NEW YORK (GenomeWeb) – A team led by researchers at the University of Montana and Cell Signaling Technology has combined large-scale protein post-translational modification (PTM) datasets with protein-protein interaction (PPI) data to profile cell signaling in lung cancer.
The research, published this week in Science Signaling, provides insights into the roles of post-translational modifications in cancer processes as well as the mechanisms of action of several cancer drugs, said Mark Grimes, an associate professor at the University of Montana and the first author on the paper.
In the study, the researchers analyzed data on protein phosphorylation, acetylation, and methylation collected in 45 lung cancer cell lines, comparing it to normal lung tissue and cell lines treated with the cancer drugs crizotinib, gefitinib, imatinib, and geldanamycin. They combined this with PPI data collected from public databases.
The PTM data was generated several years ago by Cell Signaling Technology, which, Grimes noted, has developed a number of PTM-specific antibodies for such analyses. The company built the dataset by coupling immunoprecipitation using those PTM-specific antibodies with multiplexed mass spec using isobaric labeling to allow for simultaneous measurements of protein phosphorylation, acetylation, and methylation. Adding this data to PPI information, the researchers hoped to better define cell signaling pathways of importance to lung cancer.
The basic notion underpinning the effort was that by using the large PTM dataset, the researchers could identify protein PTMs that clustered together. They could then apply that clustering data to the PPI data to identify proteins that were both known to interact and whose modifications clustered together, suggesting that they are part of the same signaling network.
"If you know that the proteins interact, and you know that their PTMs are robustly related statistically, then that gives rise to the notion that these are real signaling pathways," Grimes said.
To build the PTM dataset, CST researchers compared the 45 lung cancer cell lines to normal lung tissue across nine multiplexed mass spec runs. They also compared cell lines treated with the four anticancer drugs across six multiplexed runs. In total, they identified 13,798 modifications across 90 samples.
One issue for shotgun mass spec workflows looking at protein expression or modifications, or both, across large numbers of samples is missing data. Because the instrument selects peptides for sampling more or less randomly in each experiment, the same peptides are not sampled every time, leading to gaps in the data.
The CST dataset was missing data for 78 percent of the PTMs measured across the different samples, which, Grimes noted, presented a major challenge for the researchers' data analysis.
"Because this is mass spectrometry data, it has holes in it," he said. "We had a lot of missing values. So we had to be very careful about how we analyzed the statistical relationships among the PTMs."
Grimes and his colleagues evaluated several different approaches for clustering the PTMs, ultimately selecting a method using t-distributed stochastic neighbor embedding (t-SNE), based on its ability to identify meaningful clusters of PTMs as judged by "the uniformity and density of each cluster, and the number of prior knowledge PPIs found among the proteins within the identified clusters," the authors wrote.
"Once we knew that we could cluster the PTM data effectively, then combining it with the PPI network data [took the analysis] a step forward," Grimes said.
He noted the example of the drug-treated samples and how combining the PTM and PPI data helped the researchers map out the signaling pathways involved in drug response.
"If we know what the targets of the drugs are, then we can use the [PPI] network to trace the shortest paths from the target to those proteins whose PTMs were affected," he said, adding that their work mapped known signaling pathways involved in response to the kinase inhibitors crizotinib and gefitinib.
Applying their analysis to the heat shock protein 90-inhibitor geldanamycin, a drug with a mechanism of action that is incompletely understood, the researchers found that beyond heat shock proteins, the drug also affected a number of proteins involved in endocytosis and cytoskeletal dynamics, which, they noted, is consistent with previous studies indicating the drug alters these processes. They also found that treatment with geldanamycin affected phosphorylation of the kinases ALK, AXL, EGFR, ERBB3, and IGF1R.
"Geldanamycin inhibits heat shock proteins, but these [HSPs] connect to many other proteins in the cell," Grimes said. "They regulate not just the folding of many kinases, but also their activity, in a way that's not completely understood." Through the experiment, "we've identified networks that link the heat shock proteins to these other things," he said.
More generally, the researchers observed an antagonistic relationship between phosphorylation and acetylation.
"When phosphorylation was going down under drug treatment conditions, [often] the acetylation on that same protein was going up," Grimes said.
Previous studies have found indications of such a relationship, he said, but it has been little explored, largely because few studies have looked simultaneously at multiple modifications across a number of samples and conditions on a proteome-wide scale.
"We saw it under a variety of conditions," he said. "It needs to be more thoroughly tested, certainly, but I think it could be true."
He added that the raw dataset is available to other researchers interested in exploring it and that his lab has developed a web application that allows users to look up PTM and network information for particular genes of interest.
Grimes said he is personally interested in using the dataset to further explore the signaling processes underlying the development of cells into cancer cells.
"I got into this project by studying neuroblastoma and neuroblastoma phosphoproteomics," he said. "We're still interested in that, and we've started working with human embryonic stem cells and looking at what makes them differentiate into neural crest cells in the context of neuroblastoma. There are some lung cancers that are also derived from the neural crest, so it's relevant to a number of neuroendocrine tumors and other kinds of cancers."