Web Exclusives: Chemistry
Predicting Side Effects and Repurposing Drugs
A medicine that interacts with just its biological target is every drug developer's dream. Such a "magic bullet" would do its job without causing the side effects that often offset a medicine's therapeutic power. Unfortunately, magic bullets don't exist.
Most drugs work by binding a specific protein and blocking its activity. But whether the drug binds to a protein to lower cholesterol or inhibit the growth of breast cancer cells, it also brushes up against thousands of other molecules. These on-target and off-target interactions determine a drug's therapeutic effect as well as its side effects.
Now, a team led by two pharmaceutical chemists—Brian Shoichet of the University of California, San Francisco, and Bryan Roth of the University of North Carolina, Chapel Hill—has devised a method for finding a drug's close contacts inside the body and predicting off-target interactions.
Drug makers spend hundreds of millions of dollars developing a new medicine, partly because many compounds fall out of the pipeline during clinical trials when harmful side effects are discovered. By revealing potential unwanted interactions earlier in drug development, the new approach could offer a huge cost-saving advantage. It also could point to new uses for existing drugs.
The researchers started their work by computationally comparing the chemical structures of 3,665 drugs approved or almost-approved by the U.S. Food and Drug Administration to 65,000 natural compounds called ligands. We often liken ligands to keys because they bind to specific protein receptors—or "locks"—inside the body and trigger a biological response. For example, when adrenaline locks onto its receptors, your heart beats faster.
The massive comparative effort enabled the researchers to match drugs and ligands based on structural similarities. This method was similar to identifying a mystery key's lock by comparing its shape to other keys. If you were to find a spare a key in your house and wanted to know what it unlocked, you might compare it to other keys, like your house key, shed key, car key or your neighbors' keys.
Scientists used to look for off-targets using methods akin to trying the spare key in every lock in town-a tedious and time consuming process. Researchers also could use computational modeling to simulate binding of a drug to a receptor's three-dimensional structure. Unfortunately, this second approach only works for receptors with known structures, which represent a small subset of all possible drug targets.
The new method predicted thousands of previously unknown interactions between drugs and proteins inside the body. Roth explains, "The advantage of our approach is that targets we would never have thought to interrogate are predicted."
Shoichet says that the reason they found so many new drug-protein interactions is because their method categorizes proteins by structural likenesses among the ligands and drugs that bind them, rather than by similarities among the proteins themselves. Categorizing small molecules, like ligands and drugs, by their shapes is much easier than doing the same for proteins, which are larger and structurally far more complex.
"It turns out that similar or even identical drugs can bind to very different, apparently unrelated proteins," says Shoichet. "When we relate proteins by ligand binding, this broad range of targets is captured."
To test the validity of their approach, the researchers examined some of their predictions experimentally and found that most were accurate.
One of the confirmed predictions was the binding of a hallucinogenic compound found in toad skin to serotonin receptors, which also are targeted by LSD. This finding suggests that the hallucinogenic effect of the toad compound operates through the same pathway as LSD. The experimental work also may help explain the side effects of the antidepressants Prozac and Paxil.
As drug makers take more interest in the potential applications of the researchers' new method, Roth and Shoichet continue to search for new drug targets using the new approach. Shoichet says he hopes that in five years he'll be able to predict at least one new target for every existing drug.
Did this work bring him within reach of the goal?
"Nope," he says. "But the study suggests that, as in The Princess Bride, the goal isn't completely crazy—only mostly crazy. And we still have four years and nine months to go."