Scientist Stories: Gosia Trynka, Genetic variants associated with immune diseases

I'm Gosia Trzanka, and in my group here at the Sanger Institute, we study how genetic variants associated with immune diseases regulate the immune system. And today I will share with you results from a few projects from my group where we used a combination of immunogenomic approaches to better understand the role of genetic variants in regulating a subset of immune cells. Most of the work I will describe here has been performed with Open Targets, which is a private-public partnership here at the Wellcome Genome Campus that brings together the Sanger, EBI, and industry partners to advance the translation of human genetic findings to new therapies. In my group, we study how genetic variants associated to common complex diseases regulate the immune system. And common complex immune diseases are a collection of about 80 different diseases that collectively affect about 10% of general population. So they collectively are quite prevalent. And they have a range of manifestations. So they can have very localised manifestations, for example, lead to inflammation inflammation of the small intestine in celiac disease or the joints, the synovial tissue in the joints in rheumatoid arthritis. They can also have a more systemic manifestation, such as in lupus patients. And immune diseases have been really quite successful for— in terms of genome-wide association studies. And as a result, we currently have thousands of loci that have been mapped to risk of developing immune diseases. And although the immune diseases have this range of different manifestations, one thing they have in common is the fact that there's a dysregulated immune system. And this is really at the, at the heart of what we're doing in my group, which is to understand how exactly those genetic variants are regulating our immune system to predispose to the development of the, of the diseases. So there are many challenges when it comes to interpreting GWAS-associated variants, and in particular in immune diseases, there's a whole range of immune cell types that can be involved. And those cell types have also different cell states. For example, the effects of associated variants can manifest when a cell type is activated. So understanding which what cell types are disease relevant is really a critical first step to then take that cell type or cell state to subsequent, more detailed functional follow-up studies. And once we have this known cell type, then additional questions arise, which include what are the disease-causal genes underlying this associated loci? Which are the exact causal variants which are the regulatory variants? And are there specific cellular functions that are affected by this associated variants? There's a range of different approaches that one can take forward to try and tackle those questions. And we've summarised some of those approaches in a recently published paper. But one approach that I want to highlight in here is this a concept of taking GWAS-associated loci collectively and overlapping associated variants with chromatin marks. So when we have a resource where we've got chromatin marks assayed across a range of different cell types and tissues, we can ask a question if disease-associated variants for a given trait collectively more often overlap chromatin marks that are active, that are specific specific in certain cell types. So in this illustration, the individual loci are represented as these circles, and you can see that in this illustrative example, these three loci seem to be more often overlapping with chromatin peaks that are present in bone tissue. So through this type of approaches, we can then infer that if collectively variants are overlapping chromatin marks specific in one cell type or multiple cell types, then these cell types are more likely to be causal to to the disease. And we and others have been developing such approaches. And when a few years ago when we applied it across a range of different immune diseases, what we observed was that there was a strong enrichment across a number of different immune diseases in enhancers specifically active in CD4 T cells. And that included different subpopulations of CD4 T cells. But the presence of this strong signal in CD4 T cells was really quite striking. And that was also concordant with many immunological studies that were also pointing towards the role of CD4 T cells in development of many of those immune diseases. So let me now spend a few slides just to tell you a little bit more about what CD4 T cells are, because this is really essential to the rest of my talk. So if there's a pathogen in our body, antigen-presenting cells will pick up that there's something foreign it's alarming and it needs to be presented to the immune system to trigger appropriate response. And triggering of this response is what is driven by, is driven by CD4 T cells. So when antigen-presenting cell presents the antigen on its surface, the T cell will come across it and then be alerted that this is something dangerous and it needs to activate and mount appropriate immune response.. And that immune response is dictated by the presence of different cytokines in the environment that then polarize the T cell towards different effector cell types. So that— and these cytokines are secreted by the antigen-presenting cell, which tells what kind of antigen it is seeing and what kind of downstream immune response the T cell should help to develop. And really, depending on what kind of phenotype the T cell acquires,, it then goes on to communicate with other immune cells such as B cells and cytotoxic CD8 T cells to drive the response and clear the body of the pathogen. So this is really a critical, critical step in mounting appropriate immune response. And of course, it's also necessary that there's an appropriate braking mechanism to this to this process because otherwise our immune system would go on activated endlessly and that would be very harmful to our body. And this braking mechanism is provided by the regulatory T cells that can directly interact with antigen-presenting cell and T cells, as well as they can start secreting different cytokines that sort of all signal to the immune cells to to stop the immune response. So in my group, we study how is this T cell activation circuitry regulated through genetic variants. And today I'll focus specifically on the two components of this circuitry, the CD4 T cell activation and the role of cytokines in CD4 T cell activation and how that links to genetics of immune diseases. So, in this first project, we ask a question: what is exactly the role of cytokines in activating T cells, and are there specific cytokine-induced cell states that are more relevant to immune diseases? So, this project that was led by a very talented trainee, postdoc in the group of Blagoje Shoskic and a PhD student, Eddie Gomez, together with Debbie Smith, who has been really also instrumental in this group in generating the data. We derived this project in which we isolated CD4 T cells, and we isolated both naive and memory T cells, which are very different. So naive T cells have not been activated before, while memory T cells have already encountered in our body an antigen. They have been activated, so they hold that memory. We separately isolated both cell subsets, and we also isolated monocytes and polarized them towards macrophages. And so this is our antigen-presenting cell type. And then we exposed all of those cells to a range of different cytokines. And upon this activation in the presence of cytokines, we then profiled chromatin activity using ATAC-seq, which measures— which maps open chromatin regions. And we used H3K27 acetylation mark, which highlights active enhancers and promoters.. And we profiled chromatin activity at two different time points, and these time points were different between naive and between T cells and macrophages because they form different, two different arms of immune system. So macrophages, they fast responding cells, so we measured chromatin activity at 6 hours and 24 hours. But for T cells, we've chosen 16 hours, which is before first cell division, and then 5 days, which is more or less where where in our body T cells would acquire, would fully develop effector function. So, with that, we collectively generated data for 55 different cell states. And first of all, what we observed was that upon stimulation, there was a large number of peaks that were induced both in macrophages and in T cells. And we could observe that both when we looked at open chromatin regions as well as active enhancers and promoters. Now, what was really interesting, however, was the fact that although there was this enormous amount of new peaks that were upregulated, the majority of those peaks were actually shared between the different cell states. So here in gray, you can see that between 20 to 40 cell states shared the upregulated peak regions. And only very small proportion of cell states, a small proportion of peaks were cell state specific. But when we looked at how variable in activity those peaks were, we observed that actually there was quite a lot of variability. So, even though the peaks were shared, they were quite variable in the activity. And this was particularly present with H3K27 acetylation mark. Now, that had immediate consequences to what we were planning to do next, which was to perform the SNP enrichment analysis. And typically the SNP enrichment analysis is performed in a way where we're looking at the overlap between disease-associated variants and peaks that are present in one cell type but are absent in other cell types. And when we see this presence and we see the overlap between variance and peaks, then we inform that there's enrichment and these particular cell types are disease relevant. Now, that wasn't the case in our dataset. What we saw more was that the peaks were present across all different cell states, but the variability of the activity was different. So clearly this approach wouldn't work for what we wanted to do here. So So that motivated us to develop a new SNP enrichment method which we call CHEARS. And that method takes into account exactly the chromatin activity that's present when SNPs are overlapping peaks, and it takes that into account when calculating the enrichment score. I'm not going to go into details on this. You can read more about it in our publication, and also you can access the code codes if you'd like to run this method on our GitHub. So we applied this method to our dataset, and first of all, we observed that indeed we were able to identify cell states that were individual cell states that were enriched for disease-associated variants. And here I'm showing you IBD and different subphenotypes of IBD, as well as celiac disease, all showing strong enrichment with TH1 cell state. And the way to read this, this figure is that in purple, we're showing the strength of the association in different cell states in memory T cells and in green in naive T cells. The lighter shading shows the earlier time points and the darker shading shows the later time point. And macrophages, you can't see very well, but they are right here in yellow. And there, we don't detect much enrichment here. But the point here is that we are successful in identifying these individual cell states, such as Th1, which is very relevant to IBD and celiac disease. But more globally, when we looked across a whole range of different immune diseases, what we saw that rather than seeing this individual cell states coming up, we saw more globally that it was about the activation of early activation of memory T cells that was really strongly enriched for SNPs, for variants associated across a whole range of different immune diseases. So, how is T cell activation regulated then? As I said, T cells, upon encountering— when they encounter antigen-presenting cell that is presenting, that is signaling that there's something foreign and something dangerous in the body, they undergo activation. But that activation process is really tightly controlled. So first of all, T cell has to receive the signal through the TCR, and the TCR really says that there is something foreign. And each cell type, each T cell will have a different TCR, different receptor that recognises different combination of antigen and presented on the HLA molecules of antigen presenting cell type. But that's one signal. The T cell also needs a second signal that will tell that it's not only foreign, but it's also a dangerous signal and the T cell needs to activate and act.. And that signal is derived through the CD28 co-stimulatory molecule on the surface of the T cell. And when those two signals are met, then the T cell starts to proliferate and differentiate and start producing cytokines and essentially starts to manifest its effector function. In collaboration with David Sansom at UCL, we thought about how could we better understand this, the pathways that are regulating T cell activation through CD28 and TCR signals. And together with David, we've set up this approach in which we were able to develop a system in which we can titrate the amount of CD28 signal that we're giving cells as well as TCR signal. So we can have cells that are cultured in low amounts of both signals, in high amounts of both signals. We can also culture cells giving them a very strong amount of CD28 signal alone, or very high amounts of anti-TCR signal alone. So these are not physiological amounts of signal, but what we're trying to do here is really push cells towards one pathway or towards the other pathway and ask what happens to the cells in those states. And this project was led by Daphne Glynas, who was a PhD student in my group, and Blagoe as well. And so, what we did next was when we cultured cells in the presence of these different stimulatory conditions, we then sorted cells that were activated, and we sorted separately naive cells and memory T cells. And we also had cells that were not activated that we've used as a control. And then we performed gene expression profiling in these activated cells. And what we observed first was that there were clear differences in terms of how naive and memory cells were responding to different to different stimuli. And particularly what was interesting was how cells were responding to the CD28 co-stimulation. So you can see here that between naive and memory cells, there's a large number of genes that are upregulated and they're shared between the two cell types. But then in different stimulatory conditions, there's a different number of of genes that are upregulated either in naive cells or in memory cells. And mostly naive cells are upregulating large number of genes specifically in response to TCR. But in response to CD28, we see that there's a high number of genes that are upregulated more so in memory cells than in naive cells. And that was really interesting because that was already indicating to us that these two cell types are utilizing the signals differently. When we took those genes and ran them through a pathway enrichment analysis, we also observed that indeed there were different pathways that were enriched, that were captured by those genes in naive cells and in memory cells. And particularly what was really interesting to us was this observation that in naive T cells, genes that we classified as TCR sensitive were driving G2/M checkpoint pathway. And in memory cells, this pathway was far more significant when we looked at the CD28-sensitive genes. So that would indicate that the two sets of genes were actually driving the same pathway but differently between the two cell types. And indeed, we were able to validate these findings as well. We looked back at— we looked at at proliferation of naive and memory T cells in response to high TCR stimulation and high CD28 stimulation. And we were recapitulating what we observed with gene expression, and essentially that with strong TCR signal, naive T cells go on to proliferate, but memory T cells don't respond very well to TCR signal alone. The reverse was true for CD28 signal where it didn't trigger proliferation in naive T cells, but memory T cells were responding quite well to CD28 activation. The final point I want to make here is that we also looked more in detail into what exactly were the genes that we classified as TCR-sensitive or CD28-sensitive. And we observed that a lot of the effector, T cell effector genes, which include cytokines and receptor ligand molecules, were also CD28-sensitive. So you can see the sensitivity here on the, in the upper bars, and you can see whether they are in memory or in naive T cells. And you see that majority is genes are CD28 sensitive. And among some of those cytokines, we see really some of the critical cytokines such as, for instance, interferon gamma is very strongly CD28 sensitive and similarly IL-17F gene. So in the final part of my talk, what I want to share with you is how we are we are further trying to understand what are the genetic differences in regulating naive and memory T cell activation. And this is really exciting. This is our unpublished work yet, but we have some really interesting early results. So, in this project, what we've done is we We recruited over 100 healthy individuals from whom we isolated naive and memory T cells, and we stimulated the cells again. And then we sampled cells at 4 different time points. So, we've got cells in the resting state at 16 hours, again, before cells go on to the first division, at 40 hours, which is after cells have committed, have gone through the first cell division, and then and 5 days when cells are acquiring fully the phenotype, effector phenotype. And we, at these different time points, we performed single-cell RNA sequencing. And with that, we were able to generate a dataset of T cell activation gene expression map using over 600,000 individual T cells. And what you can see here on this UMAP is that, first of all, we really nicely capture how the T cells undergo activation through different states. So you see that these are early time points of activation, then 16 hours, 40 hours, and 5 days. And you also clearly see the separation between naive and memory T cells. So naive cells are in blue and memory T cells are in red.. So you see that they are mapping in the same regions, but they clearly are separate between— there's a clear separation between naive and memory T cells. With single-cell resolution, what we were able to capture, which we wouldn't be able to capture if we had done bulk RNA sequencing, was the ability to detect this lowly active cluster of of cells. And as I said, we sampled at 4 time points. So, we sampled cells at resting, 16-hour, 40-hour, and 5-day time point. But we saw that there was presence of cells that were in this intermediate state that we call lowly active just before cells undergo into the first cell divisions. So with this dataset, because we have large number of individuals, what we're able to do is now we can perform eQTL mapping. So we can correlate genotypes with gene expression levels. And when we have done that, we were able to detect close to 10,000 eQTLs during the course of CD4 T cell activation.. And we see that the number of eQTLs goes up. We're able to detect more eQTLs throughout the time course of activation. And we're mapping some really interesting examples. So, for instance, we see that there are cell-type-specific events. For instance, here is an example of IL-12RB1 gene in the resting cell, in the resting state, where we see that in naive T cells there's no effect, but in In memory T cells, we can clearly detect that there's an eQTL effect where AA genotypes are leading to a lower gene expression compared to the TT genotype. We're also able to now look at— to identify not only cell type-specific effects, but we can also detect cell state-specific effects. Effects. So, in this case, the GFPT1 gene, you see that there's no effect at 16 hours in naïve cells, but in memory cells, there is a clear effect at this time point. Now, of course, the single-cell RNA sequencing gives us resolution to start looking more into what kind of cell subsets are present at the different time points of activation. And indeed, we were able to look with more granularity what cell types, what cell subpopulations constitute different time points of activation, and we were able to map 38 distinct cell clusters. Interestingly, we classified the cell subsets into two distinct groups. One, a stable group of cell types that is present across all the different time points. Then the other group, which is more transient group that appears in some cell states and is absent in the other cell states. So, here's an example of this consistent subgroups of subsets of T cells that are present throughout the activation. And this includes central memory T cells, effector memory T cells, and TEMRES. And you see that they are present throughout all of the time points. And the percentages, the proportions of cells really don't change very much throughout the course of activation, which is also reassuring because this isn't something we were— we would be expecting to see. However, we're also able to identify these cell populations, such as, for example, interferon-responding cell population. And we see this cell type on this cell subset only appearing upon activation and then being present throughout the course of activation. On the other hand, we also see, for example, the NF-kappa B responding cells that we can only detect in 16 hours or in 40 hours of stimulation. So, again, with this resolution, with this zooming into different cell populations present across the present in different time points of activation, we can perform EQTL mapping specifically individually in those cell populations and cell activation time points. So, here's an example of how much more information we can get out of this dataset. Looking here at the TIC2, expression of TIK2, you see that when we take all of the memory cells together as if we didn't have single-cell resolution, we wouldn't be able to detect that there is an EQTL effect at this time point. Now, what we can do is we can look whether there are effects specifically in the different subpopulations. So in this case, we're looking at at central memory T cells and effector memory T cells. And we see that indeed there is an effect specifically in effector memory cells that is absent in central memory T cells. And when we combine those cells together into this pseudobulk of memory T cells, then it cancels out the EQTL. And that's why we're unable to detect that there is an effect. Effect. And what we can do now is we can take this map of eQTLs that we've generated, and we can ask a question whether those eQTL signals are overlapping with GWAS signals for immune diseases. And here is an example where we're showing you that indeed there is a colocalization between IBD GWAS variants and this Tick2 eQTL in effector memory T cells. Finally, what we can also do with this dataset is we can apply a trajectory inference algorithm, which is what we did here. And we observed that with this algorithm, the cells were mapping onto a continuous trajectory of cells progressing throughout activation states. So, we used this activation pseudotime and then plotted the expression of genes that are well-known markers of cell activation. So, here you can see that CD69, which is a marker of early activation of cells, has a low expression in the early or in the resting resting state and in the early stages of activation. And then the expression peaks around 16 to 40 hours of T cell activation and it goes down with time of activation. Now, IL-2RA is a marker of late activation. And indeed, this is also what we can capture with this dataset that in early time points of activation in the resting state and early time points of activation, there's low expression, but then the expression goes up. And maintains as the cells undergo activation and continue on to the later time points of activation. So, this gradual and continuous change in gene expression throughout time indicated to us that we could model time as a continuous variable. So, finally, what we're able to do with this dataset is that we can use this activation pseudotime and group cells in different windows depending on where the cells fall into this pseudotime activation, and then measure the gene expression, measure the eQTL effects in these different windows, but in a way where we can test if there's a significant interaction between a genotype and activation on a gene expression.. So, here's an example how such dynamic EQTL could look like. We could observe that in a resting state or early time points of the pseudotime activation, there would be no effect between genotypes. But then as the cells progress throughout the activation time course, this effect would be attenuated depending on the genotype. So we applied that to our dataset and we were able to see that actually for about 50% of all the eQTLs that we mapped, they were showing dynamic effect. And we could model, we could identify both linear effects as well as nonlinear effects in gene expression. And again, what we can now do with dataset, we can and overlap those effects with GWAS-associated variants to ask where those signals are colocalizing. And here's an example when we did that. Here's an example of such colocalization event for FATS2 eQTL that we see is a dynamic eQTL, and we see it colocalizing with asthma GWAS in this case. And in fact, so far we see that this dynamic eQTLs are— so far we see that there's for about 50% of all the colocalizations that we mapped between immune GWAS signals and eQTLs, 50% comprise dynamic eQTLs.. So in summary, what I showed you today is that immune disease GWAS variants are enriched in enhancers active in early activation of memory CD4 T cells. We also performed experiments to show that naive and memory CD4 T cells differently utilize TCR and CD28 signals. So they read those two signals differently and respond to they trigger different gene expression programs differently in response to those signals. And we also now get a better idea of how GWAS variants could be modulating T cell activation. And actually, we see that half of the immune disease GWAS signals that we see are colocalizing with T cell activation in eQTLs in our dataset include this dynamic effects. Finally, let me close with this message. I think that the technology is advancing really rapidly, and this is really fascinating times, and we're really starting to be in a position to get a better idea of how GWAS variants function. How do they modulate cellular phenotypes? How do they modulate cell functions? But these approaches will really require us to utilize a whole toolbox of genomic tools. But more importantly, I think it's really essential that we collaborate with experts in the fields in cellular biology, for us in immunology, to really start bridging different expertise to get a better idea to get a better understanding of the function of the GWAS variants to accelerate translational efforts and accelerate discovery of new drugs and improved therapies. And with that, I'd like to thank all of the team and past and current lab members. Especially I've highlighted here the work of Blagoje Soskic and Eddie Gamez. And also I'd like to thank very much our collaborators, and I highlighted work that we've been carrying out with David Sansom at UCL, which is really a wonderful collaboration. So thank you very much, and I'm happy to take any questions.