Scientist Stories: Paul Thomas, The Immune Response in Infection & Cancer

Welcome to the influenza webinar series hosted by the Human Vaccines Project. The Human Vaccine Project is a nonprofit profit public-private partnership with the goal of decoding the human immune system to accelerate the development of vaccines and immunotherapies. Before we begin, we are pleased to announce that registration is now open for an upcoming scientific summit on the universal influenza vaccine. The meeting coincides with the centennial anniversary of the 1918 influenza epidemic and will feature prominent researchers and thought leaders to discuss cutting-edge influenza research. Spaces are expected to fill up early, so we encourage you to register soon. A registration link is now being sent via the chat box on your GoToWebinar platform. An email with more information will be sent out to the group this afternoon. In addition, we're pleased to announce that we are now accepting applications for the 2019 Michelson Prizes. This year, the $150,000 prizes have expanded to include 3 focus areas related to vaccine discovery. Human immunology, computational biology and protein engineering, and neglected parasitic diseases. A global search is underway for young investigators who are applying disruptive research concepts and inventive processes to advance vaccine and immunotherapy research for major global diseases. You can learn more at www.humanvaccinesproject.org. Michelsonprizes. That's M-I-C-H-E-L-S-O-N-P-R-I-Z-E-S. And please be sure to pass along this information to other interested researchers. And, and now for the webinar. In order to make real progress toward the development of a universal influenza vaccine, collaboration among researchers around the world is essential. We hope this month's webinar will foster scientific foundations needed to advance influenza research. The series of webinars will be broadcast on the 3rd 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. Following the presentation, we will have time for questions from the audience. Please type your questions into the questions box in the GoTo interface. I'll read these questions following the presentation. And now it is my honor to introduce Dr. Paul Thomas. Dr. Thomas is an associate member of St. Jude Children's Research Hospital faculty in the Department of Immunology. His group studies innate and adaptive immunity to influenza with a special focus on T-cell receptor repertoires and CD8 T-cell memory. Dr. Thomas's research has been funded by several grants from both the NIH and private foundations. Today, Dr. Thomas will tell us about host immune factors that regulate outcomes and influence a vaccination. Thank you very much, and welcome to the webinar.

So thank you very much. I'm excited about this opportunity to present today. So most of the work I'm going to talk about today is actually from our studies on various either models of infection or actually infected humans, but I'll try to mention where I think this has special relevance to vaccination and where we're looking at aspects of vaccination with regard to some of these factors. I don't seem to have— let's see. Okay, so, so our view of influenza disease is that a number of different factors contribute to the particular outcomes that follow from an infection. And so these include things like your current immune status, whether or not you received the vaccination, Various environmental factors that sort of set your baseline level of health, the type of virus that you get, if you get a very severe virus, various host factors, some of which I'll talk about today, and then some host genetics that influence your immune status. And I sort of view all of these things as things that tether you to various points sort of on the edge of a cliff, and that when you get disease, you fall off the cliff, and how far you fall is determined by how well you're tethered based on these various factors. And so the first part of the talk today, I'm going to talk about various genetic factors that are a particular genetic factor that we think influence the outcome of infection. And so a few years ago, we published a result from a collaboration with Tomer Hertz at Fred Hutch at that time that was looking specifically at HLA as a contributor to influenza susceptibility and the magnitude of the influenza CD8 T cell response. And so HLA is obviously the antigen presenting molecule for both CD4 and CD8 T cells, and here we were primarily interested in HLA class I. And what Tomar had done was had developed an algorithm to determine what HLA alleles bound to the regions of the influenza virus that were most well conserved. And so this is plotting actually what he calls the efficiency score, so the ability to bind these conserved regions of the influenza virus for various HLA alleles against either the 2009 H1N1 or the 1918 H1N1, which as many of you know would share quite a great deal of homology. And there was one set of HLA alleles that were really true outliers for both of these viruses, and that was the A24 superfamily. And so this was a purely computational in silico predictive model of what HLA alleles would contribute, would be particularly poor binders, and presumably may translate into having a poor CD8 T cell response. So in a parallel analysis, what he did was, because in 2009 there was pretty reasonable country-level mortality data based on the pandemic because of the greater level of surveillance that occurred in that year, He looked at HLA carriage frequency by country and correlated it with the mortality level of the 2009 H1N1 in that country. And only one HLA actually was statistically significant in terms of its carriage frequency and the mortality levels observed across countries, and that was the A24 superfamily again. And this is the correlation. It was an R-squared of 0.37 with a p-value of 0.31. So the higher the A24 carriage frequency in that country, the more likely there was a higher mortality in that country. And so what we did in our group was we took samples from the 2009 H1N1 and looked at the magnitude of the CD8 T cell response as a function of this efficiency score that Tomer calculated based on HLA levels. And so we just took a collection of people, measured, you know, determined their HLA. He calculated an average efficiency score across the entire person's HLA cluster for A, B, and C, and we plotted that against the overall magnitude of the CD8 T-cell response as measured by the, an ELISpot assay, and there was a very nice correlation. Again, R-squared here is 0.65 with a pretty low p-value, and one thing to point out here was that this significance actually increased when we accounted for prior vaccination history, which suggests that vaccination is playing into the magnitude of CD8 T cell response and the ability to recall the CD8 T cell response efficiently. And so this is older data that's published, but is one obvious genetic factor that could be influencing CD8 T cell responses and the immunological and clinical outcomes to influenza specifically. And so moving on from that, where this, This specific genetic story I'm going to tell you ends up being about CD8 T cells, so I'll just preview that for you at the end. And we think has a major relevance to T resident memory and the ability to elicit and maintain T resident memory cells that are particularly protective during an influenza challenge. And so this started with our interest in a gene called IFITM3. The whole IFITM gene family is known to limit infection of numerous viruses, including influenza, Dengue, HIV, West Nile, pretty much any enveloped virus. And IFITM3 in particular has been shown to be antiviral by inhibiting the entry of viruses into the host cell cytoplasm by preventing viral fusion with cholesterol-depleted endosomes. So the TM part of IFITM is a transmembrane protein, so it sits on the endosome and it blocks viral fusion from the endosome into the nucleus, or into the cytoplasm. IFITM3 had been of great interest for the flu field because of a prior history of in vitro studies that had identified it as a particularly potent regulator of influenza infectivity, and, and subsequent series of papers identifying a particular SNP in IFITM3 that was associated with severe disease again in the 2009 pandemic and then in the subsequent H7N9 outbreaks in in Southeast Asia. And so that particular SNP was called RS12252, which I'll refer to as RS122 for the rest of the talk. And it has the minor allele of this SNP, which is in yellow here, has a very high carriage frequency in Southeast Asian populations. This is from 1000 Genomes, but actually has a very low carriage frequency in Western European and Caucasian populations. So that's again, the yellow here, so very small frequency. But despite that, it was originally identified in a cohort of Caucasian individuals that had severe H1N1 disease following the 2009 pandemic in the UK. And this was this Nature paper here, Ipidem-3 restricts morbidity and mortality associated with influenza. And then there were some two subsequent quick follow-ups on the 2009 H1N1 in China and the H7N9 also in China. Showing that individuals with this higher, with this risk allele, or the minor allele, RS222C, was associated with more severe disease. Several subsequent studies have failed to sustain that association, particularly in Western European populations. So the original report in the Callum paper here suggested that this could be a a splice variant because of where it fell in the gene. Subsequent analyses have shown that it doesn't appear to be a splice variant or associated with alternate splicing in the gene. And I have some data at the end if people are interested that I can show about that. And again, possibly because of the very low carriage frequency in these Western European populations, it's been hard to identify a subsequent association, although the association in Southeast Asian populations has been maintained quite well. So we set out to try to figure out what was going on with IFITM3. There was clearly very strong in vitro data that it played a major role in influenza infection. There was very strong data from the Southeast Asian cohort suggesting that it played a major role, and particularly via the SNP rs122. And so we decided to look and see if there was maybe some other polymorphisms throughout this gene that could account for the kind of contradictory results that were being observed, but also emphasize the importance of this gene in regulating disease after influenza infection. And so the approach that we took was to perform a candidate search of SNPs within the IFITM3 locus based on ENCODE data and other data that we had available to us, first filtering on SNPs that had, you know, a relatively high minor allele frequency, just so that we would be able to do the analysis that we wanted to do. We also then restricted ourselves to SNPs that were eQTLs for IFNM3. And so this is a SNP that's associated with the expression of the gene, and that's something that is also distinct in the analyses I'm going to show you from RS12252. So RS12252 is not an eQTL for IFNM3. It's not associated with expression levels. And then we were— because we were interested in trying to understand this divergence between the results coming from the Western European and Caucasian cohorts with the Southeast Asian cohorts, we decided to look for SNPs that would potentially have a differential allele frequency than RS12252, and that maybe could be in some form of disequilibrium or linkage with RS12252 that might explain some of the effects that were being seen. And after filtering on all of those screens, We ended up with two alleles, or two SNPs, sorry, RS642 and RS344. And RS642 did not look like a particularly good candidate to be a regulatory SNP for various reasons, but it wasn't in a particularly interesting region of the gene, whereas RS344 was in the core promoter and overlapped with a large number of potential transcription factor binding sites. And getting to the point of its differential allele frequency compared to RS122. This is from 1000 Genomes, the distribution of RS344. So it's the minor allele is quite common in Western European populations, as you can see here in red. Sometimes it's even the majority, so it's not even the minor allele. And then Southeast Asian populations, it's extremely rare. And so it seems to actually be really in opposition to RS122. So you sort of either have one or the other as a minor allele. And a lot of these data, though not all of the data I'm going to show, were published last year in this paper here, but I'm going to show you some of the follow-up data we've done since then. But just to go through some of the core data that were in that paper, so this is where the SNP of interest fell. So this is IFNM3, it's on the reverse strand, and so this is the 5' end and this is the 3' end. It's backwards from how you might normally look at it. Our SNP sits right here in the core promoter, which is also in the 5' UTR, and this is the ENCODE data showing a large number of potential transcription factor binding sites here that overlap with where the SNP falls. And as I mentioned, we screened on the fact that it would be an eQTL, but we also expanded that to look in a number of cohorts. This is in our cohort of influenza-infected individuals here in Memphis, and this is the homozygous for the risk allele AA and homozygous for the protective allele GG and the heterozygotes, and you can see there's a nice sort of stair-step association with gene expression. This is just from PBMC RNA levels. This is from a cohort of severely ill kids that was run out of Boston Children's Hospital, and again you see the same stair-stepping phenomenon where the pets have an intermediate phenotype. And then this is from the Broad Phenogenetic Consortium, and this is now dendritic cells derived from PBMCs stimulated with a number of different stimuli. And again, you still see across all of those stimuli the same much lower levels of expression with what we're calling the risk allele, which is the A allele, versus the protective allele, which is the genome. And so we wanted to look at actually associations with outcome, and we looked across 3 cohorts. And so we had a cohort that we had run in Memphis for 5 years called, that we called Flu09. This was a community-acquired influenza infection, mostly mild disease with some severe disease, as characterized by the need for hospitalization at the extreme end, and we rated these people based on their self-reported symptom scores, so low to high symptom scores. And the second cohort was of the— that I mentioned previously was actually run out of Boston Children's Hospital, was a national cohort called PICFLU. These are all critically ill children with no known underlying conditions. Everybody was in the ICU, and so they're really at the extreme end of the spectrum, and so they span from survival to death was our poor outcome there. And then at the other end of the spectrum, we had access to data from a challenge study from Genentech. So these were all healthy young adults that were infected with an H3N2 virus. And here we just looked for the onset of viral replication. And so we set a cutoff for delayed versus early viral replication, and that was sort of our poor versus or sorry, our poor versus our protective outcome. So we really have a really broad spectrum of disease outcomes here that we're looking for associations with our particular SNP RS344. These are the data. So this is the Flu09 cohort, the Genentech cohort, and the PICFLU cohort. On the left for all three cohorts is the allele frequencies for the mild, Versus on the right for the severe. And the risk allele, the homozygous risk allele here is in purple, the hets are in orange, and the homozygous for the protective allele are in white. And so really you're looking for sort of an increase in purple from left to right and a decrease in white. And you can see that there is an increase in purple across all 3 groups. There's a decrease in white across all 3 groups. These were all statistically significant. The odds ratios ranged from about 2.4 to 4. Depending on the cohort. And a particular note in the PICFLU cohort, nobody that died had, were homozygous for the protective allele. So everybody that died had at least one risk allele. And you can, even though statistically this isn't really appropriate, you can look at the sort of proportion of risk alleles and protective alleles between the mild groups. And the mild groups here really had, you know, really different different levels of disease as well because the Pit Bull cohort had much more severe disease as a mild group, and you can see they're already overrepresented with the risk alleles there. So this we thought was really quite strong associations. I'm not showing you the data, but there was no associations with RS122 in any of the cohorts. But again, these were all cohorts of the West with very few Southeast Asian participants. Flu09 was majority African American, and Genentech and PickFlu were either majority or wholly Caucasian. So we now had an association with our SNP. It was— we had established that the SNP was an eQTL, and so we wanted to see whether the SNP was actually causing differential binding to transcription factors. And because it fell on the core promoter and seemed to be in this region of potential binding to a large number of transcription factors, it seemed like a good hypothesis. And so we ran EMSAs, so gel shift assays, either with a probe containing the risk allele A or the protective allele G, and in a number of different cell lines with increasing concentrations of the probe. You can see Across these different concentrations, the G allele, which is the protective allele, always bound more total proteins. So this band up here is darker versus the bands in the A lanes. And then we could do something called a WEMSA, which is where you run an EMSA gel shift assay and then you blot for the particular proteins that are present in the, in the band. And so this is blotting for IRF3, which is a known inducer of IFN-M3 and of of the whole interferon response pathway, and there was significantly more IFN3 binding to the G allele than to the A allele. But the one protein that we found that bound more to the A allele than the G allele was this protein CTCF, which bound significantly more to the A allele. And so this is by WEMSA, and then we could also just take the probes and add purified CTCF that we made and look and see whether or not it bound more. And even in that system, purified CTCF bound to the A allele more than to the G allele. And so we're very excited about this because CTCF is really a master regulator of transcriptional activity because it's an organizer of chromosomal structure. And so in particular, CTCF is responsible for forming what are called topologically associated domains. And so it binds to itself and creates these little microdomains by DNA looping. And so you can imagine if this is a chromosome, each of these pinch points here represent a CTCF binding site, which will loop around and bind to itself so that all of the genes within that region— and this is basically just showing a correlation plot of the genes within that region— are correlated with each other in terms of their expression level. It doesn't mean that they are all up or down at the same time, it just means that the regulation is correlated. So A move in one will be reflected by a move in the other, either up or down. And CTCF has been studied extensively in cell biology and in cancer cell biology, especially because the destruction of a CTCF site or the removal of the CTCF site changes the chromosomal architecture and actually the long-range regulation of all the genes within that topologically associated domain. And so that's represented here. We have these two CTCF sites here in this particular individual. And if you mutate one of them, what happens is you destroy this little micro tab here and it just merges with the larger domain and you have this much larger region of sort of overall lower correlations, whereas you maintain this region on the right here that's highly correlated. And a point I want to mention here, which is sort of counterintuitive from how you may think about gene regulation is that CTCF binding is inhibited by methylation. So CTCF won't bind to a methylated site, and so the methylation of a site is sort of the equivalent of knocking out a CTCF site and potentially opening up one of these TAD domains. So we wanted to look and see if we had evidence of this TAD-like structure in individuals with a binding site for CTCF, which is that risk allele, versus individuals that we think are missing a binding site for CTCF, which is the G or the protective allele. And so here we're looking at individuals that have one of the risk alleles, and you can see we're looking across the whole IFITM3 region. So our SNP is essentially right here in the promoter of IFITM3, and we're looking both upstream and downstream of the SNP, really quite a long way away, so several hundred kb away, encompassing all the genes in that region. And then we're looking at the same genes over here. And so this is just a correlation plot of looking at in these individuals whether or not the gene expression from PBMCs, just RNA-seq data, are correlated at long range. And you can see there's these sort of microdomains where genes that are really close to each other, sort of on the diagonal, are correlated with each other. But genes that are really far away from IFITM3 in particular are not correlated with IFN3 expression. If we now look in individuals that we think are missing a CTCF binding site, so they're the equivalent of this group here where the TAD has been disrupted and you've just merged into this sort of super domain where you, where you now allow for these longer range interactions because you formed this much bigger loop. That's the homozygous individuals over here for the protective allele, and you see all of a sudden in this group of individuals we now have really a large number of long-range correlations, and they can be positive or negative. Again, it doesn't matter the direction that they go, it's just that they are correlated with each other, particularly for IFITM3. And so we're looking at genes that are now 200 to 300 kb away that are now correlated with IFITM3 gene expression, in addition to maintaining this large local group of gene expression. And there's a lot of interesting genes in here. So there's some innate immune genes and a lot of P6 SIGURR is a cofactor for toll-like receptor binding, and B4Galaxy4 plays a role in the constellation of the respiratory epithelium. So there's potentially pleiotropic effects of changes in gene expression at this locus beyond what we think IFNM3 is doing on its own. And so I mentioned as a placeholder in your mind that CTCF binding is inhibited by methylation, Actually, so in addition to, we think, the G allele being a poor binder for CTCF, which we showed in the gel shift assays and the WEMSA assays that I showed you earlier, the G allele introduces a CpG motif that could be methylated. So this is a potential methylation site. It hadn't been established as a methylation site previously though. And so we decided, well, we thought that this was potentially a further way that CTCF binding could be regulated. G, in addition to not binding CTCF as well as the allele on its own, if this G is also methylated, it would further diminish CTCF binding and potentially open up this TAD. And this result here was just so dramatic that we felt like a really powerful regulation must be occurring. And so we decided to look in collaboration with Ben Youngblood's lab, who's an expert in DNA methylation, as to whether or not this the G allele really did introduce a functional CpG site and whether it was actively methylated in relevant cell types during influenza infection. So to do that, we first went back to our in vitro assay. This is adding the purified CTCF again where we're using a methylated probe or an unmethylated probe. So this is the unmethylated A probe, methylated A, the G probe, and the methylated G. So there are two other CpG sites in the probe that we're using. And so even though we get rid of one of the CpG sites in the A, with the A allele, there's still two other sites that can be methylated that are potentially disrupting CTCF binding here. The G allele binds less actually than even the A methylated allele, and then the G methylated allele binds the least of all. And in thinking about this, we decided to start looking at what cell types we thought this would be most relevant in. Initially, we were focused on epithelial cells and monocytes as potentially the most relevant cells in the early infection period in the lung. But there was this paper in 2013 on IFITM3 in the mouse model from Linda Joaquin and Jose Villadongos that had shown that IFITM3 in the mouse model seemed to play a particular role in protecting memory CD8 T cells, and specifically resident memory CD8 T cells in the lung during secondary infection with influenza virus. And this just shows that in the bowel and the lung, if you knock out IFN-M3, there's a real survival disadvantage here between IFN-M3 high and low cells that's not actually present elsewhere in the animal. And so they hypothesized that this had to do with the ability of the virus to infect cells infected with, to infect CD8 T cells that were resident in the lung. And if IFN-M3 was present at high levels in the CD8 T cells, it would block that infection and help the CD8 T cells maintain their function. And so in addition to epithelial cells and monocytes, we looked at CD8 T cells as potential regulators of methylation at this locus. And so this is the data. Again, this is in collaboration with Ben Youngblood's lab where we looked at monocytes, differentiated normal human bronchial epithelial cells, and we're looking at individuals that are heterozygous or homozygous for the G allele., and looking for methylation at the RS344 site, which is this true site here in blue. And actually, the only cell type in which we were able to observe significant methylation was in CD8 T cells. And in addition to observing methylation at this site, we found that it was dynamically methylated during the differentiation of naive T cells into effector memory T cells. So there was a big increase in the methylation at this site as T cells differentiated into effector memory T cells. This is pretty exciting to us. And so we went and looked at CD8 T cells specifically by Western blot to see if IFITM3 levels were associated with— actually to see that it was expressed first of all at high levels in humans and whether it was associated with genotype. And you can see at the protein level, we still get that sort of stair-steppy effect where homozygote risk allele have the lowest level and homozygote protective allele have the highest level of expression. That's just quantified here. And then we had access to T cells from the site of infection, from the respiratory tract, from the FLU09 cohort. So this was from a nasal lavage that had been performed. This was on the diagnostic sample from these patients, and we just looked at the association between CD8 T cell number and frequency and genotype at that diagnostic time point. And individuals that were homozygous for the protective allele had significantly more CD8 T cells in the respiratory tract than those that were heterozygous, and we didn't have enough individuals that were homozygous risk allele to look at them. These were both highly statistically significant, and this was the only cell type we saw this association for. So we also looked at CD4 T cells and we looked at neutrophils, and there was no such association. What I've shown you in this first part is that genetic variation at the IFITN3 locus correlates with influenza outcome, and really suggests an important role for lung-residing CD8 T cells and protection. And this is all, I want to emphasize, in older children and adults. And the broader IFN-3 locus is differentially regulated across ancestral populations. And so we think that, you know, this difference between the studies that have been showing an association with RS122 and our studies that showed a very strong association with RS344 might have something to do with differential chromosomal organization, really mediated by CTCF, across these different ancestral populations. And so to look at this, we've been collaborating with Douglas Higgs in the UK to form CaptureSeq in individuals of diverse ancestral lineages with different genotypes for RS122 and RS344 at the IFITM3 locus. And we've been performing this assay called CaptureSeq which is a way of looking at chromosomal structure. And for the sake of time, I'm not going to go through the method in detail, but the way to think about it is that you pick a vantage point in the genome, and in this case we're picking the promoter of IFNM3, and you look out from this vantage point and you look at associations, chromosomal associations, at long range at a number of different sites across of the genome, and you can pick several hundred sites that you look for. And using next-generation sequencing and basically a ChIP-type approach, you can then sort of recapitulate the chromosomal structure from that vantage point. And so we've done this now for individuals that are divergent for RS344, so either the A allele or the G allele. And what you get are these tracks that look at chromosomal associations. And so again, up here is IFITM3, up here tiny on the reverse strand. So our SNP is basically right here. So this is our vantage point, and we're looking at the track for the allele. So these are the long-range and short-range interactions from this vantage point, whether you have an allele or a G allele. And this is the difference between them. And what you can see is, is there's these regions that are actually quite different between them, and I'm highlighting them here. Where there's more interactions either with the A allele or with the G allele. And we're going to blow that up here. And one particular interaction, it's actually this interaction here, which is much higher in individuals with the A allele, is with an interaction well upstream of the gene in an intragenic region that's a putative enhancer. And so this is the region that is having much higher levels of binding in the A allele than in the G allele. And interestingly, right here you can see CTCF is binding in this region. And so we think this may be one of the sort of microtads that's formed in the presence of the A allele rather than in the G allele. And this structure, we think, is really unique to individuals with Western European Caucasian ancestral history rather than those with Southeast Asian history. Okay, so that was a genetic factor that we think contributes to the burden of disease and that we think is really critical to think about in terms of vaccination for vaccination approaches that are using attempts to drive resident memory generation. And so our data suggests that individuals that have the risk allele are not going to be able to form protective resident memory populations that will efficiently control infection compared to those with the protective allele. And so it's something to consider both in trials and in targeting populations. So in the last about 10 minutes here, I just wanted to quickly go over a cell population that is frequently ignored, has been completely ignored from a vaccination perspective, and these are studies that are ongoing now in the lab. The data I'm going to show was just published today in Immunity, and it has to do with a different host factor that we think contributes to differential regulation of the immune response to influenza, and that is age, and age specifically regulating the type of immune response that's able to be mounted. And so the cell type that I'm going to be talking about for this second part here is the gamma delta T cell, and gamma delta T cells are formed very early in development, so you are born with a fairly full complement of gamma delta T cells. They have a much more restricted repertoire than alpha beta T cells, which are the T cells we're normally talking about, and I just spent the first two-thirds of the talk talking about. And they're thought to play a particularly important role in infants and neonatal individuals. They have an innate-like property in that they're able to respond very quickly to infection. And so they can be classified as producing IL-17 based on surface markers and transcription factor profiles. Or interferon gamma. As I said, they can respond very quickly, and they're thought to respond really not to peptide MHC antigens, but to stress antigens on self-targets, somewhat like NK cells, which have also recently been shown to have a sort of memory-like phenotype. And so what we found was that in adult mice, gamma delta T cell deficiency really didn't have a profound effect on the outcome of primary or secondary infection, whereas in neonatal mice, so mice that were less than 10 days old, animals that were deficient in gamma delta T cells actually were much more susceptible to death, even with an infection that would normally be survivable. And this was not associated with the control of viral titer or with the production of normal antiviral cytokines like interferon gamma. And so we had this association, and I'm gonna have a little sort of running chart of the pathway here on the left side of the panel here. With gamma delta T cells leading to a reduction in disease in individuals. And this was really pathological disease, so pathology was much higher in the gamma delta T cell deficient mice, and they would seem to be unable to resolve infection efficiently. And this was scored blindly by a pathologist, and so interstitial inflammation and bronchial hyperplasia and metaplasia were both increased in the mice that were missing gamma delta T cells. So there seemed to be a tissue repair phenotype specifically in neonatal and infant animals that was not present in adult animals. And so we wanted to figure out what the gamma delta T cells were doing after infection. So they were expanding relative to uninfected animals in the lung. And so they were activated and accumulating. And the particular gamma delta T cell type that we were generating was this IL-17A-producing gamma delta T cell rather than the interferon gamma-producing gamma delta T cell. This is very early peak of IL-17A, which is normally associated with neutrophil influx and the response to extracellular bacteria. That's just reflected here in the number of, or in the relative expression of IL-17A in the whole lung., and in a number of associated gene factors including ROR gamma. And so this just shows also here that the IL-17A is being exclusively made by gamma delta T cells, and it's a very transient effect. So by day 3, the number of those cells is back down to normal, and there's really no change in the production of interferon gamma. So this just stays at a a background level in that time. We have gamma delta T cells making IL-17A that's somehow being associated with tissue repair. And this was a surprising link to us because IL-17A is so associated with inflammation and pro-inflammatory processes and again, the bacterial response and not really a tissue repair response. So to really show that the IL-17A that was responsible for this effect is we took gamma delta T cell deficient mice here and we adoptively transferred either wild type gamma delta IL-17A deficient gamma delta T cells, and then we infected with influenza virus. We looked at outcomes and we were able to rescue animals with adding wild-type gamma delta T cells, but we were unable to rescue animals that we just added IL-17A knockout T cells to. And I don't have it on this graph and I should, but our mock transferred mice are down here with the IL-17A knockout cells. So it seems that the IL-17A was really the primary factor that was leading to the protection of, and this promotion of tissue repair. We decided to try to figure out what was the link between IL-17 and tissue repair, because this hadn't really been reported before. Most of the other inflammatory factors were the same in these animals, but one thing that was noticeably different was the levels of IL-33, which is cytokine related to the IL-1 family and had been implicated in tissue repair and tissue homeostasis previously. And so we just confirmed in neonatal mice, this hadn't been shown before, that IL-33 knockouts were much more susceptible to severe disease. And so they mirrored the IL-17 deficient mice or the gamma delta T cell deficient mice. And in the absence of IL-33, We thought that from previous reports from David Artisan and other individuals, there was an association between IL-33 and amphiregulin, which amphiregulin is an EGF growth factor that can directly influence the epithelial cells to divide and repair themselves. And so we thought this might be the ultimate link to the production of a restored lung function. And so we measured levels of amphiregulin And then performed depleting experiments for amphiregulin in wild-type animals around the peak of the amphiregulin response. And these animals were also much more susceptible to death. And so we put amphiregulin in this pathway. And I'm going quickly here just for the sake of time and because these data are now recently published. But we actually were able to show that IL-33 directly led to the enhancement of amphiregulin production. Through the recruitment of ILC-2s and Tregs, but really more the ILC-2 effects. And so each of these are direct mechanistic links in the production of this tissue repair response. And right about the time that we completed our studies, this pathway of IL-17 to IL-33 to ILC-2s to amphiregulin to tissue restoration was identified in a very different system in fat homeostasis by Lydia Lynch's group, in a Nature Immunology paper that came out earlier this year. And so this seems to be actually a really essential and maybe very primitive pathway for gamma delta T cell regulation of tissue integrity that we find is particularly important in neonates and young children. And just to show that the relevance of this in humans now, we looked back into the Flu09 cohort at the associations between IL-17 and IL-33. And this is in the nasal wash, so the same site where we were able to measure the CD8 and CD4 T cells in our individuals. And we saw a very nice correlation between IL-17A and IL-33 levels, sort of recapitulating this part of the plot, at least correlationally, and no association between interferon gamma and IL-33. So, you know, it's not just a sort of overall inflammatory increase in cytokine levels. And there was also an association between IL-33 and amphiregulin, so recapitulating this part of the plot here. And there was an association between IL-17 and severity. So this is restricted to kids under 2 years old that were in the cohort, and this is in their diagnostic nasal wash sample, looking at the levels of IL-17 and looking at severity, again using the total symptom score method. And there was no association with the level of interferon gamma and severity, but there was an association between IL-17 and severity, with kids with higher levels of IL-17 having reduced severity, or the other way of looking at it is kids that were more severe had lower levels of IL-17A. And this was really striking to us because we published previously a paper by Christina O'Shansky and the group had found various associations between cytokines and poor outcomes, and they were always positive associations. So, and this is true in most of the literature, so when you find higher levels of particular cytokines like IL-6 or TNF, they're associated with poor outcomes, particularly pro-inflammatory cytokines. Here we have a higher level of pro-inflammatory cytokine in young children, and it's actually associated with an improved outcome. And we again think that this is because we're activating this pathway that's particularly relevant and required in infants and young children for tissue repair. So what I've shown you in this last part is that gamma delta T cells can provide protection to neonates against influenza infection. Not directly through viral clearance. They do this by inducing IL-33 production by lung epithelial cells. I didn't show you that, but that is in the paper. IL-17A, IL-33 access protects neonates by elevating the levels of amphiregulin that's produced by LC2s and Tregs. And we found this association in human children as well, and it's the only positive association we've really seen between a cytokine, particularly a pro-inflammatory cytokine, and an improved outcome. And I wanted to show this today, partially because it's some of the most recent data, but also because our ongoing studies have indicated that in humans, gamma delta T cells play really an outsized role in kids under the age of 2 in terms of their lung response. And so I think going forward, thinking about vaccination in this population, we should at least track what's happening to this important cell population and how vaccination affects its functional ability to mediate this important effect. And really to keep in mind that neonates and young children are immunologically, we think, extremely distinct from older children and adults, and trials that we perform in older children and adults may not be recapitulated mechanistically or immunologically when those interventions are taken to children. So I've just shown you sort of two factors today that we think tether you to the to the outside of this cliff, genetic factor and a host factor, which was really age mediated by a particular unique immune response. And with that, I'm going to conclude, and I just want to thank the two main people that did the work I showed today. Johnny did the gamma delta work, and Katie Emma did the genetic work. This is built on studies that have been going on a lot for some time, initially started by Christine O'Shansky. I mentioned Ben Youngblood, our collaborators, Doug Higgs, and various other folks that contributed to the code. I'm happy to take questions now.

Thank you very much for a wonderful webinar. So if everyone will please go ahead and type their questions into the question box, I'll go ahead and read them out loud. Alright, we have our first question. It says, I'm sorry if I missed it, but what was the incidence of the IFITM3 risk allele in either the European or Asian populations?

Sure, I actually have a summary slide on that in case people asked about it. So this shows the two incidents of the risk alleles The RS122 that's been associated with risk in the Asian subpopulations, the risk allele here is G. It's usually reported as C, but it's on the reverse strand. And so it's somewhere in the range of 40% for the carriage frequency allele. The homozygotes are going to be a smaller proportion than that. There's some interesting things in this population as far as some of these alleles being out of Hardy-Weinberg. But it's extremely rare in the European subpopulations. In Great Britain, it was less than 1%, and that's where the Everett et al. study was performed. And that's really in direct contrast to this 344 allele, which is extremely rare in the Asian subpopulations but is very common in the European subpopulations, where actually the risk allele here is A or T, And in these two populations, it's actually in the majority. And there's some interesting inheritance associations between these two alleles that we think point to the fact that there's some relationship here in terms of how this locus is regulated. And so the risk genotype of this 122 allele, so that if you were CC homozygous, which does occur in in African American populations especially, is always inherited with the protective genotype GG of 344. And so the risk alleles are really on opposite haplotypes. And the same thing is true of 344. So the risk allele of 344, if you're homozygous, if you're AA, you're in every case that we sequenced, your TT for the RS122 allele, the protective genotype. And so this has some consequences in how actually how you interpret the eQTL data. So this is the eQTL data I showed you earlier. So for flu09 and picflu, RS244, and then this is the same cohorts looking at the eQTL for RS122. And so there's really no association here in flu09. There's actually a statistically significant association here in PICFLU, but it's the reverse of what you would expect. This is the risk allele for RS122 as reported, and there's a higher level of IFITM3 here in this population. Really, we believe that this is the bleed-through of the RS344 effect. Because if you have this risk allele, if you're CC for RS122, we know you're automatically GG for RS244, and so you have this higher level. We think this is mechanistically the actually relevant allele, and I didn't have time to show it, but we showed that using assays that this allele is actually regulating transcription of the gene. And so some of these— so you have to be careful in looking at these associations because this sort of inverse association with IFITM3 expression with 122 is exactly actually a bleed-through of this linkage between 1-2 and 3-4. There's definitely some complex inheritance structure here. We think it's probably extremely biologically meaningful in terms of how people respond to a wide variety of pathogens at the innate level and protect potentially with their resident lymphocyte populations. But the short answer to your question is over here is, um, 344 is quite common in European populations and 122 is quite common in Asian populations.

Thank you. We have two more questions right now. The first one is, what are the caveats of comparing gamma delta cells in mice and humans?

So there are a number of caveats. That's a great question. So in mice, Gamma deltas have a very degenerate repertoire, and what I mean by that is that in each mucosal organ, the lung, the skin, the gut, has a single receptor that is overrepresented clonally, and this is true in every mouse. It's not just in lab mice. So wild mice have exactly the same receptor in the lung as your B6 or your BALB/c or your 129. That's part of the evidence, really great evidence, that gamma deltas are not MHC restricted because all those mice have different MHC alleles. But so that degeneracy is a major difference between them. So that being said though, the human gamma deltas, while not as profoundly degenerate as the mouse gamma deltas, are more degenerate than alpha betas, and they have this association with tissues at the V region level. And so you'll see particular V region usages overrepresented in tissues in the same way that particular receptors are overrepresented in mice. And so the most commonly studied is the Vγ9-Vδ2 cell population because that's overrepresented in the blood and the blood is so accessible in humans. And so that's probably the major difference, I would say. The biggest caveat to studying them overall is that we are, you know, suffering for a paucity of ligands. There's maybe really, I'd say, 10 to 20 really well-defined good ligands in the mouse and the human combined. So we still don't really understand antigen recognition and TCR activation in the gamma delta population.

Thank you. We have one final question, and the question is, Is there any evidence that influenza-mediated methylation affects risk allele transcription or CD8+ T cell INF gamma transcription?

So I understand the question. Is there any evidence that methylation that I was showing specifically at the IFN3 locus regulates transcription, what we can say is at the sort of the most mechanistic piece of data we have is showing that the methylation, you know, purely in vitro changes the binding of CTCF to this promoter, and we know that CTCF binding at this promoter is largely inhibitory to gene expression, and again that's everything we can recapitulate in vitro. We can correlationally show in these data that the methylation is associated with higher levels of expression of IFITM3 in these T cell populations. The question specifically said something about interferon gamma, and I don't know if that was specifically about interferon gamma or just a typo for IFITM3. There's definitely other data about differentially methylated regions affecting gene transcription. That's something that Ben Youngblood's lab has reported on extensively. In general, methylation has been causatively shown to be suppressive for most genes. Here we're showing methylation as being activating because it blocks a suppressor, so it's a double negative.

Great. Well, thank you so much, Dr. Thomas, for your time. Thank you to everyone in the audience for joining us. I just wanted to say, as a reminder, our next webinar will be on Tuesday, October 16th. Dr. Scott Hensley will be presenting from the University of Pennsylvania on the immune history of and influenza virus susceptibility. So thank you again, Dr. Thomas, and thank you to everyone for joining us today.