Scientist Stories: Jesse Bloom, Evolution of Viruses

Welcome to the influenza webinar series hosted by the Human Vaccines Project. The Human Vaccines Project is a nonprofit public-private partnership with the goal of decoding the human immune system to accelerate the development of vaccines vaccines and immunotherapies. In order to make real progress towards the development of a universal influenza vaccine, collaboration among researchers around the world is essential. We hope that this monthly webinar will foster the scientific foundation needed to advance influenza research. The series of webinars will be broadcast on the third Tuesday of each month and posted online. For a copy of today's webinar, please visit the Human Vaccines Project website at http://www.humanvaccinesproject.org/talks/. This web URL is also posted in the chat box. Following the presentation, we will have time for questions from the audience. Please type your questions into the question box in the GoTo interface. I will read these questions aloud following Dr. Bloom's presentation. And now it is my honor to introduce Dr. Jesse Bloom. Dr. Bloom is an associate member in basic sciences and computational biology at the Fred Hutch Cancer Research Center. And an HHMI Simons Faculty Scholar. His group uses a mix of experimental and computational approaches to study viruses and their evolution. Today, Dr. Bloom will tell us about complete mapping of viral antibody escape mutations. Thank you very much and enjoy the webinar.

Hi everyone, so it's a real pleasure to be giving the first webinar in this series I'm giving this from home. We recently had a little baby, so if you're spending all your time wondering what that is behind me there, it's a pack and play, but hopefully you won't hear any crying during my talk. So anyway, I'm going to talk to you about complete mapping of viral antibody escape mutations. And since this is the first webinar in the series, I want to start off with a little bit of general background about why we care about virus evolution in the context of vaccine. See, and I think I have lost— oh, there we go. So I'm going to start off talking a little bit about this treatise, Observations Made During the Epidemic of Measles on the Faroe Islands. And this was written by Peter Panum in 1846. So when I first came across this, I did not even know where the Faroe Islands are. I don't know if some of you are more knowledgeable about me than geography, but it turns out the Faroe Islands are a small island that are part of Denmark that are sort of located halfway between Denmark and Iceland. And particularly back in the 1840s, there was very limited traffic between mainland Europe and the Faroe Islands. So when you start off reading Peter Panum's treatise, he starts off with this very sort of a colorful description of the islands, which you would certainly not see in modern scientific publications. He writes, "The winds are exceedingly uncertain and violent, making it necessary for the traveler to throw himself on the ground in order not to be carried away. But the aurora borealis often illuminates the usually pitch-dark winter nights." And he goes on like this for a number of pages, but eventually he gets to the point of why he's actually on the Faroe Islands. And he explains that there had not been any measles on the Faroe Islands until 1781. And then in 1846, an epidemic of measles broke out because measles was carried to the island from someone coming on a ship. And he does run one of the really, uh, first very careful epidemiological studies of this sort of disease outbreak. And he notes that of the 7,782 inhabitants on the island, The vast majority of them, about 6,000, were taken with measles. But the thing that makes this work really a classic of epidemiology is the next thing that he notes. Of the many aged people who were living on the Faroes who had had measles in the previous outbreak in 1781, not one, as far as he could find out by careful inquiry, was attacked the second time. So this is one of the first really clear epidemiological descriptions of what we now all know is immune memory. And this is the fact that often when we're infected with a disease once, we then have lifelong immunity that will protect us against it in the future. And this, of course, is the basis of the vaccines that now protect us from many of the viral diseases that were once, you know, really, really problematic for humanity. So then the reason we're having this webinar, of course, is we don't have good vaccines against all viruses. And so why don't we have good vaccines against all viruses? And, you know, for instance, one of the viruses that we don't have a good vaccine against is influenza. As you all know, you have to get a new influenza vaccine every year, and it's not necessarily always that effective even when you do get the vaccine. So history has given us an experiment with influenza that's a lot like Peter Panum's study of measles. We know that 100 years ago, in 1918, there was an H1 influenza virus that appeared in humans and caused a global pandemic. After 1918, this virus continued to circulate in humans for many years, undergoing a little bit of evolution every year, up until 1957. In 1957, the H1N1 virus was replaced by a new viral strain, an H2N2 strain, which then again circulated in humans undergoing a little bit of evolution every year. And then in 1968, that H2N2 strain was replaced by an H3N2 strain, which continues to circulate today. But something else interesting happened. In 1977, there was what was called the Russian flu pandemic, and this was a reemergence of an H1N1 strain that was genetically identical the strain that had been present in about 1954. So very unusually for influenza, this virus had been in about a 24-year evolutionary freeze, and it's believed that this virus was somehow, you know, accidentally reintroduced from a sample that had been frozen in the mid-1950s. So as you can see, the re-emergence of this virus is actually a lot like Peter Panum's study of measles on the Faroe Islands. You will have people who are older than 24 years old who, when 1977 comes around, will have been infected with almost the exact viral strain several decades ago. And then you'll have everyone less than 24 years old who has never seen this virus before. So what happened in this 1977 pandemic? Well, we can go to, for instance, this article, which was in the British Medical Journal in 1978.. And it's a description of influenza in a boarding school in the UK. They write, "The Easter term began on the 10th of January. One boy from Hong Kong had a transient febrile illness from the 15th to the 18th of January. On Sunday, the 22nd of January, 3 boys were in the infirmary." And then overall, 512 boys, which was 67% of the school, spent between 3 and 7 days away from class. So you can see that this, this virus really just swept through this school, almost completely infecting the boys. But they also note, of about 130 adults, all of whom would have been previously exposed to this virus around 24 years ago, who had some contact with the boys, only one, a house matron, developed similar symptoms. So what this natural experiment shows us is that natural infection with influenza actually provides durable immunity that's at least somewhat comparable to that against measles. But importantly, it's only against that exact viral strain. What was very unusual about this 1977 pandemic is the strain was exactly identical to one that some people had seen 24 years ago. Another important caveat, which is not going to be the focus of my talk, but I think will be discussed extensively in other webinars, is we think that this sort of long-term protection infection might perhaps be more a feature of natural infection with influenza than vaccination with influenza. It's less clear that the same result would have been observed if these people had been vaccinated 24 years previously. But anyway, this shows that the problem that we face with an influenza vaccine is not that our bodies cannot mount a good immune response, but that immune response only protects us against that exact viral strain. And so the reason, of course, that we're at risk from influenza virus every year of our lives is simple. It's virus evolution. And that's the reason that someone like me, who's really more of an evolutionary biologist, is giving one of the first webinars in the series. So when we think about the evolution of human influenza virus, many of you may have seen phylogenetic trees of viruses' evolution that look like this. So on the horizontal dimension here is time. This is H3N2 influenza. Here are the years from 1990 to the present. And you can see that the tree sort of has this unusual shape where there's one major lineage here with a lot of short branches coming off. And what that shape means is that, to use the sort of terminology that our president likes to use, during the evolution of this virus, there are winners and losers. So what do I mean by that? Well, all of these viruses in green, which are on these side branches, are evolutionary losers. These are strains that after a few years leave no further descendants. They die out. And then on the trunk of this tree is this orange lineage, and this is the lineage that's the evolutionary winner because it survives. And at any point when you look in the tree, you can see that all the viruses 3 to 5 years in the future are descended from just one virus that existed on the trunk. And this is sort of the basis of current influenza vaccines, which is basically to try to pick viruses from this lineage which are going to win out, and then maybe you'll have a vaccine that's well matched to the viruses that will be present in the upcoming season. Of course, this is challenging because, for instance, when we look at the more recent evolution of this virus, we know based on the past evolution that one of these lineages is going to win out. That's what's happened, you know, every time we can see in the past, but we don't yet know which of these lineages it will be.. And figuring out which of those lineages is going to prosper in the future would be very useful for selecting which strain should best go in the— would be most effective to put in the annual influenza vaccine. So another amazing thing about this sort of phylogenetic tree is that every mutation that we see on the tree originates from an infection of a single individual. So every time we see a mutation, and you can see from this tree that these mutations sweep to take over the whole globe in just a few years, there was some person and that person was infected with flu, and while the virus was replicating with that person, it created a mutation. And when we see that mutation on the trunk of the tree, it means that that virus eventually went from that person and spread across the entire world. So I still find this to be really an amazing evolutionary fact. There's probably almost a billion people who get influenza every year, but yet every year there's someone who has a virus that gets a mutation that over the course of the next few years will actually spread[No speech] So here's the structure of influenza's hemagglutinin protein, and this protein is essential because it enables the influenza virus to infect cells. It both enables the virus to bind to the receptor on cells and then to undergo membrane fusion and enter cells. And so it's been appreciated for a very long time So it's been appreciated for a very long time that infection with influenza induces very potent antibodies against this hemagglutinin protein. So one of these antibodies is shown here in red, and these antibodies tend to bind up here on the top, or what I'm gonna call the head of this molecule. And it's also now been appreciated for many decades that if you take a virus that has a hemagglutinin that's bound by a single monoclonal antibody like this, and if you grow that virus in the presence of the antibody, you will rapidly select viral escape mutations, like these mutations to the virus at these sites shown here in green, that abrogate the antibody binding and allow the virus to survive neutralization by that antibody. And then we also know that when we look across those phylogenetic trees, like the one that I was just showing you, and see where in the hemagglutinin the mutations that occurred during natural evolution tend to be, many of those mutations are up here on the top and the surface of this hemagglutinin molecule, which is where we know that many of the antibodies bind. And so this has led to the idea, which I think is almost certainly true, that these antibodies play a major role in driving the evolution of this protein. So to just sort of put that in simpler words, there are really two major selective forces on hemagglutinin during its evolution. First of all, it has to maintain this function that enables it to bind to cells and then help the virus enter cells. That's absolutely essential, so that function has to be maintained during viral evolution. Second, while the virus is maintaining this function, it needs to accumulate mutations to evade these antibodies so that it can reinfect people after several years. And so both of these are really essential to sort of the survival of influenza virus in humans. So one of the things that my lab has been very interested in doing is mapping how all mutations to influenza virus affect these two key phenotypes of protein function or viral growth and antigenicity. And the reason is because if we know how specific mutations affect these functions, we can then look at the natural evolution of this virus with a lot more sort of precision because when we see a mutation, we will have some idea what that mutation actually does at a functional level. So how do we do this? So the influenza hemagglutinin protein is 566 residues long, depending on exactly what subtype you're looking at, they're all around 566 residues long. So in the lab, we can now simply take the gene that encodes the hemagglutinin protein, and we can generate all of the amino acid mutations to this gene by some PCR-based strategy. So this gene is 566 amino acids long. Each amino acid can be mutated to 19 other things, so there's 10,754 such mutations. As you know, that's not that many mutations to make by PCR, provided you have some some bulk strategy to do it. We then take these genes that encode hemagglutinin, and we've developed techniques to efficiently incorporate them into a library of mutant viruses. So we now have viruses, each of which is carrying a different mutant of influenza hemagglutinin, and we can now do things to these viruses that allow us to read out the effects of those mutations. So the simplest thing we can do is simply ask which of these viruses are able to continue to replicate in cell culture. And so this is sort of a simple readout for whether or not the viruses have a functional hemagglutinin, because if they don't have a functional hemagglutinin, they will not be able to enable that virus to replicate in cell culture. And so obviously you could have done this experiment many, many years ago as well, but the sort of new and novel thing that we're now able to do is use ultra-accurate deep sequencing to count the mutations at each step in this experiment. So on the y-axis here, I'm going to plot the per-mutation, per-codon mutation frequency across the entire hemagglutinin gene. And so if we deep sequence this initial pool of mutations we made, we see that, as expected, we have all types of mutations present. We have synonymous mutations, which do not change the protein sequence. We have non-synonymous mutations, and we have stop codon mutations. But when we sequence the viruses that are able to replicate, we see that many of these mutations have been purged by selection. So as you might expect, the synonymous mutations are usually neutral with respect to viral replication, or often neutral, so they don't change in frequency all that much. The stop codon mutations are almost uniformly deleterious, so they go away. And some, but not all, of the non-synonymous or amino acid mutations are purged. So the powerful thing is we cannot just look at this sort of bulk statistics as I'm showing here, these average statistics, but we can perform the same analysis for every mutation to hemagglutinin. And so that yields data that look like this. So this This plot shows the effect of every mutation to hemagglutinin on viral growth, and we have such data for the entire gene. In order to make it look okay on the slide, I'm just showing that data for a small number of sites in the gene. And so the height of each letter is how often you see that amino acid after you've selected for viruses that can grow, compared to how often you can see that mutation before. So for instance, at this position, which is position 183, So those of you who are familiar with hemagglutinin might know that that's an important position for receptor binding. You see that whatever mutation you put in at the beginning, the only thing that you end up with in the viruses that can replicate is histidine. There's other positions like this one where the virus appears to prefer hydrophobic amino acids. Here it prefers aromatic amino acids, and here it can tolerate a large number of amino acids. So I think it's important to understand how these data are both similar and different from what you would get at just looking at a natural alignment of influenza sequences. So in our experiments, we've systematically sampled all of the mutations. So we're looking at the effect of every mutation sort of on its own. Natural evolution samples many mutations, but it doesn't in a very biased way. So when you see a particular mutation that's very common among natural sequences, it's hard to know if it's common because it just sort of got lucky and occurred in a virus that spread, or whether it's common because it confers some functional benefit. In addition, the natural evolution of the virus is driven by both immune selection and selection for protein function. In contrast, in our experiment so far, we're just looking at selection for protein function. That means that our experiments can measure the inherent mutational tolerance of the protein in the absence of any sort of immune selection. So for instance, if we take an H1 hemagglutinin, we can look and we can ask, is the head of this hemagglutinin more mutationally tolerant than the stalk? And you might expect that the head or the top of the molecule would be more mutationally tolerant because it changes faster during natural evolution. But as I've already described to you, natural evolution also puts a lot of pressure for mutations to occur in the head because there are antibodies targeting the head. So when we look at natural sequences, formally we cannot say if the head is more variable than the stalk because the head is better at tolerating mutations or simply because it's under more pressure to accumulate mutations. In these experiments, we can measure mutational tolerance in the absence of immune selection, and when we do that, we see for this H1 hemagglutinin, in fact, most sites on the head of the molecule are more mutationally tolerant than most sites in the stock of the molecule. Sort of very interestingly and surprisingly, we've now done that same experiment for a recent H3 hemagglutinin, the Perth 2009 vaccine strain, and there we see much less of a disparity in mutational tolerance between the head of the hemagglutinin and the stock of the hemagglutinin. We're very surprised by this result, and we're currently sort of following up on it more. So now that we have these data for the mutational tolerance of the Perth 2009 hemagglutinin, we might want to ask, can we use these experiments or these maps of mutational effects that I've just described to better understand the evolution of the virus in nature? And this, of course, is the real goal of doing these experiments because we really care about influenza not in MDCK or A549 cells but in humans. So as I've already described, over time viral strains rise and fall in frequency during influenza's evolution. So you may have, you know, new variants that emerge. Initially there may be several variants, but typically as we look over time, some of those variants will go up in frequency and then go down, and, you know, eventually typically certain variants will fully win out. And again, understanding these dynamics are very important because it's these dynamics that sort of govern what strain we would like to have in the annual influenza vaccine. If we're choosing the influenza vaccine right here and hoping it's effective over here, we would like to have chosen the green strain rather than the blue or the purple strain. So one question we can ask is now that we've measured the effect of all of these mutations, again in a very simplified system, it's a cell culture system, but there's still measurements, which are more than we had before, we can ask do the strains that rise in frequency have better mutations than the strains that fall in frequency. And so to do that comparison, we take the effect of each mutation on viral growth as we measured in our experiments, and then we compare that to the maximum frequency that viruses that have that mutation reach during natural evolution. And here I'm showing those data just for human H3N2 strains isolated after 2009. So we're asking, can our experiments on this Perth 2009 virus predict what might happen sort of after 2009, and we're also restricting the analysis to viral sequences that haven't been passaged in the lab because lab passaging can create spurious mutations. And we do see that there's some correlation between the effect of mutations on viral growth in our experiments and how well the virus does in nature. In particular, mutations that we measure to be fairly unfavorable rarely reach high frequency in nature, whereas most of the mutations that reach high frequency in nature are ones that we measure to be favorable. And we can look at this in another way. Here is a distribution of effects of mutations among the viruses that reach high frequency down here at the bottom, and among the mutated viruses are the mutations that only reach low frequency up here at the top. And you can see these bars show the median effect of the mutations, And you can see that the typical mutation that only reaches low frequency has a worse effect on average than the mutations that reach high frequency. This dotted line shows the effect of a neutral mutation. And so you can see that particularly among the mutations that reach high frequency, it looks like this dotted line goes almost to the center of the distribution. So again, this drives home a key point that I was presenting earlier in the talk. When we're just looking at how mutations affect viral growth, we don't think it's that H3N2 influenza is actually getting better at replicating in humans. It's already quite good at that. What the successful strains are doing is managing to acquire mutations that keep their replicative ability at about the same, because when I put my list of things that influenza has to do, there are two major selective forces during its evolution. It has to maintain function to enable viral growth, and what I've just shown you, that the strains that do well get mutations that don't hurt this, their ability for viral growth. So far I haven't described anything in the second factor though. In addition to maintaining function to enable viral growth, we think these successful strains are also accumulating mutations that enable them to evade antibodies. So now can we look at that part? So for the last section of the talk, I'm going to describe what was actually in the title, which is complete mapping of how mutations affect virus neutralization by antibodies. So the experiment we're using here is conceptually almost identical to what I was presenting earlier. But before I describe how we do the experiments, I just want to take a step back and ask, what would a complete map of how mutations affect antibody neutralization actually look like? Because before you do an experiment, it's always kind of good to think about what data you would like to get out. So the reason that we're interested in doing this experiment is we know that mutations can affect antibody neutralization. In this cartoon, the virus with the yellow mutation does a much better job of surviving this antibody than the viruses with the blue, green, or black mutations. So we would say this orange mutation is an antibody escape mutation or an antigenic mutant. And so the way that we normally look at that in virology is by doing something called the neutralization curve. Or sometimes a hemagglutination inhibition assay. And what those curves do is for each mutation, we ask how much of the virus is able to survive at a given antibody concentration. So when you use a very, very low concentration of antibody, all of the viral variants will survive. But as the antibody concentration gets higher, some of the viruses will be inactivated by antibody. And if the orange mutation is a resistance mutation, that means that you can increase the antibody concentration more before the virus is inactive. So this is sort of the typical representation of these data, but it's not the one I'm going to use. And the reason is that this is a useful representation if you have data for several mutations, like here. But if we're going to measure the effect of all the 10,754 mutations, that becomes a lot of neutralization curves to look at, and it stops being a very informative way to present the data. So I'm instead going to present the data in a different way. At a given antibody concentration, such as this one indicated by the dashed line, I'm going to write the height of each letter as proportional to the fraction of viruses that have that mutation that survive the antibody. So when we look at our neutralization curve, you can see that the wild-type virus in black and the L2C and S3A viruses are mostly neutralized by that antibody, so their letters are very short. But this K, or this V1K mutation, is mostly resistant, So its letter is very high, and I'm going to be using this representation because it's much easier to use when you have data for all of the mutations. So then the next question, if this is what we mean by a complete map of how all mutations affect antibody recognition, is how do we obtain that map? So one way that you could imagine doing it is to simply make each of these possible mutants and measure their neutralization curves one by one or maybe you would use a hemagglutination inhibition assay, which gives at least somewhat similar information. The main downside of doing that is it's a lot of work if you're doing it for all 10,735 mutations. And so what we would like to do is skip the neutralization assays and go directly from our virus mutants to getting this map in one experiment. And again, the way we do that is this mutational antigenic profiling approach. And how does that work? As I described in the first part of the talk, we take the hemagglutinin and we simply make all of the amino acid mutants of it in a single PCR reaction. We then use these mutants to generate this pool of mutant viruses, which will now have all of the mutations to hemagglutinin which are compatible with viral replication in a permissive cell line. We then take this virus library and we incubate it either with antibody or without antibody and put it on cells. We then collect the viral RNA from these cells, accurately deep sequence to identify the mutations, and then we can see which mutations survive the antibody versus which mutations can survive only in the absence of antibody. And that allows us to generate this sort of off map that I was describing a little bit earlier. So the result when you do this for a monoclonal antibody would look something like this. So this is an H1 hemagglutinin with a strain-specific monoclonal antibody from the classic studies by Jonathan Udell, Andrew Caden, and Walter Gerhard. So as you might expect, at most parts of the protein, mutations have no effect on how well the virus survives the antibody. There's a small number of sites here where many mutations affect how well the virus can survive the antibody. And we can— whoops. We can— ah, sorry, lost control of the computer here for a second. We can zoom in on that particular region and see in this particular part of the protein mutations are having large effects on antibody neutralization. So the first question with any high-throughput experiment is, do these results make sense? So I'm going to first validate this approach by mapping escape from some strain-specific antibodies to hemagglutinin. So these strain-specific antibodies again are from these classic studies from the Gerhard lab, including work by Jonathan Udell and Andrew Caton, and these are antibodies that bind to lab-adapted, so PR8 and WSN H1 hemagglutinins, and they bind up here on the globular head. So these are strain-specific antibodies. If you ask which strains they neutralize, they really only neutralize these H1 strains. So to facilitate visual comparison, first, rather than showing this full map of each individual mutation, I'm going to simplify it by just drawing a line that indicates the height at each site, or the average height of the letters, and that indicates how much mutations at that site have an effect. So when we take these three strain-specific antibodies, we can see that for each antibody there are mutations at a specific subset of sites that mediate viral escape. We can take those particular sites and we can map them on the protein structure and visualize the mutations in detail. So for H17L19, we can see that these small subset of mutations are the ones that affect antibody neutralization And as expected, they all mapped one region that looks a lot like an antibody footprint. And we can see that the same thing is true for the other two strain-specific antibodies. And so these antibodies are all binding to some of the classically defined sort of antigenic sites on H1 hemagglutinins. We also see when we look at these data that there's a lot of sort of mutation-level nuance. So at some sites, it looks like only certain mutations escape from the antibodies. We can validate these results in neutralization assays. So for instance, here's the data for one of the antibodies. For instance, if we zoom in at site 141, we can see that interestingly only mutations to the negatively charged amino acids aspartic acid and glutamic acid are predicted to affect neutralization of this virus by the antibody. In fact, we can see that's true. Here's the wild-type virus. It's potently neutralized by the antibody. Here's this H141E mutation. It escapes the antibody very potently. But if we make another mutation at this site, such as H141A, it has no effect on antibody neutralization. So you can see that if you really want to understand how mutations affect antibody resistance, you have to take a very nuanced approach and look at the different possible mutations individually. And we now have lots of this sort of validation data. I'd refer you to this paper So I'm not going to go through all this validation data right here. So you can now ask, what about broadly neutralizing antibodies? I've just shown you what happened with these strain-specific antibodies that bind on the hemagglutinin head, but as I'm sure most of you know, one of the more exciting things in the influenza vaccine research field recently has been the recognition that there are actually quite a few antibodies that will bind down here to parts of the molecule that are more conserved, and as I've shown you, at least for H1 hemagglutinins, inherently harder for the virus to mutate. And so you now might ask, can the virus escape from these antibodies in the same way it escaped from these three strain-specific antibodies I described? And I'm going to talk about two of these broadly neutralizing anti-stem antibodies. One, C179, neutralizes group 1 hemagglutinins, and the other one, FI6B3, is the broadest, or at least among the very broadest antibodies that have ever been described against influenza hemagglutinin. It neutralizes both group 1 and group 2 hemagglutinins. So we did the same experiment, systematically map how much each mutation affects neutralization by these antibodies. And again, this is with an H1 hemagglutinin. So to just refresh your memory, when we look at the strain-specific antibodies that target the globular head you see these mutations that have very large effects on antibody neutralization. When we do the same experiment with the broadly neutralizing anti-stock antibodies, at first it almost looks like a flat line. You're seeing no mutations that have this very large effect on the virus's ability to survive the antibodies. When you look a little bit carefully, you see that in fact it's not a totally flat line. There are some sites where mutations are selected. They're just sort of smaller in this representation, which means they're having a smaller effect on virus antibody neutralization. So we can now ask, is this just noise, or are these really mutations that are having an effect? So in fact, the small effect mutations do appear to make sense because they are, for the most part— this is an exception here— in the antibody binding footprint. So you can see that the mutations that have the largest effect in our experiment are in regions where the antibody is known to interact with the hemagglutinin. And we can also validate the small effect sizes of these mutations. So for instance, if we look at the site where the mutation has the largest effect on neutralization of this H1HA by Fi6B3, this is a mutation that introduces a glycosylation site at 278, so it puts an N-linked glycan motif here. If we look and we do a neutralization assay here, we can see that in fact both of these mutations do make the virus more resistant to the antibody., but they only make it more resistant by a couple of fold. It's nothing like the 100-fold effect sizes we were seeing with the strain-specific antibody. And again, to validate that the effect is specific to these mutations, if we put in an alanine, which is not selected by the antibody, we see that the virus does not become more resistant at that site.

So I think this sort of draws forth an interesting comparison. When we do this complete mapping of mutational effects for these narrow strain-specific antibodies, As expected, we see that there are escape mutations. There are escape mutations like this one. They make the virus essentially completely resistant to the antibody. But at least for a very broad antibody like FI6B3, and at least for the viral strain that we used in these experiments— again, we haven't tested this with other viral strains— we see that the virus can accumulate antigenic mutations that make it less sensitive to this broadly neutralizing antibody, but the effect sizes of these antigenic mutations are much smaller. And I think that's an interesting fact, and it would be interesting to investigate whether this is because, because somehow the antibodies binding to energetics are more distributed, or in some way there's just no mutation that can completely abrogate the binding. Again, I think, you know, before we get too excited about this conclusion, we need to validate it in other viral strains, because you can imagine that certain strains may have access to different mutations. But I think this is a very sort of comprehensive way of validating that While indeed any antibody can select antigenic mutations, they can't all select antigenic mutations of the same effect size. So because we're almost at 40 minutes now, I just kind of want to wrap up with some conclusions from this talk. So first, I kind of tried to make some general points that flu HA can escape antibodies while maintaining function. That's why we're all still working on a universal flu vaccine. Second, that we can now start to use new approaches which leverage sort of deep sequencing and computational biology to map in very systematic ways how mutations to the virus affect function. And then I showed you how we can actually use these data to some degree to identify which viral strains are successful in nature. I've also shown you how we can now start to map how all mutations affect antigenicity and compare how the virus can escape from different antibodies. And I think a really exciting area in the future is really trying to integrate these sort of functional and antigenic measurements to better understand and perhaps better forecast the virus's evolution in nature. And to end, I'd like to thank the people who helped out with all of this work. So Mike and Juhei, who are both graduate students in my lab, did all of the experimental work I presented here. We also collaborated with many of the computational evolutionary parts with Trevor Bedford and John, who's a graduate student in Trevor's lab. And then finally, for a bit of the antibody work that I presented, and particularly some very interesting antibody-based work that's going on right now, we have a great collaboration with Scott Hensley's lab. I'd also like to thank my support, which comes from the Fred Hutch, the Genome Sciences Department at University of Washington, which are my two major affiliations. And then HHMI, the Center for Infectious Disease Dynamics, or the Center for Inference of Dynamics of Infectious Diseases at the Hutch and the NIH. And finally, I'd just like to note that for any of you who are interested in trying to do the types of experiments that I was describing in the second half of the talk, we have some software that's designed to analyze these sorts of data that's freely available on GitHub. And I think I can now take questions if any of you have them.

Thank you, thank you, Dr. Bloom, for such a wonderful webinar. We will now open the webinar up for questions. Please use the question box in the GoTo interface to type in your questions, and we will read them out loud. So we'll give everyone just a minute to formulate their questions. Okay, Dr. Bloom, the first question is, what is the method you got the— what is the method in which you were able to generate the mutant library?

Can you explain more about that?

Yeah, so the method, and we have actually a— so for someone who's actually interested in doing it, we have a webpage, you can send me an email. If you Google my name, you'll come up with my email. I can send you the link that has like a very detailed protocol. But the basic idea— and I apologize that I do not have slides in here that have this— is we order off— if the gene is 565 amino acids long, we order 565 primers, each of which sits down at a different position in this gene. So you can just order, turns out, primers like this from IDT, and they come resuspended spend it in a 96-well plate already for you. So it's a couple thousand dollars. And then with each of those primers, we do PCR. So each primer puts an NNN mutation in a different codon going in one direction. Then we have a single primer going in the other direction. We do this all in one reaction. And this makes a series of staggered fragments that each have a mutation at a different point. And then we take those fragments and do PCR back in the other direction. And it ends up sort of making this library. So it's essentially, several cycles of just a bulk PCR with all of these 565 primers in there at once. So that's probably the best I'm going to be able to explain it without a slide, which unfortunately didn't include. But like I said, if you either go to our lab GitHub page, or if you can't find it there, if you just send me an email, I can send you a link to a detailed protocol as well as a computer script that designs the primers.

Thank you, Dr. Bloom. Our next question is, how does your in vitro mutational analysis compare with diversifying selection analysis naturally evolving influenza strains?

Yeah, that's, that's a really great question. So, and so something that hopefully was clear in the talk, although I didn't really draw it out, is so, so far you can only obviously do the analysis of the natural strains or strains that are sort of contemporarily circulating. So the experiments I was describing were done with the WSN strain, which is a lab-adapted strain of influenza, and then the Perth 2009, which is a contemporary H3 strain of influenza. So. So we can really only do very powerful evolutionary analyses with the PERC 2009 strain because the WSN strain is not all that much like any strain circulating in nature. So when we do these analyses, what we see is that broadly the effects of mutations we measure on viral replication and cell culture describe the rates of evolution at different sites in nature fairly well. But we also can identify, and I think this makes sense, because most of what hemagglutinin has to do is fold and help the virus enter cells. But then you would expect that there are some selection pressures that are present in nature that are not going to be present in our experiments. So the virus in nature is infecting a much more austere, complex environment. It's our actual airways where the cells are a lot more complicated than a cell line.

And also in natural evolution, there are clearly pressure from antibodies.

So one of the things that we've been doing is working on methods to identify sites that are evolving differently in nature, for instance, faster in nature than we would expect from our experiments. And we anticipate that those are then the sites where the immunity is strongly targeting the virus. So in a way, this is very similar to doing like a dNBS analysis for which sites are changing rapidly in nature, but we're able to do a more powerful analysis because we've already used our experiments to say how fast we think the site should evolve. So anyway, that's sort of how we're trying to look at this question, but we have gotten the data for the recent H3N2 strains sufficiently recently that I can't yet really tell you any conclusive results for now.

Thank you, Dr. Klum. The next question is, how do you foresee determining what the mechanism is by which each amino mutation that is not tolerated in the absence of selection affects the virus biology or HA structure?

Yeah, I mean, that's a great question. I mean, I don't think we'll probably ever determine the mechanisms of all of them. I mean, as As everyone probably knows, protein, particularly proteins like hemagglutinin, are extremely complicated. And even with all of the amazing work on the biochemistry and structure of the protein, there's still a lot of things that we can't fully rationalize. But I think a tremendously interesting area, and I'm always sort of excited when people who really sort of do biophysical and modeling work and also structural biology are interested in these questions. I think there's a lot of work to be done to sort of take these functional measurements and really integrate them with structural data and try to figure out why specific mutations are having specific effects.

So some of it, when we just look at our proteins, we can rationalize.

So often, you know, we'll see strong selection for hydrophobic amino acids in the core of the protein. Sites that we know are important for receptor binding tend to be sites where there's pretty strong selection to maintain the amino acids. But I think a very interesting thing that could be done, and I mean, we're trying to do it, but I would encourage anyone who's interested to do it because this data is all out there, and really, if you know a lot about hemagglutinin, it's a great opportunity, is you can take our data and you can look at it in the context of the structure of hemagglutinin, and if, for instance, you see a region of the protein where mutations are strongly selected against and it's not known why, you could hypothesize there's some unknown functional constraint on hemagglutinin, in that region. So I think those sorts of analyses are very, very interesting to do, and right now I would say we have not gotten nearly all the mileage as possible to get out of these data.

Thank you, Dr. Bloom. The next question concerns— it says, concerning the narrow and broad antibody, are they region-specific? For example, the narrow prefers the head and the broad prefers the stalk.

So for, I mean, again, I would be hesitant to totally overgeneralize what we did there, but in general, antibodies against the stalk tend to be much more broad because the stalk is both well conserved in nature and, you know, as our data showed, at least for the H1 hemagglutinin, inherently less tolerant of mutations. So it tends to be the case that more broad antibodies are against the stalk more of the antibodies against the heterostrain-specific, although there have been some, some interesting exceptions that have been described in the literature recently.

Thank you, Dr. Bloom. Our last question currently is, how easy would it be to do similar experiments with multiple antibodies, or even better, with a mixture that's somehow representative of an antibody repertoire like the one the viruses would encounter in vivo?

Yeah, so that's a great question. And Juhye in my lab, who's the grad student working on this, is really sort of working on this question right now. So she's taking antibodies and mixing them together, and she's also taking polyclonal sera. And a lot of this is work that we're kind of doing in collaboration with Scott Hensley. And I think it's a really fascinating question to think about. So it's very clear how a virus escapes a monoclonal antibody. The antibody binds somewhere, the virus gets mutation and gets away. It's much less clear how the virus escapes a highly polyclonal response. So for instance, during his postdoc with Jonathan Udell, Scott Hensley showed that when you select at least certain strains of influenza with very highly polyclonal responses, you no longer get mutations where any given antibody binds, you just get mutations that sort of make the virus a little bit more resistant to all antibodies by helping it enter cells faster. However, something else that's come out from a number of labs, including again much of it from Scott Hensley's lab, is that certain people have actually fairly immunodominant immune responses that are focused on fairly specific epitopes. So you could imagine that, you know, human serum may span some range from being very highly polyclonal, where it's very hard for the virus to get a mutation that fully escapes from, or even appreciably escapes from any of that serum. But some people may have very focused antibody responses where the virus may be able to get a mutation that escapes from a large fraction of that, that serum. And then you could even imagine that the natural evolution of the virus may not sort of occur symmetrically in all people. There may be certain people with highly immunodominant immune responses where escape mutations look more— this is all wild speculation. I should caveat it with that. But anyway, so right now we are starting to work with both some of these antibodies and some of these different sera and really see what these sort of comprehensive escape profiles look like. And so hopefully in a year or so we'll be able to give a similar answer to that question similar to what I gave, but rather than waving my hands and speculating, actually show you some real data.

Thank you, Dr. Bloom. We would like to just give everyone one more minute to write in any more questions that they may have. And while everyone is formulating their questions, I would like to remind everyone our next webinar is scheduled for August, um, Tuesday, uh, the 21st at 11 o'clock Eastern. We would encourage everyone to join us again next month. If there are no further questions, we'd like to thank everyone for the participation. If anybody has any lingering questions, feel free to email either the coordinating center, the Human Vaccines Project, or Dr. Bloom himself. And we thank everyone for a wonderful first webinar and Dr. Bloom for your your, for your presentation.