Computer Algorithm May Speed Drug Discovery

In 2007, researchers noticed that the anti-HIV drug nelfinavir, developed in the 1990s, had a surprising second benefit. It reduced tumor size in a variety of cancers. Now, scientists think they have figured out how nelfinavir and other drugs work their double magic. Instead of strongly inhibiting one enzyme like most drugs on the market, they act weakly against many targets.

"This is part of a paradigm shift in how people are thinking about drug design," says Vijay Pande, a biophysicist at Stanford University in Palo Alto, California, who was not involved in the study. "[It] will open our eyes to compounds that we might not have found before," he says.

To identify the mechanism behind nelfinavir's cancer-fighting effects, pharmacologist Philip Bourne of the University of California, San Diego, and colleagues assembled a database that contained all of the known shapes of the proteins that make up the human body. Then the team created an algorithm, based on the three-dimensional structure of nelfinavir, to search for the proteins to which the drug would bind. In all, the authors identified 92 binding partners, 85 of which belonged to a family of proteins called protein kinases.

Kinases tack phosphate molecules onto other enzymes, which can slow down or speed up their action. Some protein kinases have been linked to tumor growth because they speed up division and prevent the cells from sending out signals that would cause them to stop dividing and die. Although these kinases have very different functions from HIV enzymes, their shapes are similar enough to allow nelfinavir to bind and disrupt their activity. Existing anticancer kinase inhibitors have the same side effects as nelfinavir (fatigue, rash, diarrhea), which suggests the drugs bind to similar proteins.

The authors then tested how strongly nelfinavir binds to the 12 protein kinases predicted by their model as being the best fit. The experimental assays showed similar binding strengths as their calculations. This weak action against many kinases effectively applies the brakes on uncontrolled cell division. Improper kinase activity strongly dampens the signals that tell the tumor cell to divide. For shrinking tumors, a small decrease in many kinases or the complete inhibition of just one can be equally effective, Bourne says.

The approach can be used to identify numerous drug targets more cheaply and easily than traditional experimental methods, the researchers report this week in PLoS Computational Biology. Scientists usually begin by selecting an enzyme to target, says Bourne. Then, they screen tens of thousands of compounds for any that might inhibit the enzyme. The process can take decades and cost hundreds of millions of dollars. In contrast, a computational approach would take just a fraction of that time and money, says Bourne.

With the record-breaking profits of the pharmaceutical industry, the need to fundamentally change how we design new drugs isn't always obvious, Pande says. "It's the mentality that if it ain't broke, don't fix it." But the pipelines for new drugs are beginning to run dry, as all of the easily discoverable drugs have already been identified, he says. "In this next century, [drug design] is going to be considerably harder than in the last 50 years." Computational biology approaches like Bourne's, says Pande, show that weakly attacking many targets can be an effective tactic in fighting disease.