Skip to content

2026-04-20-am-plasticity-cancer-progression-clarkson-symposium

Date: 2026-04-20 Primary topic: single-cell-spatial-omics Cross-referenced from: single-cell-spatial-omics Word count: ~12,810

Auto-generated captions

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

Morning, everybody. My name is Bill Heit, and I'm honored to serve currently as the AACR Chief Scientific Advisor, but it's really my even greater honor to introduce today's symposium in honor of my friend and colleague, Barney Clarkson. And we welcome you to the AACR Bayer D. Clarkson Symposium on Plasticity in Cancer Progression: Mechanisms and Consequences. You probably know that this symposium has been held annually for many years, and it recognizes Dr. Clarkson's contributions to cancer research, and especially in the areas of cancer stem cells. Dr. Clarkson always attended this symposium, and even in more recent years when he became more frail, he always watched it on live stream. He always maintained his curiosity and passion for cancer research, and that's reflected in his really unbelievable scientific legacy. He was an expert in the field of heme malignancies for more than 50 years. His laboratory studied the growth and differentiation of normal leukemic and cancer progenitor cells. His research has added significantly to our understanding of hematologic malignancies and has been pivotal in the development of improved treatment strategies. He and his colleagues were among the first to develop curative chemotherapy for adults with various types of acute leukemias and lymphoma. They were also the first to show that an intensive chemotherapy regimen could induce remissions in patients with chronic myelogenous leukemia. In more recent years, Dr. Clarkson's research focused on characterizing and understanding leukemic stem cells with the goal of identifying specific drugs or treatment paradigm to target these quiescent cell populations that are often not eliminated by traditional chemotherapeutics. Barney was a member of the AACR for more than 60 years. He showed an unwavering commitment to advancing the organization's mission, and he served the AACR in numerous influential capacities. In fact, Dr. Clarkson is the only individual to have ever held four major AACR leadership positions. He served as a member of the AACR Board of Directors, he was AACR president, and he was AACR treasurer for an unprecedented five terms. I, as the immediate past treasurer for AACR, have a full appreciation of the importance of this role, and I'm grateful, will be forever grateful for Dr. Clarkson's leadership. And finally, Barney was president and founding chair of the AACR Foundation Board of Trustees, and he spearheaded the initiation of the AACR Foundation and served as its president for many years. In 2007, in recognition of his immense lasting impact on cancer research, patients, and the AACR, the AACR Board of Directors decided to name a symposium in his honor. Dr. Clarkson died on December 30th, 2025, and so today's AACR Bayer D. Clarkson Symposium, which is titled Plasticity in Cancer Progress: Mechanisms and Consequences, will honor his remarkable life and legacy. Chaired by one of our AACR Annual Meeting 2026 program chairs, Dr. Lynn Marie Postovit, this session will explore the mechanisms underpinning induced plasticity, with particular emphasis on stress-induced plasticity and the role of epigenetic dysregulation. The speakers will also discuss how plasticity may be a therapeutic target for the treatment of advanced cancers. We thank our chair, Dr. Postovit, and additional speakers, Dr. Matthew Loupiain, Hongjun Ji, for presenting their work and for honoring the member of Dr. Clarkson. Thank you for your attention, and I'd now like to invite Dr. Postovit, to begin the session. Thank you for the introduction, and thank you all for being here today. It's my pleasure to begin this session and to introduce, first, some concepts around plasticity. So in the background here, you see an African michaela zene butterfly. And why would we start a cancer session with a butterfly? It's because this particular butterfly undergoes a pretty dramatic phenotypic plasticity. And so in response to its environment, whether it's a dry or it's the wet season, in Africa, it'll actually change its wing patterns in a fairly dramatic way. This isn't genetically regulated. It's an epigenomic and transcriptionally regulated process. And so plasticity, in the case of the butterfly, is really the ability of the organism to change its phenotype in response to the environment. And today, we're going to be talking about cellular plasticity in the context of cancer, where cancer cells can also change their phenotype non-genetically, and most often in response to the microenvironment. In cancer, we can assume that plasticity should confer a tremendous selective advantage, should allow cells to evade the immune system, to spread, and to evade therapy, and so it's a very important problem in cancer.So what does plasticity look like, and how do we understand it? The first is really histologically in a loss of structure function. It's first seen well over 150 years ago now with the advent of the first microscopes by Kannheim and Geronte, and then Virchow. It's really the loss of structure function. And so this would be the loss of glands, tape structures, et cetera, and the acquisition of what looks increasingly like embryonic-like structures within the tissue. We see this, and we'll see it today with adenosquamous type conversions, as well as a number of malignancies or transformations during malignancies, such as neuroendocrine prostate cancer. Below, I'm showing you a picture of something I'll be talking about, dedifferentiated endometrial cancer, where you can see in stark contrast to the fairly well-differentiated areas here, we have very poorly differentiated areas right next to it. In addition, there's functional plasticity, which is more of a coordinated program wherein cells can actually behave like other cell types. One of my favorite examples of this is from my previous mentor, Mary Hendrix, in work that was done in 2002. In this particular study, they took melanoma cells, so very aggressive melanoma cells, labeled them with an RFP protein, and then injected them into the ischemic limb of a mouse. And so in an ischemic limb model, there's neovascularization, as you can see here, with a green dextran and beads that can move through it. And what is truly remarkable is that the red melanoma cells were forming chimeric vessels with the endothelial cells. So these weren't melanoma cells becoming endothelial cells, but rather co-opting certain features of those cells, in this case, the expression of VE-cadherin. And finally, we have the movement between cellular states and fates, which is sort of where we are right now with concepts of lineage plasticity. This was described many years ago now with EMT and then MET, so epithelial-mesenchymal transition and then the opposite, where it's thought that cells can actually acquire mesenchymal phenotypes, spread and metastasize, and then colonize with epithelial phenotypes. And as you can see in this cartoon here, overlaid on a Waddington's epigenetic landscape, it can also be simply heterogeneity and cells moving through different epigenomic patterns in order to display plasticity. And so we do consider and understand plasticity in the context of epigenetics. So here I'm showing a cartoon of Waddington's epigenomic landscape that was co-opted by Dr. Yamanaka several years ago now. And so within this, we have what would be considered a totipotent cell at the top of the hill. And during development, in response to developmental cues and microenvironmental factors, there's cell fate specification until you eventually lead to the various cell types within the body. It was considered very difficult to transdifferentiate or move between these epigenomic states, and even more difficult to move back up the trajectory. And of course, Dr. Yamanaka showed that this is actually quite easy and can occur with just three or so factors. And since then, we've shown that this can occur with cocktails within the microenvironment. Superimposed upon this, I added my small addition, which are the lightning bolts, which is a context of cancer wherein we have mutations. And certain mutations, like the loss of P53, can actually dramatically increase the ability of cells to reprogram and move up that hill. There's something about the interplay of genomic instability and mutations and epigenomic instability that occurs in cancer. And so I bring you to our session today. We're going to start with Dr. Lupien, who's going to be talking to us about this interplay and the heterogeneity in genomic mutations that can occur concomitantly with alterations in the epigenome, and how these work together to enable plasticity as well as heterogeneity within tumors. I'll then be talking about dedifferentiated endometrial cancer and trying to get at some of the mechanistic underpinnings by which cells can go from something that looks fairly well differentiated into something that looks embryonic-like. And finally, Hongbin Ji will be talking about his work related to lung cancer and adenosquamous conversions, and trying to understand signaling pathways that may mediate this so that we can inhibit this particular process, and notably looking at how factors in the microenvironment, such as inflammation, can help to promote these conversions. And with that, I'll invite Dr. Lupien to the stage. All right. So, thank you for the organizers, for the chair, for enabling us to come here and present our work. So today, I'll be talking about epigenetic heterogeneity or epigenomic heterogeneity and how it enables tumor evolution, how it assists in the process of tumor evolution. And when I use the term epigenetics, epigenomics, I want to be specific about what I'm talking about. I'm actually specifically talking about the chromatin itself, from the policy in the lab or whatever the mindset in the lab, which stands for that, yes, DNA might hold the words in our genome, but it is the chromatin that tells the story. Okay, good. I have no conflict to report for this presentation. So what do I mean by chromatin so that we're all on the same page? Going back to Biology 101, chromatin is this complex of DNA and proteins that actually is found in living cells, where you have specific proteins known as histones that come in, form an octamer around which 147 base pairs of DNA will wrap, and that will define what we call a nucleosome. And the nucleosome is the basic unit of the chromatin. There's multiple nucleosomes in cells, and the spacing between nucleosomes will define how DNA sequences are actually interpreted in a given cell as opposed to another. And so the nucleosomes can be very close to each other, forming a highly compacted structure, in which case there's not going to be any stretch of DNA readily available for anything else that could read DNA, such as DNA-binding proteins, transcription factors, or DNA repair machinery. But you can also have other areas of your genome where the nucleosomes are sparse, far away from each other, such that the DNA in between, the linker DNA, is accessible for any protein that can read the DNA sequence. Again, transcription factors, DNA repair machinery, DNA replication machinery, and so on and so forth.So we're clear on this. This is where epigenetic lies. It's in this complex of DNA and histone proteins that make up the chromatin. And why do we care so much about the chromatin? Well, there's two specific aspects, and both have to do with the notion of cell division. When a cell divides, if it wants to ensure that it maintains its identity, it has to engage in a replication process that will provide for the daughter cells equal amount of the critical information that carries the material, the information that's needed for cell identity. And while we're all very familiar with the notion of DNA replication, which happens in vitro in a test tube, what actually happens in the cell is chromatin replication, such as when the DNA gets replicated, the histones themselves are also going to be involved in the process of replication, such that from the parental DNA, the chromatin, the histones associated with the parental DNA will split in half, and half of it will go to the lagging strand, and half of it will go to the leading strand such that the information present in the parental strand is still passed down from these histone proteins onto the de novo synthesized strands. And so this is how a cell ensures that the material provided for the daughter cells does not only have a good quality copy of the DNA sequence, but also has a reproducible copy of the chromatin template from which it came. The second aspect that's critical about cell division is that when it's time to divide, there's a Mendelian segregation that happens, such that you do end up with equal copy of your DNA in the two daughter cells. This happens with the chromatin as well, right? Remember this, the chromatin is also segregating in a Mendelian fashion. And so that's why of all the different molecules that one can study in a cell, that's why we pay so much attention to the chromatin, because in terms of lineage fate commitment, if we talk about potency, plasticity, and the likes, if you want to interfere with cell identity, ultimately you have to be able to alter the DNA sequence and/or the chromatin state of that DNA sequence in order to change the identity, the function of the cells that come out from that replication process. And so what my lab has been focusing on for a number of years now is in specifically identifying areas of the genome in cancer that diverge in chromatin state compared to, let's say, normal tissue, or when we compare primary to recurrent metastatic tumors. And so what we do is that we look for what we call chromatin variants. These areas of the genome that may or may not harbor mutations, but that we'll see a given stretch of DNA sequence that can be measured at nucleosomal resolution change its chromatin state. So in this given example that I'm portraying here on the right side of the slide, what you can see is that for a given cell type, whatever you want it to be, the given piece of DNA that might be in this case about a KB long in sequence, like one KB in nucleotide sequences, would adopt a specific chromatin state where there's a high nucleosome density. And under such conditions, this type of compacted chromatin would prevent transcription factors, DNA repair machinery, DNA replication machinery from reading the DNA, and therefore, these types of DNA sequences are typically inactive in a given cell. If that same DNA sequence adopts a different chromatin state in a different cell type, then we will be able to have a different output from that same DNA sequence. So in the bottom graph here, what we're showing is, for instance, an instance where the chromatin state implies nucleosome depletion, a piece of DNA in between nucleosome that's accessible for transcription factors, DNA binding proteins, different modifications to the histones that also pair up with that difference in compaction that defines a completely different chromatin state. So any given region of the genome that you find that diverges in chromatin states between a reference and your treatment condition, a reference and your disease of interest, so normal versus primary, primary versus recurrence, would define that piece of DNA as a chromatin variant. Okay, good. Now, to be able to give you specific case examples of what I mean by this, I'm going to use prostate cancer as an example. Prostate cancer has been exhaustively studied over the ages to be able to identify, for instance, the drivers of phenotypic change, either from normal to primary, from primary to recurrence. In this given figure coming out from a review from Rob Bristow, what I'm portraying is the suite of drivers, cancer drivers that have been identified by looking at mutations, drivers of phenotypic change. And so a lot of these are going to be genes. So the SBOP genes, for instance. There's fusion transcripts. They're coming out from an ETS fusion. PTEN is involved. MYC, RB1, AR are burdened by mutations more than expected by chance along this process of evolution and tumor for prostate cancer. These are all coding cancer drivers. And in addition, the field has looked at mutations in the non-coding space and has identified a series of non-coding cancer drivers from mutations. So for instance, the binding sites for the androgen receptor, which is a transcription factor, as well as for FOXA1 and OCTB13, are burdened by mutations more than expected by chance, defining them as cancer driver cistromes. For FOXA1, for PLCBC4, the regulatory elements controlling the expression of these genes are burdened by mutations more than expected by chance. And so we call these units of gene regulation regulatory plexuses, so you have cancer driver regulatory plexuses. But these have all been identified in the primary tumor setting.Moving forward with work that followed from my group and others, we then started looking at chromatin variants in the same way that people look at genetic variants to figure out where in the genome there are areas that could be cancer drivers. And in the process, what we discovered was that the binding sites for the androgen receptor, FOXA1, and OCTB13, were not only affected by mutations, but they were also burdened by chromatin variants, by changes in chromatin states in primary prostate tumors. So this is picked up here as AR binding sites being cancer drivers picked up from chromatin variants. Another interesting case example comes from this ETS fusion gene, which is referred to as T2E, so a TMPRSS2-ERG fusion, where the TMPRSS2 upstream elements is considered to be hijacked to serve as a promoter for the expression of the ERG fusion gene. What actually happens in cancer is that not only is there that structural variant picked up from a genetic standpoint, but in addition, upstream of that fusion, there's a massive engagement of regulatory elements that change chromatin states to be turned on in instances where the fusion is positive, where the fusion is present, so that we have these clusters of regulatory elements upstream of the TMPRSS2-ERG fusion that fuels the overexpression of that gene. Those of you familiar with this notion of a cluster of regulatory elements might be more familiar with the terminology of super enhancer. Now, the work that had been done and showed in those resume slides had been focused primarily on identifying the drivers of cell state transitions in the non-repetitive genome, which accounts for about 45% of the sequences in the human genome. In recent years, thanks to technological advances, we've now been able to start mining the repetitive genome for such events where chromatin variants might change the way the genome is being utilized. In the repetitive genome, there are a lot of these repetitive sequences that are transposable elements, so research often focuses on transposable elements when mining the repetitive genome for features that can drive phenotypic change. And from that standpoint, again, published just recently, based on chromatin variants, we identified a series of transposable elements active in the primary prostate cancer setting, ERV2s, SINEs, LINE5, and solo LTRs, that promote a more stem-like phenotype within the primary prostate tumor. Now, the problem is that we haven't really done much in characterizing the chromatin variants and how they enable cell state transition in the course of progression to metastasis. And so a few years back, we had the opportunity to start collaborating with colleagues in California that were involved in the Stand Up to Cancer West Coast Dream Team, specifically focused at characterizing metastatic castration-resistant prostate cancer, the mechanism enabling that transformation and that progression in patients. Luckily for us, those collaborators, so Felix Fang, David Quigley, had already processed the samples from patients for whole genome sequencing. So structural variants and mutations were known. Gene expression had been characterized. We came in and helped out to better understand genome topology and the 3D structure of the genome, and then more recently started profiling the chromatin to identify chromatin variants within that advanced prostate cancer setting. In this case, using ATAC-seq as the assay of choice to identify regions of the genomes that gain accessibility in metastatic castration-resistant prostate cancer. And so the story that I'll show now is unpublished, but we'll cover all of these data sets to be able to demonstrate the degree of heterogeneity at the epigenetic level that is conferred in this context of metastatic castration-resistant prostate cancer. But at the same time of the interplay that exists with genetics and gene expression to be taken into account in the process. Okay. So first result of interest is showcased here on the left side. We're looking at unsupervised clustering of our patients, of our 71 metastatic castration-resistant prostate cancer, classified according to the degree of chromatin variants falling over repetitive sequences. So each column is a patient, each row is a different repetitive sequence, a different transposable element. And what came out from this unsupervised clustering is four different subgroups of metastatic castration-resistant prostate cancer. These are labeled at the top in the color code. In dark red, we have the patients that we ultimately call the LINE1+, because most of the repeats that were found in chromatin variants within these patients were of the LINE1 family of transposable elements. We had a group of patients that we labeled as constant because they had very few repetitive sequences found in chromatin variants. Another subgroup of patients were called SETLs, because these had chromatin variants within different types of transposable elements, SINEs, ERVs, transposons, and LINE1s, or LINEs in general, sorry. And then our fourth subgroup, which we labeled SINE1+, because most of the repetitive sequences found in chromatin variants within these patients were SINE1 positive or SINE1 transposable elements. Okay, so moving on to the relationship between the classification we got from chromatin variants over repetitive sequences to the classification that's been established in the field based on gene expression. So I'll draw your attention to the color code that's portrayed in this upper right panel. The color code corresponds to gene expression-based classification. And if you go all the way down to this right side panel, what we're portraying is the proportion of cases that were classified according to the expression pattern within the LINE1+, the constant, the SETL, and the SINE1+. And the key take-home message you observe here is that there's no agreement. Okay? There's not a single color found within the LINE1+, there's not a single color within the constant, there's not a single color within the SETL, and so on and so forth. So there's disagreement between the classification that comes from gene expression versus the classification that comes specifically from the repetitive genome, okay, not the non-repetitive genome.And so is our classification meaningful, or is it a fluke? And so to assess this, we then turn our attention to clinical data from the same patients that we had been able to classify as LINE-1+, constant, settle, or SYNE-1+, and ask whether any of these subgroups behave differently in terms of clinical outcome. And so in this figure, what we're portraying is the Kaplan-Meier curve for survival probability in patients treated with RC androgen receptor selective inhibitors. So drugs specifically targeting the androgen receptor, which is standard practice. And what you can see from this result is that these LINE-1+ are intrinsically resistant to androgen receptor selective inhibitors. We had responders in all other three subgroups of metastatic castration-resistant prostate cancer, some progressing faster than others, but still all responding for a certain amount of time. So then we turn ourself to genetics. Is our classification linked to any specific genetic aberrations picked up in these samples? So again, from whole genome sequencing, we could call copy number gains in red, copy number losses in blue. Intensity of red representing more or less the degree of copy number gains one could observe. And we did an analysis to figure out if there was any preferential association between a genetic event and our classification based on chromatin variants. And the only one that came up significant was this association for the androgen receptor gene locus being highly amplified within our LINE-1+ cohort. This information is further portrayed on the right side image with copy number gain on the Y-axis in the LINE-1+ versus the three other subgroups. You can see that in some patients we have 25, 50, all the way up to 150 copies of the androgen receptor locus within patients part of the LINE-1+. Those of you familiar with such high levels copy number gains probably already have in mind that this is not going to be derived from a linear amplification, but that there's a likelihood that actually this is coming out from extra chromosomal pieces of DNA that are enabling the amplification of the AR gene locus. And that's exactly what we see. So from whole genome sequencing data, we could call ecDNA positivity, so extra chromosomal DNA, these circular forms of DNA that replicate but segregate in a non-Mendelian fashion. So they're often referred to as cancer gangster in the lab. And what we could see is that there was definitely a bias for having these ecDNA positive tumors within our LINE-1+ population. But again, this was not a one-to-one relationship. Some ecDNA positive cases were picked up in the other subtypes, but to a lesser degree. So another take home here is, again, if one was to classify patients purely based on genetics, the classification would end up being very different than if we did the classification based on chromatin. So the need to integrate all these different classification when wanting to better understand exactly what are the drivers of phenotypic change and of particular phenotypes, to account for this inter-patient heterogeneity. Now, of interest to us was the fact that in these LINE-1+ patient, we had systematically an overexpression of the androgen receptor gene. Whether they were ecDNA positive or not didn't really matter. As long as they were within the LINE-1+ cohort, AR expression was much higher. So the chromatin variants were predictive of a high expression for the androgen receptor. So then we decided to inquire what these repetitive sequences could be doing. Were they in any given way influencing AR expression directly by serving as regulatory elements, for instance, and the likes? And so the first analysis we did was to assess genome-wide where these repetitive sequences were found preferentially accessible in the LINE-1+ cohort. And here I'm showing you a depiction of two chromosome, chromosome eight on the left side, chromosome X on the right side. Eight was selected because of the MYC locus, which is an interesting locus for most of us in prostate cancer and other cancer types. And the chromosome X was selected because that is where the androgen receptor gene locus resides. The red color is depicting the density of accessible repetitive sequences within different patients. And as you can see over the chromosome eight, that's fairly distributed across the chromosome. But then when you look within chromosome X, everything is piled up on top of the AR gene locus. And so we had, from this standpoint, hypothesized that there was a high density of de novo activated regulatory elements working at that locus to enhance the expression of the androgen receptor. And so what we did next was to try to assess what biology could be happening there. Since we were identifying these sites from ATAC-seq assay, chromatin accessibility, we decided to investigate if they could serve to recruit specific DNA binding proteins. And so we did this by tapping into a database of over 8,000 different ChIP-seq data sets that have been profiling various proteins and various cell types to see if we could pick up any specific protein from that ChIP-seq data preferentially able to occupy these repetitive sequences in prostate cancer. The list is shown here on the right side. For those of you familiar with prostate cancer, this will be reassuring, but at the same time maybe a bit disappointing because we keep seeing the same culprits over and over again. FOXA1, AR, OCTB13, amongst the top four transcription factors able to occupy these accessible repetitive sequences to potentiate a specific phenotypic switch towards a LINE-1+ identity. The next thing we could do was to assess the extent to which these elements found at the AR locus were actually in a capacity to physically interact with the promoter of the androgen receptor and therefore influence its expression. So for this, we took advantage of Hi-C data, which is measuring chromatin contact frequencies between an anchor of interest, in this case, the androgen receptor promoter, and distal areas that can be 20, 30, 40, 50 KBs away from that promoter.And what we depicted here is a case example in a LINE1+ patient versus a case example in a non-LINE1+ patient, a LINE1+ to be specific. The black lines are indicative of interactions between the androgen receptor promoter and distal sites that do not have repeats but that are accessible. So there's a similar pattern in the two cohorts. However, the red lines depict physical contacts between repetitive sequences accessible in the LINE1+ and the androgen receptor promoter, and so you can see that they have the ability to directly interact with the promoter of the androgen receptor, despite being far and away from the promoter itself in the linear space. And so this brings us up to the model of choice, which is that in this subset of patients, the heterogeneity that we can identify within metastatic castration-resistant prostate cancer has a lot to do with the activation of these chromatin variants that specifically give rise to this LINE1+ subgroup, which is classified based on having all these repetitive sequences that gain accessibility in these patients in order to drive the overexpression of the androgen receptor. And they do so in part by actually recruiting that protein as it's being transcribed, since they're permissive and able to recruit the androgen receptor itself, FOXA1, and others. And so this creates a positive feedback loop to bolster the expression of the androgen receptor, which can account for the resistance to androgen receptor selective inhibitors. Because if you overwhelm the system with the target, it can supersede the abundance of the drug and therefore ensure that the system can still work through androgen receptor signaling, despite some androgen receptor molecules being antagonized by the drug that's provided to patients. And so we're here now, where in the metastatic space, we actually have discovered this heterogeneity as it pertains to repetitive sequences and how they're activated by chromatin variants with the LINE1+ subgroup requiring deep investigations to identify new, better alternative therapeutic strategies to move forward. So I'll conclude here just by highlighting the necessity to consider epigenetics in the space of prevention. We need to figure out why chromatin variants arise. Is it mutation? Is it environment? Is it stochastic detection? There's so many tools that are being developed, and we need more of those to be able to extract features from the chromatin to effectively track the disease before, during, and after treatment. And on the intervention side, epigenetic therapy is there. The first passed attempts have been made with small molecule inhibitors, but there's definitely a need for a new generation of approaches that relies on epigenetics to guide treatment selection. And so hopefully we'll have the chance to discuss this. So with this, thank you for your attention, and here are the people that helped in the work. Great. We have time for a few questions. I invite people to come to the microphone. And while they're moving, we do have one online, so I'm going to put my glasses because I can't see a thing. Does the increased accessibility of LINE1 elements also promote ecDNA generation, given the chances for HR between LINE1 elements? Yeah. Excellent question. We have no evidence that that's the case, but from a hypothesis standpoint, it's an excellent hypothesis that would need to be properly assessed in a system where we can actually manipulate the degree of accessibility within repeats and see if that can induce ecDNA formation. And so we haven't had the chance to play with those, but definitely a very valid hypothesis that we agree with. Yeah. Other questions? Hi. Great talk, Matthew. It was super interesting. I was curious about whether with the increased accessibility, you're seeing increased transcription of lines as well and potentially retrotransposition. So at this stage, I will say I cannot give you the answer yet. So we're in the process of assessing, and I would justify that actually discovering or measuring the degree of expression from repetitive has been a challenge for us. So if you have tools that work, we have the data, and we'd be happy to sit down. We've tried a number of tools, and the data is not conclusive from our perspective. Okay. We can talk more. And in that LINE1 positive group, what was the p53 status? Was there a difference in terms of p53 from the others? Right. So it was not significant. So there are instances of p53 mutations in there, but it was not a significant association. The only, from a genetic standpoint... In our pool, we have 71 cases, so each of the groups is around 20 patients or so, a little bit less. And so maybe we're not sufficiently powered, but we could not detect any statistically significant association with p53, although in the field, there's definitely arguments in favor of p53 playing a role in the repeat activations. Yeah. In different cancers. Yeah. Thank you. Yeah, exactly. One more. Wonderful talk. So I was just curious, how large are these extra chromosomal elements? Oh, yeah. Tens and tens of 50 KB plus. So per extra chromosomal element, how many unique LINE repeats are you seeing per extra chromosome? Yeah. The actual number of repetitive sequences that we see within an amplicon is in the hundreds. I don't remember the exact number, but in the hundreds, under a thousand, but in the hundreds. Okay. Yeah, that's really interesting. Yeah. Okay. Well, one more. We'll do one more. Yeah. Sure. Sorry. Yeah, no. Nice talk. Well, I'm curious. You introduced the chromatin heterogeneity during the prostate cancer progression. Yep. Well, and you introduced different outcomes after the treatment. I'm curious about what did cause this chromatin heterogeneity. It's the genetic background in the patients or the treatment or what? Yep. So, in these specific instances, we don't have an answer, but from the field, we know that mutations elsewhere can induce chromatin variants in regions of the genome that have no mutations, so there can be an indirect effect.We know as well that it can be an environmental stressor that would lead to these. So for instance, methionine cycle is tightly linked to the production of S-adenosylmethionine, which is a universal methyl donor. And so if you reduce your methionine consumption or for whatever reason, the methionine is less available, SAM levels will go down. The system will have a hard time maintaining DNA methylation, will have a hard time maintaining histone methylation, which can then change the chromatin compaction state. So definitely environment can be a player. And then stochastic events. In the literature, it's already known micronuclei formation is sufficient to bring upon epigenetic changes with no known specificity. So we can't predict where those changes will happen, but in the process of micronuclei formation and resorption, that will lead to chromatin variant acquisition. So in our system, we don't know because we don't have a living system. We have a static system, so it's harder to predict. But definitely in the field, that's a question that's of constant debate and assessment. And so those three mechanism for sure exist to cause chromatin variants. Thank you. But from a prevention standpoint, if you specifically want to prevent prostate cancer, we have to figure out what those are in prostate. Yeah. Thank you. Thank you very much. Thank you. Perfect. All right. So it's a pleasure to be here again, and to talk to you a little bit about a project that we have going on in the laboratory related to dedifferentiated endometrial cancer, and I have no conflicts to disclose. So dedifferentiated endometrial cancer, and we've seen this picture in the introduction, is a histopathology that's characterized by the well-differentiated areas that are in juxtaposition to very poorly differentiated areas. Now due to this rapid morphological change, it's theorized that these types of malignancies are occurring via stochastic mechanism, most likely a mutation. Now DDEC, in contrast to endometrial cancers in general, has a very, very poor prognosis. Over 80% of patients with DDEC will progress and succumb to their disease. This is characterized by extensive metastases as well as resistance to therapies. And it highlights the need to better understand the process by which these cells become dedifferentiated and enable progression. So to do this, we think of this cancer within an epigenomic context and relating it to development into embryonic stem cells. So in this particular cartoon, we can see embryonic stem cells on the left-hand side, and what would be our DDECs on the right-hand side. As we discussed in the introduction, during development, or when we're doing directed differentiation, in response to factors in the microenvironment, embryonic stem cells are gradually specified to the various tissues in the body. And so in terms of this, it would be the epithelium within the uterus. We think that during cancer, this can be undone, leading to these very embryonic-like phenotypes, and that it's very likely that this is due to a mutation. I'll highlight some similarities and differences. So one is in the microenvironment and features that may help to promote this dedifferentiation. This can include features such as low oxygen or hypoxia, which has been shown to both enhance induced pluripotency, as well as the acquisition of stem-like phenotypes and dedifferentiation in a variety of cancers. I'll also highlight a major difference with that. On this side, we have a case that's highly unregulated. So while these cells may display an embryonic-like phenotype, they're really unable to get cues from the microenvironment and to properly redifferentiate. And that's of course, in contrast to the stem cells, which yes, can self-renew, but then also can listen to their microenvironment to execute appropriate programming. And so we were really asking the question as to what kinds of mutations may actually be driving this dedifferentiation in these endometrial cancers. And so this work was initiated by Mackenzie Cotham here in close collaboration with two pathologists, Changhon Lee and Martin Koebel, at the University of Alberta in Calgary. And so what was done initially was to take these lesions, and to microdissect the adjacent regions, and then to compare the well-differentiated components to the dedifferentiated components using exome sequencing, but on a panel of epigenomic marks. And in doing this, the finding was remarkable that in over 80% of cases, there was a loss of one or up to two proteins. The first one being SMARCA4, as you can see here, and these are adjacent regions. And the second one being ARID1B. And I'll note that in the well-differentiated components of these tumors are usually ARID1A low. So these would be ARID1A low, ARID1B deficient tumors. So why is this important? It turns out that we have 80% of cases that are associated with this, and these hits are both components of the SWI/SNF chromatin remodeling complex. And so SMARCA4 is an ATPase within this complex that's used to generate the energy to help to move nucleosomes. And ARID1A and ARID1B are mutually exclusive proteins that bind the DNA. So if you lose both of them, you actually just disrupt the entire complex. So what we're seeing in the dedifferentiated components of these cancers is a loss of the SWI/SNF complex. And so we wanted then to see whether this may be driving dedifferentiation. So we took cells that would represent the precursor lesion, so they're ARID1A low, and did CRISPR genetic manipulations, in order to genocopy what we were seeing in our patients.Now, the ARID1A, ARID1B double loss, we were actually never able to bring cells through the process, but we were able to do this with SMARCA4 in a couple of cell lines in a couple of ways. So here we are genocopying what we're seeing in patients, the complete loss of SMARCA4 with a clonal resolution. And so we thought, well, if this is the switch, then the loss of SMARCA4 should increase the tumorigenicity of these cells. And what we saw initially in vitro was really the opposite. And so you're seeing across the top sphere formation assays, and so this is just a serial dilution of that. And at the bottom, soft agar assays, the light gray would be the knockouts. And it's just clear in all the models, we in fact are seeing an initial reduction in tumorigenicity and in sphere formation. And so we went back to the drawing board and thought about plasticity in general. And so this is another paper that's well over 100 years ago, done 100 years ago by John George Adami. And Adami was a pathologist that again was looking at tissues, and this is the paper that he wrote, so making a beautiful observation. But it's essentially saying that perhaps in cases of plasticity, you don't have proliferation, and that so as cells are changing or gain the ability to change and acquire another cell state, they aren't in fact proliferating. And we know from lineage plasticity studies, as well as a number of others, that the microenvironment is also very important. And so we have Paget, the seed-soil theories, as well as Dr. Fidler, who has these sort of sequential acquisition of metastases, particularly in melanoma model, showing that the microenvironment is very important and can also help to select for different types of phenotypes. And so we thought that we would take this model and move straight to the mouse. And so what you're seeing here is what we did. So we took cancer cells with knockout or wild type SMARCA4, and then did serial dilutions through the mouse. So grew a tumor, broke it up, grew another tumor, et cetera. And what we found in the first passage, so the F0 passage, was the knockouts, very much like in vitro, had a much lower take rate. And then when they did grow, they grew slower. But this was overcome over time till eventually the knockouts and the wild types had similar growth patterns. When we remove the cells from that F2 or F3 tumors and we put them in vivo, you can see that this ability to form tumor spheres was restored and that in fact the tumor spheres themselves looked quite different. So we were hopeful that perhaps we had induced this dedifferentiated phenotype. And indeed, the tumors themselves had characteristics of a dedifferentiated cancer. And so this is just one example where we're showing that like the human patients, the tumors became resistant to chemotherapy, in this case carboplatin. So what did these cancers look like? So I'll remind you that when we do our CRISPR workflows, we're actually selecting for clones. And in cancer biology, we normally don't want to do this, but here for hypothesis, we actually did. And what we found blew our minds, which is that these tumors weren't undifferentiated, which is what we thought, but were rather properly dedifferentiated. So you can see on the left-hand side are model system. On the right-hand side are actually tissues from patients. This would be the well-differentiated and undifferentiated components of the same patient. And I'll just highlight here we have SMARCA4 knocked out on this side. And you can see by the hatched lines in H&E they begin to dedifferentiate. They start to lose markers such as E-cadherin, as well as lineage markers such as PAX8. But the notable thing was that we saw both well and poorly differentiated components within the same tumor. Which caused us to ask how this was actually happening. So this could have been another stochastic event that would be indicated by a mutation or perhaps a hierarchical reprogramming that we should be able to see with epigenomic changes genome-wide. And so this is just showing that it is very unlikely to be a genomic alteration. So here we're looking at pooled tumors, and this is repeated. We took cells before they go into the animal and then cells at F2 or F3 and did these comparisons. So on the left-hand side is just a summary of exome sequencing that was conducted, really with just a focus on highly mutated genes. I'll note that the background of these cancers is MSI high, so they have a high mutational burden. So we would expect some noise and variability. On the left-hand side are the wild type cells and tumors, and then the right side, the knockout cells and tumors. And I'll just note that there really isn't anything popping up at a high frequency in the knockout tumors. That would suggest to us that this is driving dedifferentiation. And I will note that at the time that we take these tumors, they are about 50% undifferentiated. So we have a very low frequency of some mutations and really nothing that is different from what we're seeing in the wild type. We think this is actually reflective of the MSI phenotype. This is DNA methylation array that was also conducted. And what you're seeing here are cells in vitro before they go into the tumor. And then wild type on this side, knockout on this side. In here, we have cells that we've taken out of the tumor and regrown in vitro, and then cells right from the tumor. And here we have cells right from the tumor and then grown in vitro. This is just a principal components analysis because it's just a little easier to see. What you'll note right away is just moving cells from in vitro to in vivo causes a marked change in methylation. So we see a hypomethylation in general. The next point is that we do get a separation of the knockout and the wild types, suggesting that at least DNA methylation is at least playing some role probably in the dedifferentiation that we're seeing. So we wanted to know then, how are these cancers evolving? So it looks to be something that's regulated by the epigenome. So in order to do this, we went back again to the beginning. We barcoded the populations in vitro, so these would be naive populations in vitro, and then took them through the entire course of the serial passaging again, conducting single cell multi-ome sequencing at each step.So the first thing that we did was look at the barcode variations. So this is really confirming our exome sequencing. So if you were to have a mutation or stochastic event that allowed an expansion of clones in the knockouts, you would expect to see that in the barcodes, and we saw really no difference between wild types and knockouts. And in addition, we did mitochondrial entropy and confirmed again that there's really no difference between wild types and knockouts. So we're not selecting a clone during this particular process. This is a single-cell ATAC sequencing that I'm highlighting here, and I'll point out a few things. These are the wild types here, and these are our knockouts. So the chromatin occupancy is different between wild types and knockouts. I'll also note that while we go through the passaging, the wild types remain rather uniform sitting here, whereas the knockouts start to drift over time. And I'm calling this a drift quite intentionally because I think it's a rather random drift. I've also overlaid on here what we'll be looking at in a few seconds, which is the RNA clusters. And what you can notably see is that those clusters are actually shared between the knockouts and the wild types, albeit in different frequencies. So what we're seeing is, yes, a different chromatin state, so different chromatin occupancy, but in fact the same transcriptional states between the wild types and the knockouts. And that can be seen here. So this is just the integrated multi-ome single-cell sequencing that you can see in the various clusters, wild types, and knockouts here. And albeit the ratios are different between the two, we really aren't seeing a difference in the states that they're existing in. If you recall, we barcoded everything, so we can look in each tumor over time at the frequency of the states for each cell. This is showing that all cells actually have the capacity to exist in all of the states in any given tumor. So this is truly plasticity within that. Whereas the wild types remain fairly constant between passage, suggesting that this is probably in response to location within the microenvironment. We see that the knockouts are erratic in terms of the frequency of expression of the various states. So is there anything stem cell-like or embryonic in these cancers? The answer is startlingly, and to my chagrin, absolutely not. So we have no embryonic signature. We have no dedifferentiated signature. And in fact, the only pathway that came out positively was hypoxia in all of this. And so we have a difference in some mesenchymal genes and other types of genes. But in general, there definitely is not a stem cell or developmental signature that's emerging. So we didn't quite believe it, and we've used a number of other types of assays to look at this. So what I'm showing you here is an archetype analysis. So this is really looking at extremes of phenotype or states that can exist within a population. Again, showing that, in this case, we have these four states. And this is different in that it's also using machine learning to extrapolate which cells are sitting in a non-committed state. So it's not clustering out everything. So what you can see is, again, wild types and knockouts are existing in all of the states. And the notable difference is that the knockouts exist in an uncommitted state much more frequently than the wild type. So they're not in an embryonic state. They're more often in an uncommitted state. And we've confirmed this with other experiments using, for example, pseudotime and velocity analysis. So on the right-hand side, we have wild-type tumors. On the left, I guess it's the other way when you're looking at it. On this side, we have the knockout tumors, and you can appreciate with the lightness of color, then the knockout tumors, while they can hit a final state, are existing in a far earlier state. So they're much less committed within these transcriptional states, in contrast to the wild types, which are existing in rather static states within the tumor. So collectively, those results suggest to us that dedifferentiation isn't actually dedifferentiation. It's not the acquisition of an embryonic phenotype, but rather the inability of a cell to sit in a set state. And so we thought if this was the case, then we should be able to go back to the human tissue samples, look at them, and be able to find characteristics of a epithelial cell within those dedifferentiated areas. And so we used a bank of tissues that we had and looked to see simply whether these areas had E-cadherin within them. And you can see with IHC staining, so in red, that we did in fact see this. So if you didn't have the staining, it kind of plays a trick on you. You would assume that these were dedifferentiated regions. But what we discovered was E-cadherin sort of heterogeneously expressed in these regions, depending on the patients. And we have this quantified down here to really highlight the heterogeneity that is existing. And so, in summary, we have shown that SMARCA4 loss in endometrial cancer cells does induce a DDEC histopathology. And we're careful about this because we don't think it's inducing dedifferentiation per se. We think that rather SMARCA4 enables cell state disorder. And this can be exemplified with our human tissues, showing that E-cadherin staining can occur in even differentiated regions of SMARCA4-deficient tumors. And in fact, when we're looking at the expression of markers like this across the different states, they're not defining the states per se. All cells, both wild type and SMARCA4, can manifest plasticity in transcriptional states. However, only the knockout cells exhibit a random transition between states, and we actually think this may be a detachment from being instructed by the microenvironment. And that finally, histological dedifferentiation in the case of DDDC is likely not always due to the reprogramming towards a stem cell or more embryonic fate. And so I'll just end thanking the people that did this work. I've mentioned Mackenzie and Chenghan Li and Martin Kobel. We've had a lot of help with the informatics by people such as David Knapp, and Hannah and Julia have taken on this project to sort of understand how this is actually occurring.I'll take questions. Hi, great talk. I noticed in your cartoon model for your in vivo models that use NSG mice- Yeah ... do you expect these same epigenetic changes from the chromatin remodeling knockouts in any immunocompetent mice? That's a great question. I think the immunocompetent mice would likely have a pressure from the immune system. The components such as HLAs also change in this particular model, and so we don't have that pressure on there, so we're not seeing the consequences of that. This is really probably cell intrinsic and due more to features such as hypoxia and nutrient availability or location within the tumor. But it's a great question and I think it's also part of the reason why in patients, over time, the well-differentiated component, it doesn't exist in the final tumors. It's undifferentiated. I think a lot of that has to do with the immune system. So it's a great question, and that interplay is really important. We are missing out in our models, unfortunately, with that. Okay, thank you. Yeah. Yes. So, really interesting talk. Jeff Wahl from the Salk Institute. So I have two kind of related questions. One is how you are looking at the embryonic state. So are you looking at transcription factors or genes that are expressed in embryonic stem cells, or are you looking at transcription factors that are specific to the early embryonic stages of this tissue? We have done both. Mm-hmm. And neither are popping up in particular. So my first inclination was that we would-- I have a background looking at Nodals and NANOGs and that type of thing, so I'm very stubborn thinking that was what was happening. That's not emerging, nor is a dedifferentiation back to the origins within the endometrium. So I assume you've looked at SOX factors that are important. Yeah. And this kind of blew our mind as well, and when we looked at the patient tissue, we found the same thing. So something like SOX2, as an example, can show up in these tumors. So you do get pockets here and there. It's highly heterogeneous, and it doesn't necessarily correlate with the dedifferentiation. We have found the same thing in our model, and so the stem cell factors are almost randomly distributed, especially when we're comparing the well versus poorly differentiated regions. Yeah. Yeah. The other thing is that what's been found in many tumors, which would agree with what you're saying is kind of a cell state instability or confusion, is that you get markers that are consistent with multiple cell types within the same cell. So it's kind of a hybrid state phenotype. Do you see that? Yes, we do. Okay. So really random, like hair, neuron, et cetera, will pop up in the individual cells, but not enough to drive a cluster, if that makes sense. So I think as bona fide heterogeneity and instability, it might be noise. Yep. So, yeah. Nice talk. Thank you. Yuan Zhu from UT Southwestern. And so, echo the questions. I'm thinking about, have you looked at injury? So the tissue injury potential can reprogram the cells, and because I see one of the signature you pop up is hypoxia, so is that somehow related to injury-induced reprogramming? And Hannah's sitting there, so she's particularly interested in what types of things may be enabling this and driving it. So far, there probably is an interplay between that. The caveat is that something like hypoxia itself pushes the epigenome and changes it, and then also remodels greatly things such as protein synthesis. So we're trying to understand this coordination within that, and it's probably playing a little bit of role. So something about going into a stressful environment sets this chain off, right? I would say in vitro before they go in, almost everything is changed a little bit, but probably not driving anything. But the big changes we're seeing is when you put it in vivo. So, yes, we're trying to understand it, and I don't have an answer aside from waving my hands. Any evidence between inflammation, like injury to the endometrium, to the incidence of those cancer? Not as of yet. These do tend to occur in an MSI-high background, so it's quite possible that there is an inflammatory component to it. Thank you. And I have time for maybe one and a half questions. Yeah. Awesome. Amazing talk, and I've thoroughly enjoyed the discussion thus far. Stacy Lloyd from Tuskegee University. My question extends beyond what you presented today because I was interested in your earlier data where you showed how the methylation sequencing was very different once you put the cells with the in vivo cells. So I'm just wondering if there was anything done with those set of cells and what that data may have looked like. So we do have the cells straight out of the tumor, and then back in vitro and that typically-- And I will say, even with the single-cell ATAC, and I didn't put it because the scale is so crazy, it moves everything. There's nothing that changes things more than going in vitro to in vivo. We're now trying to understand whether that's retained or if you put them back in vitro, do they go back to kind of an in vitro type of state. But it was alarming how-- So the differences from going in vivo are greater than taking out a SWI/SNF complex. Wow. Yeah. Thank you. Yeah. Hi. Gaoyang Liang from Salk Institute, and I'm trying to be quick. Do you find any transcription factor programs stabilize this uncommitted state? Nothing's really popped upAP1 a little bit, but really nothing, mostly because of the fact that they're transcriptionally moving between those states. So the short answer is no, maybe AP1, but nothing that's concretely popping out at us. Okay. So with that, I'll invite Dr. Ji to come to the stage. Thanks for the organizer to give me this opportunity to present our work. So, this one example for plasticity. I was like most of you wearing glasses yesterday, but today I should have. But I got lost during yesterday's poster session. So if you guys or your friends can find or know that, please let me know or go to the room 21 and help me, okay? Thank you. So I think I'm also plastic. Okay. So I will try because I cannot see very clear, but I will try, see if I can go through smoothly. So this is my disclosure. I think Lynn has talked about the pathology. So you can look at the adenocarcinoma and squamous, they look quite differently. And also I think in clinical, the treatment will be quite different. And my lab has worked on the animal model. We work on the LKB1 in the KRAS G12D mouse model, and we have found LKB1 loss can drive the adenocarcinoma to squamous transition, we call AST, spontaneous transition in the animal model. But if we treat this model with some inhibitors or drug, we can kill the majority of the adenocarcinoma, but all the leftover tumor will be squamous. So we think, okay, maybe it's very likely this transition called AST can link to the drug resistance. So we're really dying to know if this really happen in clinic. So fortunately, I think that Charles Swearingen and Helena Yu has reported in 2020, and about the squamous transition in the osimertinib-resistant lung cancer patient, this is EGFR mutant lung cancer patient, in about 7 to 9%. And in contrast, 4 to 6% which has small cell lung cancer transformation. So if we look more, Andy from Dana-Farber work with Passiani has found two out of nine KRAS G12C lung cancer patient after develop the resistance to the aggressive, they begin to showing up the squamous pathology. So it's really two out of nine, it's close 20%, it's really very high amount, high percentage of these kind of events. So we try to ask because this all a transition to become resistant. We want to ask if before the transition, we can detect something and we can tell if tumor really become resistant or not. So we contact Andy and Kwak and work with Moratti. We collect the samples from the CHRISTOPHER-1 clinical trial, and we collect six to eight lung cancer patient samples with ONCOSEC data. So among these six to eight, only one is squamous. All six, seven samples are adenocarcinoma. So we really want to ask, if this lung adenocarcinoma, they really showing up something like a squamous signature, if their signature correlated with the drug response. So very fortunately, we can see here, some are really showing up the enrichment of squamous signature. And with the increase of the enrichment of squamous signature, we begin to see shorter treatment duration indicate of the poor patient response. So this really link to the drug response to squamous signature even in adenocarcinoma. So also we found very interesting, this kind of correlation was only observed in LKB1 mutant, another name STK11 mutant lung cancer subgroups. So Andy also found this. Two sample I just mentioned undergone the AST. Among these two sample, one really showing up LKB1 mutation. So with this, we really establish the link between the drug resistance and also the AST in both EGFR mutant lung cancer and also KRAS. Here talk about KRAS G12C mutant lung cancer. So we will try to ask what are mechanism. So we show the spontaneous transition and also drug treatment-induced transition, they looks very similar. That's why we want to go through the details mechanism. Okay, so we also identify KLK6a as a biomarker can predict the response to aggressive therapy. But of course, we cannot say all these samples with high KLK6a will go through the AST or not. And also in another independent cohort, we can see the high KLK6a level really correlate with poor patient survivalSo our previous data working the KRAS LKP1 model has shown the LKP1 loss can cause the excessive accumulation of ROS level. But we don't really know how ROS can trigger the AST. So with this, we try to establish system using organoid, try to see we can establish the AST model in vitro. So we take tumor from KL model, lung adenocarcinoma is early time point, and we take lung adenocarcinoma from KL model and culture in organoid system. So fortunately, we found over half of this organoid really showing up the squamous pathology after long-term culture. So from passage one, we can see the tumor was display like luminal morphology. With time, we begin to see a solid sphere, and we can see the p63 expression in nuclear. And also we inject these two organoid into the lung orthotopically, and we found the tumor become squamous. So with this system, we try to dissect the detailed mechanism behind this kind of transition, AST. So we collect tumor from the multiple tumor, lung adenocarcinoma, squamous carcinoma from KL model. And also we did the collective different passages of the KL organoid, which can go through the AST. And also we studied the human lung adenocarcinoma squamous specimen. And we did bunch of the single-cell RNA-seq and also the bulk RNA-seq and ATAC-seq, and tried to establish the trajectory of the transition AST evolution. So basically, we identify three stage. One is early stage is like adenomatous state with class one to three, and we then go through the double positive state, and then eventually go to the basal state, called squamous state. So we can clear see the response to oxidative stress here, because just mention, we don't really know how this stress can cause the drive trend AST happen. So if you look at the detail about class two, we can see the increase of the RELA, the NF-kappa B activation. And this also confirmed in the organoid system and also the lung adenocarcinoma human samples. So we try to ask is RELA was activated by ROS or not. So we treated KL mice with NAC, try to remove the ROS level. So basically, we can see the in vehicle group, control group, RELA was really high level, but a dramatic decrease in the NAC treatment group. And then we can treat the organoid with the NF-kappa B inhibitor using BAY compound. And basically, we can see the increase of the p63 expression, indicated by p40 staining, indicated this transition AST was accelerated by the NF-kappa B inhibition. So, we also tried to do one more thing. It's like in the normal organoid culture, we use with or without NAC. And surprisingly, if we deplete NAC in this organoid culture, and plus NF-kappa B inhibitor, we don't see any organoid growth. So, this really indicate NF-kappa B even transient activated, they could be very important for the lung adenocarcinoma survival under the oxidative stress condition. So we also did animal model work. So in KL model, we try to treat the tumor model with a PL to induce high amount of ROS can cure the adenocarcinoma. So we can see adenocarcinoma was gone, but the left of tumor was like a squamous, really big. But NF-kappa B inhibitor treatment can cure most adenocarcinoma, but we can see more like a squamous was like here. But if combined PL and BAY compound, and we can see most of tumors gone, both adenocarcinoma and squamous carcinoma. So this really indicated at early time point, like A state. So probably NF-kappa B activation, transient activation was really important for maintaining lung adenocarcinoma lineage and also same time try to maintain the tumor survival. So we can combine NF-kappa B inhibitor and maybe TKI to treat those patients. One way to prevent the transition happen, and another way probably can really cure the cancer. So we go further, try to look the if inflammatory signaling was involved in late time point because we can go through the different clusters. So I just mentioned class two. But with class three, four, five, we begin to see even interferon signaling was quite high. So we check compare the non-squamous conditioned medium and to the adenocarcinoma. We found the chemokine like CXCL1, CCL5, and also cytokine IL-1 alpha was dramatically up-regulated in the squamous conditioned medium.So this indicated the tumor cell, once they become P63 positive, they begin to secrete these different cytokines. Especially interleukin 1 alpha, which we feel very interested in. So we culture the conditioned media from the squamous, conditioned media with the CAF fibroblast, because there are a lot of fibroblast in the squamous. So we can see the dramatic increase of the markers of the leukocyte-related CAF, like S100A8, and CXCL2, and also CCL1. So this indicated the squamous cell or P63 positive cell secrete IL1alpha can promote the reprogramming of CAF into the LCAF, leukocyte-related CAF. And this reprogramming can be broken by the IL1 receptor antagonist. So further, we try to ask, because we also find the interaction between the tumor cell, CAF, and also neutrophil. We see quite a lot of neutrophil in the squamous microenvironment. So we try to ask is, what does this LCAF do? So it secrete IL1beta. Not IL1alpha, IL1beta. And we culture this conditioned medium with the neutrophil. We can see the dramatic increase of the NETs formation. NETs is neutrophil extracellular traps, was dramatically increased. So if we check the mice animal model with the IL1 receptor antagonist, we can see a dramatic decrease of the NETs formation here. So to summarize this part, during the double positive stage, late stage of the transition, we found P63 positive cell can secrete IL1alpha, and they can reprogram CAF into the LCAF, which can secrete IL1beta, try to trigger the formation of the NETs. So we also try to confirm the NETs level in the early time point, like adenocarcinoma, and also late time point, like squamous carcinoma. We can see dramatic increase amount, level of the NETs in the squamous tumor. Then want to ask, what if we block this NETs formation? So we treated the KR model with the GSK484, which can block the NETs formation. We can see very clear NETs level was down. And also, I think we see dramatic decrease of tumor burden. But we don't see much difference between the adenocarcinoma number, but we see dramatic decrease of the number of the squamous here. So this really indicate the NETs can promote the squamous transition and also the growth. So we treated this model also with the RAS oncogene inhibitor, because recently we have this IMCS6236. So when we treated this RAS oncogene inhibitor, actually we can see the majority adenocarcinoma can be cured, but not the squamous in KR model. So once they transition to squamous, they are become resistant to the RAS oncogene inhibitor. So we'll go through detail. We can see the NETs formation was really dramatically increased, even increased after treatment. So this reason we don't really know why this up-regulated, but we really see dramatic high level of the NETs here. So we think maybe we can combine with IL1 receptor antagonist because that was on the deferment in clinical. So we combine the IMC6236 plus the IL1 receptor antagonist, and we can see the squamous number was inhibited, and also we see the NETs level was dramatically suppressed here, too. So try to summarize this unpublished data. We see story. So we think during the transition AST, early time point was like A state, because the LK born loss can accumulate quite high amount of ROS, and that can trigger another stress, like environmental stress, like NF-kappa B activation. So this activation is quite transient, but they can do two things. One is to maintain the tumor survival, and another way is block the transition. They will maintain the adenocarcinoma lineage. And with time and the late stage, they became P63 positive, double positive state. And they begin to secrete cytokine IL1alpha, and can reprogram CAF to LCAF, leukocyte-related CAF, and secrete IL1beta to trigger the neutrophil NETs formation, which is very important for the squamous tumor growth. So we can do the multiple ways, blocking or intervening to prevent or to treat this kind of transition or this recent tumor. So, this is my last slide. I will thank my collaborators like Huaking Wang and Andy, and also my lab, Xinmian Tong, who did this job. Okay, I will stop here. Thank you. We have time for questions. So nice talk. I have two questions. Well, actually, I have three questions. Okay. First one is, you didn't say, why do you think that there's no correlation between the squamous... It's only in the squamous transition that you see the correlation with... What was that? I forgot already. In the early part of your talk, you said there was a correlation of survival only in- Yeah ... LKB1. Is that the same? The human part, right? Right. Yeah. So we see a high amount of the keratin 6A, so we think keratin 6A was really important for keratin stemness, but we don't really know that samples that can really go through the squamous transition or not, because that was like people take a dissection, take the tumor out, and then we don't know the consequence, what happen. Right. But it's only in the LKB1 that you see the correlation. Oh, yeah. So that thing, that's subgroup, right? Yeah. Yeah. So we don't really know why, and I think that the LKB1 mutation also quite closely linked to the KEAP1 mutation. So we don't know why, but the data's there. That's not... Okay. In your last data slide, you showed that if you treat with RMC plus IL1B agonist, you block the production of the squamous transition and the NETs. Yeah. You didn't say whether you block the tumor growth. So do you get enhanced tumor control? Yeah. We also see the tumor growth was a problem because during short of time, so I didn't go through all the data. Actually, I have a lot of data- It's kind of the most important data. Yeah. Okay. Thank you. And the third question is... What was it? Okay. Why don't you ask your question, I'll think of the third one. I'm getting old. I can only remember two at a time. Yeah, can come back later. Yeah. Thank you. Sort of related question. So you showed the correlation with AST, with the LKB1 loss. Yeah. Why do you think, A, it only happens in a subset? You showed that it happens earlier. It also happens- You mean human data, right? Yes. Because it also happens in a subset of the wild type. Why do you think it only happens, in a subset of the LKB1 loss and only after targeted therapy, with the G12C inhibition? Are either one alone not sufficient, or do you think that there's a threshold required for AST, thinking of your sort of three-model mechanism, of the three states? Yeah. So, from the mechanistic standpoint, we think LKB1 loss can cause the... One way can promote tumor malignant progression or increase stemness, but I think they can also create the high amount of the ROS level that could probably push the transition. So once they become translated to the squamous, they have an up-regulation of NF2, they will up-regulate a lot gene, can move the loss level. That probably explain why LKB1 mutation is so important for the transition, that's linked to transition in clinical. Yeah. But detailed mechanism, we still need work out. Yeah. And then slightly related question, I can ask one more follow-up. So as you know, germline STK11 loss, you get Peutz-Jeghers at risk of multiple- That's right, yeah ... adenocarcinomas, and many of those are KRAS driven and RAS driven. Do you think that those patients in those settings are more at risk for AST in their tumors, given that a lot of them are RAS driven and by definition of LKB1? Yeah, I think so. Yeah. From the clinical observation, that should be the case. Yeah. I don't know if anyone's looked at that specifically, but it sounds like it would be the case. Yeah, so I think that should be... I think LKB1 mutation was quite frequent in the Caucasian cohort, but in China, we only see like 5%, 7%. I think maybe Quag or Andy can look at that details. Yeah. Right. Thank you. Thank you. So Jesse, I remember the last one. So do you have... It went by a little quickly, but in the double positive state, are they making neutrophil trophic chemokines like CXCL- Yes ... they are? Yes, because I didn't mention. Yeah, CXCL. Yeah, they're quite a lot. Yeah. Right. So they have quite a lot amount of the neutrophil infiltration into the tumor microenvironment. Yeah. We have time for one more. Thank you. Great talk. Do you have any idea if the inflammatory phenotype is retained at resistant to RMC 6236, and therefore it can be a possible target? We didn't check other chemokines. No. Okay. Yeah. Okay. Thank you. Yeah. Thank you. So with that, thank you. Thank you. And with that, I want to thank everyone for sitting through the session and-