Science Editor

The modern way of classifying proteins into families is molecularly based: Sequence comparison reveals which proteins belong together. But it is not the only way to classify proteins - in fact, it is not how most of them were originally classified.

Receptor families, for example, predate molecular biology and were originally determined by testing which molecules would bind to them. And "that view of biology led to classifications that . . . often crossed major protein boundaries," Brian Schoichet told BioWorld Today.

For example, "there is no relationship, genetically speaking, between [the serotonin receptors] 5HT-3 and 5HT-4. But they both bind serotonin."

In the Nov. 1, 2009, online edition of Nature, co-corresponding author Schoichet and his colleagues used the method of classifying proteins by the small molecules that bind to them to predict currently unknown targets of known drugs.

The paper, corresponding author Schoichet said, represents a "chemistry-based approach that goes back to a classical pharmacological view of biology," supplemented by modern informatics methods. "This method has . . . returned to that view not because it is a superior view, but because it is such an orthogonal, and such a forgotten view."

In their paper, the authors first classified roughly 250 targets by the known small molecules that bind to them. Such binding data are collected in a database, the MDL Drug Data Report, or MDDR database. They then used computational modeling techniques to predict which of nearly 3,500 FDA-approved and investigational drugs would bind to each of the targets.

The approach generated thousands of predicted new drug targets; the team then went on to test 30 of the strongest ones experimentally. Twenty-three of the predictions were confirmed, with five showing binding that the authors described as "potent," occurring at concentrations of less than 100 nanomoles.

To Schoichet, the most interesting findings of the paper are of "established drugs that crossed major domain boundaries." Perhaps the most striking of those is delavirdine (Rescriptor, Agouron Pharmaceuticals Inc.), a reverse transcriptase inhibitor, which the study found also binds to the histamine H4 receptor, a G-protein coupled receptor.

The method is constrained by the protein targets it uses, and Schoichet acknowledged that there are "still huge gaps" in our understanding of everything that can be a target. "Those will be areas of darkness for this method," he said.

But in those areas where it works, he added, it should be useful both for finding new uses for drugs and for predicting side effects. Of the two, Schoichet believes that finding new uses for drugs is the more exciting one, but also has further to go, though the initial results are encouraging. "Who knows whether it will ultimately work," he said.

Using the method to predict side effects, on the other hand, "is something that people could be doing right now." In their paper, Schoichet and his colleagues showed that antidepressant SSRIs also bind to beta-adrenergic receptors with very high affinity, which may explain their known cardiovascular side effects.

Beyond the specifically new, the authors believe that their work also shoots holes into the concept of the magic bullet more generally.

"Drugs are not as selective as we once thought," co-corresponding author Bryan Roth, who is at the University of North Carolina at Chapel Hill, said in a press release. "It turns out that the most nonselective drugs are frequently the most effective for complex diseases. Rather than 'magic bullets,' we need to come up with 'magic shotguns' that hit more than one molecular target at a time."