A new computational method that searches an enormous database of protein structures could allow researchers to predict a drug’s potential side effects without breaking out a single test tube. The technique, developed by researchers at the University of California, San Diego (UCSD), could also be applied to existing drugs to explain known side effects or to identify additional uses.
“The approach that they’re taking is different and certainly way more computationally intensive than other approaches,” says Bryan Roth, a pharmacologist at the University of North Carolina, Chapel Hill, who was not involved with the work.
Drugs work by latching onto very specific receptor sites on protein targets, much like a key fitting into a lock. If a drug also happens to latch onto another, unintended target, it may produce unwanted side effects. The new method analyzes the shape of the intended lock and then combs through the structures of other proteins to look for similarly shaped locks. If such a lock is found, an algorithm rigorously determines whether the drug fits snugly into it. If it does, the technique has probably identified what’s called an “off-target.”
The researchers, led by UCSD pharmacologist Philip Bourne, didn’t stop there. Once they had identified an off-target, they worked backward, looking for other drugs known to bind to it and then using their technique to determine whether those drugs could also bind to the original target. If so, the researchers had further evidence that the two proteins–the intended target and the unintended target–had highly similar drug receptor sites. Then, by assessing the biological function of the off-target, they could postulate possible side effects of the drug being tested.
In the November issue of PLoS Computational Biology, Bourne and his team described applying the new method to a class of drugs called selective estrogen receptor modulators. These drugs–including tamoxifen, the most widely prescribed drug for breast cancer–latch onto a protein called estrogen receptor alpha. Tamoxifen is known to cause side effects such as cardiac abnormalities, retinal degradation, and blood clots. All these side effects have one thing in common: they involve disruptions in the ways that calcium ions normally flow through cells, regulating the balance of electrical charges in various cell compartments.
“That would imply that maybe what’s involved here is some protein that involves calcium, as part of the normal calcium flow in the cell,” says Bourne. And sure enough, the team found that in addition to the lock on estrogen receptor alpha, tamoxifen could fit into a similar lock on a calcium-binding protein known as SERCA. Bourne speculates that when tamoxifen latches onto SERCA, it physically prevents calcium from doing so–potentially leading to the drug’s calcium-related side effects.
“The prediction is certainly plausible and interesting,” says Roth, but he cautions that it’s purely computational and must be validated biochemically.
Bourne’s new technique inverts the approach ordinarily taken to computational drug design. Most methods focus on the drug rather than on the protein that it binds to. Pharmaceutical researchers routinely sift through databases of small molecules looking for drugs to match a particular protein. Bourne’s team, by contrast, is looking for proteins to match a particular drug.
While other groups have also used computational methods to identify off-targets of known drugs, the technique has never before been deployed on such a colossal scale. The new method searches the entire known “druggable genome,” a set of proteins with the potential to bind to drugs that were culled from the Protein Data Bank, a database containing the structures of more than 10,000 proteins. Crunching through so many structures required a major dedication of computing resources. Lei Xie, senior scientist on the project, originally planned to develop the technique for a pharmaceutical company. When he was not granted the resources, he brought the project to Bourne’s lab.
Previous work on off-targets was limited to small clusters of proteins grouped by functional or structural similarity. Searching the whole known druggable genome opens the door to the discovery of unexpected drug-protein relationships that narrower searches would miss. Indeed, that seems to be what’s happened with tamoxifen and the calcium-related SERCA protein.
But although the Protein Data Bank holds an enormous number of protein structures, it is by no means comprehensive. Bourne estimates that the set of proteins his team worked with represents approximately 40 percent of the true druggable genome. Many drug receptor proteins either haven’t yet had their structures elucidated or are not amenable to current methods of determining protein structure. “There are a lot of very interesting targets that we have no structural information about, and this approach is not going to be useful for those,” says Roth. However, he adds, “if you have a three-dimensional structure for a target, then it’s a great way to go.”
Bourne hopes that this kind of computational screening will be adopted by the pharmaceutical industry. By screening in silico–using computers–for potential harmful side effects, companies may be able to eliminate drug candidates before they undergo expensive animal testing and clinical trials. In addition, as Bourne demonstrated with the selective estrogen receptor modulators, a drug can be modified so that it binds more tightly with its target protein and more loosely with off-target proteins, increasing its effectiveness and reducing its side effects.
“In an ideal situation, this would become part of the drug-discovery process,” Bourne says. Roth agrees that drug design would benefit from such an approach.
The technique could also be used to identify ways to repurpose existing drugs, an application that Bourne’s group is currently exploring. Not all side effects are bad: just as the antidepressant bupropion is also used as an aid in smoking cessation, many drugs could have more than one beneficial use. Computational screening for off-targets could identify such alternative uses of known drugs.
How Facebook and Google fund global misinformation
The tech giants are paying millions of dollars to the operators of clickbait pages, bankrolling the deterioration of information ecosystems around the world.
DeepMind says it will release the structure of every protein known to science
The company has already used its protein-folding AI, AlphaFold, to generate structures for the human proteome, as well as yeast, fruit flies, mice, and more.
Inside the machine that saved Moore’s Law
The Dutch firm ASML spent $9 billion and 17 years developing a way to keep making denser computer chips.
This is what happens when you see the face of someone you love
The moment we recognize someone, a lot happens all at once. We aren’t aware of any of it.
Get the latest updates from
MIT Technology Review
Discover special offers, top stories, upcoming events, and more.