Scientists from Abbott Laboratories' pharmaceutical-discovery division have released kinomics screening data about how 3,800 different inhibitors affect 172 protein kinases.
In a study published last month in the online edition of Nature Chemical Biology, the researchers showed how they tried to group these kinases based on both sequence and pharmacological relationships and by their interactions with various inhibitor chemotypes.
They also developed a system for statistical interpretations of these relationships, which, according to their paper, could provide a framework to help researchers better predict drug leads, [and] improve drug makers' ability to identify compounds "with a high likelihood of targeting disease-related kinases while avoiding kinases associated with clinical liabilities."
The Abbott team was unavailable to comment on the work, but David Simmons, chief scientific officer of biotech firm Cellzome, which has developed its own kinase-binding matrix technology called Kinobeads, called the study "a tour de force."
"They've done a very impressive job," he told ProteoMonitor. "What's new about this paper is the sheer number of different kinds of analyses they've done coupled with the totality of the dataset. A lot of other people have done either single analyses or used a smaller dataset, but I think this is the biggest dataset with the biggest number of comparisons."
The study is "quite an impressive feat," agreed Steven Pelech, president and chief scientific officer of Kinexus Bioinformatics, which offers a variety of tools related to protein-kinase research, including its PhosphoNet KnowledgeBase featuring information on 93,000 phosphorylation sites in about 14,000 human proteins.
"It's quite significant in that they tested so many protein kinases side by side against so many different compounds," he told ProteoMonitor.
The research offers insights in two main directions, Simmons said. On the one hand, the data shows which chemotypes tend to be the most promiscuous inhibitors – binding to a wide variety of kinases – while on the other it demonstrates which kinases are the most promiscuous – binding to a wide variety of inhibitors.
Ideally, target kinases and potential inhibitors will both exhibit low promiscuity, allowing researchers to avoid off-target binding that could result in unwanted side effects. This makes having good selectivity data on protein kinases and their inhibitors essential to drug development.
In the paper, the Abbott researchers cite targeted polypharmacology – the development of compounds that can selectively inhibit multiple protein kinases – as their ultimate goal. This approach, they said, "holds significant potential in the identification of treatments for highly complex diseases" where it may be desirable to modulate more than "a single, isolated molecular target."
However, because this aim carries the risk "of serious selectivity issues that can translate into toxic side effects," more effective ways of predicting kinase-ligand binding are needed.
In addition to the kinase values dataset presented in the paper, the researchers describe a statistical framework they've developed to better assess the relationship between various protein kinases. "Data incompleteness and the significant changes in interaction networks that can occur as new data become available … represent a serious limitation to productive use of pharmacology networks in drug discovery," they wrote.
When evaluating pairs of kinases, the addition of new data can often change their perceived relationship, revealing that proteins thought to be similar actually behave quite differently, or vice-versa. Currently, the Abbott team noted, "no criteria have been described for understanding the information content in such relationships."
By measuring the information content contained in the pK values of kinase pairs using scaled Shannon information entropy, the researchers developed a threshold for what levels of data are needed to make robust claims regarding the relatedness of kinase pairs, proposing in the paper a minimum level required before "kinases can be confidently classified as 'related.'"
Kinexus' Pelech questioned the researchers' stated goal of targeted polypharmacology, suggesting that it would be too difficult to anticipate all off-target effects generated by such an approach.
However, he said the database did provide "a strategy for drug discovery" by offering leads to "structures that you may want to try testing first to cut down on the number of different compounds you have to screen to find a hit."
The data also allows researchers to identify particularly promiscuous compound chemotypes they might do well to avoid, as well as promiscuous kinases that could prove difficult to target, Cellzome's Simmons said. He added that he thinks the dataset "will be a very valuable tool."
"If you look at the supplementary figures, there are all the files someone would need to pretty much put the data together," he said. "It will take a fair amount of work, but it's a publicly useful thing they've done to put this out there."
Simmons noted that last December, Bristol-Myers Squibb published a similar kinase dataset in the Journal of Medicinal Chemistry. That paper included information on more than 20,000 compounds screened against 402 protein kinases. He suggested that there would likely be more releases from pharma firms of large kinase-screening datasets.
"Any large pharma company has been doing a lot of kinase screening over the last 15 to 20 years," he said, "and now they’re probably all pulling together this kind of data. They've probably kept hold of these for several years and mined the utility out of them for their own internal programs, and now the publications [based on the work] are starting to come out."
One thing such datasets don't do, of course, Simmons said, is illuminate the actual biological roles of the kinases being studied. Some of the more prominent failures in kinase-based drug development have come from inhibiting particular proteins without fully understanding their role in the networks being targeted, he noted, citing the example of p38 MAP kinases.
"Of all the vast money spent on p38, some of it has come to nothing because it turns out that if you inhibit p38 you actually take off the brake it normally provides on the pathway [being targeted]," he said. "That actually comes out of cell biology experiments, not trolling through the kinase selectivity or sequences. So there's a sort of biological database that's orthogonal to this that's necessary to tell you whether a kinase is a good kinase or not."
Have topics you'd like to see covered in ProteoMonitor? Contact the editor at abonislawski [at] genomeweb [.] com.