BioEd Online

Researchers have identified thousands of new targets for existing drugs using a computer program that compares the molecular structures of drug compounds and chemicals that occur naturally in the body. The technique can be used to uncover new applications or reveal potential side effects for drugs already on the market.

"It's a new approach, and it's a totally different from what everyone else has done," says study author Bryan Roth, a pharmacologist at the University of North Carolina at Chapel Hill. "That's why it actually works."

The most common methods for predicting whether a small molecule binds to various drug targets involves either high-throughput laboratory screening or virtually simulating whether a particular compound fits together with proteins like a key in a lock. The experimental approach, however, is tedious and time consuming, whereas the computational method relies on the existence of high-resolution protein structures, which are hard to come by for many drug-sensitive proteins.

Tantalizing targets

Last year, Peer Bork and his colleagues at the European Molecular Biology Laboratory in Heidelberg, Germany, developed a new approach for finding novel drug targets for existing medications. They showed that drugs with similar adverse side effects often share a common target protein, even when those drugs are chemically quite different¹. Now, a team led by Roth and Brian Shoichet, a computational chemist at the University of California, San Francisco, have succesfully identified new uses for marketed drugs by comparing existing drug compounds with different ligands — biologically active molecules that naturally bind proteins. Their hunch was that if a drug and ligand have similar three-dimensional structures, then there's a good chance that the drug will bind to the same protein as the ligand.

The researchers' suspicions were proved correct. They produced chemical 'fingerprints' of more than 3,600 drugs that were either approved or in late-stage clinical trials and some 65,000 ligands that together bind to around 250 protein targets. They then developed a statistical technique to compare the two types of molecules and singled out nearly 7,000 pairs of predicted drug-target interactions, thousands of which had never been shown before.

The technique provides "a way of giving you decent molecular-based hypotheses for side effects of drugs and a way of looking for new targets for these very special molecules", says Shoichet.

The authors validated 30 of these connections experimentally and showed that 23 of them were bona fide associations. These unanticipated interactions included hitherto unknown effects of many well-known drugs, the researchers report online today in Nature². For example, the antidepressants Prozac and Paxil, which work by boosting serotonin levels in the brain, also bound the ß-adrenergic receptors, the researchers showed. This ß-blocking activity could explain some of the drugs' side effects which include nausea and decreased libido.

Trip or treatment?

"The pharmaceutical industry should take note of this work," says Jeremy Jenkins, a chemoinformatician at the Novartis Institutes for BioMedical Research — the drug company's research arm — in Cambridge, Massachusetts, who was not involved in the new work. "This could really help us improve on preventing safety issues, which are one of the major contributors to drugs failing."

The authors also discovered a target for the hallucinogenic compound N,N-dimethyltryptamine (DMT). This target had been unclear until earlier this year when researchers at the University of Wisconsin School of Medicine and Public Health in Madison showed that the drug — which is found in toad skin, the psychoactive tea Ayahuasca, and the brain's pineal gland — binds to s-1 receptors, but with poor affinity³.

Shoichet and Roth ran DMT through their predictive algorithm and found that the psychedelic drug binds much more strongly to serotonin receptors, including the one associated with the well-known hallucinogen LSD. The researchers confirmed the computational findings in cell assays and in live mice.

Andrew Hopkins, a computational chemist at the University of Dundee, UK, who wrote an accompanying commentary to the paper, was "impressed" by how the authors went from computer predictions through to experimental validation. "You can really see that some of this data makes biological sense," he says.