In Shanghai, a woman shivers with fever; her throat is raw and her body aches. She goes to the hospital, where doctors diagnose her with the flu. After treatment, they send a genetic sample of her virus to the Chinese National Influenza Center in Beijing, part of the WHO’s flu network.
Each year, approximately 5000 viruses from all over the world are shared with the WHO, and each virus is just a little different. The flu virus evolves rapidly to evade the human immune system—since 1968, the virus has exchanged up to a quarter of its genome—and, as a result, the WHO must constantly update the strains used in the annual flu vaccine.
But with so many strains to choose from, which should be put in the shot?
A few weeks ago, scientists convened at the WHO headquarters in Geneva to decide which flu strains should go into next year’s vaccine for the northern hemisphere. Scientists chose strains for the vaccine that were most different from those in the current vaccine, in other words, the ones our immune systems theoretically would recognize the least. They merged these data with information on viral prevalence and circulation patterns to determine which strain should make up the shot. This method is an informed guess, one that’s usually (but not always) fairly accurate.
Marta Luksza and Michael Lassig from the University of Cologne in Koln, Germany have come up with a way to more deliberately choose flu strains for inclusion in the vaccine: a predictive model that projects the fitness of a given viral strain and then compares the fitness of all viral strains in circulation to each other to determine the one most likely to infect people the following year (1).
Three strains of flu are typically put into a vaccine: influenza A subtype H1N1, influenza A subtype H3N2, and influenza B. The new model deals with influenza A subtype H3N2, focusing specifically on a protein found on the flu viral surface called haemagglutinin (HA). HA is recognized by human antibodies and evolves to increase the ability of the virus to avoid our immune systems.
Every few years, HA evolves in a so-called “cluster transition.” A 2012 Genetics paper (2) by Lassig and Natalja Strelkowa at Imperial College London showed that this process is basically a race between viral strains that have developed different beneficial (for the virus) mutations of their HA epitopes. “Given there’s this race between beneficial mutations, then, can we predict which one will win?” asked Lassig. “Can we make evolution a predictive science?”
So Luksza and Lassig set out to find which flu virus strains would be the “fittest,” the most able to outcompete other viral strains and infect more of humankind. The most prevalent strain should be in next year’s vaccine.
Searching GenBank for sequences of flu viruses collected from people all over the world over the past several years, they first looked at HA working with the assumption that most epitope mutations give the virus an advantage over other flu viruses because they make the virus harder for the human immune system to detect.
The second component of their model took into account deleterious mutations outside the epitopes. Lassig’s 2012 paper (2) showed that background changes in the protein are also key to flu evolution. These are typically under negative selection, because they affect basic biological processes such as protein stability. Any such mutation here will likely lead to a less fit virus. For any given viral strain, the researchers can calculate fitness with mutations within HA considered positive and others considered negative. “Every strain has a sum total that we use to rank the strains in terms of fitness,” said Lassig.
Incorporating the effects of deleterious mutations makes this a particularly interesting model for Katia Koelle, a researcher at Duke University not affiliated with the study. “That genetic background hasn’t really been looked at by the WHO,” she said. “The Luksza and Lassig paper compared the WHO selected strains [going back to 1994] to what their model would select, and both were relatively good, but theirs was an improvement.”
A Better Vaccine
According to Lassig, the model still needs fine-tuning and testing, as well as the incorporation of additional data, like HA inhibition and geographical distribution. In addition, the authors used many assumptions that could be problematic.
Lassig assumed that most mutations in the epitope region contribute to fitness, yet prior work has shown that some mutations in the epitope have a large effect on immune system recognition, while others have a negligible effect. The model also failed to account for epistatic interactions—how the effect of a genetic change might depend on what other mutations have happened—and the circulation patterns of viruses across the world.
But incorporating all these variables is a complex undertaking, and researchers don’t yet know how to integrate all of this information into a single model. “Everyone kind of has their pet idea of what’s important, and it gets simplified away in a model,” Koelle said. Still, Lassig and Luksza’s model is proving to be effective.
“It does a good job in terms of being predictive, despite ignoring all of that stuff. We’re going from a good situation to a better situation,” said Koelle, who proposed not scrapping the current method of vaccine selection, but instead trying to integrate the methods for future predictions and vaccine strain selection.
Lassig hopes the idea can be used beyond influenza evolution. “Cancer growth is a theoretically related system: you want to know what sequence of somatic mutations makes the cancer evolve. That’s also a rapid evolutionary process,” said Lassig. “So there’s a number of things that are conceptually related to our research. I certainly would be interested in exploring these broader implications.”
 Luksza, M. and Lassig, M. A predictive fitness model for influenza. Nature. 2014. 507: 57-61.
 Strelkowa, N and Lassig, M. Clonal interference in the evolution of influenza. Genetics. 2012. 192: 671-682.