The Model Approach to Drug Development

May 6, 2013 | In a presentation at the Bio-IT World Expo last month, Anna Kondic of Merck & Co. highlighted some recent successes and shortfalls of computer models in the development of cancer drugs.

The statistician George E.P. Box famously summarized a common complaint of modeling and simulation, saying “All models are wrong, but some are useful.” Successful models, according to Kondic, Principal Scientist of Merck Modeling and Simulation Division, should be able to mimic how a system acts when it’s unperturbed, and then “predict how it will react to a drug.” While many bioinformatics models have been successful in determining the mechanism behind the effects of certain drugs, few have yielded clinically-actionable results. 

“Where we have not been successful in the modeling and simulation world is making decisions,” Kondic said.  

Systems biology is particularly concerned with creating computer models that replicate properties or cells, tissues, and entire organisms. These models, when made correctly, are especially useful for drug developers. When current estimates place the cost of developing a new drug in excess of a billion dollars, a process that can span a decade or more, models that help scientists make smarter decisions during drug development can mean saved time and money to pharmaceutical companies.

These models can potentially be applied to many of the steps in drug development. As Kondic put it, “We need models that can tell us the right pathway” for the drug to target, the “right target” within that pathway, the “right molecule” to target it with, the “right dose” for that molecule, and the “right patients” for that drug to treat. 

In her talk, Kondic presented three models that were successful in various stages of drug development. In one example, Kondic spoke about the development of imatinib—Gleevec—Novartis’ blockbuster drug for treating chronic myelogenous leukemia (CML).

Imatinib was a miracle for many patients with CML when it became FDA approved in 2001. CML is caused by a mutant tyrosine kinase known as BCR-ABL. Normally, tyrosine kinases signal cells to divide, however in CML, patients with the BCR-ABL mutant have increased white blood cell counts due to uncontrolled division of these cells, causing the cells to accumulate in the blood. Imatinib is a tyrosine kinase inhibitor that specifically blocks the action of BCR-ABL. In one 2006 study, CML patients treated with the drug for 60 months showed an 89% survival rate. 

What researchers didn’t know, though, was what stage of cancer the drug was targeting. Leukocytes undergo a series of differentiations from hematopoietic stem cells before becoming fully developed white blood cells. If imatinib was targeting the leukemic stem cells, the “roots” of the cancer cells, patients could discontinue the drug once these cells were eradicated. But if the drug was targeting differentiated cancerous cells, leaving the stem cells free to make more cancer cells, patients would need to stay on the drug indefinitely.

In a paper published in 2011, researchers from Novartis constructed computer models that tested three hypotheses, each differing based on which stage of cancer cell differentiation imatinib targets, against experimental results. By comparing each model’s predicted effect of the drug on each cell population with patient data, the researchers concluded that “patients treated with imatinib exhibit a sustained, gradual reduction of [leukemic stem cells],” suggesting that imatinib may be a cure for some patients with CML.

While successes such as these have served as a proof-of-principle for systems biology, Kondic still thinks there’s a great deal of work before quantitative pharmacology has reached its full potential. 

She hopes that in pre-clinical development, models will “map the disease” on the pathway level and “identify target populations likely to benefit from therapy.” For clinical development, researchers should be able to develop simulations to “understand the time course of disease progression” as well as the “dose response to interventions.”

Yet one major obstacle in developing these models is incorporating differences between individual patients. In models of protein-protein interactions, for example, researchers are able to account for the effects of specific protein mutations on a system within their simulation. Currently, “it depends on the situation,” said Kondic, “but we’re incorporating it where we can.”