NEW YORK (GenomeWeb) – Researchers at Harvard Medical School have generated a protein interactome based on the proteomic profiles of 41 breast cancer cell lines.
In a paper published this week in Nature Biotechnology, they used this data to identify established breast cancer subtypes and predict the sensitivity of specific lines to drug treatment.
The study also demonstrates the large-scale use of protein co-regulation analysis to establish protein-protein interactions and associations, which could prove a higher-throughput alternative to interaction mapping approaches like affinity purification mass spec (AP-MS), said Wilhelm Haas, assistant professor of medicine at HMS and senior author on the paper.
Interest in protein interaction mapping has grown among proteomics researchers as the field has come to explore not just the detecting and quantitation of discrete proteins, but their behavior within the larger complexes and pathways in which they exist and through which they function in vivo.
A primary method for studying protein interactions is AP-MS, wherein target proteins are pulled down using antibodies or another affinity reagent and then they and attached proteins are identified via mass spec to determine interactions. As Haas and his co-authors noted, however, this is a laborious process, and generating a proteome-wide interactome can take years. Additionally, such an interactome provides only a static look at protein interactions in one sample under one set of conditions. Generating comprehensive AP-MS-based interactomes looking at interactions across a number of conditions "currently seems out of reach," they wrote.
Instead, Haas and his colleagues applied protein co-regulation analysis to determine interacting proteins in the 41 breast cancer cell lines they analyzed. Such analysis relies on the fact that interacting proteins are co-regulated to maintain appropriate stoichiometry for the complexes and pathways they are a part of. If expression of a particular protein is depressed, for instance, levels of its interactors would likewise be downregulated to maintain the proper proportions.
This phenomenon allows researchers to identify likely interacting proteins by looking for molecules whose changes in expression are correlated.
Using an Orbitrap Fusion and 10-plex TMT tagging measured at the MS3 level, the HMS researchers quantified the 41 cell lines in duplicate, making for 82 total samples, which they ran across 11 mass spec experiments. They quantified 10,535 proteins across the 11 experiments, with an average of 9,115 proteins quantified in both replicates of each cell line and a total of 6,911 proteins quantified in all of the cell lines. They used this latter set of proteins for their co-regulation analysis.
Clustering the cell lines based on this analysis, they found that they grouped into the known breast cancer subtypes — luminal, basal, claudin-low, and nonmalignant. They also found that the protein-based subtyping matched subtypes determined based on mRNA data.
Comparing their interactome to the AP-MS-based Bioplex protein interaction database, they found that 18 percent of the known and 3 percent of the unknown interactions identified in their study matched the Bioplex dataset. Comparing a different large-scale AP-MS interaction dataset to the Bioplex database, they found 22 percent of its known interactions and 4 percent of its unknown interactions matched those in the Bioplex set, suggesting, Haas noted, that their study achieved similar overlap in terms of identified interaction as do typical large-scale AP-MS studies.
Interestingly, a comparison of known protein interactions in the CORUM and STRING databases, which compile high-confidence protein associations, to interactions predicted by the protein co-regulation analysis versus interactions predicted by mRNA data found significantly higher overlap for the protein co-regulation-based predictions.
"Protein co-regulation analysis identified known protein-protein associations in around 40 percent of cases, a relatively high overlap," Hass said. "And this dropped to about 4 percent when we looked at mRNA."
This, he noted, suggests that something is happening at the post-transcriptional or post-translational level to co-regulate protein levels. That matches with findings from past studies including by the National Cancer Institute's Clinical Proteomics Tumor Analysis Consortium that have found that while correlation between gene copy number variations and RNA levels is relatively good, this correlation "goes pretty much to zero when you look at proteins," Haas said.
One theory, he said, is that proteins that are not part of complexes will be degraded at higher rates, which would serve to remove excess protein from the system. He added that this could apply not only to proteins that are physically interacting, but also to proteins that exist in a pathway where different ligands passed from protein to protein transmit signals throughout the network.
Genetic mutations could dysregulate protein function by inhibiting protein interactions, which could create changes in protein levels that could ripple throughout a larger network. For instance, if a given protein is mutated such that one of its interactors no longer binds to it, then that interacting protein would be more susceptible to degradation, resulting in lower levels in the cell.
Typically, Haas said, cells are able to compensate for such perturbations by using alternate pathways. "But when you have more and more dysregulated protein interactions, the more you will run out of compensating pathways," he noted.
To see if they could use information on dysregulated protein interactions to predict drug response in the cancer cell lines, Haas and his colleagues ran a GO analysis on the dysregulated protein interactors in each line to identify functional modules particularly affected in different samples. They then screened the cell lines against 195 drugs and identified six agents, all known to target elements of cell cycle regulation, that had a significantly stronger effect on cell lines with dysregulated cell cycle interactions.
The results suggest that such analyses could be useful for predicting treatment based on dysregulated protein interactions, Haas said, adding that he and his colleagues are currently planning experiments to test this idea in primary tumor tissue from actual patients.
Such work will be more complicated than the cell line study due to issues like tumor heterogeneity and stromal tissue that could add noise to the analysis, he said. But, he added, "the data we have so far is promising."