mammalian cells, however. In order to identify protein interactions, the proteins first have to be tagged, and then pulled out. The intrinsic interacting partners of the proteins are then identified by mass spectrometry.
It is “very easy” to replace yeast genes with a tagged version, Coulombe told ProteoMonitorlast week, “so this tagged protein will be expressed at physiological levels in yeast. … It’s expressed the same way as the endogenous [gene].”
But in human cells, that cannot be done, he said. To express proteins in human cells at normal levels, an inducible system in which plasmid can be added to allow expression of the protein is necessary “because you cannot make replacements in your human cell as easily as with yeast,” Coulombe said.
The expression system used by his team involved cloning human cDNA into an ecdysone-inducible expression vector that encoded the tandem affinity purification tag. Transfection in EcR-293 cells and selection of expression clones were then performed, followed by induction with ponasterone A.
The researchers identified interaction partners of proteins, tagged them and sent them through the same inducible system “thereby enriching the interaction dataset,” they said in their report. Thirty-two tagged proteins were used, leading to 170 successful affinity purifications.
Algorithm To Predict Interactions
To maximize the quality of their dataset, Coulombe and his team also developed a new computational algorithm. The first thing the algorithm does is filter out non-specific interactions “caused by proteins that bind nonspecifically to our columns, and very abundant cellular proteins that may have remained as contaminants after affinity purification” the researchers said.
The algorithm also allowed the researchers to assign a reliability score for each interaction, making it possible for them to filter out true interactions from likely false positives and false negatives.
According to the researchers, the algorithm achieved a specificity and sensitivity of 83 percent each.
“This is very good for mammalian cells,” Coulombe said. “It’s a very low level of false positives and false negatives.”
Through mass spectrometry, he and his colleagues were able to find 2,008 putative protein interactions. By applying their reliability score, they were able to shave that number to 805 interactions that they deemed “highly reliable,” involving 436 proteins.
“We’re at the beginning of [understanding what happens when proteins interact with each other]. What we see is a kind of snapshot, a picture of the average of the interactions that are going on in cells. It’s much more complicated than we imagine,”
The researchers say that with their method, many proteins of previously unknown function can now be assigned putative functions based on their associations. To confirm that physically interacting proteins are also functionally related, they further analyzed proteins they found in their network and discovered an enzyme, which they named MePCE, that regulates the stability of small RNA molecules.
The existence and importance of the enzyme, Coulombe said, had been known in the scientific community for more than a decade, but had been virtually impossible to isolate in protein complexes.
“Our results indicate that siRNA-mediated silencing of MePCE expression reduces the steady-state level of cellular 7SK [small nuclear RNA], suggesting that MePCE protects 7SK from exonucleolytic degradation,” the researchers said.
Coulombe and his team are now working on a network of 100 proteins compared to the 32 they analyzed in the Molecular Cell article. Included are regulatory viral proteins.
“We want to know how these viral proteins interfere or connect with this network of interactions,” Coulombe said. They have also further refined their computational algorithm and raised the specificity and sensitivity levels to up to 95 percent.
They will be ready to present their findings in about a year, Coulombe said.
Just Like Google, For Protein Interactions
Guiding the work is Coulombe’s goal of the Human Proteotheque Initiative. By mapping protein interactions, he hopes to build a technology platform that he compares to the Google search engine. Just as Google uses a complex algorithm to weed out irrelevant search results while gathering relevant information, Coulombe wants to create a protein interaction search engine that will present only relevant results.
Coulombe acknowledges HuPI is an ambitious project whose undertaking could outlast him. Little remains known about protein interactions in humans due to their complexity.
“We’re at the beginning of [understanding what happens when proteins interact with each other]. What we see is a kind of snapshot, a picture of the average of the interactions that are going on in cells. It’s much more complicated than we imagine,” Coulombe said.
While he foresees his research having use in drug development, Coulombe said he is not interested in pursuing that line of research for now.
There are people “much more qualified to do that than I am,” he said. “My plan is to give them the data.”