Scientist Stories: Nikhil Joshi, Immunotherapy & Tumor Microenvironments

3, 2, 1.

Welcome back to the Early Career Immunology Seminar Series. For those of you returning, thank you for your continued support. If this is your first time watching this live, allow me to introduce myself. My name is Tim O'Sullivan. I'm an assistant professor of immunology at UCLA. The purpose of this seminar series is to increase the exposure of early career faculty in a time in which we have not yet achieved equal representation at scientific conferences and seminars despite equally innovative discoveries. We're really excited to have Nick Joshi with us today. Nick received his PhD from Yale working with Sue Keck. His graduate work focused on the mechanisms that control how effector T cells become memory T cells. And so Nick's work largely defined how short-lived effector T cells develop during acute infection and how the environmental factors drive effector T cells to become short-lived effectors. He then went on to postdoc with Tyler Jacks where he described how Tregs present in tumor-associated tertiary lymphoid structures inhibit anti-tumor T-cell responses. After all of these exciting discoveries, Nick returned to New Haven to establish his lab at Yale, where his group has created a number of immunogenic GEM models to study how T-cell function differentiation in different cancer subtypes and healthy tissues. If you have any questions during his talk, go ahead and type them into the chat box, and I'll ask these questions at the end of the talk. Nick, thanks for being here today, and it's all yours.

Yeah, thank you so much for the invitation and for setting this up. It's been really fun to watch these talks. So today I'm going to talk about some of the work that my lab has been doing, focused largely on understanding the mechanisms, some new mechanisms of how immunologic tolerance and specifically peripheral tolerance for T cells might be being mediated. I'll talk a little bit about some of our cancer work as well, but as I mentioned, most of the focus will be on tolerance. And so I wanted to, since this is kind of a broad audience that could be watching this talk, I wanted to start off with kind of a maybe a more philosophical discussion about what tolerance is. And so I like this quote, just the definition of tolerance is the ability or willingness to tolerate something that one does not necessarily agree with. And this is definitely the problem a problem that the immune system has to deal with, and specifically that T cells have to deal with. They have randomly rearranged receptors on their surface that can recognize antigens from a variety of different sources. These could come from many harmful things such as bacteria, viruses, even tumors. And of course we want to make a robust response against those types of things. But we also have a number of things in the body such as self-antigens, we have commensal bacteria, even allergens that we don't want to make an immune response to. We want to have tolerance to these things. And so this is a very fine line that the immune system must walk to try and make this happen. So we're very interested in how this balance is established. Of course, there's a lot known about this subject. Nothing I'm going to start telling you is very new. It's been known for decades that in the thymus you have central tolerance where you remove autoreactive thymocytes against the majority of your self-antigens. And that's done through the expression of antigens in the context of these medullary thymic epithelial cells that then give off antigens or directly present antigens to autoreactive thymocytes. And you either get death of those cells or you might get the formation of Tregs for CD4s. Of course, this is imperfect. We know from work, from beautiful work from Mark Davis and others, that there are autoreactive CD8s and CD4 T cells even in healthy patients. Healthy people, and that these are present in the circulation. So it's an imperfect process. So we have this secondary process we think of called peripheral tolerance, where T cells could get a number of signals that help them to not make responses against self-antigens. And specifically, we think of T cells as requiring 3 antigens associated with becoming— 3 signals associated with activation. So canonically, we think about them getting TCR signals through the T cell receptor, getting co-stimulatory signals through CD28, and then also getting cytokine signals. And if they don't get these signals, if they say just get signal 1, they might go into an anergic state, which means that they never acquire functional capacity, they don't become very proliferative, they can often die, we see clonal deletion. And by contrast, if they get these 3 signals, now they can have much more function, they become effector CD8 T cells, Those cells can produce lytic molecules, they can also produce cytokines. And those cytokines and lytic molecules can help them to attack the, say, the infection or the cancer and to mediate destruction of that. In the context of chronic challenges now, it's been very appreciated by work from many, many labs now that's focused on how these responses are resolved and really how we sustain responses under chronic conditions. And what we know is that CD8 T cells become shut down, they become, terminally differentiated. We think of this, some people call this exhaustion. This is really a way by which T cells lose some of their functions. They're not completely nonfunctional, but they turn off a lot of their functions in order to preserve homeostasis and not to cause immunopathology. We see that through their ability to secrete lytic molecules, their ability to produce cytokines, which is often diminished, their ability to proliferate, which is really, really often very diminished, and also their expression of these inhibitory receptors that provide inhibitory signals that counteract, say, the positive signals they would get from T cell signaling. And many of these are famous now because we know that in the clinic we can provide, in the context of cancer patients, we can provide inhibitors to these molecules, we can block these interactions, and now we can get back some of these functions. And so this just shows one patient, an example of a patient who has metastatic melanoma, And you can see this is a liver metastasis that's being tracked here. And when they get CTLA-4 and then they get anti-PD-1, so blockade of two of these molecules sequentially, now they get a very robust response and they reject all of the tumors throughout their body and they become tumor-free. So it's a really remarkable sort of therapy that can happen in patients. So this has been an ongoing story for the past 10 or 12 years. We're really starting to understand that these immune checkpoint receptors are having an important role in regulating the anti-PD-1 response. Antitumor responses by T cells that exist when people come into the clinic, but are inhibited because of these receptors. And when we block these receptors with antibodies, now we see these really great responses in a very small fraction of patients. And that tells us that these receptors have an active role in preventing the response up until the point at which the antibody's given. However, what has become very apparent and will be really a large focus of the talk today is this idea that these after the treatment with these inhibitory receptors, especially in the combination therapies where you're getting blockade of CTLA-4 and PD-1, you get a lot of these localized autoimmune-like side effects. We call these immune-related adverse events, or IREs. And these happen in the majority of patients who are treated with combination therapy. What these are, they're like sort of localized autoimmune reactions. They can be rashes, they can be colitis, they can happen in a variety of different tissues, and often it seems like at random.. And it's not really associated with whether there's a good anti-tumor response. There's a lot of questions about what's driving these things. We know that you can treat them often nonspecifically with steroids or with some more targeted agents like TNF blockers, but we are not clear about how those treatments really affect the anti-tumor response. And also, what's probably most problematic is that the threat of these IRAEs is now one of the biggest barriers to giving combination therapy to a large fraction of patients who might benefit from it. So that's what's happening clinically. What's happening sort of in our understanding of the immune response has been that these adverse events have really taught us something that was a little bit unexpected. And that is the idea that these immune checkpoint receptors have a very important role in terms of maintaining homeostasis in the body. In terms of potentially regulating self-reactive T cells. Because when we block them now, we're getting some amount of autoimmunity in patients, and this is happening in 80% of patients, so it's not even people who are necessarily autoimmune-prone who are getting this. So these seem to be important in normal healthy people. Today I'm going to tell you about two stories, one is published, so I'll go through it kind of quickly, and discuss some of our work on trying to understand in the context of developing tumors, kind of following up on that first role. How are T cell responses maintained over the course of tumor development. This is a seminar series which I think is, one of the best parts about it is we highlight some of the people in the labs who are doing the work and who are really taking a risk by coming into our labs. One of those is Kelly Connolly, who's a very, very talented postdoc who's in my lab who spearheaded the first project, and she's been working with two very talented graduate students, Manik and Arthi, who work in Smriti Krishnaswamy's lab, who's just a phenomenal bioinformaticist who's here at Yale. We also, I'm gonna talk about in the second part of the story, second part of the talk, an unpublished story that's really going into some of the mechanisms by which we think IRAEs might be happening. And this has been, oh, I'm sorry, I forgot to change the name here. This is actually Martina Damo, who's a postdoc in my lab. She's just phenomenally talented as well. She's done a tremendous amount of work to help solve this. I apologize that her name is incorrect there. And she's been assisted in this process by Noah Hornek, who's a dermatologist in the lab, and also by Sam Sui, who's done some of the bioinformatics work. And we've had some really good clinical collaborators, John Leventhal and Bill Dansky, to do some of the human work. And so what we were interested in, as I mentioned, is trying to understand this balance between how T cells participate in anti-cancer responses and peripheral tolerance. And we want— we kind of early on realized that we needed some animal models where we could study these two different processes in the context of normal healthy tissues or in the context of developing tumors. And so when I was a postdoc, I helped to start developing this mouse model we call Ninja. And the Ninja model is really a very useful tool for making inducible neoantigens in the setting where you have really good spatial and temporal control over when and where the antigen is created. Expressed. And in addition, we've shown through a lot of work that these antigens are not expressed at any, even at a very low level before we turn them on, and this helps us to avoid central tolerance in the mouse. And so the way that we do this is that we've created this allele that has two modules. It has a neoantigen module which encodes the inducible neoantigen, and it also has a regulatory module that encodes a gene called Flipo, Flipo is the recombinase that's going to act on this inducible neoantigen and turn it on. And so the way this works is that we took an antigen substrate that had peptides for both CD8 and CD4 T cells. This is peptides from lymphocytic choriomeningitis, which most people will recognize, GP33 and GP66. We encoded these into DNA, and then in the DNA that encodes the substrate, we added splice sites. So that we create this central exon, and then we inverted this in such a way that in the off state of this allele, you have the production, sorry, the splicing skips this central exon, you have production of incomplete peptides, and so you don't actually make anything immunogenic. When you want to turn it on, you need to flip this exon around, and for that we use these non-compatible frit sites, which respond to that recombinase I mentioned, Flipo. Flipo causes a permanent inversion in the substrate, and now you get these splice sites that line up, and you get production of these antigens. And so we also wanted to make it so we could read out this neoantigen, and so the way that we did that was we embedded it within a GFP molecule, does something very useful, where in the off state you get an early form of a stop codon that truncates the protein, so you get half the GFP, so it's non-fluorescent. When you now turn on the neoantigens, now you get the fully formed GFP. GFP and encodes the antigen. So we can use the GFP as a metric for which cells are expressing the antigen and how much they're expressing it. We also have this regulatory module that I mentioned, and this is what allows us to have that very good spatial and temporal control over activation. So this includes the FlipoER. FlipoER in this case is broken in half and flipped around, very similar to what we did on the neoantigen side. And the way this works is that we provide pre-recombinase, CRE is one of the first elements, it's absolutely required to turn this thing on. That causes an inversion, a permanent inversion of the allele, it lines up the FlipoER, and so now you can produce it, but it still requires doxycycline and tamoxifen to actually be turned on. So we can give this CRE, and then we can wait a little bit, and then we can give the doxycycline and tamoxifen. That allows us to actually initiate in-engine expression based on when we give this doxycycline and tamoxifen, but to control where we express it, based on the expression of where the Cre is. So in the context of our cancer models, we use these genetically engineered mouse cancer models where we initiate a tumor by turning on an oncogene and removing the tumor suppressor p53. So you can see here we use Cre to actually activate these by turning on oncogenic KRAS and removing p53, and that also poisons these cells to now respond to doxycycline and tamoxifen and turn on antigen expression. Specifically within the developing tumor cell. That's how we study cancer. When we want to study tolerance, we actually use a different trick, which is that we use, there are a number of different tissue-specific Cre recombinase genes. We can cross our mouse to this, and then when we want to induce, test the effect of the peripheral expression of those antigens, we now just provide the mice doxycycline and tamoxifen and turn on the antigen. So the first story I'm gonna tell you about is one, as I mentioned, that was recently published, that focuses on understanding how T cell responses are sustained in the context of a developing tumor. And so I'll show you that we have some sort of idea about the cell types that will respond to immunotherapy when we give immunotherapy, the specific subsets of CD8 T cells. But what we were particularly interested in is this idea that tumors develop over the course of months or years in the context of a person. And we don't really have a good understanding of how these T cells that are necessary to maintain the response or to elicit the therapeutic response, we don't understand how those cells are maintained over the course of tumor development. How do they persist in the tumor microenvironment over many months or years? And so this is what's become kind of the paradigm for how T cells might function in the context of immune checkpoint therapy. And the idea that when T cells see exhaustion, or sorry, see chronic antigen, they become exhausted, they lose the function as I mentioned earlier. And that this is really the majority of the cells that might be present in a tumor. However, there's also these presence of these stem-like T cells, and these are a less differentiated subset of T cells that express a marker TCF1 in their nucleus, it's a transcription factor. They also express low levels of PD-1, and they express often the chemokine receptor CXCR5, And they express this receptor called SLAMF6 that seems to be very important for identifying these cells. So we think that chronic antigen is driving these cells in a large part in the context of chronic infection and in tumors towards this exhausted cell state. But if you give antibodies to PD-1, often these cells are thought to now proliferate and then give rise to these functional effector cells. And these are cells that have granzyme, they make cytokines and can cause killing of tumors. So this is how we think that they're actually driving therapeutic responses, by turning into these functional cells that can be seen after checkpoint therapy. And so work from just a number of really very great labs has shown that these cells are present in a lot of human cancer types. I highlight this one from a group in, in Italy that has really shown in human lung cancers, which is the cancer that we focus on. Several years ago, they showed very nicely that these cells are present within the lung tumors, so they're present. Work from Nick Haining and others has highlighted the idea that these— the presence of these cells in human tumors correlates with better responses. And also this really fantastic paper from Warner Health, where they actually showed in animal models that the presence of these cells is necessary inside of tumors to actually get therapeutic responses. So we know these cells must be in the tumor to give out the therapeutic response, at least that's what we think. We want to understand how these cells are persisting in the context of our tumor models. And so we initiate tumors in what we call the KP-Ninja mouse, which has the Ninja allele to elicit the T cell response against, and then it develops tumors. Slowly over the course of several months. And then we're gonna look at those tumors at what we call an early time point, which is about 8 to 10 weeks, and then a late time point when the tumors have matured into something that's considerably larger and more advanced, and that's around 16 to 20 weeks. We use an intravascular labeling technique that many people use, where we inject antibodies, and because the lung is so highly vascularized, what this does is it allows us to determine which cells are in the lung because they show up as antibody negative. You can see that in a naive mouse, a B6 mouse, or in a mouse that doesn't have KRAS, we've given it all the same inducers, but it doesn't form a tumor. You can see that these cells are largely absent, and that allows us to at least feel confident that these cells that are present within the lung are there because of the tumor, because they're really only there when there's a tumor. Tumor. We can gate on those cells based on MHC class I tetramers, and we study the endogenous T cell response against these. And what we saw was that using the marker TCF1, we were able to see that both at early time points and at late time points, there was a population of cells that is gated on these, they're antigen-specific, everything I'll show you is antigen-specific. They are in the lung tumors, and that they express TCF1, a small fraction of them do. And all the cells in the tumor are expressing a moderate to high amount of PD-1. If we waited for later time points, we saw that these cells were still there, so they're not going away. But what was really striking to us was the fact that there was a large number of these cells in the tumor-draining lymph node that were tumor-specific, that expressed high levels of TCF1, and that were expressing PD-1. So we knew that they were activated there. And what was really also very surprising to us was that these cells were present kind of at early time points, but also they were maintained at late time points. And we had kind of predicted that over time we would lose these cells, so we were very surprised that they would be still there in the late tumors, in the lymph nodes associated with late tumors. Just in terms of phenotyping, we did a little bit of phenotyping by HACS to show that these cells that are PD-1 high, TCF-1 high in the tumor, not a lot of them expressed this marker SLAMF6,, but a large fraction of the cells that are in the draining lymph node are SLAMF6 positive. So, just in terms of the markers that they express, they're SLAMF6 positive, we think a fraction of them express CXCR5, they express other hallmark genes that would be associated with the stem-like cells. So it looks like these are the true stem-like cells that are present in the lymph nodes. We also looked at their ability to function by giving them peptide in vitro. And what we saw was that only the cells in the draining lymph node are the ones that produce interferon gamma after stim. So they seem more functional as well. Now we weren't the only people who were starting to think about what these cells are doing, and there's been a real flood of papers in the human space focused around what is this, what is thought to be maybe a more global response to immunotherapy. And this concept that when you give immunotherapy in patients now, you start to see the introduction of T cells into the tumor microenvironment that weren't there before. So if you look early and late, or if you look for clones that are shared with the blood, you start to see that there's an increase in these cells that you wouldn't have normally seen before. So this is suggesting that maybe some of the therapeutic response is coming from outside of the tumor. There's also these really great papers that came out late last year that focus on this concept that maybe you can elicit T cell responses with immunotherapy by targeting the tumor-draining lymph node. And so this, they use very, very clever techniques to actually deliver the antibody specifically to the draining lymph nodes and focus on this idea. So it really seems to be a hypothesis that maybe you could stimulate T cells before they leave the lymph node by targeting them. We wanted to understand more about the biology of these cells, so we performed single-cell RNA-seq. We took the draining lymph node and the tumors from the early and the late mice. We sorted out the antigen-specific. CD8 T cells using MHC with class I tetramers, and then we did single-cell RNA-seq and paired TCR-seq on those. And I'm just going to show you the highlights of that data, and namely that when we compare with this, this approach really allowed us to do a number of different types of comparisons. If you want to look into more detail, it's in the paper. But what I was most interested in was this idea that we could compare both spatially and temporally what were the tumor-specific T cells, what were their phenotypes in kind of high-dimensional space. And so what we were able to do was to compare here, looking at the T cells that are in the early tumors and in the late tumors, we can see whether or not they have overlapping gene expression. And we're using this technique called PHATE that Smita's lab has developed. This is kind of like a UMAP, only it allows a little bit better visualization of the data., and what you can see here is we have these light blue cells and dark blue cells. The light blue cells are from the early tumors, and the dark blue cells are from the late tumors. And you can see that what we've done, instead of giving gates, we've actually just marked where the prototypical stem-like cell would be, and where the prototypical exhausted cell would be. And what you can see is that majority of the cells that are in that, near where the prototypical stem-like cell are, those cells are coming from the early tumors. And then when you look for the exhausted cells, a lot of those are coming from the late tumors. We also did unbiased pseudotime analysis to show that, kind of the same thing, that in pseudotime, the early cells from the tumor were earlier developmentally than the ones that come from the later tumor. So that was the tumor. We also looked at the draining lymph node in the same way. And what we were really surprised to see was that the majority of the cells that were present in the lymph node really have fairly overlapping gene expression between the early and the late tumor time points. And this is shown both by the fact that a large fraction of them are near this lamppost that marks where the stem-like cells are, but also between the fact that when we look at pseudotime, the majority of the cells that are in the lymph node and at the early and the late time points are overlapping in pseudotime. And then, so this was a temporal comparison. Now we were able to do also a spatial comparison, between the lymph node and the lung. And when we did that, what we saw was that the cells that were present in the lungs— excuse me, in the lymph nodes at early time points and at late time points were largely falling into this category of stem-like cells. And again, the ones that are in the tumors are falling into this category that's more differentiated. And we can really see this nice transition as cells would move from the lymph node, they would differentiate in the middle here as they were entering the tumor, and then become these more exhausted cells. You can see that also in the pseudotime analysis. So one question you might have is, how do I know these are the same cells? And for this, we were able to utilize the fact that we had done TCR sequencing, and we could look at the same T-cell clones, because they have the same alpha and beta pairs in the mouse. And so when we do that, what we can see is that this is just showing the top clones. You can see that they are actually shared between the cells that are in the, in this stem-like state, and then also in the exhausted state. We did a number of other types of comparisons in the paper that would support this concept. So what we were then interested in was this basic question of how is this whole system set up? What we thought was maybe that the lymph node was acting kind of like a reservoir of these cells, and that the tumors themselves were driving the terminal differentiation of these cells, So maybe in order to maintain the few stem-like cells that were present in the tumor, it was necessary for a fraction of these cells to migrate from the lymph node to the tumor. So this was an easy thing for us to test. We just provided these mice with FTY720 to block migration and trap the cells in the lymph node. And when we did that, we did this over a period of 3 weeks before the early time point. When we did this in the lymph node, what we saw was that— sorry, when we did this in the mouse, What we saw was that on the right here with the FTI-720, if you look at total number of T cells in the lymph node or the tumor, you don't really see much of a difference. When you look specifically at the TCF1 high cells, the cells that we think are the cells that would be most recently migrated, now you see a pretty big drop-off in those cells, specifically within the FTI-720 tumors. And when you look at the cells in the lymph node, there isn't— any change, so that suggests that these cells really are just getting trapped in the lymph node, and that they aren't able to migrate into the tumor, and therefore that's why we're seeing this drop-off. So the concept that we came up with was this idea that the tumor-draining lymph node is really the site where these stem-like cells are maintained within the context of a developing cancer, and that this site may be a good site for maintaining them, because We think that this is a site where there's not a lot of TCR signals. We don't think that there's a lot of dendritic cells that are potentially presenting antigen to these T cells, and so therefore, when we look at things like TCR signaling, they seem to be very low for TCR signaling in the draining lymph nodes. By contrast, in the tumors, almost all the cells are seeing TCR signals at any given moment when we looked at them. We also see that they seem to be differentiated. Initiating, probably as a result of seeing that chronic antigen. So we think that just having these cells off-site may be, may be very important for actually protecting them and for maintaining the response over long periods of time. There's been a number of studies, we had a tiny bit of data on this in our paper, but really some way better studies are coming out now looking at this same phenomenon now taking human tumors and paired lymph nodes and actually looking for these. I really want to highlight this, this fantastic study that came out on BioArchives a couple weeks ago by a group in Memorial Sloan Kettering, Matt Hellman's group, where they really have looked at comparing the T cells in both the lymph node and the tumor using tCR-C, kind of the equivalent of what we did in the mouse, and showing that there really is good overlap there between the cell subsets. So it really supports this idea that the tumor-draining lymph node contains these cells., and that may be useful. And so one of the big questions that we have, and I think a lot of people have in this field now, is that these cells aren't just in the tumor, but they're in the lymph node. Is this a therapeutic opportunity? Should we be designing drugs that actually target these stem-like cells within the draining lymph node and try and either do a couple of things? One would be to actually stimulate them to see if you can drive them into the tumor. The other, which would be to try and prevent them once they get into the tumor from undergoing exhaustive processes. So I think these are kind of a combination of two different things that have been thought about for a while, but now this kind of provides a new angle on how it might be done. Okay, so now I'm going to switch gears and talk a little bit more about the other topic that I wanted to cover today, and that's trying to understand the mechanisms by which PD-1 and the PD-1 receptor is involved in maintaining peripheral tolerance. Tolerance. And as I mentioned, and I really do apologize because I copied this very quickly last night, this is Martina Damo. Let's make that very clear. Martina is the postdoc who's led all this work. She's actually going on the job market, which is one of the reasons I specifically want to highlight her. She's a fantastic postdoc and going to make a very good PI going forward. So she— Martina came to my lab, and she was very interested in tolerance. She had actually worked on tolerance, so she really brought this approach of trying to ask what's going on in our models using the models to study peripheral tolerance. And what I wanted to highlight was this idea that we've known for a long time that PD-1 seems to be involved in maintaining T cell homeostasis, but it's never really been quite clear what the role of PD-1 is in peripheral tolerance. And I'll go through some of that data to highlight where we think that the hole actually is. Is. So PD-1 provides an inhibitory signal that counteracts the signals that are coming from the TCR and from co-stim. So not surprisingly, PD-1 knockouts and the functions of PD-1 seem to have a lot of different roles. So there was very nice work early on from Arlene Sharpe and others highlighting the role of PD-1 potentially in thymic selection, specifically in positive selection. Some nice work focused on the role that PD-1 might play in the tolerance of recent thymic immigrants when they leave the thymus through nonspecific processes of activation, i.e., seeing TCR without necessarily looking at the peptide that they're seeing. Also, the role that PD-L1 might play in terms of suppressing cells as they go through the endothelium, so another sort of nonspecific process by which it could it could be involved. And then finally, this idea that it's involved in helping Tregs to have their function. So these are a number of different mechanisms that were described to explain how PD-1 might be involved in homeostasis. And this explained maybe some of the findings that were coming out of the mouse model when you looked at the PD-1 knockout mouse. And the PD-1 knockout mouse had an autoimmune phenotype, it was fairly subtle, they would at some low rate get lupus, or they would get cardiomyopathy, depending on the type of genetic background that they were on. And so this has been known, you can see this is from 1999, so this has been known for a long time, that this could be involved in receptor, or sorry, in homeostasis, but the function of it has never really been that clear, and most of the studies that were done on the role of PD-1 in tolerance has really focused on, this idea that if we get rid of it in an autoimmune-prone setting, we predispose these mice to getting really rapid onset autoimmune disease. So it seems to be important in restraining autoimmune disease, potentially after the point when you've initiated it. And I'll come back to that point in a second. However, this was a little bit different than what we're seeing in patients who are relatively normal up until the time that they're treated with these checkpoint receptors. So this is showing you data from a Cell paper that came out recently that focused on the cells that are observed after dual checkpoint therapy in the context of colitis. And what you can see is that these cells, which are tissue-resident memory cells, those cells, they showed, were clonally related to the cytotoxic lymphocytes that came out and seemed to be mediating the disease, suggesting that the cells that were giving rise to the disease were actually present before the treatment, and they were in the tissue. The other thing is that they— there were some studies in cutaneous dermatitis, which is what I'm going to focus a little bit on, showing that you could get CD8 and CD4 T cells into the tissue. You would get these really strong anti-skin responses, and that in some cases these would have monoclonal peaks of these— of CD8 T cells— I'm sorry, of T cells, alpha-beta T cells, suggesting that there was a very specific specific antigen these T cells were responding to. And then this dovetails with another series of studies that have been done on the role of PD-1 in terms of its role in peripheral tolerance. And this is what most people may be more familiar with, which is how is PD-1 involved in maybe the induction of T cell energy. And so going back many decades now, this is a fairly well-known study where Linda Sherman's group took a mouse where it expresses antigen in the pancreas, they transferred in antigen-specific T cells into that mouse, and they showed you get this really nice energy that's elicited that way. And that you could break that energy if you gave the mouse a nonspecific TLR signal, they gave TLR ligands, or if you just give influenza, you could now break the energy and now you would get T cells that make interferon gamma. Whereas in the context of no TLR signal, now you don't get that. Arlene Sharpe's lab used this very similar model where the T cells were getting energy and showed that if you get rid of PD-1 on these cells, now you'll get T cells that proliferate more, they have more function, and you also get disease in this model. So that suggested that PD-1 was very important in blocking the process of energy. But Viverdeses and Brian Feif had this really great paper looking at T cell responses in the small intestine, and they showed that while this was true very early on, if you give a blocking antibody to T cells transferred into this mouse, this is a mouse which expresses antigen within the small intestine, if you give T cells to that mouse and you give blocking antibody right away, those T cells become activated and will have effector function, they make gamma, they make TNF, they make granzyme, but if you wait a little bit and they get tolerized, now those T cells no longer have this ability to respond to PD-1. So this was an irreversible energy. And this fit with kind of our previous notions of how T cell tolerance works, in that once you've established T cell energy, now you can't break it associated with therapy. So we wanted to study whether this was true using the ninja model, and specifically we were able to now do something that was a little different from what had been previously done in that we could turn on antigens peripherally and then study endogenous T cell responses against those antigens just using MHC class I tetramers. So this is showing you turning on antigens in a variety of tissues, and I'm going to focus on the skin here, but we've done this in the pancreas, we've done this in the colon, we've done this in the skin, and we've done this actually in the liver. I won't talk about the liver, but this is really the outlier where we don't see T cell responses, but in both the pancreas and the gut, we see pretty robust antigen-specific T cells by the T cells that are specific for the antigens once we turn them on. When we look in the skin now, this is using a ROSA26 Cre, where we're painting on tamoxifen, we get a fairly robust T cell response, and that'll be the focus of what I'm going to talk about today. So we set this up so we could paint tamoxifen on the skin of the mouse, and then we would study what happened to the T cells, either if we did or didn't give checkpoint inhibitors, checkpoint therapy. And what was really very gratifying to see is if we just turn on antigen and we look and see what happens, there really wasn't much of an effect. We just give checkpoint therapy without turning on antigens, now we don't see much of an effect. And when we give the combined therapy, we're turning on antigen in the skin in this little square, we're also giving checkpoint therapy, now we get a skin disease that looks a lot like what we're gonna call a lichenoid dermatitis forms in the context of patients. And this is just quantification, and this was done, we have a dermatologist, a practicing dermatologist who's in the lab who did all our quantifications for it, and he and some of the other dermatologists we have who see patients with lichenoid dermatitis associated with IREs, they were very nicely willing to help us to actually compare these directly. So this is just quantification using a pathological score that he created based on the disease features that he would see in the patient. This is showing— this is a nice comparison showing some of the features that may be relevant in these diseases. I won't go into this, but because it's recorded, if people are interested, they can go back and look at this. But we did see a number of features that you would expect to see in the human disease that are present within the mouse. We also looked at the histology at the level of the tissue, and what we saw was that in contrast to the controls where we either didn't turn on antigen We just gave a checkpoint receptor blockade, or we just turned on antigen. We didn't really see much of an effect in thickening of the epidermis, but when we turn on antigen and give checkpoint blockade, now we get a very marked thickening of the epithelium. You can't see it here, but there's a lot of lymphocytes that are lined up kind of at this interface, and those end up being T cells. We've quantified it here a little bit to show that there are these features that are associated with lichenoid dermatitis. Again, I'm not a dermatologist, but this was made by dermatologists, including keratinocytes that are— and this is the human on the right and the mouse on the left. You can see these necrotic keratinocytes, which are very prominent in the epidermis. There's also spongiosis, which you can see as these kind of gaps in between the the keratinocytes themselves. So we see both of these features that are thought to be conditions that arise in this very specific IRE after therapy. So we were pretty excited to be able to see those. We also showed another feature that's been associated with the human disease, and that is this idea that the disease is really induced by PD-1, or by checkpoint disease, dual checkpoint therapy with PD-1 and CTLA-4, but not really with CTLA-4 alone. For it on its own. So that seemed to fit nicely with what people had seen. The next thing we wanted to do was to ask, you know, we've never had a model of IRAEs like this where you don't get disease or you do get disease when you give checkpoint therapy. So we wanted to understand some of the mechanisms behind it. The first thing we did was just look basically at what was going on with the T cells that were infiltrating the tissue. Previously it's been shown that both CD8s and CD4s infiltrate the tissue.. We saw that our infiltrate was predominantly CD8s. You can see here there are some CD4s, but not a large fraction. And that the CD8s were present in the skin whether we gave checkpoint or whether we didn't. But there was a marked expansion of those cells, about a 10-fold expansion of those cells when we gave the checkpoint therapy. We showed that if you deplete the CD8s and not the CD4s, that you could get rid of the disease, which fits with the idea that at least CD8s are sufficient to cause that. It doesn't say that in humans that CD4s don't have a role, they just don't seem to be playing a major role in this particular model. And we also used the transfer of antigen-specific T cells, so everything I've shown you before was endogenous cells. We used this transfer of antigen-specific cells to allow us to track things like the clonality of the cells, or sorry, the kinetics of the cells, and also to track the location within the skin of the antigen-specific cells. So we have these really nice luciferase-expressing cells. These are P14 cells that are specific for those antigens. We also have DS red cells. We transfer them in, and then we do the same thing. This changes very little about the model in terms of the outcomes. The disease is a little bit more severe when you have these cells, but overall, it's not like it changes the disease that much. What's critical is that this allowed us to track what happens to the antigen-specific both in the presence or in the absence of therapy, where you can see kind of early on they're located in the lymph nodes and maybe the spleen. Later on now you're getting the localization of cells, even in the absence of checkpoint therapy in the skin, but it's much more pronounced when you have checkpoint therapy. Additionally, now we could see where in the skin these cells were getting, and what's probably harder to see on this slide, but maybe hopefully you can see, There's a lot of cells in the dermis here of these red cells. These are all the antigen-specific cells. There's a lot of them in the dermis if we don't give checkpoint. Now when we give checkpoint, now these cells are really highly invasive of the epidermis. And that seems to be associated with where the disease is actually occurring. Because we have put antigens into the system that are fluorescent, we can see which cells are expressing the antigens. And what you can see is that in the absence of checkpoint receptors blockade, you can see that we're inducing GFP-positive cells. And I'll show you in a second, these are mostly epithelial cells. But when we give the combination therapy, now these cells are being ablated. And we know this is done by CD8 T cells because if we give anti-CD8 beforehand, we can block this destruction. And as I mentioned, the majority of these cells that seem to be turning positive are epithelial cells in this case. Skin, we think these are keratinocytes for the most part, and that those are the cells that are then being destroyed after we give checkpoint therapy. Okay, so we wanted to understand, again, more about the biology of what's going on here and really try and understand what's changing about the CD8 T cells after we give these therapies. And so what we did was we treated these mice, the mice that have this dsRed T cells in them, with either the tamoxifen and doxycycline to turn on the antigen, or we're going to give a checkpoint therapy on top of that. And just so that everybody knows, we also gave doxycycline and tamoxifen to our ninja mice, so they're getting the same treatment, they just don't have the recombination, the recombinase that's necessary to turn on the antigen within the tissue. So that we felt like was the best control. We did, we digested these and then we performed whole tissue single-cell RNA-seq and TCR-seq to try and understand what was different between the groups. And so this is just showing you kind of what I call like the Godzilla plot because this kind of looks like the monkey that— this is King Kong and this is Godzilla. People can disagree with me if they want. But so over here in King Kong we have epithelial cells that are very prominent when we have not turned on antigen. And when you give the checkpoint therapy, you can see that a lot of the epithelial cells are disappearing. We also see changes in the T-cell compartment and the myeloid compartment that I'll talk about in some detail. So just focusing on the myeloid cells and the changes that we saw in the myeloid cells, what had previously been noted by investigators looking at the disease that comes out, is that there are a lot of these CD14 positive, CD16 positive, maybe some sort of macrophage. I'm going to try and be careful about what I call them, because I don't really know a lot about which myeloid cells are which. And I know that there's a lot of thoughts about this, so if I call them the wrong thing, I apologize. But we were very interested in the fact that we were seeing pretty profound changes in the myeloid compartment when we just turned on antigen. These mice are treated the same way, they have the same, they're both treated with doxorubicin and tamoxifen, they both, you know, we're taking the tissue out the same way at the same time, but only in the case where we don't have the recombinase, we're seeing that Langerhans cells are really predominant in those no-antigen mice, but when we turn on antigen, now we see this influx of, or this change in the population, so now we get a lot of these cells that we think maybe are some sort of dermal macrophage. We also see that there's this increase in CD8-positive DCs. And this seems to be retained when we give antigen and checkpoint therapy. So it seems to be something that's happening largely when we just give the antigen. So what's driving this? Well, what we looked at was, first we looked at whether or not these cells were consistent with these cells that had been described before. And the cells up at the top here, these Group 0, Group 1, CD2 cells, they express lots of CD14 and CD16. They also express PD-L1, so we thought maybe these cells could be important in delivering the PD-1 signal to the T cells. That's some work that's ongoing right now. But what was really striking to us was that when we looked at PD-L1 globally, these were the only cells that were consistently expressing it in the mice without checkpoint therapy and with checkpoint therapy. So we looked by FACS and we saw the same thing, that this was a C11B, positive CD11c positive population that was increased after we gave antigen, and then those cells were the ones that were largely upregulating PD-L1 on their surface. We also showed that if we got rid of interferon gamma or if we got rid of CD8 T cells, that these processes didn't really occur. We mostly focused on the upregulation of PD-L1 on these cells, and that was a gamma-dependent process and also a CD8 T cell-dependent process. So that suggests that something surprising to us is that when we're turning on the antigen, we're getting the influx of CD8 T cells, that's causing a little bit of a change in the myeloid compartment, but that doesn't seem to be associated with the disease. So the next thing we asked is what's happening to the T cells that are actually present in the disease setting. So this is just showing you all of the T cells, and what you can see is you can see a decrease in gamma delta T cells when we turn on antigen, and that's further, same when we give antigen plus checkpoint. There are some increases in Tregs, but the really predominant change is this increase in CD8 T cells, which is expected. Of course, 'cause these mice had these DsRed positive p14 cells, we could gate on the antigen-specific cells and try and understand more about the biology of these cells and how it had changed in the context of just either giving antigen or antigen plus checkpoint receptor. And so this is, so I just wanted to remind you of the difference in pathology that we saw between these cells. This is the checkpoint therapy treated mice and the mice that get just antigen alone. In both cases we're seeing these CD8 T cells that are present within the setting. But the first thing we did was we just lined up all the genes that these cells expressed and we asked what are the differentially expressed genes? And it was really shocking to us that there were very few differentially expressed genes between these two settings. So this is just a correlation plot where you can see all of the genes they express, and on the y-axis you can see the genes that are higher in the cells that are given checkpoint plus combo, and on the bottom you can see the genes that are higher in the cells that get just checkpoint. What you can see is there's a handful of genes that are differentially expressed, but for the most part the genes are fairly similar in terms of their transcriptional levels. And in order to try and convince ourselves that this is true, we looked at heat maps of these genes if we just focus on this panel here, you can see that for the majority of these effector genes here, granzyme B, Fas ligand, perforin, other molecules, for the majority of them, the cells that just got antigen alone are actually as high or higher in terms of their gene expression level of these genes. Only granzyme A seemed to be increasing. We created something called an effector score and showed that these these genes are really not very changed between antigen alone and checkpoint plus antigen. We also showed that if you just remove granzyme A, this one was slightly higher, but that granzyme A seems to be the main difference in that. So this was really confusing to us because we were expecting to see a lot of differences in genes. We really spent a lot of time trying to focus on this and trying to understand what was different, but ultimately, We were also in parallel looking at these T cells using antigen stems and out of ex vivo stems looking at protein level expression between the checkpoint therapy treated mice and the controls. And what we saw there was quite a difference in terms of the expression of the production of proteins associated with these conditions. So when we just turned on antigen, we saw that the majority of the cells that we were getting out didn't produce interferon gamma. And there was a small fraction that did produce interferon gamma. But this was much lower than what we were seeing when we looked at mice that got gamma— sorry, antigen and checkpoint therapy in the skin. We also saw this increase in granzyme A, but if we looked at granzyme B, another classic marker of T cell function, we saw that these cells were expressing high levels of that in both cases. So there really did seem to be a few genes that were changing, especially these cytokines, but we really— at the transcriptional level, there wasn't much of a difference. So we wanted to understand, you know, maybe we thought that it's probably these changes at the protein level that seem to be maybe driving the disease. And we wanted to understand whether this was something that was acquired because we gave checkpoint therapy, i.e., we had activated cells in the lymph node, and now when we gave the checkpoint therapy, we were doing something different, we were gaining a property. Or whether, conversely, these cells were functional before they entered the tissue and they were losing function as a consequence of going into the tissue. And maybe that checkpoint therapy would block this loss of function, and that would be how we got the pathology. In line with this, we looked at where PD-1 was expressed on the cells, and what we saw was that the majority of the cells in the tissue were expressing PD-1, whereas in the lymph node it was considerably lower, about 3 to 4-fold lower in the lymph node. And that when we gave the antibody,, we could see that the antibody was binding, because it would block, you can see this shifts down, you can see in vivo this is blocking antibody. The effect is that it's actually decreasing the amount of surface expression of PD-1. So we think that the antibody can bind to the receptor in the tissue. We also looked at the function of the cells when we looked at the lymph node versus the skin, and what we saw was that in contrast to the mice, when you give antigen plus antigen plus checkpoint, where there's not really much of a difference between the lymph node and the tissue. When you look at the antigen alone mice, you see that the T cells in the lymph node tend to be more functional than the ones that are found in the tissue after, in the absence of checkpoint therapy. So there really does seem to be both for the change in gamma, also in TNF, when they enter the tissue, they seem to be losing their effector function, at least at the protein level. We also showed that, you know, for the cytolytic proteins that I showed you earlier, they were expressing these proteins, but in order for them to express them and actually function, they have to be able to externalize them. They have to be able to secrete them as lytic granules. And when we look at their ability to do degranulation, we see that in the lymph node they can degranulate very well, but when we look in the skin, most of the cells that are present there really can't degranulate, and that's why they can't upregulate CD10. CD107a. So even though they're making these cytolytic proteins, we don't think that they're actually able to do killing, and that's something we're actively trying to show right now through ex vivo killing assays. This phenomenon was not as strong in the context of the mice that get antigen plus combination therapy, which suggests that these cells, maybe at the post-transcriptional level, can't kill because they can't degranulate, they can't produce cytokines at the same level. So they really are tolerized, but they're tolerized largely at what we think is the post-transcriptional level. So that takes you to the model of how we think this works, and this is one of my final slides. We think that in the context of at least the skin, that when we turn on antigens, and that this may be part of the natural regulation of the skin, that there are some activation of T cells, within the lymph nodes, and that these T cells do have effector functions. So this is not like classic tolerance, where the T cells are not functional and therefore they never acquire it. These T cells seem to be largely functional. When they go into the tissue now, we think that they're secreting maybe a little bit of interferon gamma, and this is important for altering the dermal myeloid cell subsets in the tissue. And this is something we're fairly actively working on. We think that this is resulting in PD-L1 upregulation, Feedback loop is what's keeping these cells from causing disease. But when we give checkpoint therapy, now we block this interaction, and now we get these T cells that get up into the epidermis, secrete cytokines and cytolytic molecules, and cause a lot of pathology. So just in summary, you know, we've— as Tim mentioned earlier, that I used to work on acute infection. This is sort of how we thought that this process worked in acute infection. Where T cells had functioned kind of both as memory precursor effector cells and as short-lived effector cells, that the functional capacity was there, and that cells could make cytotoxic and cytokine genes in acute infection, like kind of all the cells could do this. In the context of chronic infection, we think that there's a stem-like cell that really doesn't have all these functional qualities that resides in the lymph node, and that when it goes into the tumor, it can either acquire these, especially in the context of PD-1 blockade, and cause destruction. What I'm describing in the tolerance setting we think is something a little bit different. Classically we think about that as being T cell energy, and I didn't have time to tell you this, but we see this in the liver model, that when you look at the liver model, you get a lot more of this anergic T cell phenotype, so it's not like that doesn't happen. But in the skin, we seem to think that there's a different pathway whereby now you're getting these non-pathologic effector cells that when they migrate into the tissue, they really become tolerized at the post-transcriptional level, and therefore they're unable to mediate effector function and not cause pathology. When we give PD-1 blockade in this case, we think that now we can unleash these cells and they'll cause disease, and that's ultimately why now we're seeing the disease that comes out. And that we think, I think most importantly, we think that's the mechanism for how patients are getting IREs. That they have T cells in their skin and maybe other tissues that are being completely held in check by the checkpoint receptors, but those are functional effector T cells that when they give blockade, now those T cells will attack the skin. So this is the most important slide. I mentioned some of the people along the way. I really want to highlight Martina, since her other slides didn't actually say the correct name. Noah, who's a dermatologist who's been working in the lab. And then Kelly Connolly, who's been very important in the first story, and the funding sources that we have.