But using metabolomics as a predictive tool has never been done before, says Rima Kaddurah-Daouk, a research scientist at Duke University in Durham, NC. “We had already determined that metabolic signatures vary with individuals,” she says. But no one had proved that this variation could be used predictively, she says.
Working with researchers in Sweden, France, and at the pharmaceutical company Pfizer, Nicholson’s group analyzed the metabolites in the urine of 65 rats, both before and after they received a lethal dose of acetaminophen. This was done using nuclear magnetic resonance (NMR) spectroscopy, a diagnostic technique that rapidly quantifies the presence of organic compounds by applying powerful magnetic fields to detect the spin in any hydrogen atoms present within these molecules.
The data derived from this technique was then used to create a predictive model based on as many as 1,000 different metabolites. The researchers found that certain metabolic signatures could indeed be used to predict the severity of liver damage caused by the drug.
In the United States, a large proportion of the drugs withdrawn each year by the Food and Drug Administration are removed because of side effects, which often affect only a small proportion of people. Now it may be possible to screen out these few people, so the majority can benefit from the drugs, while at the same time the minority will be protected against negative reactions.
“In many ways this is ground breaking,” says Kaddurah-Daouk. The fact that it captures both environmental and genetic factors makes it highly attractive – and it’s likely to prompt a huge amount of interest from drugs companies, she says.
Others agree. A lot of companies have invested in proteomics and genomics in the belief that they will help to identify biomarkers of disease, says Mike Milburn, chief scientific officer of Metabolon, which runs a metabolite screening and analysis service in Durham, North Carolina. “Now many of them are already looking to metabolomics as a way of complementing that data.”
There is still a long way to go, however. First, a new predictive model would need to be developed for humans, says Nicholson, although it would not be like starting from scratch. “We are not there yet, but this is the first step to personalized health care,” he says.