Predicting flu evolution

Each year, influenza circulates globally, evolving and adapting as it goes, and its arrival in any given geography usually coincides seasonally with wintertime. This steady evolution changes the face that the virus presents to our immune systems and is what enables it to stay one step ahead of us and to keep coming back. For a vaccine strategy to work, scientists must try to anticipate sufficiently far ahead what forms of the virus will emerge in the forthcoming season, to give healthcare systems enough time to mass produce and administer an appropriate cocktail to protect the population. We get it right about 60% of the time. So why not 100%? Where are the shortcomings in the system and what’s it going to take to push up our hit rate? This is what Richard Neher, at the University of Basel where he works on predictive models of flu, has been wondering. Speaking with Chris Smith, he’s hoping to shed some light on where the weaknesses are, and how surmountable they may, or may not be…

Richard - We know that influenza viruses adapt very rapidly to human immunity, meaning they mutate such that antibodies don't recognise the virus as well anymore as they used to. And these variants that have these immune escape mutations, these adaptive mutations, that tend to increase in frequency and our primary means of predicting future population would be that you look at the ones that grow in frequency, and extrapolate these frequency growths further into the future. But it turned out that many things that grow at some point stop growing and then just meander about this sort of curious growth is what triggered this work.

Chris - Is that because it grows and turns itself into a mutational disaster? Or is it because it grows flourishes but then we catch up, our immune response catches up and heads it off so then it it is kind of stuck?

Richard - Yeah, we believe it is the latter. So, as these viruses grow in frequency and circulate, whoever gets infected by them generates immunity against them. So, in a way they are using up the susceptible populations that allows them to circulate and thereby their growth advantage ends up being diminished as their frequency increases.

Chris - Why do they not though just stay one step ahead of us continuously and just keep on changing? So, that they can continue to grow and we are always playing catch up?

Richard - To some extent they do, but sometimes you have a mutation that endows a virus with a particular advantage, but from there it's stuck in a place where it can't really go further. But some other viral lineages that also circulate, they tend to have gone a different path than mutational space that offers them more future opportunities to further adapt. So, there is sort of a way in which viruses can back themselves into a corner and while they have a transient advantage, then they're sort of stuck in a place while other ones that were initially not as successful then overtakes them.

Chris - There are obviously lots of different families and types of virus. Many of them share some characteristics with flu, but they are different from the flu. We've just been through a pandemic caused by a different group of viruses, the coronaviruses, but they seem to show a very similar pattern to the one you are describing where we saw jumps of the virus to adopt a new configuration. It thrived in that form for a bit and then a new one came along and then it got as far as the omicron variant and it seems to have stalled there and we are getting lots of offshoots of that, but we haven't seen these dramatic shifts again. So, is COVID showing a very similar pattern to the mechanism you think is at play with flu?

Richard - Yeah, COVID is a very, very interesting example of, you know, the main difference with COVID is none of us had seen COVID five years ago. Now we have all gone through multiple exposures, multiple times being vaccinated or infected while flu has been around as long as we've been living. And people that have been born in different decades have been exposed to different flu histories in a way. They've seen different flu viruses throughout their life and thereby have more diversity in the immunological makeup. We believe that these differences in this immunological makeup affect the virus host coevolution. And that for, if you have a very heterogeneous immunological makeup, there's lots of sort of different niches that some viruses can exploit and others can't. While in the more homogeneous case with COVID, it's more likely that you have one variant that escapes a larger fraction of the population and they then have a growth advantage that carries it through the entire population and it ends up wiping out all other variants

Chris - With things like the flu then, if you can now work out mechanistically what is going on, does this mean we are in a stronger position to make predictions about the future and therefore make better judgements about the sorts of vaccines that we're compiling to try and head off what the virus is going to do as it goes through the population?

Richard - I certainly hope so. I think the main insight from this work is actually exploring the limits of predictability. What aspects of a future population are predictable, which ones aren't, and how much faith should be put into this kind of predictions versus another kind of prediction. So this equal evolutionary dynamics that we see inherently limits the prediction horizon, if you wish, so that the amount of time we can run these predictions forward is limited by the sort of intricate interaction between host and virus.

Chris - How far ahead do you think we can look then? Because obviously people - in lots of other laboratories and rival groups to your own - they're spending a fortune trying to answer this very question they're bringing AI in to look at past flu behaviour, future flu behaviour and so on. Are you saying then that in fact we are gonna have to narrow our horizon?

Richard - Well, there's certainly a lot of, you know, molecular features that we've discovered over the last decades. So, mutations in particular places changes in the viral protein. These are features that often make a virus successful and I think these predictions are going to get better and better. This is absolutely crucial for this vaccine optimization vaccine selection process. And then there is sort of another question as to how, you know, how confidently can we predict the relative proportions of different variants three months out, six months out, twelve months out? And that sort of was the main insight from our work here is that prediction of these explicit measures of population makeup of the future. These are going to be much, much harder. Currently, it's sort of one season that over which these predictions can be meaningfully made. But making now sort of actually frequency predictions for the next winter season, the 2025-2026 winter, that currently I don't think is in reach and sort of our paper here highlights some sort of intrinsic limitations that predictions of these kind face.

Chris - If we know what the limitations are, though we understand something about the mechanism of those limitations, which means we might be in a position to do something about it. So, does your paper also shed some light on where we need to direct our efforts to try to become better at doing this in the future?

Richard - Yes. So, you know, my take home from, from this work is that to actually make those predictions better, we need a much more fine-grained understanding of what immune systems, of what part of the population recognize which viruses, how well. So, we would want to be able to measure this kind of immunological makeup for, you know, children and teenagers and adults and older adults and and, and the elderly in different parts of the world. So that we, you know, we really kind of understand if there's a new virus coming in, sort of who is how susceptible to this new virus and who is protected from this virus that will allow us a) to, you know, understand the future frequency or trajectory of the virus for some time. And it also might help us to identify variants that are of particular concern because, you know, vulnerable groups are less well protected than they should be.