Computational systems biology is giving oncology research a boost at Johnson & Johnson Pharmaceutical Research and Development, where scientists are using technology from Gene Network Sciences, Ariadne Genomics, and GeneGo to elucidate the mechanism of action of potential therapeutics and point the way toward biomarkers related to drug efficacy and response.
Timothy Perera, a principal scientist in the oncology research early development group at J&J PRD Europe, presented early results of this work at the annual Cancer Progress conference in New York on March 8.
Perera said that the J&J PRD oncology team has been using Affymetrix arrays to study gene expression patterns to help characterize the biological behavior of kinase inhibitors, but realized that this approach is limited because traditional gene expression analysis tools can only identify sets of co-regulated genes — not causal relationships. While mapping these genes onto known pathways provides additional insight, these pathways may not even be relevant to the mechanism of action of a novel compound, he noted.
In addition, the J&J PRD team is primarily working with multi-targeted kinase inhibitors, "which means that it becomes difficult to really understand what the true mechanism of action is — which kinase is the most relevant kinase," Perera said.
|
"One thing we are finding is that when [computational systems biology] is used on top of the standard techniques that we use, it's almost like a lens that allows us to look down at the mechanism." |
In response, Perera and his colleague Andrew Stubbs, principal scientist, functional genomics, turned to systems biology with the goal of "building a model to identify those pathways that differentiate compounds."
Working with GNS, the team ran a series of gene expression experiments for a library of compounds that were also against a panel of 200 kinases to generate data that they used to reverse-engineer a network model of compound action. Singling out one compound, dubbed JNJ1, Perera and his colleagues then used the completed model to predict JNJ1's mechanism of action in a series of in silco experiments.
These simulations were also used to identify genes that appear to play a role in the compound's ability to impact proliferation, which could serve as drug efficacy biomarkers. Likewise, the model identified genes that predict the response of a cell to therapy, which could be used as prognostic biomarkers to identify responders or non-responders to particular therapies.
So far, J&J PRD has used the GNS simulations in combination with pathway data from Ariadne and GeneGo to winnow an initial list of around 900 genes to 57 potential efficacy biomarkers and 28 potential prognostic biomarkers.
In a phone interview following his talk, Perera stressed that the work with GNS "is still an evaluation — it's a proof-of-concept study," and that J&J PRD will have to confirm the model's predictions experimentally and statistically before it decides where computational systems biology fits within the company's broader discovery toolkit. Nevertheless, he and Stubbs expressed optimism for what they've seen so far.
"One thing we are finding is that when this is used on top of the standard techniques that we use, it's almost like a lens that allows us to look down at the mechanism," Stubbs said. "We may find many genes that change with a compound — that are related to the compound effects — but this allows us to focus in on those genes that are directly related to the mechanism and its effect on that particular phenotype."
|
"We may find many genes that change with a compound — that are related to the compound effects — but this allows us to focus in on those genes that are directly related to the mechanism and its effect on that particular phenotype." |
In addition, Stubbs said, the ability to predict dose effects could prove to be another advantage, since it's possible to predict the effect of compound treatment on gene expression entirely in silico prior to wet lab experimental validation.
Perera said that J&J won't judge the usefulness of these methods solely by the downstream success of the compound, but rather by how well the compound's behavior in the wet lab matches its predicted characteristics. "I think the technology itself has been validated to a certain degree," he said. "If the compound fails because it's got poor bioavailability or whatever, it's not the fault of the technology that we used."
The researchers also see promise for computational systems biology in areas outside of oncology. "We don't see any reason at all why it shouldn't be applicable to other areas as well," Perera said. "If the information content is out there or can be generated, then I don't see why it can't be applied to any other disease area at all."
— Bernadette Toner ([email protected])