2026-04-18-pm-fibroblast-dynamics-heterogeneity-tme

Date: 2026-04-18 Primary topic: single-cell-spatial-omics Cross-referenced from: single-cell-spatial-omics Word count: ~13,917

Auto-generated captions

This transcript is Vimeo auto-generated English. Expect misheard names and technical terms. Use the ripgrep/search approach for exact quotes.

All right. Welcome to our educational session on fibroblast dynamics, how heterogeneity shapes the tumor microenvironment. I'm going to be talking a lot about fibroblasts. And you have this QR code up, so this is how you can go online and type in questions. We also have two microphones. You are more than welcome to come up and ask questions at the microphone. But if you'd like to type your question online, that's the QR code, and then it will pop up on the computer here so we can read out your question. All right. Great. All right. Do you want to come up? No. Okay. All right. My name is Stephanie Duggan. I'm an associate professor at Dana-Farber Cancer Institute and Harvard Medical School, and we'll be talking to you today about fibroblasts and anti-tumor immunity. These are my disclosures. All right. So when we think about fibroblasts, this is one of the most fundamental cell types. It's part of every normal healthy tissue. These are the cells that are making your connective tissue, performing lots of structural functions, and although this session is about fibroblasts and cancer, I do want to iterate that these are normal cells that are very important for homeostasis. And they're not just one kind. And so if you look in various tissues, the fibroblasts are poised for separate functions in those tissues. All of them make extracellular matrix, some more than others. But they also have a lot of metabolic functions. They can secrete a lot of proteins. These are cells that are really good at protein biosynthesis that's important for making extracellular matrix, for remodeling extracellular matrix. But they can also do things like secrete growth factors and cytokines that can call in immune cells as needed or restrain them as needed. And so this is a picture form, but it actually comes from a lot of transcriptional profiling. This is a really nice study from Shannon Turley's group, where they looked at healthy tissues, and you can see the tissues here. And a lot of them were public data sets where they subsetted out fibroblasts and just asked, "What does the fibroblasts look like in lung, in bone, in intestine?" And they're slightly different. And so you see that there's some heterogeneity. They're all capable of making extracellular matrix, but they have slightly different gene transcriptions that make the normal functions that are needed in those tissues. A huge function, of course, is to produce the extracellular matrix. We think of this in terms of collagens, but there's also a lot of proteoglycans in there. Huge amount of ECM components. This is an incredibly diverse set of proteins that can be cross-linked in long fibers, that can be cross-linked in a more chaotic way. And a lot of the remodeling of the extracellular matrix is also performed by fibroblasts. Macrophages can contribute, too. But in terms of both building and breaking down collagen, making matrix metalloproteases to make sure that the extracellular matrix is appropriately remodeled. This is a cartoon, but also some really nice work. This study was led by Richard Hynes' lab, asking what proteins are there? And I think proteomics has really led the way in this, where you can take the extracellular matrix from a tissue, you can decellularize it, and ask, "Well, what is left?" And it's a surprisingly rich set of proteins, some of which are the core matrisome that are detected across tissues, but some of them are also more tissue-dependent. And asking what does a tissue need in its extracellular matrix can be very revealing about the function of the fibroblasts that secreted it in the first place, and that tissue in particular. And so we talked a bit about healthy homeostasis, but of course, tissues can become inflamed. And one of the huge important roles of the fibroblasts is to sense tissue damage and to help mediate the wound healing and remodeling of the extracellular matrix. And so the same subset of fibroblasts that we just surveyed across healthy tissues also has a counterpart in diseased tissues that look slightly different, but are oftentimes recruiting in innate inflammatory cells, so they make a lot of chemokines. These are your CCL and CXCL family of proteins, which form gradients that are deposited in the extracellular matrix that allow for things like monocytes and neutrophils to come in. Monocytes, of course, can differentiate into macrophages and help with that wound repair and tissue healing. All right. And this is going on in cancer. It's going on at a much higher rate. And so cancer was originally described as a wound that doesn't heal. And so many of these activation of fibroblast pathways, of course, are also happening in cancer. They're happening chronically. They're happening for a long time. And many of the processes that we think about in terms of cancer directly act to activate fibroblasts. And so certainly, these are contact-dependent signalsWith the cancer cells and extracellular matrix, DNA damage, factors like TGF-beta, which we're going to talk a lot more about, and stress induced by hypoxia or reactive oxygen species, can certainly all activate these sort of constitutive fibroblasts that are present in whatever tissue the cancer is arising in and turn them into this so-called CAF. This cancer-associated fibroblasts. And so we've appreciated these for a long time, that they're in all cancers. They seem to be really bad or at least complicated and sort of appreciating that they have an origin with the normal tissue is an important point to note. And so although we have CAFs in every cancer type, we are today going to use as an example pancreatic cancer. And so our speakers today all are sort of active in the pancreatic cancer field. And the reason that pancreatic cancer has kind of led the way in thinking about fibroblasts is because it is highly desmoplastic. I have a cartoon that I drew here, but you can also see from the H&E that you have these tiny nests of cancer cells that are the dark purple, that are kind of in this giant sea of extracellular matrix. And so you have these activated fibroblasts making this huge amount of stroma that can, by its sheer biomass, outweigh the actual cancer cells themselves. And so this has been appreciated for a very long time. And so how can we therapeutically intervene? And I would say that we're going to talk about all of these strategies to either deplete fibroblasts or the extracellular matrix, and this kind of makes sense. You look at something like pancreatic cancer, it has a very horrible prognosis. It's got all of this desmoplasia. And so the obvious thought is, okay, well, maybe that stroma is contributing to progression of the tumor, and so what can we do about it? And so these are strategies that have been tried. We'll talk about Smoothened inhibitors, pegylated hyaluronidase to enzymatically digest the extracellular matrix, blockade of two different cytokines, and CAR T cells. All right. And so what is Smoothened? This is a cartoon from Sunil Hingorani's lab, who's worked on this quite a bit, where the tumor cells themselves can make these Hedgehog ligands that will bind to Patched1, which is a natural inhibitor of the Smoothened protein, which can then prevent it from signaling. And if you block this pathway, then Smoothened will naturally activate these Gli transcription factors, and this can lead to proliferation and activation of these fibroblasts. And so there was a seminal study published a while ago now by Ken Olive's group, when he was in Tev Tuveson's lab, looking at, okay, well, maybe can we use these Smoothened inhibitors to try to kind of debulk this extracellular matrix with the observation, as you can see in these pictures, that if you have a pancreatic tumor growing in a spontaneous so-called KPC gem mouse, it's this dark spot in the middle. And what is this dark spot? Well, this is looking at sort of a fluorescent gemcitabine trying to get in, and you can see that the drug is able to get everywhere in the pancreas but not the tumor. Right? And so it was sort of obvious, like okay, it's very dense. You try to cut into this, it's very hard, like a rock. Maybe if we got rid of that extracellular matrix by sort of inhibiting these fibroblasts, could we get more drug in? And so gemcitabine is one of the chemotherapies used in pancreas cancer, and you can see in the mouse model that combination of the Smoothened inhibitor with gemcitabine is able to extend survival in an incredibly difficult mouse model. So this had some unfortunate results in humans, though. And I do want to kind of show you the mouse and the human data. And so the initial trial that was started right on the heels of this pre-clinical study was using a Smoothened inhibitor, that's listed here, where the trial had to stop early. This was not actually published, but it was released by the company, that the Smoothened inhibitor plus gemcitabine actually made the tumors progress faster than with chemotherapy alone. A second trial was done where you can see that there was not really much of a difference, but I think that this early press release really was a huge wake-up call to the field saying, "Wow, we really don't understand fibroblasts." We show that we can deplete them or sort of quiesce them, and this idea that we just get more drug in ended up making the patient's survival worse. Similarly, there have been strategies to try to deplete the extracellular matrix, and so pegylated hyaluronidase, this is an enzyme that kind of cleaves up hyaluronic acid, sort of pictured here in yellow. And if you look by immunohistochemistry, you can see that there's a range of hyaluronic acid in pancreatic cancer in humans, some of which can be quite, quite high. And so if you are able to digest this enzymatically, then maybe you'll sort of dedensify that extracellular matrix. The blood vessels are going to be less compressed, and you'll be able to get more drug in. And so the initial Phase II trial with gemcitabine and paclitaxel, listed here as AG, was promising. Although in Phase III, you can see that there was really no effect of using the pegylated hyaluronidase to increase chemotherapy delivery to the tumor. And very strikingly, if you just change the chemotherapy regimenAnd this is now FOLFIRINOX, which is used very frequently in first-line pancreas cancer. You can see that these labels are not mixed up, that in fact, depleting the extracellular matrix with pegylated hyaluronidase worsens overall survival. And so this again sort of had a moment of reckoning for the field where we thought, well, maybe there's something about this, that there's fibroblast heterogeneity. Are there some fibroblasts that are actually good, that are sort of tumor restraining, and maybe we need to understand this better. And so this led to some initial discoveries on sort of bucketing fibroblasts into multiple subsets. And one of the first ways of thinking about this was looking at transcriptional profiling. And you could see, this is work from Julia Biffi when she was with Dave Tuveson, that the fibroblasts that are next to the cancer cells, that are sort of surrounding these little cancer cell nests, receive a lot of TGF-beta signaling, and they become these sort of myofibroblastic-like CAFs or myCAFs. These make a lot of extracellular matrix. They have direct contact with the tumor cells. Whereas the fibroblasts that are sort of further away are not getting as much TGF-beta signaling, and they can see more sort of IL-1, as it induces this phenotype. And they're making some extracellular matrix, but a lot of cytokines and chemokines that are interacting with the immune system. So IL-6 is hugely produced by these fibroblasts, as is the Th2 alarmin IL-33, as well as complement proteins. And so these were termed inflammatory fibroblasts or iCAFs, thinking there would be an interface with the immune system. And when we said, "Oh, okay, look, there are two kinds of fibroblasts. Maybe there's the good kind and the bad kind." And so this has now been tried in humans. Okay, we know that there are cytokine signals that prevent or preclude development of these sort of iCAFs and myCAFs. And so I'm going to show you data from a blockade of TGF-beta, which should deplete the myCAFs, not the iCAFs, and IL-1 beta blockade, which should have the converse effect. And so this is the mouse data. This was done in my lab, where we looked at TGF-beta blockade in a mouse model of pancreas cancer, showing that it nicely prevents signaling through phospho-SMAD2. And if you look at fibroblasts, you can see that indeed, the sort of alpha-SMA positive myCAFs were depleted. That was statistically significant. So the drug sort of does the thing we thought it was supposed to do. But the trials in humans have been very disappointing. In fact, kind of in the same vein of, wow, this backfired. This is showing the published data, which is primarily lung cancer and colorectal cancer, but the trial of TGF-beta blockade with gemcitabine plus paclitaxel in pancreas cancer was actually stopped in phase three for a worsening of overall survival. So this is not the path to go down. Similarly, we've looked at IL-1 beta blockade. Most of the data that's firmly established is in lung cancer, where there was a very promising signal that sort of early blockade of IL-1 beta could prevent the formation of lung cancer. I think that's still true, but if you have patients who have established lung cancer, these were three very disappointingly negative trials where it just didn't do anything. And this is data we were involved in, in pancreas cancer, where although this waterfall plot looks a little better, it really doesn't beat out chemotherapy alone, suggesting that there was not a big effect of IL-1 beta blockade in pancreas cancer. So perhaps a little more promisingly, I don't want to leave you with such a depressing historical run through our clinical trial history. CAR T cells have been developed that target fibroblast activating proteins of FAP, which is expressed on many fibroblasts. This is the mouse model showing that, in fact, you can deplete FAP-positive fibroblasts. Alpha-SMA goes down, so the CAR T cells have their mechanism of action, and they can control pancreas tumors quite well, sort of maintaining stable disease for a long time. And I think encouragingly, the CAR T cell doesn't even necessarily have to deplete the fibroblasts. And so this is a very interesting study where there is a CAR T that was developed to target a fibronectin splice variant that is unique to the extracellular matrix of many cancers, including pancreas cancer. In this case, it's actually a melanoma example, where you ask yourself, why would you make a CAR T cell that doesn't even kill a cell, right? It's targeting a protein in the extracellular matrix. That seems like a ridiculously futile effort. And yet you can see that the tumors are controlled quite well in these mice. And when we looked by immunohistochemistry, you can see that you get this massive influx of activated T cells, a reopening of CD31 here, so these are nicely patented blood vessels. And the effect is likely due to just localizing the CAR T as a major source of interferon gamma that reprograms that whole tumor microenvironment. And it doesn't really matter that it doesn't actually kill any cells.All right. So I've sort of left you with, well, maybe there's some fibroblast heterogeneity. Are there any clearly good or bad fibroblasts? And we've talked most about how to bucket them by transcriptional states, where you have kind of this iCAF, myCAF. This is really only one way to think about them. And I would say that my colleagues here, Klaus and Mara, have contributed a lot to understanding, are there nuances of how you can define these cells better? This is a figure from Mara's recent review, which goes into all the different ways of looking at cell of origin, surface markers, with the hope that maybe if we just figured out who were the really, really good fibroblasts, that maybe we could keep those and reprogram the other ones. And so to that end, this is a study looking at sort of epidemiologically, whether there was a signal for overall survival by subdividing cancer-associated fibroblasts and sort of promoting CAFs, these pro-CAFs or restraining CAFs, which mirror similarly the iCAF, myCAF designation, but are slightly differently based. And this is kind of the biggest effect you can see in shifting those overall survival curves, which, I mean, all of you are looking at this as P value significant, but pancreas cancer is really bad, and so what are we actually going to succeed in if all we're doing is shifting the fibroblast subsets? So is this the best we can hope from it? I do want to mention briefly a sort of minor population of fibroblasts that came along with the sort of iCAF, myCAF designation that there was appreciated to be a smaller subset, more rare, that express MHC class II. I love this. I'm primarily an immunologist, and so this is how cells present antigen to CD4 T cells, your helper T cells, and so what is it doing on a fibroblast? And so this, I think to understand it fully, we need to go back to sort of what are these fibroblasts doing in the normal pancreas. And this is a really nice study looking in diabetes models, and actually in humans with pre-diabetes, where you can see that you have the islets here, right? And these islets are surrounded by these class II positive fibroblasts. And what are they doing? They're there so that you can present antigen to a CD4 T cell, but turn it into a Treg, a regulatory T cell. And this way, you can prevent autoimmune diabetes and also pancreatitis. So I think this is a very important natural process by which the pancreas promotes immune tolerance rather than destruction. So summarized in this title here, Tregs provide a critical buffer against diabetes and pancreatitis. But of course, if you have the misfortune to develop pancreatic cancer, you're doing so kind of on this background of a lot of Tregs, which can inhibit anti-tumor immunity. Whoops. All right. So where do we go? Are fibroblasts sort of cause or consequence in cancer? I think we've seen from trial after trial that just depleting fibroblasts is maybe not the way to go. So I've listed some kind of fibroblast-adjacent strategies, which involve sort of either blocking downstream signaling in the tumor cells. It's known that the tumor sort of relies on focal adhesion kinase for support from those fibroblasts, and so these trials are actually looking pretty promising. And then kind of going after the immune profiling angle, can we deplete Tregs? We actually have these new Fc enhanced CTLA-4 antibodies that do this pretty well. So this I'm actually really excited about. And can we sort of shift that cytokine balance by either delivery of Th1 polarizing cytokines, or there's some hint that CD40 agonism also acts by massively reprogramming the extracellular matrix. All right. A few take home messages here that I'll let you read, and open it up to questions. All right. Thank you. First question. Is this on? Can you hear me? Yes. Okay, great. Great talk, great overview of all the clinical space and pre-clinical space. Probably not an unoriginal question, but thinking about all the experiments we did pre-clinically, and then trying to translate them into clinics, is this us not understanding the therapy and how it works, or is this a limitation of the mouse models that we've been trying in, and really it's just not reflecting the clinical space that we don't really understand in humans? Yeah, I think that's a great question. I mean, how do we model this in order so that our models give us the correct answer that will be what we see in trials? And I don't think this is just a mouse-human difference. I think a lot of this is a time course difference. And so fibroblasts are very plastic. They can reprogram incredibly well. And so if it's one kind of fibroblast today, it doesn't mean it can't shift its phenotype. And so I think in the short course of the mouse models, we don't necessarily capture that. How are we going to do this better? There's certainly human sort of organoid fibroblast co-culture data that's coming out. Does that really tell you long term? Probably not. I think at this point, we should just look at what the trial data that we have done are telling us and say we had enough patients go through these efforts that perhapsWe need to have just learned our lesson by now, and say, what are the downstream effects of fibroblasts? And are any of those individually targetable that might kind of be more selective way? Yeah. Edie. I want to put you on the spot a little bit, because I learned something from you that I think that most people would like to learn as well, I hope. You enlightened me to understand what the reticular fibroblastic cells do in lymph nodes. And I wonder whether you think that the local fibroblasts in the pancreas initially, the ones that are making sure that the precancerous lesions don't go to cancer in most of our cases, whether those are similar and whether that's what we're calling maybe the AP-CAFs. Yes. I mostly focused on the CAFs in the tumor microenvironment. Of course, fibroblasts are everywhere. They play a critical role in the lymph node. But I think when we're talking about Class 2 positive CAFs, which I do try to use that term, AP-CAFs or antigen-presenting CAFs has also been thrown around. But that implies you're priming a T cell to go do something, right? And the fibroblasts universally never have co-stimulation, so they don't have that second signal that tells the T cell, "Oh, this is an infection. You should go fight it." And so it ends up polarizing towards these regulatory T cells, which is an important and critical role to say, okay, if it ends up on Class 2 on a fibroblast, this is in quiescence. You should not make an immune response. You should tolerize. Yeah. Thank you. You just answered my question. It's about the AP-CAF. So I have another question. It's about the extracellular matrix, because we think more in biophysical properties. But there are more data that are actively involved in immune response. The proteoglycans, they are activating different kind of receptors. So what do you think about that ECM, like immune component that is giving maybe immunosuppressive effects or no? So your question is about the cytokines or chemokines that are in the extracellular matrix? No. Or I don't think I quite heard you. Just the ECM itself acting like active immune effector in the T cells, the microphages also. Because there are a lot of proteoglycans that are agonists of different, TLR4 and the other. So ECM is now produced by myocaf, but also have active component in the microenvironment, right? Yes. So certainly if you're talking about immune cells crawling around in a tissue, they do so on the extracellular matrix. And so CD44, which is on activated immune cells, does bind hyaluronic acid on extracellular matrix. So it's a critical part of how cells crawl through the tissues. And as they're doing that, they of course, can see whatever else is deposited in the matrix. And so I really love the proteomic analyses to ask, what proteins are there? Which you can miss by transcriptional profiling, because it doesn't necessarily correlate with the RNA of the cells you're identifying. Thank you. When you're studying fibroblast-mediated signaling and immune exclusion, specifically, let's say in PDAC, are there specific spatial scales where there might be the most strongest effect size in signaling? Considering contact dependent versus maybe global tissue architecture. Are there two? That is a great question, and certainly, I think we're still understanding how far things diffuse. So a lot of the contact dependent mechanisms are obviously it has to be the cell touching it. And so you get a lot of integrin signaling happening between fibroblasts and cancer cells, fibroblasts and immune cells. But the secreted molecules can diffuse. Now, they don't diffuse like it's empty space because it's a lot of dense pressure there. And so, they sort of deposit in these gradients with some cytokines being very localized in growth factors, and some kind of extending further out. And so, really nice work on looking at cells that are receiving the signal, like through phosphorylated transcription factors, and can you ask sort of from the cell that produces it, how many cell bodies away can you still see the response to that signal? And I think it does differ for different molecules. But it's a great area of research. Maybe we'll take one more question, and then do we have any online? Okay, great. Okay. Thank you. Great presentation. And my question is regarding the type of cancers you have, because you started the presentation saying that pancreatic cancer is highly desmoplastic. But does the exocrine pancreas, if you have a neuroendocrine pancreas tumor, it's not so desmoplastic. So the different type of cancer drive different responses in the same fibroblast. Have you done something like that, seeing which signals are given by one cell to respect to another, and not treating the fibroblast as a unique species that respond always the same? Yes, that is a great question. I'm glad you bring up pancreatic neuroendocrine tumors, because again, it tells you same organ, very, very different histology, very different results in sort of activation of fibroblasts. And so certainly there is crosstalk even with the neoplastic cells as well. And so a lot of really good work done, I'd say senescent cancer cells and sort of how they produce a lot of cytokines that could activate fibroblasts and give you more of the sort of iCAF phenotype. And the neuroendocrine tumors also make a lot of different factors. I don't know if my panel wants to contribute more towards PNETs. But you see this giant inflammatory reaction, and yet, the tumor's progressing, right? So not all inflammation is good. Just like not all fibroblasts are good, but they could certainly act very differently. Thank you. Okay. All right. Maybe last question. Yeah. Thanks. I have a question. Since you say that depletion might not be the right way to go, what do you think about reprogramming and programming, let's say, from yeah protumoral to anti-tumor fibroblasts? Yes. So I think this is a very active area of research. Can we reprogram a fibroblast? We know they're very plastic, so is there some kind of signal we can give therapeutically that will keep them in whatever the good state is? I think we want to be really sure we know what that good state is, first, and how well or how permanent that reprogramming is. I think you're also stuck with, as soon as you stop using the drug, you can get it to revert. So, similar ideas have been proposed for macrophage reprogramming as well, and I think it goes down a slippery slope of, is this really the effect you want, and how long can you sustain that? So I think there's a role for this fibroblast reprogramming, but I think it's in combination with other therapies that are either directly killing tumor cells or inducing an immune response that can make that sort of transient nature of the fibroblast reprogramming permanent. Thanks. Yeah. Okay. Thank you. I will introduce my colleague, Mara Sherman. All right. Well, thank you, chair extraordinaire, Stephanie, for the opportunity to participate in this session and for starting us off so nicely. So my lab studies cellular hierarchies in the tumor microenvironment, and our goal really is to understand how this immune-suppressive, chemo-resistant microenvironment gets established, but also how it might be dismantled for therapeutic benefit. No disclosures. So in this vein, we spend a lot of time thinking about mesenchymal elements such as fibroblasts. And really our motivation is evidence from context of development, health, and inflammation that fibroblasts provide critical instructive cues that regulate epithelial cell state and tissue function. So for example, in the embryo, the splanchnic mesenchyme critically instructs alveolar airway epithelium formation in the developing lung, and that epithelial mesenchymal crosstalk remains critical for lung function in early life. In healthy adult tissue, there are positionally defined populations of fibroblasts that secrete distinct developmental cues that program epithelial heterogeneity in a way that's critical for tissue function, and this is really nicely exemplified along the length of the intestinal crypt. And in the context of inflammation, there's a beautiful recent example from the mighty reindeer showing that anatomically distinct populations of fibroblasts can respond to similar wounding insults with totally different secretomes, and therefore different impacts on the epithelial compartment. So specifically, a wounding insult to a regenerative organ, in this case, specifically the velvet or the antlers, causes those fibroblasts to secrete regenerative factors, stem cell niche cues that ultimately promote epithelial cell proliferation, regeneration, and tissue repair. Whereas a really comparable wounding insult on the back skin causes fibroblasts really to enter an inflammatory state and ultimately result in epithelial cell death and a scarring reaction. And importantly, really comparable fibroblast transcriptional states are found in human skin in the fetus versus the adult. And so, of course, we know that there are these intrinsic differences among these positionally or anatomically defined fibroblast populations, and we're motivated to understand how about this epithelial mesenchymal crosstalk and fibroblast heterogeneity in the context of cancer. So there has long been evidence that fibroblasts can really have heterogeneous functional impacts on cancer cell growth. Almost 60 years ago now, Michael Stoker's lab demonstrated that normal static fibroblasts could suppress the growth of polyoma transformed cells. This was really an early example that components of normal tissue architecture could serve as barriers to tumor genesis. About 20 years later, pioneering work from Mina Bissell's lab in the chick embryo comparing tumor progression in an uninjured versus a clipped or wounded chick wing showed that a local fibroinflammatory or wounding reaction was permissive to Rous sarcoma virus-mediated tumor formation. This was really a seminal example of a principle that's been shown many times since, that a fibroinflammatory microenvironment can promote tumor genesis. So we study these networks in the context of pancreatic cancer really because, as Stephanie so nicely described, these really are the poster child of a fibroinflammatory malignancy. So pancreatic cancer certainly exhibits heterogeneity from patient to patient, but these tumors generally share two common features. One is activating mutations in KRAS, something you'll no doubt hear very much about at this meeting, and the second is that during disease progression, the epithelial compartment coevolves with this progressively dense desmoplastic stroma. And the stroma, of course, includes abundant immune cells and extracellular matrix components and is really set up by a prominent population of cancer-associated fibroblasts. So it stands to reason that during this dynamic coevolution of epithelium and stroma, these fibroblastic cells may very well have totally different molecular profiles and functional consequences over time with respect to tumor progression. So as a field, we've really learned a great deal about factors or functions that fibroblasts gain during tumor progression. As a community, we've seen really nice studies showing that a CAF can produce immune suppressive cytokines, chemokines, matrix components, and other factors that can promote tumor genesis. And in enacting this perturbed wound-healing response in the context of cancer, they can also engage functions that are tumor restraining, including engaging Hedgehog signaling, as Stephanie talked about earlier. And there's also some really nice evidence that collagen I production can have some tumor-restraining consequences as well.But what we've really learned less about as a community is factors that are expressed in normal tissue mesenchyme and that are lost over the course of tumorigenesis, which may potentially be barriers to tumor progression. So really our motivation to think about this comes from these pioneering studies from the labs of Laura Wood and Marina Pawlikowska-Magliano that sort of took our collective breath away in reporting that most adults, including donors with no evidence of cancer, walk around with pre-invasive lesions called PANNs in our pancreas. So just as many of us walk around with BRAF mutant moles on our skin, many of us walk around with KRAS mutant lesions in our pancreas. The good news is that these lesions vastly outnumber cases of invasive pancreatic cancer. But the frequency of these lesions, this new normal or newly appreciated normal, really compels us to ask why? What are the mechanisms active in many of us that are keeping these lesions from progressing to invasive cancer? And can we learn from those mechanisms about biomarkers for those at risk of progression and/or for preventative strategies? So I think it's very likely that tissue-level tumor suppression takes three forms. One, of course, is epithelial cell intrinsic mechanisms, expression and activity of p53 as a key example. Surely, there are roles of the immune microenvironment in keeping those lesions from progressing. But what I'll talk about today are the non-immune microenvironmental mechanisms that are at play in the pancreas. And really, these tumor suppression mechanisms likely vary from individual to individual, and probably some mathematical modeling is needed to infer temporal dynamics among the three. So we really wanted to look in the mesenchyme to see if it could teach us about any new obligate barriers to tumorigenesis. And this was a project led by a wonderful graduate student in the lab, Catherine Onyate, together with her lab mates, Chet and Shahini. So before I get into the data, I just want to talk about what's in the pancreatic mesenchyme. There's a kind of cell in both the pancreas and the liver called a stellate cell. These are lipid-storing cells that can sense and respond to the presence of cancer or a wound-healing insult and give rise to activated fibroblasts. It's not the whole story, though, and we know this just based on how we isolate these cells in the lab. So if you want to isolate stellate cells from normal, healthy pancreas or liver tissue, you make a single cell suspension, you run it through a density gradient, and because these stellate cells are such robust lipid-storing cells, they'll stay trapped at the interface of this density gradient. What you see sinking into this dense layer on the bottom is a bunch of these conventional PDGF or alpha-positive tissue-resident fibroblasts, which are also sitting in the tissue, also poised to sense and respond to the presence of cancer and give rise to CAFs. Just showing you a little bit of evidence that these cells are different. Right after you plate these stellate cells in culture, you can see the cytoplasmic lipid droplets, even evident by bright fields. Within a couple of days, these cells will sort of spontaneously undergo activation to sort of a myofibroblast-like morphology. But within those lipid droplets, they store esterified vitamin As, retinol esters, and they have this characteristic blue-green autofluorescence that you can even see in your tissue culture microscope using the DAPI channel. And so, these are different populations that are poised to give rise to CAFs. I'm really not going to talk about differences between these populations in normal tissues or tumors today. But I did just want to make one point, given that reading abstracts for this meeting, it's just so clear that this theme has emerged with respect to the importance of spatial niches in regulating cell function, molecular profiles, and tumor outcomes. And I think this idea that there are different mesenchymal lineages that live in our pancreas and give rise to CAFs is really consistent with that. So specifically, stellate cells in liver and pancreas are in the perivascular niche, whereas these conventional fibroblasts generally accumulate in peri-ductal and peri-acinar spaces. Okay. So in thinking about how to study these, there's a really nice Cre-based mouse model, the FEBP4 Cre model, that's been used to study adipocytes in vivo. So needless to say, this is not a stellate cell-specific Cre. But what's nice is that it robustly induces recombination in sort of professional fat-storing cells in the body. And this includes stellate cells in the pancreas and liver, but does not include conventional fibroblasts. So it's a way to kind of pick apart the two. This can be combined with lineage tracing tools, in this case, the Rosa MTMG allele. So we can indelibly label stellate cells and the CAFs derived from them in green. And so in this study, we really wanted to understand the temporal dynamics of mesenchymal activation during pancreatic tumorigenesis and really understand what gets lost. And in this case, instead of looking at all the fibroblasts in the pancreas, all the fibroblasts in the tumor, we thought maybe we'll learn something new if we look at a more focused population and track it over time. So we're going to focus on this lipid-storing origin and the CAFs derived from them in this example. Okay, so what we did is we took this reporter model. We took stellate cells from a healthy pancreas tissue, CAFs derived from them from tumors, and did single-cell RNA-seq on these GFP-positive cells. So just as sort of a sanity check, we looked at these cells in both normal tissues on the left and tumors on the right. We see that they sort of express all the classic mesenchymal markers. They pervasively express vimentin. They're along this continuum from PI16 positivity to collagen 15A1 positivity. This is the trajectory of universal fibroblasts that was reported in the study from Shannon Turley and Matt Buechler that Stephanie mentioned earlier. And most of them express one of the universal fibroblast markers, collagen 4A1 or HSPG2 or perlecan. I did mention that these are in the perivascular niche. In the liver, stellate cells have been called the pericytes of the liver, though I'll just mention they seem to lack robust expression of conventional pericyte markers.Okay, so then what are the differences across tissue states? So in the context of tumorigenesis, these cells gain expression of the expected players, immune suppressive cytokines, CCL2, IL6, and IL33, just as a few examples. So we looked at what gets lost, and we really wanted to look at factors with the potential to function in a paracrine manner and influence other cell states in the tissue. And a gene that really leapt out from this analysis was a gene called Kit ligand or stem cell factor, which was clearly expressed in normal pancreas tissue but really seemed to be lost in tumors. And this is consistent with a pseudotime analysis that we performed on these cells. Okay, so what is Kit ligand? Generally, Kit ligand is a paracrine factor that is expressed by so-called sender cells in either a membrane-associated or a secreted form. It signals through a receptor c-Kit on the surface of receiver cells and generally is thought to activate mitogenic processes and pathways, Ras MAP kinase signaling, Jak-Stat, and PI3 kinase/AKT signaling. There is certainly a precedent for thinking about Kit ligand in the context of the mesenchyme. For example, there was a really nice study some time ago showing that there are leptin receptor-positive mesenchymal cells that indeed express Kit ligand, and these are specifically hepatic stellate cells in the fetal liver. And this Kit ligand expression by hepatic stellate cells is actually critical for normal hematopoiesis in the fetal liver at the appropriate stage of embryogenesis. Now, this idea that leptin receptor-positive mesenchyme expresses Kit ligand is something that's been reported in other anatomic contexts, including the bone marrow stroma. So we asked, okay, are we looking at actually a leptin receptor-positive component of the pancreas? And so what we did is we looked at quiescent stellate cells taken fresh from mouse pancreas, saw that these actually express really high levels of the leptin receptor. But when these cells are just spontaneously activated to a fibroblastic state in culture, they lose the expression of this leptin receptor, suggesting that indeed, this is part of this kind of anatomically conserved population of leptin receptor-positive mesenchymal cells that for some reason express Kit ligand. Okay. So we were motivated to ask what is the functional significance of this factor in normal pancreas tissue? But first, we really wanted to validate that it was expressed in mice but also in human pancreas. So we did a number of experiments doing RNA in situ hybridization. Kit ligands, like a number of other secreted factors, just cannot be reliably detected by conventional immunohistochemistry using commercially available antibodies. So we did a series of RNA fish experiments showing indeed that the mesenchymal component of mouse pancreas expresses Kit ligand. And importantly, we used floxed alleles, this is a model I'll get back to later, to show that our RNA fish experimental approach was specific. So we then went on to look at human pancreas tissues. We costained for vimentin to mark mesenchyme and Kit ligand, and we indeed saw expression of Kit ligand in human pancreas tissue. And importantly, though there is, of course, an exuberant mesenchymal population in human pancreatic cancer, we saw a significant reduction of Kit ligand-positive mesenchyme in invasive cancer. And this was, of course, conserved in mice. Okay, so we wanted to really confirm this further in the human pancreas, but also really assess the relevance of this mesenchymal population we were studying in mice. And so we leveraged a protocol we're employing where we're taking pieces of benign pancreas tissue from humans. These are people who come into Memorial Hospital for resection of a primary tumor, so I can't call this normal pancreas tissue. This is benign adjacent. That's a meaningful caveat there. But we've worked out a pipeline similar to what we do in mice, where we can take this benign pancreas tissue, subject it to density centrifugation. We can isolate fibroblasts. We can isolate stellate cells and study those on their own, or either co-culture them with organoids or culture them on collagen-coated plates together with cancer cell-derived secreted factors in the form of conditioned media to make CAFs from them. And so we see that the stellate cells look like stellate cells, kind of similar to what we see in mice with these abundant cytoplasmic lipid droplets. We see this in every patient that we've analyzed so far. And when you culture them, both the stellate cells and the fibroblasts very readily give rise to really what looks like a conventional myofibroblast or a CAF.And so we looked in these cells to see if they express Kit ligand, and the good news is that they do. When we take these fresh out of human pancreas tissue, they express this marker. But when we activate them, either with conditioned media or just with recombinant TGF-beta, we see a significant reduction of this factor. So it seemed like this was a thing. So we then wanted to ask, okay, we've looked in normal, we've looked in cancer. What happens in between? And so we wanted to look in the context of pannens. And so we went back into our mouse models, and we used a dual recombinase genetically engineered mouse model that allows us for fate mapping of stromal evolution in the context of stepwise tumorigenesis. So this is something that's missed when we do orthotopic transplantation of invasive cancer cells. But these genetically engineered mouse models go through these pre-invasive stages that many of us harbor. And so when we looked in pannen lesions from pancreas tissue, we saw that really this GFP positive mesenchymal seems to be sort of at a steady state of abundance in pre-invasive lesions versus tumors. And when we looked at Kit ligand, we saw really negligible signal in regions of invasive cancer. But we saw some retention in the GFP positive stroma in the context of these pre-invasive lesions. And this is something we really want to look into further. The last piece before we went forward with functional studies was, is its partner there? Does it have someone to talk to? And so we looked at expression of c-Kit in normal pancreas tissue, and we did see expression of both c-Kit itself, as well as the activated phosphorylated form, in both human and mouse pancreas. I will mention that the c-Kit and Kit ligand, it's rare, and it's in these defined niches where both the receptor and the ligand are in the same region. That's something we can maybe talk about during Q&A. And so together we thought, all right, this is a thing. This is active in normal pancreas. What does it do? And so we addressed that by doing transcriptional profiling and loss and gain-of-function systems, and also perturbing Kit ligand in conditions of health, injury, and cancer in vivo. So first, to do transcriptional profiling, we used stellate cells in culture. They really give rise to CAFs nicely when you culture them. But the approximation of stellate cells in culture really is something in between what's normal and what's a CAF. Because in culture, of course, they're not really exposed to the rich milieu of factors that's in an intact tumor. And so we reasoned that probably to really understand what Kit ligand is doing here, we should do both gain and loss of function experiments to knock down Kit ligand to where it is in a tumor and to overexpress it to where it is in normal pancreas. And so we did both and did molecular profiling by RNA-seq to understand what Kit ligand does. I'll just briefly mention that subpopulations of these cells express c-Kit, so the Kit ligand had someone to talk to in these cultures. And so what we found is that Kit ligand seems to positively regulate a network of genes that we would predict is associated with maintenance of normal tissue architecture. So this includes a number of components of basement membrane and matrix components, really, that are associated with keeping our tissues intact. But we also saw negative regulation of factors that are involved really in detrimental CAF functions, including LRRC15. This was a marker described by Shannon Turley's lab that may or may not be functional itself, but marks a CAF state really associated with immune suppression and resistance to immunotherapy. It also negatively regulated expression of a number of immune suppressive cytokine and chemokine genes. And generally speaking, our loss and gain-of-function experiments agreed with one another. So from here, we moved into functional studies. We started in just healthy mice. We conditionally deleted Kit ligand in the pancreatic mesenchymal. We did H&E, saw no differences. But we were undeterred, so we crossed in a reporter. Of course, still no difference by H&E, but really thought to take a look under the hood, and we did some multiplexed imaging analysis using CODEX in collaboration with Sung Kim's lab at Stanford. And so really, there were two differences that caught our eye and motivated us to keep going. The first is that when we conditionally deleted Kit ligand, we saw this significant increase in CD45 positive cells or leukocytes that were recruited into pancreas tissue. And the second, which we think is related, is an increase in the abundance of perivascular mesenchymal, or these cells that are touching endothelial cells. And these cells in those positions really are spatially poised to secrete factors into the bloodstream that promote leukocyte trafficking or that impact endothelial cells to regulate leukocyte trafficking themselves. So from here, we moved into an inflammatory setting, and we engaged a protocol that's been used heavily by our colleagues in the field using this secretagogue called cerulean that induces damage of the acinar compartment and pancreatic inflammation. And in wild-type healthy mice, you can treat with cerulean. You get an inflammatory reaction in the pancreas, and that inflammation resolves within a few days. This has been really nicely benchmarked by many of our colleagues in the field. I will mention, though, that a combination of cerulean and a K-RAS mutation gives something very different. In that case, the inflammation does not get resolved and actually really drives the formation of pre-invasive lesions in mouse pancreas. And so what we found, really to our surprise, is that when we induced pancreatic inflammation in the absence of Kit ligand, we saw really much more inflammation, as well as a clear increase in what looked like an acinar-to-ductal metaplasia within the pancreatic epithelium. And this really aligned with what people have reported in the context of a K-RAS mutation in the pancreatic epithelial compartment, motivating us to ask whether there's some sort of paracrine inverse relationship between Kit ligand in the stroma and K-RAS signaling in the epithelium.And so we wanted to look a little bit more carefully at this idea that kit ligand negatively regulates epithelial plasticity. This acinar to ductal metaplasia is an early step of pancreatic tumorigenesis. And so you can see that in our control mice, you see that most of these acinar cells express amylase. The ductal cells express pan cytokeratin. Generally, these are in different cellular compartments. But when we conditionally knocked out kit ligand, we really saw fairly extensive co-expression of these markers consistent with this metaplastic state that's an early stage of tumorigenesis. And when we looked at the immune composition of these tissues, we saw a significant increase in CD45 positive cells and CD68 positive myeloid cells when we had this conditional kit ligand deletion. And finally, we looked in the context of cancer. And what we've now seen in three different implantable models is that when you lose kit ligand in the stroma, you significantly increase tumorigenesis. And we're totally aware of the limitations of the implantable models, but the one point I want to make here is that in these cases, these are insta-tumor, right? They've already lost P53. They are invasive cancer. But even in this case, kit ligand has some sort of tumor restraining impact at the tissue level, something we certainly want to look into further. And so to conclude, I hope I've convinced you that normal pancreatic cancer is not the barrier, but a barrier to tumorigenesis. We nominate stromal kit ligand as one factor that seems to limit epithelial plasticity and inflammation. The real relevance for epithelial sulfate is something we've only just started to look into and think about. In human pancreas tissue, we really need to dig deeper and understand where and when kit ligand gets lost during tumorigenesis. And of course, we want to really better understand what's happening in this dual recombinase genetic mouse model to understand what happens to PanIN progression in these mice. And the ultimate goal really is to think about downstream mediators that may serve as targets for interception or as biomarkers of progression. And so with that, I'll conclude. I'll thank all the wonderful people in my lab, especially Kathrina, who did this work. Thanks to our wonderful collaborators and funding sources. And with that, I'll thank you all for your attention and gladly take any questions. Very nice talk. And you mentioned there's a key role of the kit ligand during the pancreatic cancer progression, right? So I'm curious about the loss of the kit ligand expression in CAF. Is it a consequence or trigger of the poor cancer progression? Thank you. Yeah, no, it's a great question. So I sort of glossed over it, but TGF-beta alone was sufficient in our, at least, experimental culture models to induce loss of kit ligand from the mesenchyme. So TGF-beta gets induced in the pancreatic epithelium pretty early in tumorigenesis. So I think a possibility is that kit ligand sticks around kind of in the early stages of tumorigenesis, but once there's enough TGF-beta, then kit ligand gets downregulated, and it's presumably part of a program that serves as a barrier to tumorigenesis. I think probably there are other signals in an intact tumor. We did the is TGF-beta sufficient. We haven't knocked out TGF-beta in the context of conditioned media, for example, to ask whether it's necessary with all those other signals around. But certainly, that's a key factor that we think probably mechanistically is related to kit ligand regulation and expression. Got it. Thank you. Thank you. So great talk. I'm Yue Fan from Johns Hopkins. So I noticed that you used like a PanIN as a model study. So I'm wondering whether like IPMN to PDAC progression share the similar pathway you mentioned, or there's some other different mechanism. Thank you. Yeah, no, it's a great question. One I don't have an answer to right now. We were motivated by these totally striking human studies from Laura's lab and Marina's lab showing how frequent PanINs are. But these cystic lesions like IPMNs are reasonably common too, and they all do have their own so-called microenvironment, right? There is a modest stromal expansion around all of those lesions. I have no idea what they're comprised of. I have no idea what the kind of kit ligand expression is, but really, this was a very mousey work, and we're really motivated to go into humans and try to understand expression patterns in pre-invasive lesions of different stages compared to cancer. And also, we're so fortunate that we have now nice genetic models that actually enable us to study IPMNs, to study cystic lesions in the lab. So it's something we haven't touched yet. My sort of naive thinking is there's TGF-beta around in all of them, so perhaps similar rules would apply, but we have to do the experiment. I think it's a very valid question. Thank you. Thank you. Thank you. Very nice talk. I'm impressed by the stellate cells. Do you know if there are any other organs where something like that happen? Because you mentioned two endodermic organs, and there are other organs that need fatty acids too, so. Yeah. So, liver is really the original site where stellate cells were discovered a century ago. There's a really, really nice Cre-based mouse model, the L-RAT Cre model, that's been used for studying stellate cells in liver. Unfortunately, pancreatic stellate cells use something else to esterify retinoids. They don't express L-RATs. We can't use that in pancreas. Fabp4 Cre is a tool that we can use in both organs, but we don't have an L-RAT Cre of the pancreas. But one thing I'll say that's interesting, bona fide stellate cells have only been found in pancreas and liver. But what' CAFs cool is there was a sort of meta-analysis, one of these kind of harmonized single-cell RNA-seq studies in sort of a pan-cancer study that came out in cancer research, I think last year, that actually showed what they called Fabp4-positive pericytes that they inferred to be major origins of CAFs across tissues. And so I have to wonder whether this idea of a fat-storing perivascular cell of origin is actually a more anatomically conserved feature than we think, if that makes sense. Exactly. Back to the thing, pericyte is supposed to be a reparative cell, and now you're saying that you have a fat-saving pericyte cell now, which is great. Exactly. Thank you. Yeah. Exactly. Thank you. Very interesting data. So other than TGF-beta, were you able to recognize anything else upstream that gets induced? Because if you think about therapeutic potential, it will be easier to down-regulate a target rather than overexpress it, right? And TGF-beta, we already saw the failures. So were you able to recognize something else, or that's something you're trying now? Yeah, it's definitely something we're trying now. I think ultimately, one thing I didn't talk about, just briefly covered in the background about Kit ligand is that conventionally, c-Kit is an oncogenic signaling mediator. We were certainly surprised by our results, but I don't think we would want to specifically agonize the Kit ligand c-Kit pathway as a preventative or therapeutic strategy. So we're really motivated to look really mechanistically in both the sender and receiver cells to think about, okay, probably we don't want to target this specific pathway, but what's associated functionally? What underlies this apparent link between Kit ligand and this kind of negative regulation of inflammation and tissue plasticity? Some of the things that popped out are some of these basement membrane and matrix components, some integrins, laminins, other factors that, generally speaking, are associated with normal tissue architecture. And if we can understand the molecular regulators of those, I think those would be really appealing targets. In thinking about the question that came up during Stephanie's talk of how do we reprogram or favor a tumor-restraining fibroblast phenotype, and I think one that promotes a normal, if you will, matrix and basement membrane architecture, that would be appealing. We don't know the specific molecular mediators to chase yet, but we're motivated to look. Yeah, I think instead of switching between good and the bad, because there's so little known, bringing them back to the quiescent would probably be the best bet. Yeah, for sure. All right. Thank you. Okay. The timer's already started countdown, so I better get going. Thank you for inviting me. It's been a great session. It's going to be completely downhill from here. So what I'll talk about is some of the work going on in the lab. Firstly, I have no disclosures. So essentially, I think as a starting point, just to reiterate the complexity of the tumor microenvironment in pancreatic cancer, that we have a multitude of heterocellular interactions that are present and that really drive tumor progression. And that we've had, through many years, several studies that have really identified these key signaling pathways that mediate these intercellular interactions. And similarly, as was just reiterated earlier, a number of preclinical studies have really demonstrated quite clearly that, at least in the preclinical models, if we target some of these interactions, we can sensitize these tumors to standard care chemotherapy, and in some cases, we can even make them sensitive to immune checkpoint inhibitors. But also, as we were just discussing, other studies have really illustrated that it's not just a free target for all to target this tumor microenvironment. There's a certain level of buyer beware that if you go after specific subsets, such as with a smoothened inhibitor and sonic hedgehog targeting, or specific depletion of certain fibroblast subsets or T regulatory T cells, you actually see an acceleration of tumor development or tumor progression. And actually, where that leaves us is in this scenario where we have multiple heterogeneous interactions in the tumor microenvironment, and that we probably have the coexistence of both tumor-suppressive and tumor-promoting interactions between tumor cells and cells in the microenvironment. And so, my co-chairs or co-speakers here have already gone through this in much greater detail than me, but just really to emphasize that we have multiple subsets of fibroblasts in these tumors. And a lot of effort have really gone into classifying these based on the transcriptional and cell surface marker, and we have an emerging understanding of their function. And a lot of them seems to have really pro-tumorigenic functions, at least the way that it's been described. And so the question is really, where does the anti-tumor function sit, and how can we think about this? But also, more conceptually, I think as we move forward and start thinking about targeting fibroblasts in these cancers, both for pancreatic cancer and also in other cancers, I think there's just some considerations that are worthwhile just going through. So we have these subtypes, and we know that they're plastic, there can be switches between them. But when you start overlaying lineages, it's worthwhile just really thinking about whether the particular function that's being interrogated is really restricted to the lineage or whether it's a compounding effect of the lineage plus the subtype. And so, this creates some thinking in and around how do we actually identify these subsets, and what's the right way to target them?We know that there's a spatial relationship between tumor cells and fibroblasts, and other cells as well. So how does this play into how we target the fibroblasts? What about tissue specificity? We have a much better understanding of the fibroblasts and the fibroblast function within the primary tumor, in particular in pancreatic cancer, compared to what happens in metastatic sites. And if we are thinking about targeting these fibroblasts, there's a bit of an underlying assumption that they do the same thing across all tissues, and I think emerging evidence suggests that that's probably not the case. Also, in terms of the cellular interactions, are we going after interactions between fibroblasts immune cells, fibroblast tumor cells, and is the consequence of targeting those, is that the same? And finally, I think it's also worthwhile thinking about whether the interaction that we're going after is actually a dominant effect that's really being driven from the fibroblast, or whether the change in fibroblast phenotype or transcriptional output is really a consequence of the environment in which it's sitting. And so therefore really being very careful in terms of how we select our targets and how we test them mechanistically before we go forward. And so just to illustrate some of these points, I'll just go through three examples in a bit more detail. So the first thing is just sort of related to the relationship between lineages or more stable fibroblast functions, for lack of a better term, versus the sort of subset, more plastic function. So we described in the lab a couple of years ago, this marker CD105, that sort of denotes two different subsets or perhaps even lineages of fibroblasts. And both of these subsets can adapt the transcription profile of myofibroblasts, inflammatory fibroblasts, and antigen-presenting fibroblasts. And so the way we actually ended up identifying those was using a slightly orthogonal approach to the single-cell sequencing, where we used mass cytometry to probe cell surface receptors, and this was in collaboration with Jen Morton from Scotland Institute. That really allowed us, based on the KPC model, to identify the CD105 positive and CD105 negative cancer-associated fibroblasts. But the sort of experiment that got us really excited about this was when we take the fibroblasts out from the normal healthy pancreas and grow them in vitro, the fibroblasts that are CD105 positive remain CD105 positive, and the ones that are CD105 negative stay CD105 negative, giving us sort of an early indication that this is a stable marker expression. And actually, when we treat these cells with either TGF-beta, interleukin, or interferon gamma, again, we don't see the sort of shift in terms of between positive and negative CD105 fibroblasts. These fibroblasts population subsets or lineages were also defined by a sort of a differential response to common signals present in the microenvironment. So if we take these fibroblasts and treat them with a number of different signaling factors that are known to play important roles in the tumor microenvironment, and then we use mass cytometry to profile single-cell phospho-specific readout, we can see that the sort of response to these common signal differs between the positive and negative fibroblasts, suggesting not only that they express this differential receptor, but actually the way that they interpret the same signal from the microenvironment is actually inherently different between these two populations. And what we could also see is that they have a differential function in terms of how they regulate tumor development. So if we do a subcutaneous implantation of tumor cells and then co-implant CD105 positive fibroblasts, we see a slight acceleration of tumor development or tumor growth. But if we implant CD105 negative fibroblasts or even co-implant CD105 negative and CD105 positive fibroblasts, we can see that there's a suppression of tumor growth, and this is actually dependent on a functional immune system. What we don't know yet, and this is something we're actively exploring, is whether these sort of functional effects is not only encoded within the CD105 positive and negative subsets, but actually whether that's dependent on them adapting a particular myofibroblastic or inflammatory fibroblastic subset, trying to link the distinction or the interplay between the sort of hard-coded functions and the more plastic functions of fibroblasts. But what we also noticed is that when we start probing these fibroblast subsets across a larger number of models here using different genetic engineered mouse models, we can see that not only are the CD105 positive and CD105 negative fibroblasts identifiable across a number of models and also in human tissues. But what we can see is that when we look more broadly across a number of the sort of very well-described markers of fibroblast subsets, that the abundance of these markers differ highly depending on the tumor origin, right? So if we look at something like alpha-SMAOr a very useful marker of myofibroblasts. We can see that that's very well expressed, abundant in fibroblasts from the pancreas, the colorectal models, and breast cancer. But in other models, it's just much less abundant, even though the fibroblast's there. And so it led us to speculate, when we're thinking about fibroblast subsets, their presence, and the different origin of these fibroblasts, how can we start getting a handle on this in terms of the different subsets and whether there are tissue-restrictive or tissue-specific functions? And so the way we've started probing this is using implantation models, and coupling that with mass cytometry readouts. So in this particular case, we take cell lines, we implant them into the pancreas, the liver, and the lung, and then we probe that. And the benefit of the system is that we can control the cell line, so it's not an evolution of the cell line. When we see differences, it's the same cell signal the fibroblast and the tumor microenvironment is exposed to, but we are essentially probing the tissue-specific response to get sort of an experimentally refined way of interrogating this. And when we do this, and just look at the sort of mesenchymal marker composition, so this is just using a panel of about 32 cell surface markers, and looking in the normal healthy pancreas, the liver, and the lung. And I don't think we need a lot of statistical analysis to really say that the fibroblasts based on these 32 markers between the pancreas, the liver, and the lung, are just fundamentally different when we probe them in this sort of high-dimensional space. But what we can also start looking at is what's the response to tumor growth? And so we have the pancreas, the liver, and the lung, the healthy and the tumor-bearing tissue. And when we then take and cluster these individual fibroblast subsets that have been identified, and then look at their abundance in response between the healthy tissue and the tumor-bearing tissue, and we cluster this in a hierarchical manner, in an unsupervised hierarchical manner, what we can see is that the sort of clustering really resembles the tissue of origin, suggesting that the response of fibroblasts to tumor growth is really restricted based on the tissue of origin, and that there's sort of a hardwired, encoded way that these fibroblast subsets respond to tumor growth depending on the tissue organ. And the control experiment here is to do the same thing, but looking at immune cells. And essentially, if we do exactly the same set experiment, we don't see the same sort of clustering suggesting that there are sort of an inherent cell-specific way that these mesenchymal versus immune subsets respond to tissue damage or tumor growth. And so to really nail this down a bit more, we then isolate fibroblasts from the different organs and expose them to conditioned media from various tumor cell lines. And when we then start again probing, use a mass cytometry against these cell surface receptors, we can see again that the way that the fibroblast responds is really encoded depending on the tissue of origin, suggesting that the tissue of origin might dictate the fibroblast response to tumor growth. And so I think it causes a bit of pause in terms of thinking about, well, if we target one of these interactions in one organ, we really need to be sure that we see the same or at least a non-antagonistic effect of that in a different organ. And so the benefit of this could be that maybe actually we can use this as a way to think about doing tissue-specific targeting of fibroblasts, but we need to think about the sort of broader consequences of that. So, the sort of last example I'll highlight here is sort of like, so how do we think about probing this in a system whereby we can really control, in a very detailed mechanistic manner, the interactions between the cellular subsets to understand what happens within the cells, but also ultimately how the tumor cells respond to this. And so this is some work that's been ongoing in my lab, and this is really based on collaboration, in particular with Linda Griffith, who's a fantastic bioengineer out of MIT. And the sort of system that she's helped us develop, is a fully synthetic PEG gel system that allows us to do co-cultures between patient-derived organoids together with primary pancreatic fibroblasts or fibroblasts from other organs. And we can then include in vitro differentiated immune cells as well. And so when we start doing this, we can sort of start putting the model system together where we have the direct co-culture between defined cellular subsets. And this is just really to show that we have cellular interactions between the tumor cells, the immune cells, and the sort of fibroblasts you can see with the phalloidin stain here. And we got decent viability. But the sort of like, I think really exciting outcome here is that when we take fibroblasts, and when we take this model system and we start probing it by using single-cell RNA sequencing, even though we, in this particular experiment, take fibroblasts from the normal healthy pancreas and then co-culture them with the tumor cells and also the immune cells, we can see that the fibroblasts from the same sample start adapting these sort... subset transcriptional profiles. So we've got cells starting to look like myofibroblasts, MHC class II positive, CD74 positive, APCAFs. We've got ICAFs and complement secreting CAFs as well, suggesting that we can start rebuilding the tumor microenvironment using these bottom-up approaches. And what we can also see is that we have a tumor cell response that differs. So when we're looking in the tumor cells here, cultured by themselves together with fibroblasts and together with immune cells and fibroblasts, we can see that we have loss of transcriptional subsets and gain of other transcriptional subsets, and they actually correspond to more EMT, a more basal type transcriptional subsets, suggesting that when we rebuild these sort of interactions, we can start recapitulating some of the more aggressive features that we see in patients. But also that different patient-derived organoids respond differently to these interactions, suggesting that we can also start recapitulating some of the heterogeneity that we know or struggle to recapitulate in these preclinical models. So we've now started using this to probe a bit more detail what actually happens in the tumor cells as they are co-cultured. And as sort of a starting point, we focused on looking at RAS, just because it seems to be important in pancreatic cancer. But here, what we've been doing is we've taken a TurboID TAC to RAS, and then we can start probing is the number of proteins that are proximal to RAS as a consequence of the organoids being cultured by themselves in the presence of fibroblasts and the presence of fibroblasts and immune cells. And what you can sort of start seeing emerging here is that we have a vast expansion of the proximal protein network around KRAS as we increase the cellular heterogeneity in these models, suggesting that the way that the tumor cells respond to a high degree of cellularity really impacts the subsequent signaling dynamics within the tumor cells. And so what I've hoped to touch upon a little bit is that fibroblasts, they can regulate the tumor microenvironment as a whole in multiple ways. I think we're still trying to really work out as a community which of these effects are really dominant to tumor cell function, to tumor development, and identifying those, I think, is a very important first step to really thinking about how we can target these interactions. Some of the other things that I think would be very important to understand, and certainly there are ongoing efforts, is to really understand how the evolution of the interactions between the fibroblast microenvironment and the tumor cells is during treatment, and also how the sort of spatial interactions impact CAF biology and how that might apply to therapeutic response. So with that, I want to thank the people in the lab who've been instrumental to doing this work, the funders, and I'm happy to take any questions. Thank you. We actually do have one online question, but it's for Mara, sorry. Asking about your modulation of Kit ligand and the induction of CD45 positive cells. Do you have a rationale for why that happened? Our guess for now is that when we inhibit Kit ligand in this population, we increase expression of a number of inflammatory cytokines and chemokines in the lineage. So I think even in normal tissue, when these fibroblasts aren't in an activated state, it's still an increase in various chemoattractant cues that are presumably bringing the immune cells in. But another possibility is they're sitting right next to endothelial cells. So it's totally possible that they're also changing adhesion molecules on endothelial cells that are impacting leukocyte trafficking. So those would be probably my two guesses. Great. And then one for Claus. Do we have any insight into the tumor cell cues that impact CAF expression of CD105? Well, so CD105 is a co-receptor of the TGF beta receptor. So we know that if we treat fibroblasts with TGF beta, we see an increase in the level of CD105. But the sort of fundamental expression where we-- So we haven't identified any signals so far that turns on CD105 in a CD105 negative fibroblasts. So the ones that are negative seems to stay negative. Okay. Hi, Claus. Bobby Norgard, Boehringer Ingelheim. Nice talk. Thank you. Kind of odd question. As I think about where you're pulling fibroblasts from different tissues and showing how they're very different depending on where they come from. Yeah. The idea of making a fibroblast normal is not an unoriginal idea. But to my knowledge, fibroblasts are usually not a good thing in cancers. To your knowledge, is there a cancer type where fibroblasts are a good thing?And should we be thinking about not turning CAFs normal, but turning CAFs to a different tissue of residence? So I need that again. Sorry. Yeah. It might be the jet lag. It was kind of an odd question. It's kind of thinking about changing CAFs, reprogramming CAFs, not to a normal state- Mm-hmm ... but to a different residence state. Yeah. And I don't have a great example for this, but in fibroblasts in breast cancer, they're usually bad, and fibroblasts in pancreas cancer, they're usually bad. But if you took breast cancer fibroblasts and put them in pancreas tumors- Yeah ... would that now be a better situation? It's just a different way of thinking about reprogramming instead of pushing them back to normal, which I don't know what normal actually means. Yeah. Well, the short answer is I don't know. So I think it's an intriguing question. So in fairness, I think a lot of emphasis over the years have been to try and target fibroblasts because they are there, and they are abundant. And so I think that has led a lot of efforts to try and show that they are pro-tumorgenic. I think the sort of efforts to try and understand how might they do the other thing, how might they be anti-tumor development, I think that's lacking behind. So, I think we still have a lot to learn, in particular, as Mara alluded to, the sort of early stages of tumor development, what's actually happening there. So how does a fibroblast function differ between organs? That's a good question. I don't have the answer to that, but yeah, a good one. Thank you. Hey, Jen. Hey, Claus. Question for you and everybody, really. Given Laura and Marina's really nice work on the abundance of PANNs- Mm-hmm ... how likely is it, do you think, that some of the stuff that we can't translate from mouse to human is because we're having differential effect, the stromal targeting therapies are having differential effects on the primary versus the residual PANNs that are kicking about in those models and in the people? So I think that's very likely. Right. And I think it might even extend beyond that, because a lot of our current way of probing fibroblast biology is really focused on primary tumor. So yeah, we can target those. If they have a different function in pre-cancer lesion, that's obviously an issue. But also when we go in clinically, it's typically in metastatic disease. And so actually, that's compounding the effect there if the function is different between different organs. And so I think the emerging evidence suggests that that is the case. And so I think it's just critically important, as Stephanie alluded to, is really to learn from the clinical trials that have already happened, to really try and probe some of these questions before we rush into the next trial. Because it's not good for the patients, and it's not good for the community to shut down therapies and the development of therapies just because we're not doing the right translational workup. She claims you already answered. Okay. You're welcome. What about fibroblasts going together in clusters to the metastatic site? Mm-hmm. Is that going to now mingle and mix the- Yeah ... heterogeneity and the functions again, or is this really truly imprinted? Can we change them? Is it really truly imprinted in what's going to happen with the travelers? Yeah. So, and again, that's a good question. So, I think a lot of it is hardwired. So if we take the fibroblasts out and culture them in vitro, we see that the response to these cues are really hardwired into the fibroblasts. Sure. Right? And so that would indicate that even if they travel with the tumor cells to a site of metastasis, that the way they would respond at that metastatic site might be similar to what they would respond to in the primary tumor. I think where we still don't know enough is how many of the fibroblasts we're finding, either in a primary tumor or metastatic site, are from the tissue of origin versus the ones that have actually either traveled with the tumor cells or being recruited from the bone marrow, such as fibrocytes, et cetera. Right. And one of the challenges is really having robust markers that do not change over time depending on the tissue that they're in. So how can we track this with epigenetic markers, for example? Right. So once we track them, then we need to study their reciprocity because now they're going to be companions. Yeah. Absolutely. I think really picking apart the sort of underlying mechanisms of those interactions, I think is going to be critical. And I think we haven't really discussed this in much detail here, but in particular, the structure in and around the ECM as well, I think is just going to be so important in terms of how those interactions drive the tumor phenotype and immune phenotype ultimately. Yeah, I made you said that. All right. Oh, you have your question? Yeah. So maybe this is for the whole panel. So, given that all of these CAF targeting approaches work great in mice and then they fail in the clinic or make it even worse, should we start thinking about other animal models in looking at these therapies? Or should we be comparing the human setting with the mouse setting a little bit more to figure out what are the exact differences that account for these phenotypical differences we see with the different species? Yeah. So I think that's a very important-- If I may start. Yeah. So I think that's a very important question. So I would be very careful with discarding the mouse model, right? Actually, I think some of the fundamental challenges we've had in terms of translating from mouse into human, I think some of it is just because we don't understand the underlying biology well enough. And I think that some trials have probably been started too early, right? Where we really haven't understood what we're looking for in the biomarker. I think it's worthwhile really thinking about having other models. I think the best model to study this sort of thing is ultimately the human patient, right? And so actually having access to tissue from clinical trials, even if it's not a stromal targeting agent, but for any other agent, right? Would really help us understand the interaction network and how they respond under those conditions. I think that the sort of fundamental biology is really where we're lacking, if you ask me. Yeah. I'll maybe bluntly say what Claus is too humble to say, that I think his approach really is such a critical way forward to, yes, use mouse models that incorporate lymphatic vasculature and inter-organ crosstalk and things that we really need in vivo models for. But to really have as a critical intermediate step, these sort of the brilliant engineering platform that you've developed or other kind of patient-derived models to complement those and at least understand whether some of the paracrine signaling mechanisms we've seen in mice hold true in patient-derived models, whatever we choose to make them before moving forward and translating. It does seem like a critical intermediate that hopefully will be incorporated more kind of across the field. Great. Okay, with that, we will close our session. So thank you. Thank you. Awesome talks, guys. Yeah.