Millions of people may one day have better odds of fending off the flu as a result of new research that could improve the choice of viral strains included in each year's vaccine.
Princeton researchers Joshua Plotkin, Jonathan Dushoff and Simon Levin analyzed the genetic sequences of flu strains from the last 16 years and found patterns that could be used to predict which strain is likely to predominate in the following year.
Each year, the scientists at the World Health Organization, the U.S. Centers for Disease Control and the National Institutes of Health analyze pre-season reports of flu cases around the globe and select which of the constantly evolving strains of influenza virus to include in the 75 million doses of flu vaccine that are distributed around the country.
These predictions have proven to be largely accurate and the resulting vaccines are credited with saving millions of lives. In some years, however, the vaccine has not targeted the strain that turned out to be most active.
In a paper published in the April 23 online edition of the Proceedings of the National Academy of Sciences, the Princeton researchers proposed a mathematical method for predicting the coming year's flu strain based on the genetic sequences of the strains from previous years.
Applying the technique to each of the last 16 flu seasons, the researchers concluded that, for a few of the years, their approach probably would have chosen a more prevalent strain than the one that was actually included in the vaccine. In at least one year, however, their mathematical technique failed to identify the most active strain, which public health officials did correctly identify.
"Certainly we don't think this should replace what the CDC and World Health Organization are doing," said Levin. "It is just another tool that we think can help inform the decision making."
In the United States, the flu affects 35 to 50 million people and is linked to 20,000 deaths and more than 100,000 hospitalizations each year, according to the Centers for Disease Control.
The virus' annual spread is made possible by constant mutations in a gene that makes a protein on the surface of the viral particle. These small changes can be enough to disguise the entire virus and prevent people's immune systems from recognizing it. Even people who had a flu shot in previous years or who fought off an earlier strain fall victim to new strains that arise through this genetic drift.
The Princeton researchers studied the character and rhythm of flu evolution by using mathematical tools that ecologists use to study ecosystems and the ebb and flow of parasite-host relationships.
"The whole idea of thinking about disease dynamics as an ecological problem is a relatively new approach," said Levin, noting that other researchers initiated the idea about 20 years ago.
The researchers started with a database that contains the genetic sequences of 560 samples of influenza A virus taken from people between 1984 and 2000. They devised a way of measuring the genetic "distance" between any two examples in the database -- a number on a scale of zero to 10, with 10 being the greatest genetic difference.
They discovered that, although some viruses from the database were very similar and others were very dissimilar, the 560 examples were not randomly distributed across the scale. Instead, the virus samples tended to fall into nine clusters in which the genetic distance between the viruses was relatively uniform -- about two on the zero-to-10 scale.
Further study showed that the viruses that constituted each cluster all arose in the same few years, corresponding with the observed emergence of new flu strains.
"We were very excited when we saw that," said Plotkin, a Princeton graduate student who also is a member of the Institute for Advanced Study. Plotkin noted that the results give the first thoroughly mathematical definition of what doctors mean when they talk about a flu strain. Rather than a precisely defined genetic sequence, a strain really is a "swarm" of genetically similar viruses that arise and decline together, Plotkin said.
The researchers plotted the rise and fall of these clusters and used the trends to predict which cluster would dominate from one year to the next.
In a final step, the researchers looked at the genetic sequences that changed most during the transitions from one strain to the next. They found that the mutations that characterize these transitions never occur twice in a row in the same region of the viral surface. "It can't repeat the same trick for evading the immune system," said Levin, noting that this information also may be useful in predicting the course of flu evolution.
The researchers noted that the strength of their technique depends greatly on whether the viruses recorded in the database are a representative sample of the actual viruses that infected most people. The researchers believe, for example, that the database tends to include the most unusual cases that doctors were encountering at the time of the samples.
"If there are problems with this analysis, a lot of them have to do with what sorts of data are or are not available," said Dushoff. "It was a godsend that people put all these sequences in one online database. We hope that now even more people will add to it."
Proceedings of the National Academy of Sciences