Scientist Stories: Max Krummel, Discovering Spatiotemporal Archetypes of Cancer Immunity

Hello, everyone. I'm delighted to be opening another Global ImmunoTalk. If you're joining for the first time, I'm Elina Zúñiga from the University of California, San Diego, and I'm here with one of the co-organizers, Dr. Matteo Iannacone, and our terrific global immunospeaker today, Dr. Max Krumer from the University of California, San Francisco. So just a brief announcement that next week Global ImmunoTalks will be given by Dr. Akiko Iwasaki from Yale University. And this is kind of a special talk next week because it happened to coincide with the 9th annual meeting of the International Cytokine and Interferon Society. And so they are featuring Global ImmunoTalks in their program. I think that you need to register for the annual meeting for the, of the Cytokine Society. However, as always, the Global ImmunoTalks will be free and the link will remain the same. So I also want to remind everyone that in this Global ImmunoTalks format, the questions are via Twitter. And so please ask questions after the talk via Twitter for So that you can find in the account Global ImmunoTalks a tweet that is already there that says, ask questions for Dr. Max Grummel here. And you can reply to that tweet with the hashtag GlobalImmunoTalks, and Max would use his own personal account to answer the questions. And remember, you can ask questions after watching the talk in YouTube in an asynchronous manner. Okay, with that, I will pass it to Matteo so he can introduce our speaker today. Thank you.

Thanks, Alina, and it's truly a great pleasure to introduce today's Global Immuno speaker, Max Krummel. Max obtained his Bachelor of Science at the University of Illinois and his PhD at UC Berkeley. And during his PhD, Max worked with Jim Allison and contributed to the discovery that engagement of CDLA-4 by either antibodies or by its ligand resulted in the dampening of T cell responses. And as you all know, this work led to the approval of anti-CTLA-4 antibodies for cancer immunotherapy and formed the basis for the 2018 Nobel Prize in Physiology or Medicine. Max then did postdoctoral training with Bill Heath and Ken Shortman in Australia, as well as with Mark Davis at Stanford. And after that, in 2001, he was recruited at UCSF, where he quickly rose through the ranks, and he now is the chair of the UCSF ImmunoX Initiative, and he holds the Robert E. Smith Endowed Chair in Pathology. His contribution to science has been tremendous, and I don't have time to go through all of them, but his lab specializes in using real-time imaging to launch and test hypotheses related to how the immune system processes information and makes decisions. And key recent discoveries from this, from his lab include determining features of T cell membrane biology and motility that govern how they efficiently survey for antigens, as well as the discovery of archetypal collections of immune systems in cancer, notably those involving networks of cells built around stimulatory dendritic cells.. And I like the fact that Max's work spans scales from, you know, the membrane organisation to cellular biology to whole organism and systems biology. In addition, Max drives collaborative science. He founded the Microscopy Collaboratory at UCSF, which unites shared technical personnel, and he developed a novel industry consortium-funded project called Immunoprofiler, which unites studies of over 15 cancer indications to understand the biology of individual patients. And together with other UCSF faculty, he founded the ImmunoX Initiative, which is a radical collaboration platform focused on methods and data sharing as a means to accelerate discovering cures. Besides co-discovering anti-CTLA-4 drugs, he— Max co-founded two biotechnology ventures, pioneer immunotherapeutics and found innovations to develop novel cancer immunotherapies. Max has received numerous awards for his work, including the Pediatrics Flag Mentorship Award from UCSF, a career award from the Leukemia and Lymphoma Foundation, an Investigator Award from the Cancer Research Institute, and he is a fellow of the American Asthma Foundation. Now, Max, thanks for accepting our invitation. And before you start, It is a tradition of Global ImmunoTalks to ask speakers a question to get to know them better and to inspire the next generation of scientists. So the question we would like to ask you is, how do you think cancer immunotherapy will look like in 5 or 10 years from now?

Well, this is a good segue to the talk. And I think one of the things that I'll talk about today is the idea that cancer immunotherapy won't just be one thing, and it will be very much focused on these things like archetypes, these— the biology of cancer. And I think that's a good segue into the talk.

Fantastic.

We very much look forward to it. You can now share your screen.

All right. Let's see, where is it going? Give us a sec.

Sure. Max's talk is entitled Discovering Spatio-Temporal Archetypes of Dominant and Reactive Cancer Immunity.

Okay, Max, take it away. Let's go. All right, I'm going to take it away. So I will, I will enter with a little bit of a background on where my lab is coming from. And, and really the thing that we focused on in the last 20 years really is this concept of trying to figure out emergent behavior in immunological systems. And one way of thinking about this is that all biology is just controlled chemistry. At some point, the spatial and the temporal organization of molecules is what gives rise to things that are at the subcellular and cellular level. It's really compartmentalization of chemistry that gives us biology. So if we watch in real time how molecules move, we basically can build up the understanding of how subcellular systems and cellular systems work. So we image a lot of that, and I want to use an image here. This is, this is going to play as a movie in a moment. In green is a T cell, and it may be not exactly how you see a T cell in imaging, but this is a method called lattice light sheet imaging that we use, and in it you'll see a lot of T cell behaviors of palpation. So T cells are palpating the world at large, and as you play, in red, the cell is a dendritic cell, And what I want you to see is that, that while the T cell is palpating, looking for antigens, the antigen presenting cell is wrapping itself around the T cell to be seen. So rather than T cells sort of in this kind of world of immunology just acting on their own to find targets, you can see at the membrane how much the dendritic cell wraps itself into the grooves of the T cell and essentially allows itself to be scanned. And I really bring this up because as we're going to move forward into today's talk, I want you to be thinking about the idea of partnerships amongst cells, where the cell types are really aligning both in terms of their gene expression and their actual sort of cell biology to be partners. So if you think about subcellular organization and cellular organization, that gives rise to the next level of emergent behavior, which is how multicellular assemblies form. And I'll show you a movie like this in a moment, but essentially if you think about then, how those form, that gives rise to the clinical, the clinical level of emergent behavior. And if we just take checkpoint blockade, which Matteo introduced, I've done for quite a long time, and you look at patients and you look at their overall survival on this Y-axis versus years, you can already see that there's at least two kinds of cancer patients, those that respond to a therapy and those that do not. And I'm going to argue quite— I think it's pretty obvious that that represents different multicellular biology, that the way that the immune system is arrayed in those patients has to be different. And so this has been sort of an origin story for me in terms of going back to this. So the other thing that I kind of need to bring into this talk today is to remind everybody that you will have seen this a lot in global immunotox in general, but there's a laser-like focus on cancer and cancer immunology and how to make checkpoint blockade better. To treat those people. But there's a— there's been an orthogonal development in immunology over the last 20 years, and that is that we've started to realize that the immune system is not what we thought. It has a lot of functions that are not about defense, and it has a lot of functions that are not about tolerance, which are the two ends of the spectrum that really have defined immunology over the last 100 years. So just as an example of this, the work that's been done in microglia in the brain tells you that you will not remember this Global Immuno Talk. You may not remember it anyway, but you're not going to remember it for sure if you don't have an immune system. And the reason is that your immune system is busy right now pruning your neurons, pulling off synapses that are not to be used, allowing other ones to reinforce, and that is memory. That's the circuitry of your main brain remodeling. Similarly, just one example, there's a bunch of them on this slide, and I'll just call your attention to the fact in the last 10 years, the microbiome is becoming really apparent role, both in terms of immunologically, but because we understand that microbiota bring us genomes that we don't have. So evolutionarily, the immune system takes a stance of quarantine and curation rather than elimination of bugs, which is sort of the dominant idea we had about what the immune system should do maybe 10 or 20 years ago. It's really— we've had a really strong reason to shift and understand that the immune system needs to curate these. This is a good thing. And how do we keep them in our system and yet keep the bad ones out? Right? So I'm going to make an argument to start us off today that understanding tumors that don't respond to immunotherapy starts with understanding the spectrum of the goals of the immune system. So on the left here would be kind of a classical spectrum of immune reactivity that you will have seen in textbooks. It goes from tolerance, basically ignorance and lack of reactivity at one end, to destruction at the other end of the spectrum. And what I just showed you shows you essentially that there's this emerging archetype or behavior— I'll define archetype in a moment— that we call accommodation. So accommodation— I'm going to— can I mute you? I can maybe mute him.

I can mute him.

Thank you. Sorry, Mattel. So the point being here that you can find collections of immune cells sometimes the same, at least ostensibly the same names, that are involved in different immune function. And it's really how they combine with other cell types to define that function. So for example, we understand that Tregs together with macrophages can manage tissue metabolism, but the same ostensibly Tregs are also involved in commensal symbiosis, that sort of curation I was mentioning. So it's not the single cell type that defines this collection that has a function. But what we call an archetype. And this defined as these collections of cells that are in linked states, they work together, almost certainly following an evolutionary design for how recognition of an event can lead to one of these accommodation phenotypes, for example, healing wounds or managing tissue repair, et cetera. And so these are shown here. So again, understanding that that might be the way that the tumors are growing is by accessing these. And, and to that extent, I'm going to posit that some of these are the mold for dominant immune systems in tumors. Different tumors basically being able to pull in the immune system to do some of these functions, to serve them, is really in opposition to the idea of a reactive immune system, which is the one that we've all understood for years in this destruction area. And so in that area, I'm going to show you some data that we've generated over the last 5 or 6 years that defines two of the major reactive immunotypes that are— that we want to access. So these are the things you would like to have in cancer. And in fact, these are the ones that you have. These are the dominant and these are the sort of modest reactive ones. So one way of showing you this, and this is actually how we first got into it, was a look-see experiment, a look and see. And the look and see was to look and see how immune cells were behaving within the tumor microenvironment. And develop hypotheses on these. So sometimes when people say you're going in with just no hypothesis, I think sometimes the hypothesis is there's something interesting here, and that often is good enough. That's curiosity. Okay, so if I, if I play this movie for you, what you're going to see is the tumor is in the area in black. Green and yellow and orange are various forms of macrophages, and blue are some T cells that are specific for an antigen that's on the tumor. And this is a really common behavior right here. This is the swarming of CD8 T cells onto macrophages. And we'll hear a little bit more from a postdoc in my lab at the end of this talk on that. But this is another region within that exact same tumor where T cells are actually activating on very red dendritic cells here. They actually change their color of blue a little bit as they activate. That's because the pH changes that happen with T cell activation can take place. And this is what we're going to presage. This is what we're going to discover as reactive immunity. This is the kind of things that are necessary to get a T cell response. And down to the left, what we saw before in the first sort of region is the much more common dominant archetype behavior that we see in, in this particular tumor model. This is a— that was a breast cancer model, PYMT CherryOva. So the first thing that I want to bring up is this concept of reactive immunity. And for us, it started with a study in 2014 by Miranda Brose, who was a graduate student in the lab, that isolated all the different possible antigen-presenting cells in tumor microenvironment and essentially characterized them and asked which ones might stimulate T cells. Because at the time, the theory was that the immune system was entirely suppressed, that the tumor microenvironment was completely suppressive. And what I'm going to tell you is to the— au contraire, that there are in fact cell types that are available to stimulate T cells, but they're very rare. So here they are, this— they're CD103, CDC1s. At the time there was a nomenclature discrepancy, so our first paper actually called them CDC2s, but they're distinguished from the CD11B population of dendritic cells that are over here. So if you mix CD103 dendritic cells with T cells, you get NERF77 upregulation and CD69 regulation when the T cells are specific for tumor-specific antigens. And in fact, those cells, as Miranda showed, are really the only cells that make IL-12. They are very good at cross-presenting. They keep the antigen in a low pH environment so it doesn't degrade. There's really quite a lot of important studies in Miranda's paper. But she also showed that they're super rare and they're typically tumor-excluded in the tumor microenvironment. And in fact, she also showed that they're predictive of overall survival across multiple cancer cell subtypes. So if you look at a stimulatory dendritic cell signature in metastatic melanoma, people that have that have a distinct survival advantage, shown here, time versus overall survival, compared to the people that have low frequencies of that signature. And this has really expanded a ton. Lisa Cousins and Brian Riffel showed also that these make IL-12, and they've targeted them using TIM-3. Mary Murad and Sancho also used BATF3 knockouts to show that these are required for anti-PD-L1 responses. And Tom Kajewski and Caetano also showed associations of CDC1 with outcome. So we've been looking at this from the standpoint of using biopsies to understand how checkpoint blockade works. And I'm just going to briefly introduce the immunoprofiler. Essentially, we're taking hundreds and hundreds of biopsies, dissociating them, doing flow, VAX, RNA-seq, and getting all the clinical data to make the associations. So when we looked at human melanoma and we looked at the frequencies of CDC-1s, the good antigen-presenting cells, we saw this really curious finding, which is that inside the tumor, the more CDC-1s you had, the more of the putative sort of cytokine that drives CDC-1s we found, which would imply that there was a cell type in tumors that had CDC-1s that are letting them grow. So FLT3 ligand is the cytokine that makes CDC-1s grow. The more of that you had, the more of these we had in the tumor. So it suggested that there was another missing player in the tumor microenvironment. So making this a long story short here, if you make a knock-in of FLT3 locus and you put a reporter there, you find that really there's very little expression in the stromal cells, in the tumor themselves, in the myelin compartment. But the key cell types that have expression of the FLT3 transcript are NK cells and then to a lesser extent T cells. And that's quantified down here just on some graphs so you can see the large amount of NK-produced transcript. And again, making that story relatively short, that turns out that that's essential. And on the work that we've done on that has been complemented by work from a number of groups showing that there's a chemokine attraction between NK cells and, and CDC1s that allow them essentially to amplify and create this little niche. And the niche is important, and the way we know the niche is important is that if we profile patients that are treated with anti-PD-1 before they're treated and then after they're, you know, we treated, until we collected all their data, we can categorize them into responders to PD-1 or non-responders. You can see that the number one cell type that has a strong association with responsiveness is, are the CDC1s. And the second most sort of defining feature are NK cells. So you see all this red here in the responder categories, and then there's a lot of blue in the non-responders. So this is essentially what I'm gonna call a class I archetype, mainly these are class I presentations to CD8s, and you're getting these NK cells and the CEC1s to work together. But we also noted even in this paper, this is from Kevin Barry, who's now at the Hutch, we noted that there was a popular small population of responders that also don't have that archetype. And their second most associated phenotype was that as the CDC1s go down, the left-hand side of this graph here labeled D is showing you the decrease and as sort of a waterfall plot in CDC1 frequencies in responder patients, non-responders are here on the right. And as you lose CDC1s in those, you actually find that another population of dendritic cells, CDC2s appear And if I just were to circle this for you, you'll notice that the second most associated set of cell types are CD4 T cells for responsiveness. And they're only in these ones that really don't match the archetype I've just told you about. So really kind of summarizing some of the work we've done over the past 5 or 6 years in this area of reactive T cell immunity, there appear to be sort of two ways that you can get antitumor immunity. And it's not really surprising when you put it all together is that on the one hand, you can have a CD8 response that we call type 1 reactive immunity. And that's essentially the CTC-1s that in the tumor microenvironment can be stimulated by NK cells. They ultimately migrate to the lymph node where they can activate T cells there. And then when T cells come back to the tumor, they find— if they find CTC-1s, they can be reactivated, leading to tumor rejection. And really, all of these parts have now been shown, many by us, but also by, again, a few others that I've mentioned. And on the other side of the equation are type 2 responses. That involve CD4 T cells, which although this is not famous about them, can quite easily kill and many tumors upregulate class II. So this seems to be a story in which CDC2s can migrate again to the lymph node and they can activate CD4s to come back. And this little story that I won't get into detail here but is in Mikhail's paper shows that this is really heavily regulated by regulatory T cells so that normally CDC2s, though they come to the lymph node, they stimulate CD4s to become relatively kind of duds. Okay, so I'm going to use this as a way to jump into the question of what are the dominant, or what is the dominant, if you think singular phenotype of the immune system in cancer. If these are little minor reactive things that we can access and we'd like to have more of, what is the immune system actually doing in cancer? Is it doing one thing or is it doing different things across different tumors? So to introduce this, I want to remind you how cancer has been thought of clinically for a very long time. So going back hundreds and hundreds of years, doctors would specialize in certain kinds of cancer. So a melanoma doctor saw cancer differently than a liver doctor. In fact, these have been subtyped molecularly over the years so that breast cancer, for example, has 4 distinct pathological subtypes. Probably more. And when you then think about how you're going to treat cancer, you aim your therapeutics and your diagnostics at the tissue of origin features. So for example, Herceptin is a drug that is used for HER2-positive breast cancers, but really not very much anywhere else. You can also look at cancer as we started to do, people started to do in the '80s, '90s as oncogenes were discovered, as cells that are driven by oncogenes. And so if you wanted to treat a cancer, you found out what oncogene expressed and then you made a drug against that. And just to give you a fantastic example of human ingenuity that hasn't really done all that much is vemurafenib is a fantastic, I'll back that out a little bit, it's done a lot, but vemurafenib is a fantastic drug. It's an amazing example of specificity and affinity in a drug against the mutant form of BRAF V600E, which is present in about 50% of melanoma patients. And it kills cells. It blocks this like nobody's business without blocking anything else. And so it gives rise to a drug that's very much specific to, for example, BRAF-driven tumors. The issue with that, of course, has been that over the years, we've discovered that targeting single molecules like oncogene drivers is really just asking tumors to mutate. And that's one of the best things that tumors are good at doing. So this has been— it's sort of pitfall. Okay, so if we're focusing on immunotherapies now and we're realizing, hey, this is a real angle in which we can use the immune system to help redefine a tumor viability, then we're actually defining cancer as some form of immunopathology. In other words, the immune system has done something wrong that we want to apply a therapy to. So the question that I think we needed to ask ourselves, and we started asking about 2014 or so, maybe a little earlier, was how many kinds of immunopathologies exist. So to define that for you in sort of more layman's terms, if you think of all the balls here that represent different things the immune system might be doing wrong in a tumor, how many tumors have which of them and which ones are combined with each other? So for example, you all know this concept of T-cell exhaustion. Not every tumor has T-cells that are exhausted in it. Not every tumor has a clear overabundance of T regulatory cells. So which ones pair with each other and why? And when do you have more than one? What are the markers and how do they, how will we treat this? So this was a process that we started again quite a while ago. It's taken a very, it's, it's been a labor of love of a lot, in a lot of ways to discover the dominant immune systems in cancer. And to do this, we didn't want to just take one cancer type. We wanted to see this across the spectrum of cancer, guessing that there will be themes that were repeated. So we took cancer from about a dozen different indications, and we subjected all the tumors that we could get from the clinics at UCSF. This is a huge collaboration, a huge kind of local effort to take this thing apart and literally take apart the biopsies, do flow cytometry, a little bit like I showed you in the, in the story with Kevin Barry, Kevin Barry's work, and isolate all the cell types and quantify them and sort of say, what are some of the really fundamental patterns here. So at the bottom is a— maybe a trivialization. You can take every tumor type and do waterfall plots where you show relative frequencies and rank them from highest to lowest for the frequency of T cells, the frequency of myeloid cells, frequency of stromal cells. And if you look at these, you'll see that it's possible to do that, but that the order really has no connection to one another. It's not that you're always T cell rich and then you're always correspondingly, you know, stromal rich or stromal poor. It turns out those connections just don't exist. And I think one of the most important things is you see how much variability there is. Maybe the best one to point you to if you want to look at this would be look at the frequencies of myeloid cells in PDAC. PDAC is pancreatic cancer. It really shouldn't be shown on a box and whiskers plot by all virtues of the fact that it's clearly non-Gaussian. There are some PDACs in which there are tiny numbers of myeloid cells, and there's some PDACs in which there's just tons of myeloid cells. And in essence, immunologically, these probably shouldn't be considered the same kind of cancer. And yet we do. We call that pancreatic cancer. That's the historic lens of origin. Okay, so if we take all the frequencies that we got from immunoprofiler— now, this is the work of Alexi Combs in the lab, and this is a work that's hopefully just about to come out. We can do essentially the equivalent of a tSNE or UMAP clustering using Levan clustering and isolate tumors from different origins based on simply 3 frequencies. And that is, here we do T-cell frequencies, myeloid, and stroma. So this is equivalent to the concept of immune desert versus immune hot or versus cold. And you can see that there are combinations that come up very frequently and these cluster together. We've done cluster analysis, and this is an optimization. And so you can find things where there's lots of T cells and lots of myeloid cells but low stroma, or you can find ones we call immune stromal where they have high of everything. Just for sort of to give you a little sense of this, there are two kinds of immune deserts that come out. Basically, the difference between them is both of them don't have T cells or myeloid cells in high frequencies, but some of them are really high in stroma. So there's a A fundamental lesson here, and as you do clustering, I'll remind you of something that's really important here, is that if you cluster automobiles by color, you'll probably find that they cluster very nicely into red, blue, green, but that doesn't tell you much about their performance. If, on the other hand, you cluster by features that are important, you'll find other things. Like, for example, if you cluster by motor size, you'll mean your frequency size, find wheel size or transmission, etc. So there's aspects of performance that are more important than some of the other things that you could that you could cluster. So as we go through clustering, we've been really conscious of the question of whether the clustering based on a certain number of features predict other features or find other features that we think are important. And, and that's, you know, to think about that, maybe think about clustering in, in more detail. But here we take the clusters that we find and we just ask whether there's regions that have lots of chemokines, which would be predicted. That is, if you have more immune cells, you would predict more chemokines. And that's definitely true if we look on this plot here. These are the— the left-hand side are the immune-rich tumors, and these are the immune-poor tumors. So you can see that in general, as you get to these super immune deserts, very few chemokines are expressed. And in fact, the chemokines are distinct. It's not just chemokines broadly, but certain immune-rich tumors bring in certain kinds of— are associated with expressions of certain kinds of chemokines. And immune deserts, you can see, really don't have very many at all. What's important about this is that the method that we're using actually spans from our dataset, which we call IPI here, to TCGA. So if you look at all the chemokine expression across tumors, you can quantify from RNA-seq, you find that if we quantify and we qualify tumors into these— here are 6 compartments based on just those 3 parameters. We find that there's a really good lining up of which chemokines are expressed in which tumors. In other words, tumors that have lots of immune cells down here have particular chemokines associated with them based on which immune cells they have. Okay, so we've been going through this data more and more, and I'm going to tell you in the next 5 to 10 slides how, as we've added in additional data to the clustering algorithm, you begin to split some of those first 6 archetypes that I've told you ultimately into what looks like a kind of a fairly stable 12. And I'll be the first to admit that we're so early in this process that even with 400+ tumors, we have a sense that there may be additional subsplits and that sometimes due to resolution, we may have a single tumor that maybe belongs in one category or the other that's mixed. But in general, what I'm showing you here is what are called alluvial plots that as we add in, we go from the 3 features that I just showed you, T cells, myeloid, and stromal to 6 that have features of how frequently do you find Tregs, CD4s, and CD8s. For example, the immune-rich compartment easily splits into those that are CD8-biased and CD4-biased. Other things split based on whether they have regulatory cells. And if we start to add into the 6-feature clusters that we get the features of where they have CDC1, CDC2s, monos, and macrophages, there begins to be some quite stabilization. For example, the immune stromal CD8-biased compartment really is unchanged by adding that in, as is, for example, most of the cells— excuse me, it's the phone— most of the cells in the immune-rich CD8-biased are largely unaffected. So as we get into these higher sort of dimensional spaces, we start to find some really interesting things. And I'm going to show you how that all pans out. So on the left here is taking all 300+ cancers and clustering them by the features that I just mentioned, those 10 features that included myeloid cell, et cetera. And on the map on the left, we're color coding based on the call-down archetype. So for example, I want you to note there's a red composition versus a pink composition here that are immune-rich CD8 macrophage versus immune-rich CD8 monocyte. If you look on the right map here, we map by indication. And you can see that both of those are very rich in kidney cancers. So I'm going to take you now to the plot here. If we, again, plot these two particular archetypes, one versus the other, you see a lot of red, red representing kidney cancer. Conversely, if you look in the bottom left here, you find populations that are T-cell-centric. That means they have lots of T cells, not so many— not so much stroma, not so much macrophage. And in there, if you go across to indication space, you find a lot of melanoma and lung cancer. And that's shown in this region that's right in here. So this T cell macrophages is the browns and the blues represent these indications that have a lot in common. So on the one hand, you can find some things that are starting to become evident across many studies. But you can also find— and for example, you find a lot of gynecological cancers up here in the top right. These are the oranges. That in archetypal space are called out as immune desert CD4 macrophage bias. That's the region again on the top right. So you can find this propensity for certain cancers to lie in certain quadrants of the archetypal space. But I want you to note that even though melanoma tends to be CD8 T-cell rich and sits down at the bottom, you will find melanomas that are immune deserts. This is not one disease. This is immunologically different kinds of diseases. And similarly for this, for the red kidney cancers that I showed you at the beginning, they largely lie in this CD8-rich macrophage monocyte-rich compartment here. But you will find kidney cancers that, again, are in the top right that are immune deserts. And that kind of has been shown in this distribution, which I don't expect you to read in the context of this, but just to appreciate that tumor indication is not the same as immune composition at all. So one of the ways, again, that I want to make sure to emphasize that I think we're looking at something that's real isn't just a feature of computational clustering. Is that as we get into 10 features and we look at that chemokines that I showed you before here, this is a heat map for the chemokines expressed in clusters that are formed by just 3 features of the immune microenvironment. If you bring in 10, it really sharpens it so that the red and the pink that I was showing you before is really where you find a lot of CCR5, XCR1, CXCR3 chemokines and their ligands. And for example, in these green clusters here that are more stromal rich, you find a whole different collection of chemokines and their receptors. So there's a specificity of the chemokine systems that are being used that represent the cells that end up being in the tumors. And this is where we sort of get into one of the sort of Steve Jobsian moments where you say, and there's one more thing. The phone also tracks you all the time. So in the context of these archetypes, one of the one more things is that having just looked on those 10 features, T cells, myelids, fairly simple ones, it turns out that the clusters that are formed predict a bunch of other things. And so I'm going to show you those over the next few slides. So for example, if you look here at NK genes, plasma genes, B cells, and mast cells, these are not cell types on which this data was clustered. You find that all of the NK genes are basically in these, the pink and the red clusters, the immune-rich, CD8-biased. Conversely, all the plasma cell genes are expressed in this one called T-cell-centric macrophage biased. All the B cells, naive B cells, are in this one that are called T-cell-rich. And all the mast cell genes are in this one called immune stromal macrophage CD4 biased. So it turns out that a certain collection of cell types will help— they'll create the mirror image, or they'll create the sort of the shadow of all kinds of other cell types. And just to back up what I'm showing you here on the left, this is all using gene expression from our dataset. We went back to the flow cytometry. And indeed, the pink and the red as shown here are where all the NKs are by flow cytometry. And for example, this light blue is where all the B cells were found by flow cytometry in different archetypes shown on the right. Okay, so now another one more thing is that if we go into gene expression within the compartments, so if we take a series of genes that are defining Th1 cells, this is just from the literature of T-reg, TRM homing and exhausted cells, M2, these various things. And we color code those genes and they're named down at the bottom of this slide. You can find that individual archetypes cluster those genes and they are actually pretty much what you predict. So in other words, if we take the CD8 biased, these ones that are really heavy, for example, in kidney cancer, we find most of the Th1 genes. The Th1s are these pinks and of course the NK genes that I already mentioned. But other ones come together too. So in this gray archetype that's through here, you find a lot of the green, which is Th17 genes and associated suppressive T cells. And for example, in the regions over here, this is where our B cell genes are that are in blue, et cetera, et cetera. So the archetypal designation is starting to look like it's not only descriptive, but also in some ways predictive of other things. And that's a good thing. If your clustering actually finds things that you didn't cluster based on, that, that at least it suggests that it has solid meaning. And another thing is that if you take the clusters and you map over them the frequency that you find Ki-67 proliferating cells in the tumor microenvironment, you find that all of the Ki-67 is very heavily clustered in immune deserts. So the immune deserts are shown over here on the, on the right, and you can see that they are— that's where the KI-67 high occurs. So that prompted us to look one level further at these and essentially ask if we took collections of genes that we know are associated with certain kinds of tumor biology. So for example, senescence or fibrosis or cell cycle, as I just showed you, what do the archetypes say about that? And there's again a partitioning so that. For example, if you look down at the bottom, these are— there are 3 forms of immune deserts that we find. Really only one of them is really high in fibrotic genes. That's what's down here. And there are some fibrosis genes associated with some other kinds of tumors as well. And fairly uniquely, you can see fibrosis is really— there's 2 archetypes that have a lot of fibrosis genes in them. EMT is pretty heavy in this monocyte bias, but not in the CD8 macrophage bias. So you start to find features of the tumor biology that is mapping to the sort of coincident immune biology. And to take this again yet one more step, I think what I'd really like you guys to take away from this is that if you use this collection of gene expression cell type archetypal distinction, you can find all kinds of things that match, like I said. So for example, you can find the genes in the tumor compartment that tend to be most associated with those archetypes. So this is a unbiased screen for those, whereas what I showed you in the last slide was we took genes of particular origin. We can do that from the live compartment, creating signatures, and then we can throw that at something like TCGA. And over here on the right is showing you that if you look in the CD8-biased archetypes, the presence or absence of macrophages— that's the pink versus the red, or the green versus the yellow— there, macrophages versus monocytes makes a big difference. But if you look in ones that are CD4 biased and you add in or take out macrophages, it really makes very little difference. And the point I want to make from this is, A, that these archetypal distinctions are predictive. Great. But also that they're predictive in a way that any of the single cell types, if you just called out Th1 or macrophages alone, you don't get that because the key part of this is it's the collections of cells that are defining this biology. Biology. Okay, so one last bit before I turn this over to some of my lab members, I'm going to do something sort of different today, is to point out that you can look at the question of whether the dominant immune archetype predicts the reactive one. So here's the map that I showed you earlier of the various different archetypes and their color coding by names. If you look at CD45 fractions, you can see what I've sort of shown you before, These bottom left guys are all the immune-rich tumors, so-called hot, and the top right are the so-called cold tumors. And it's true then, if you overlay onto that same map a signature for stimulatory dendritic cells, the immune hot tumors is where you find a lot of the SDC scores. So it implies that people with a lot of immune— you know, the richness of the immune system is going to be generally good for their response to immunotherapy. But I also wanna point out that that's not universal. You can find high STC scores that are present in the immune deserts. And I think this is important because as we partition patients into what we're gonna give them, this goes to the question that Matteo asked in the future about the future. And in the beginning of this, of the talk, I'll sort of answer this now, is that it's really gonna be about the biology that you have a drug for that matters the most. And as we get into away from just trying to do me-too checkpoint blockades, we will find that there is biology that we're trying to access that is very specific, not to the necessarily personalized patient, but to this precision aspect of what immunobiology we're trying to affect. So we really need to understand this. Okay, so here's what I'm going to do, something different. I'm going to make the argument to everybody that's on the line right now that there's no reason why in virtual world We should just be talking on behalf of our labs. So I'm sitting here with Kelly Kirsten, who is a postdoc in the lab. In the last 10 slides, we're going to do something kind of cool and different, and that is that I'm going to let the postdocs tell their story. In particular, I'm going to show— or let two folks from my lab that are just finishing postdocs tell their vision for the future a little bit, and also the work that they've been doing in the lab that ties into this. And I want to emphasize that they don't have— in the interest of time, they don't have a ton of time to tell their whole story. But a lot of it fits in with this theme, obviously, because we've been working as a lab. So I'm going to phrase this as a question for introducing Kelly, and then Kelly will introduce Ken, who in the lab— so Kelly has been working on the question of what is the relationship between cell types that define these dominant archetypes? That makes them part of an archetype. Why do they parse together? Is there a biological reason for that? And to introduce that a little bit, I'm going to point out that as we looked at the feature of exhaustion across the archetypes that I've just been showing you, we noticed this really clear correlation, which is that when you look between, let's say again, those red and those pink archetypes, one which is macrophage-rich and one which is monocyte-rich, we see this distinction in the exhaustion score. So although there are exhausted cells in the pink, they're higher in the macrophage-rich red than in the pink. And for example, an immune desert macrophage is one of the highest ones too. These are— these contain macrophages. The one next to them contain monocytes. So you can see this distinction. And that extends, as Kelly will tell you, into single indications. So if you take kidney cancer, for example, and you plot various features in a T-cell compartment of exhaustion, like T cell A4/P1/CD38, you find that individual patients, which are shown in the columns here, that have high exhaustion scores are also the ones in which the macrophages are most— or the myeloid compartment is most distributed to macrophage identity. So we're seeing this relationship between macrophage differentiation and regulatory T cell. So I'm going to introduce Kelly. Kelly came to us from Netherlands. And she's been here for 4 and a half years. And she has a really great story that she's just submitted on exactly what she's going to tell you about. So I think this will be a really fun way to go, and I hope other people follow on this because again, there's really no reason why in this era of not traveling we can't sort of include the entire group. So I'm going to rotate to her. I'm going to make sure that she shows up. I got to put this there. Sorry, this is Yeah, you're good.

Thank you so much, Max, for this opportunity. So like Max just mentioned, I became very interested in understanding the biology that exists between, or the correlation between, the presence and abundance of these macrophages in tumors with exhausted T cells. And it actually has been shown up in many different single-cell RNA sequencing approaches, but nobody's really looked at the biology underlying these interactions between these two different cell types in the tumor microenvironment. So that is something that I've been working on in Max's lab and is also something that I want to continue doing in the future is to better understand the molecular mechanisms that these different T cells use to kind of co— like the co-evolution of these cells in the tumor microenvironments. So first, of course, I mentioned that there's a correlation. There seems to be a correlation between the abundance of macrophages and exhausted to T cells in the tumor microenvironments. But this data, of course, is correlative and doesn't say anything about the causality of the presence of these cell types. So what I did is I used a mouse melanoma model that Max introduced earlier of the B78 cherry ova mice. And I treated these mice with anti-CSF1R antibodies targeting the CSF1, CSF1R pathway, which is very important for macrophage survival and differentiation. So what we found, if you look in panel B and C, is that if we deplete mice with anti-CSF1R, we can significantly reduce the proportion of tumor-associated macrophages. But what was super interesting is that if we look at the expression of exhaustion markers PD-1, CD38, and TOX on antigen-presenting— sorry, antigen-specific T cells, OT1 T cells in the tumor microenvironment, you can see that the expression of these exhaustion markers is significantly decreased upon macrophage depletion. Moreover, we found that the ability of these cells to produce effector cytokines like TNF-alpha and interferon gamma is significantly higher if we deplete macrophages. So this data actually shows that there's a causal relationship between macrophages and exhausted T cells in the tumor microenvironment. So to get a better look at the molecular mechanisms underlying the crosstalk between macrophages and exhausted T cells, what we first did is transcriptional profiling of exhausted T cells isolated from the tumor microenvironment, and we did this at different time points. So we have in the graph here on the left, we have exhausted T cells that have been in the TME for 14 days and thereby reflecting a more terminally differentiated exhaustion state. And in blue, we have exhausted T cells that are early arrivers in the tumor and reflect a more progenitor-like exhausted state. And if we look at the expression of different genes in these exhausted T cells, we can see, like, in line with the literature, is that upon infiltration in the tumor, these cells lose the expression of markers that are characteristic for a progenitor-like or naive phenotype. And they start to highly upregulate markers that we know are associated with exhaustion, including PD-1, LAG-3, TIM-3, and TOX1. But what caught our attention is there was also a large set of genes that are mostly known for their involvement in myeloid biology, but they're highly expressed by these exhausted T cells. And we also performed epigenetic profiling, which showed that the chromatin accessibility near the transcription start sites of these myeloid genes is indeed— the chromatin is much more accessible in exhausted T cells compared to naive T cells. And I just want to mention that the ATAC-seq for this project was done in collaboration with the lab of Anshu Swasthpathy at Stanford. So based on this, we wondered what is the functional significance of the expression of these myeloid-related genes by exhausted T cells. So I adopted an in vitro transwell system of chemotaxis where I plated T cells with different activation states in the bottom well. And I quantified the recruitment of mouse monocytes over this transwell. And you can clearly see in the quantification here in the bottom right that exhausted T cells have a much greater ability to recruit monocytes compared to T cells T cells in any other activation state. Moreover, if we look at the phenotype of these monocytes, we can clearly see that specifically exhausted T cells kind of pushed the monocytes in their differentiation towards a more antigen-presenting phenotype. And this is in vitro. So we also wanted to look in vivo how that works. And we did this by using systemic depletion of CD4 and CD8 T cells. In tumor-bearing mice. And then if we look at the expression of these different markers for antigen presentation and myeloid phenotype on macrophages in these tumors, we can clearly see that specifically upon depletion of CD8 T cells, we lose the antigen-presenting capacity of or capability of macrophages in terms of expression of MHC class I, MHC class II, but also CD11c. And interestingly, we find that upon CD8 depletion, we have an increase increase in the pro-tumorigenic marker CD206 on macrophages. So these data— together these data show that exhausted T cells are specifically able to recruit monocytes to the tumor microenvironments and then kind of shape their differentiation trajectory favoring antigen presentation. So we were wondering what this means for the actual antigen presentation of macrophages to T cells, and we took a closer look at the direct interactions between TAMs and antigen-specific T cells. And we use lattice light sheet imaging to look at these interactions. And we actually found that TAMs form these unusual antigen-specific synapses with T cells that are ultra-stable. And this is also characterized by the clustering of the TCR, which you see here in green in this movie, on the T cell at the macrophage interaction site. But despite this synaptic interaction, we find that macrophages that do present OVA peptides do not support proliferation of T cells, while our positive control of CD103-positive dendritic cells, who are, Max mentioned, are known to be potent stimulators of T cells, they are able to induce proliferation and TEMs do not. So we adopted an in vitro co-culture of macrophages with T cells, and we actually found that macrophages prime T cells to become more exhausted. So we find that macrophages actually induce the expression of PD-1 on T cells and then fail to support proliferation. And we found that this induction of exhaustion programs in T cells is exacerbated in hypoxic conditions. And this is actually in line with a recent paper that was published by the lab of Greg Delgove earlier this year that hypoxia is required to establish a full-blown exhausted phenotype in T cells. So this data kind of shows that there's this causal circuit going on between macrophages presenting antigen and then inducing exhaustion in T cells. So next, I want to— oh, I want to introduce another postdoc in the lab, Dr. Kenneth Yu. And we were wondering, like, if this axis between macrophages and exhausted T cells, how it is controlled spatially in tumors. So now I'm going to hand it over to Ken because he developed a new technique called ZipSeq in which we can study just that.

Thank you, Kelly, for the introduction. Yeah, so a major focus on my work has basically been to kind of link the high wealth, you know, the big wealth of high-dimensional data that you can get from these single-cell omics kind of assays, like single-cell RNA-seq, with some kind of spatial context. So I think we can all appreciate that in tissues, especially tumors, there's significant spatial heterogeneity that you can see in this colorectal biopsy. And what we really want to do is to be able to understand what is going on in these different regions, as you can see, like region 1 versus 2 versus 3, where you might see more collagen deposition or more clustering of T cells or myeloid cells. So really what we're trying to do is translate basically spatial dimensions to transcriptomic dimensions. So to this end, we developed this technique, ZipSeq, which you can find in our Nature Methods publication. It really starts with just labeling cells in a tissue section with this antibody or lipid conjugated oligo, which will look very familiar to anybody who's done some kind of like cell hashing. And then on the equipment side, what you can see is, you know, we start with what is a pretty standard epifluorescent microscopy setup. You can image your sample with some excitation wavelength, but the real key here is that there's a spatially modulated illumination module here, this mosaic DMD, which basically you can create arbitrary patterns and illuminate your sample with those patterns, as you can see here with these 5 squares. And then the second part of the equation basically is these photocaging groups that prevent basically DNA from hybridizing. So by basically alternating a sequence of specific regional illuminations and additions of oligos, you can get this kind of printing of barcodes along basically arbitrary user-defined regions, as you can see here. So we have a cycle of these kinds of square illuminations, and then by cycling in 3 different fluorescently labeled oligos, you can see the formation of this kind of pattern. So this just represents one application of this approach towards kind of understanding the co-evolution of the CD8 T cell in the myeloid compartment in the tumor. So what you can see here is a section from a B78m cherry ova tumor. This is day 14 post-injection into a basically CD206 reporter mouse that Arjo Ray in our lab has been characterizing. And what you can see is based on the kind of the venous to cherry ratio, this green to red, and see that there's different regions naturally evolved. And we can ask the question, what's going on in these different regions with the ZIP-seq? And by doing so, what we can actually see is, first of all, like in the CD8 T cell compartment, you can see this movement kind of from this more stem-like or naive CD8 T cell to this exhausted CD8 T cell part of the UMAP. You can see as we move from out out to in. You see this red density here move, green, and then finally blue here for the inner region towards that more exhausted state. And what this really, I think, represents is as we're moving through physical space, we're actually also moving through transectomic space. And concurrent to that, you can also see in the myeloid compartment a similar movement. You can see there's a concentration of red here, and then as we move mid, and then finally inner, and see the concentration of cell density moves towards this end of the UMAP towards this MS4A7 and proliferative population. And you can also quantify this using different kind of signature scores such as here for CD8 T cell exhaustion and the CD8 T cell glycolytic score. And you can see that gradually increases as we move from outer to inner regions. And with that, I'll pass it back to Max.

I first and foremost want to thank Ken and Kelly for trying this experiment today. I hope it's useful to those at the other end. And I also encourage you to, if you are in the position of looking for folks, this hopefully gives you an idea about what both of these folks are doing in the lab. And some of their really unique approaches, both in technology and, and I think also in, in questions to look into this issue of how multicellular biology really works. So I want to thank them extensively for, for trying this experiment together of co-presenting, and hopefully this is something that we can continue to do as a community. And then I really want to thank the immunology community as a whole. I think we have one of the best biology communities. And this slide, you know, apart from Global Immunotox could be quite a lot longer if we thanked everybody for all the various different things that they've given to our projects, and hopefully vice versa. I do also want to call out briefly Alexei for his work on the archetypal model. He's going to join the faculty here at UCSF and a few others, Kevin, who's down and now at the Hutch for some pioneering work in this area. Vincent Chan and Bushra Samad, who really drove a lot of the immunoprofiler work over the years and, and really a huge component of our community. So thank you everyone for, for listening. I hope this was a good experiment and, um, and maybe others will want to build on it. So thank you very much.

Thank you so much, Max. It was a terrific experiment and it worked. Uh, thank you, Kelly and Ken. That Great. And congratulations on your research. Yeah, bravo. Yes. So as always, questions are via Twitter. And so if you missed the explanation upfront, let me share again that slide just in case. And so people can look for the tweet in the Global Immuno Talks account that says, ask questions for Dr. Max Trummel here, and reply to that tweet with your questions, the hashtag Global Immunotalk, and Max will use his personal account to answer the questions via Twitter. And maybe Ken and Kelly could help with that. And so please go ahead, ask the questions. If you watch this in YouTube, you can ask questions in an asynchronous manner. And so thank you again so much, Max, Ken, and Kelly, for sharing your research in this forum. And so everyone, I hope you can join again for the next week Global Immunotalk by Dr. Akiko Iwasaki. Okay, thank you. Bye guys, see you.