2026-04-17-decoding-tumor-evolution-human-tumor-atlas-network

Date: 2026-04-17 Primary topic: single-cell-spatial-omics Cross-referenced from: single-cell-spatial-omics Word count: ~12,795

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. It is on the hour, so I will go ahead and kick off this session. For me, this is not just kicking off the session, but it's also kicking off the entire meeting. This is probably true for many of you. So today's session is entitled Decoding Tumor Evolution through the Human Tumor Atlas Network. We have one introductory talk from a program officer at the National Cancer Institute who will start out by telling you a little bit about the Human Tumor Atlas Network, and then we have three scientific talks representing three of the 10 atlases in HTAN. There's no time for questions after the first introduction, but there will be time after each one of the scientific talks. So without further ado, I'll introduce our first speaker, Dr. Indu Kohar, a program officer at the National Cancer Institute, who will tell you a little bit about the Human Tumor Atlas Network. Thank you, Hunter. A very good afternoon to all, and I thank the chair and the organizing committee for giving an opportunity to present and introduce our NCI's Human Tumor Atlas Network program. It's not moving. Okay. I have no financial relationships to disclose, and I'm here representing the entire NCI HTAN team. Members span across the various divisions of NCI as well as within the NIH. We know that traditional hallmarks are the basic guiding principles for malignant transformation of cancer from pre-cancer to advanced disease. But these relationships between these hallmarks are evolving, and the HTAN program is addressing such evolving relationship between the cancer hallmarks, starting from pre-cancer to late-stage disease. Again, the deep understanding of the biology of the carcinogenesis cannot be done without the advancement in the spatial omics technologies, and I'm here listing some of the recent technologies which include spatial transcriptomics, spatial proteomics, spatial metabolomics, spatial multi-omics, and 3D/4D spatial profiling. The NCI's Human Tumor Atlas Network program is a collaborative initiative to map the tumor evolution at a single-cell resolution. It aims to build highly dynamic three-dimensional atlases of pre-cancer and tumor. The main goals include the integration of molecular, cellular, and tumor tissue composition and architecture involving the tumor microenvironment and immune milieu, describe the transition across the cancer spectrum, premalignant lesions to malignancy, locally invasive to metastatic cancer, responsive to resistant cancer type, with the ultimate goal of enabling predictive modeling to refine detection and therapeutic choices. HTAN program was funded under two cycles, under Phase I, which was from 2018 to 2024. It encompasses various tumor types. Six of them are listed here: lung, pancreas, colon, breast, skin, and pediatric. The highlights from the HTAN Phase I is published in a collection in a "Nature" journal. It is a publicly available set of spatially resolved tumor maps describing transitions in cancer for use by the cancer research community. Some of the important discoveries include one study which examined the interaction among cancer as well as the non-cancer cells in six different tumor types. The study identified distinct substructures, which are called microregions, within the tumors, which have been associated with different activities and interactions, as indicated by a specific metabolic state or a T cell profile. Several studies also focused on colorectal cancer, including two that defined early molecular events that drive the transformation of precancerous lesions in colon to cancer. Two studies focused on breast cancer. One used single-cell sequencing, RNA sequencing to profile metastatic cancer from 60 individuals, while the other described the underlying genetic networks that define the breast cancer subtypes. Some of the common themes that came across these tumor types and analysis include the recurring pattern of cellular spatial patterns that is indicative of co-evolution of tumor and stroma, which are emerging drivers as progression of progression and resistance. Additionally, discovery of rare transitional dedifferentiated or plastic tumor cell types often conserved across tumor subtypes and correlated with the poor outcome. In addition, the study addressed the deep exploration of stroma immune and immune cell interactions. So far, HTAN has led to more than 200 publications and preprints, with 20-plus primary atlas publications and 50-plus new analytical and experimental tools. So the findings of the HTAN Phase I drives the implementation of HTAN Phase II, which was initiated in 2024 and is still ongoing. It is complementary to HTAN Phase I with respect to the diversity of the tumor types. It includes PCA as well as HTA center. PCA is the Pre-Cancer Atlas centers, which include skin, pancreas, brain, gastric, and myeloma, whereas the HTA centers include prostate, colon, ovary, lymphoma, and pediatric solid tumors.This is the general structure of the NCI Human Tumor Atlas Network, comprising of five BCA centers, five Human Tumor Atlas centers, the HTAN Data Coordinating Center, which manages the data, as well as the HTAN Steering Committee, which governs the program. In addition, there are several subcommittees and working group, including Precancer Atlas Subcommittee, Advisory Committee for Data Coordination, Methods Working Group, Visualization Working Group, Translational Working Group, and Outreach Working Group. HTAN is built on four core principles. One is the community resource, complement approaches, open communication, and data and resource sharing. Here, I'm showing how the HTAN data sharing and community engagement takes place. All these atlases, along with certain cross-network projects, generate huge amount of data. This data is then ingested by the Data Coordinating Center that uses their data platform and provides standards, validation, management, and governance. The data sharing is facilitated by HTAN portal, Cancer Data Service, ISP's Cancer Gateway, and Seven Bridges. Further, the DCC, the Data Coordinating Center, facilitates the community engagement through user supports, manuals, workshop, working groups, and data jamboree. This is a snapshot of HTAN data portal. Here is the QR code, which you can scan if you want to get more information on the various data and resources and tools available in the HTAN portal. As you see, there are 14 atlases now, spanning across 20 organ types with 2,400 cases and approximately 10,000 biospecimens. HTAN is not alone. It is part of a broader atlasing ecosystem. It is collaborating actively with the Human Protein Atlas, Chan Zuckerberg Initiative, Human Cell Atlas, HuBMAP in various ways, through a working group, workshops, cross-consortium metadata, as well as spatial biology meeting. How can you be part of HTAN? So HTAN basically offers researchers which are not currently funded by HTAN the opportunity to apply to join the program as non-voting associate members. If you want to be associate member, you please contact the HTAN Data Coordinating Center, email is listed here. And that way you will be able to engage in HTAN consortium activities through working groups, annual meetings, and developing collaborations. Overall, in summary, the NCI's HTAN program is addressing tumor evolution to accelerate cancer discovery. It is built for scale and collaboration. It's a NCI-funded network comprising of tumor atlas as well as precancer atlas centers, built by multidisciplinary teams, unified standards via Data Coordinating Center. It answers the questions beyond the scope of single study, addresses tumor heterogeneity and microenvironment at single-cell resolution, builds multidimensional atlas from precancer through metastasis, help identify biomarkers of detection, prognosis, and treatment response, additionally generates new biological insights and hypothesis. It also provides open data to the community, which they can use today through HTAN data portal. With that, I'll end my presentation here. Thank you so much. Thank you, Indu. And as I said, we did not build in time after Indu's talk, but if you do have questions, you can come up and introduce yourself afterwards. So my name is Hunter Shane, and I am one of the co-PIs of one of the phase two atlases, the precancer atlas of skin cancer. And I'm thrilled to also chair this session and kick off the scientific talks for today's session. So I will walk you through the journey from a normal type of skin cell known as a keratinocyte, to a common form of skin cancer known as cutaneous squamous cell carcinoma. These are my disclosures. And if you want to understand the evolution of cutaneous squamous cell carcinoma, you need to first start by understanding what are the somatic alterations in normal skin cells. And I'll pose the thought experiment to the audience, and that's how might you do that. Your reflexive reaction is probably to just sequence normal skin, and this is not a bad idea. But the problem with this approach is that it gives you an average signal from the complex mixture of cells that are there. Now, to be sure, this has been used to great success in the past. You can go all the way back to 1994 when Doug Brash's group used this approach to show that there are p53 mutant patches of keratinocytes in normal human skin. And then, of course, Inigo Martin-Carrena and colleagues really turbocharged this approach in 2015 in a landmark paper, to show more generally the mutational landscape of clonally related patches of keratinocytes in human skin. But I would argue the ideal resolution to do this would be at single-cell resolution. In my opinion, single-cell DNA sequencing technologies have come a long way, but they still suffer from low levels of sensitivity and specificity in their mutation calls. And so to get around this challenge, our laboratory developed the following workflow, where we start out by extracting a bulk population of skin cells, and then we single-cell sort them so that there is now one cell per well in a 96-well plate. We clonally expand these cells as much as possible. These are cells from normal tissue, so they have a limited replicative ability. But it is easier to genotype from a starting point of, let's say, 100 cells than one cell. But to finish the job, we would lyse these small colonies of cells, releasing their genomic DNA and mRNA, and then further amplify these nucleic acid populations in vitro.And between the combination of ex vivo clonal expansion, in vitro nucleic acid amplification, and a heavy dose of bioinformatics that I'm not going to go into, we are able to call somatic alterations in individual skin cells at nearly 100% sensitivity and specificity. I think the value of this data is best exemplified with an example. And shown here is a skin biopsy that measures three millimeters in diameter. I hope you can appreciate that for all intents and purposes, this is physiologically normal human skin. It came from the back of a 75-year-old woman. She did have a previous history of skin cancer, as well as self-reported high levels of cumulative UV exposure. But this is normal skin from which we genotype 10 individual keratinocytes. And the first thing I want to point out to you is that many of these so-called normal keratinocytes actually have mutations known to drive keratinocyte carcinogenesis. This trio of cells has a biallelic loss of function NOTCH1 mutations, as well as loss of function mutations affecting p53, FAT1, PBRM1, NOTCH2. And then two of these cells further acquired yet another NOTCH1 mutation. And then one of these cells acquired yet another NOTCH2 mutation. And so if you're keeping count, these cells have up to eight pathogenic mutations in the same cell. Now, if you had done bulk cell sequencing, you might have gotten lucky and detected some of these mutations, but you would have never recognized that they were co-occurring within the same cell. Cancer is thought to arise after the accumulation of four to five pathogenic mutations, and so we were somewhat surprised that these cells can tolerate up to eight driver mutations, and yet that is insufficient to form a discernible neoplasm. Now, the second thing I want to point out is that many of these cells are clonally related. I cannot say with any degree of certainty how these clonally related cells are spatially organized, but we have drawn a schematic here of how they may have theoretically been organized. And this is drawn to scale if you assume that the size of each clonally related patch of keratinocytes is proportional to the number of cells within it. A couple things to point out here. It's perhaps not surprising that the cells with pathogenic mutations have formed a patch, but you'll see that there are also these two patches of keratinocytes without any apparent pathogenic mutations. And not only that, they are of a similar size, and I'll circle back to that point in a couple of slides. Now, the last thing I want to point out to you is the extraordinary range of mutation burdens of these cells. And you might not have initially recognized this because I had to introduce a break in the y-axis in order to plot all this data on the same graph. But these three cells with pathogenic mutations have 40 times as many mutations as their neighboring cells without pathogenic mutations. So what's going on here? What's shown here is more generally the landscape of somatic alterations from individual keratinocytes. And exactly as I showed you in the previous slide, somewhat paradoxically, most keratinocytes actually have somewhat low mutation burdens with relatively low proportions of UV radiation-induced DNA damage. But there's this tail of keratinocytes with extraordinarily high mutation burdens. And when you ask what's notable about them, every single one of these has an underlying pathogenic mutation, typically in the p53 or Notch signaling pathway. So this is another example where had you done bulk sequencing, you would have correctly deduced that the average mutation burden per cell is around five mutations per megabase, but you would have missed the fact that this is actually a mixture of cells with one mutation per megabase and some with up to 50 mutations per megabase. Now, this is a classic example of a mutator phenotype, meaning that these p53 and Notch pathway mutations are causing those cells to accumulate mutations more rapidly than their neighboring cells. But are they doing something else, such as inducing a proliferative advantage? And I'm not so sure that they are. So shown here is the inferred surface area of each clonally related patch of keratinocytes that we could detect in the study. And you'll see here in red, some have pathogenic mutations, but actually most of them don't. And the ones with pathogenic mutations do not appear to be any larger than those without pathogenic mutations. You can go back to the Martin Krainak paper, and they too noticed with a completely different technology that clones carrying driver mutations do not appear to be much larger than those without them. So I'd be curious to hear anyone's take on this, but my take is that these p53 and Notch mutations are conferring a persistence advantage, but not necessarily a strong proliferative advantage, meaning that when a keratinocyte has one of these mutations, it is less likely to undergo apoptosis in the face of high levels of UV radiation-induced DNA damage, and that causes them to accumulate mutations more rapidly. Maybe they'll passively expand in the space occupied by neighboring cells that do die out, but they don't really have a strong proliferative advantage relative to those other cells. So I've brought you up to this point, what I like to call the pre-neoplastic phases of progression. What follows next is the premalignant phase of progression, and in the setting of skin, this is a benign neoplasm known as an actinic keratosis. And in rare cases, an actinic keratosis can further progress into a cutaneous squamous cell carcinoma. And so in order to better understand these phases of progression, we assembled a cohort of informative cases where we could identify a cutaneous squamous cell carcinoma at that moment in which it was ostensibly growing out of a preexisting and still intact precancerous neoplasm, such as this actinic keratosis shown here. We would microdissect and sequence each region, allowing us to infer the order in which somatic alterations accumulated. And much to our surprise, more often than not, these neoplasms were not phylogenetically related. So that's indicated in the scatter plot.You can see that there's these mutations hugging the X-axis here. These are exclusive to the actinic keratosis region. And then you'll see these mutations hugging the Y-axis. These are exclusive to the squamous cell carcinoma. If these were related, we would have seen some mutations in the middle here with high allele frequencies in both the actinic keratosis and the squamous cell carcinoma. So this is what we call a collision. These are just two neoplasms that just so happen to be near each other, but they are not genetically related. But we scraped and scratched and clawed and spent a lot of money on uninformative neoplasms, but eventually we started to find a handful of cutaneous squamous cell carcinomas that did actually evolve from the adjacent actinic keratosis. When these two neoplasms are next to one another, it's only about 10% of the time that they are actually related. And this is one of those examples where you can see here clearly that there are mutations with high allele frequencies shared between both the actinic keratosis and the cutaneous squamous cell carcinoma, as well as some mutations that are exclusive to the cutaneous squamous cell carcinoma. So you can draw a phylogenetic tree based on this distribution of shared and unshared mutations to deduce the order in which they accumulated. And here the actinic keratosis is driven by loss of function NOTCH1 and 2 mutations, as well as a loss of function p53 mutation. This is not necessarily surprising. We know these mutations are common in normal skin, and they would, of course, likely be inherited in actinic keratosis that grow out of them. But we also saw a TERT promoter mutation and a loss of function CDKN2A alteration. These are not so common in normal skin and probably are responsible for the formation of a discernible neoplasm. As for what drove the transition to cutaneous squamous cell carcinoma, that appears in this case to be driven by a loss of function ARID2 mutation, as well as a hotspot mutation affecting the SIBL oncogene. This is a well-established RASopathy mutation, meaning that it is known to activate the RTK RAS MAPK kinase signaling cascade. So that was just one case, but if you look more broadly across all the bonafide cases of cutaneous squamous cell carcinomas that we could identify that did evolve from a neighboring actinic keratosis, these patterns largely held up. p53 and Notch pathway mutations occur early and often and are present at all stages of neoplasia. TERT promoter mutations and mutations that disable the G1S checkpoint, typically loss of function CDKN2A mutations, these seem to emerge in actinic keratosis and are retained in cutaneous squamous cell carcinoma. And then this is not obligate. Cutaneous squamous cell carcinoma does not have to have a MAP kinase pathway mutation or a loss of function SWI/SNF mutation. But when they occur, these most often mark the transition to cutaneous squamous cell carcinoma. Now, because it was so difficult to find these informative cases, as part of HTAN, we're now layering on multi-omic assays whenever we do. And this is one example where we've done spatial transcriptomics on a case where we know that the cancer was phylogenetically related to the pre-cancer. And I thought this was an interesting example because you can see here in blue, there's this band of immune infiltrate that's beneath both the cutaneous squamous cell carcinoma and the actinic keratosis. But these immune cells could not be in more different states. As you can see here, the cutaneous squamous cell carcinoma cells are expressing high levels of checkpoint ligands, and concomitantly, the lymphocytes in their vicinity are expressing high levels of checkpoints, whereas there's not much activity in the immune cells and the actinic keratosis cells above them. And so this suggests that one major tumor microenvironmental change that marks the transition to malignancy is the engagement of immune checkpoints. And our hope is that as we sequence more of these cases, we can begin to connect these microenvironmental changes to pathogenic mutations that coincide with them. So I've brought you up to this point from the evolution of a normal skin cell to a primary cutaneous squamous cell carcinoma, but these neoplasms are not done evolving. Approximately 3% of patients who have a primary cutaneous squamous cell carcinoma will eventually develop metastatic disease. And I think you can look at this number from two different perspectives. On the one hand, this is a relatively good prognosis compared to many other cancer subtypes. On the other hand, this still leads to a lot of sleepless nights where patients are left wondering how aggressively should they treat their tumor in order to avoid being in that 3% who will eventually develop metastatic disease. And regrettably, we do not have high-quality biomarkers to predict who will progress to metastatic disease. And so to address this gap in knowledge, I began a collaboration with a wonderful group from Erasmus University and SkylineDX in Europe. They had assembled what is called the DeSquam study. This is a nationwide cohort of nearly 45,000 primary cutaneous squamous cell carcinomas with detailed clinical and histopathologic information, as well as outcome. And if you study another cancer, maybe this doesn't impress you, but in the United States, we made the regrettable decision to not include cutaneous squamous cell carcinoma in our SEER cancer registry. And so I don't even think this study could have been done exclusively in this country. And we took a subset of these cases for molecular analyses. We constructed this as a nested case control study where half of the tumors or cases defined as primary cutaneous squamous cell carcinomas that over the next 10 years eventually progressed to metastatic disease and half were controls, meaning that they did not progress to metastatic disease. And we made sure that the controls were matched for clinical and histopathologic risk factors, thereby ensuring that if we do see molecular differences, they are likely to be prognostic above and beyond common staging systems. We performed whole exome sequencing and whole transcriptomic sequencing on these tumors.Now, the first thing we did is we established the landscape of driver mutations in cutaneous squamous cell carcinoma. Another regrettable decision was the Cancer Genome Atlas project did not include cutaneous squamous cell carcinoma among its cancers that were profiled. Also, cutaneous squamous cell carcinoma has an extraordinarily high mutation burden, which makes it difficult to recognize driver mutations amidst the numerous passenger mutations within each tumor. And so our field had not yet even settled on a consensus set of driver mutations. So this is now the largest analysis available. We used Intogen to identify genes under selection and propose here this list of approximately 40 genes, which we believe are the main drivers of cutaneous squamous cell carcinoma. But the main advantage of our cohort is that we have metastatic outcome data on these tumors. And so we could then ask, "Are any of these mutations associated with poor metastatic outcome?" And indeed, many were. Loss-of-function mutations affecting SwitchSNF, gain-of-function mutations in the MAP kinase signaling cascade, and loss-of-function mutations in the TGF-beta signaling cascade. These were all associated with poor metastatic outcome. Now, in addition to whole exome sequencing, we also performed the largest transcriptomic analysis of cutaneous squamous cell carcinomas, and after consensus clustering, there were three sample clusters and six gene expression clusters. It was encouraging that the sample clusters did correlate with outcome, one with poor outcome, one with favorable outcome, and one with intermediate outcome. And when you ask what is driving this clustering, there's a few different signals, but the main signal is the differentiation status of the keratinocytes. As a secondary signal, there was a change in metabolism. The favorable outcome tumors relied on lipid oxidation, whereas the poor outcome tumors were more likely to rely on glycolysis. We were somewhat surprised that we didn't see an immune signal. Now, if you look at individual tumors, there are definitely differences in the states of immune cells, but it did not drive, at this higher level, the unsupervised hierarchical clustering. So it's encouraging that we can risk-stratify tumors with this thousand genes, but that's an unwieldy signature if your goal was to bring this into the clinic. And so we distilled the signature down into a simplified version of 23 genes using a Cox net model, and you can see the list of the 23 genes here. Generally speaking, these genes are involved in keratinocyte differentiation, which is not necessarily surprising because this was the strongest signal observed in the unsupervised hierarchical clustering. But the million-dollar question is: Is this better than our alternatives? And the answer here is a resounding yes. When we compared the ability of this 23-gene expression signature to risk-stratify tumors in an independent validation cohort, that's a key detail. This is not just circular reasoning on the same discovery cohort. It dramatically outperformed Brigham and Women's and Health in the AJCC staging system, and even outperformed a more recent EMC model that was developed by my colleagues from Erasmus University that takes into account more clinical and histopathologic factors. Not only that, it works well in biopsies and excisions. That's always an important detail to ask for these gene expression predictors because you don't want a gene expression predictor that is tied to surgical procedure or microdissection technique. So in conclusion, we performed whole exome and RNA sequencing on the largest cohort to date of cutaneous squamous cell carcinomas linked to outcome. I said there was no TCGA project for this cancer, so we're hoping that this will effectively serve as that TCGA project that never happened. We cataloged the driver mutations and found that there are some that correlate with poor prognosis. We also performed transcriptomic analyses showing that tumors with poor differentiation gene expression signatures have poor metastatic outcome, and this is captured quite well with a 23-gene expression signature that outperforms common staging systems. As part of H-Stand, we're now layering on additional omic assays, including spatial transcriptomics, hoping that we can find yet even more biomarkers within this rich data set. So with that, I'd like to acknowledge everyone that made this work possible, particularly Barbara Renteria-Pacheco, the trainee from Erasmus University, who led these studies, as well as my European colleagues, Luz Holestein, Marlies Swakhuis, and Domenico Bellomo. And then for the earlier work on normal keratinocytes, Vishal Tandekar and Delaney Diveltern. So with that, we have time for questions if there are any, and if you do have a question, please come up to one of these mics. Could you explain the mutational phenotype a little better and whether or not it could apply to many other situations, particularly the MSI high phenotype where the MSI is acquired or is a second mutation if one is inherited where you don't know which came first? Yeah. The MSI is driving mutation, the high mutational burden, maybe with a mutational phenotype, is actually producing the second MSI hit. Yeah. So I'd love to do a similar study in the set of normal colonic epithelial cells. But if you forced me to ask what's coming first, the chicken or the egg, I'm going to stay in my lane and say, I think that the p53 and Notch pathway mutations, and this is probably true for the MSI genes too, that those occur first, and then these cells accumulate mutations more rapidly. And in my particular case, I think that's justified because the neighboring cells without the p53 and Notch pathway mutations tended to have low mutation burdens. But it'd be fascinating to do that in a colon. Thank you. Ricky Peters, Fred Hutch. You showed that your signature with 23 genes performed better than the other models, like the EMC, but also you said... I didn't see the difference so much with the EMC, with the other ones, it showed better performance. Yeah. Maybe I missed something. No, you have a very keen eye. In the slide I showed you, there was a slight difference, but what I was not able to convey is that it was also complementary to the EMC, so ideally you would do both. But the other thing with the EMC model is that it cannot be applied to biopsies. It needs to be applied to excisions. It takes into account a large number of factors that sometimes if you just have a small shave biopsy, that you're not able to actually use that model. So we do think that for that reason, that there will be more utility. But I'm a big believer in both the EMC model and the Jebb signature. Heinz-Josef Lenz from USC. I think these data are incredibly exciting. I was wondering because at least in colon cancer, when you have one mutation of KRAS in one cells, it changes the tumor microenvironment dramatically, one out of a thousand cells. Did you have a chance to look at if certain mutations or mutation signatures changes the microenvironment? Because it looks like it's all patches and regionally determined, but is there a field effect on certain mutations, particularly in the MAP kinase signaling pathway? Yeah. As part of HTAN, that's the next frontier. Unfortunately, the assays in the data I showed you were destructive, meaning that we had to take the skin and disaggregate the cells in order to measure the mutations. And so then it's hard to go back and see what was their local microenvironment. But we've cooked up some ideas, I think, that we could in situ see where these mutant cells are residing. So, it's a sharp question. Thank you. Hi. Farah Abdullah from Caris Life Sciences. I know the point of the work was to look at normal versus premalignant versus malignant. How does this or are there any plans you have to compare to hyperproliferative diseases like lupus or reactive actinic neoplasms that are not cancers or precancerous? Yeah. Again, I would love to do that. But I'm probably personally going to stay in my lane, which is skin cancer. But if we can get a phase three of HTAN, let's do it. Yes. Patrick Meeker, Uppsala, Sweden. I'm a pathologist, and therefore, I was surprised that you did not find a relation between the actinic keratosis and the squamous cell cancer. That seems for me, totally surprising and not logic. What is your resume? When I first saw it, I was surprised, but people who get squamous cell carcinoma, they have so many actinic keratosis, and I think that once your skin just gets so sun damaged, it's almost like your whole face is just one big neoplasm, and they're all fighting and competing against one another. I'm starting to see this more in other epithelial tumors. I've been paying attention, for instance, to the pancreatic cancer, HTAN, and they too have seen that a lot of the PanIN lesions that are next to a PDAC don't have necessarily the same KRAS driver. And so maybe that's an important takeaway for those of you that are not necessarily studying skin cancer. If you see a cancer and it is adjacent to a pre-cancer, it's not safe to assume they're related. They might be, but you need to genetically test that. Thank you for the question. Okay. I think we're running up on the question time, but really appreciated all those questions from everyone. So our next speaker is Li Ding, right? From Washington University, who will tell us about her HTAN atlas on prostate and breast cancer. There you go, Li. Thank you very much, Hunter. Okay. Good afternoon, everyone, and glad to see you all here. Great turnout. And today I'm going to talk about from pre-cancer to metastasis: evolution and microenvironment of breast and prostate tumor. Mm-hmm. Going? Yes. First, I will provide a quick overview of genetic drivers of tumor evolution discovered by bulk sequencing. Then I will touch on advances in spatial technology and the computational pipeline for allowing us to study 2D and the 3D tumor evolution and the pre-cancer to cancer transition. Lastly, I want to tell you about our recent work on metastatic breast and prostate tumor in the bone marrow microenvironment. So back in 2010, we launched a study to understand the tumor evolution in acute myeloid leukemia to identify genetic factors driving AML relapse. So we conducted a whole genome sequencing of eight primary and relapse pairs and discovered several patterns of tumor evolution. In this highlighted case here, this small orange subclone in the primary tumor survived the chemotherapy and became the dominant clone in the relapse tumor by gaining additional driver mutations in genes such as ETVC.So the beautiful fish plot became a standard illustration for tumor evolution studies in the later years. So this study, along with others, prompted us to propose a model. Tumor evolution is a stepwise process driven by mutations in the oncogenic pathways, including the introduction of the initial driver mutation, then the addition of cooperating driver mutation, leading to the formation of the initial tumor mass, and then the addition of progression mutation leading to the formation of tumor sub clones. So with the advancement in single-cell technology and the spatial technology, we now can study tumor evolution in 2D and 3D space, and at single-cell resolution. So the Human Tumor Atlas Network was launched in 2018 to build a 3D tumor atlas from pre-cancer to advanced disease. So in October 2024, the H10 Nature Bundle was released. At WashU H10 Center, we analyzed over 130 transcriptomes from six different cancer types to study tumor evolution and the microenvironment interaction in the 2D and 3D space. This work was published as part of the bundle. So from this work and other studies, we formed a consistent bioinformatic workflow for studying tumor evolution. From isolating the tumor micro region using MORF, detecting spatial sub clones using Calico ST and other tools, and studying tumor evolution by using a variety of software tools, including Mushroom, a 3D neighborhood analysis tool. So tumor micro region represent spatial cancer cell clusters that are separated by the stroma component. So a group of tumor micro regions sharing the common genetic alteration, as Hunter was talking about, then will form spatial sub clones. So we were able to detect spatial sub clone mutations using a variety of technologies, including Visium and Xenium. On the left-hand side, we were able to map this LDHB mutation to the green subclone two using the Visium sequencing data. On the right, we were able to map the LDHB mutant probes carrying the mutation to the spatial subclone two again, and the wild type probe to both sub clones. So how about spatial sub clones in 3D? So we build a 3D structure using multiple 2D sections, as illustrated in this cartoon on the left. So we look at the two topological matrices, mostly, connectivity and the loop. On the right, you are looking at an experimental 3D model for breast cancer liver metastasis. So here we are looking at the spatial subclone one in pink with three different volumes, but with maximum connectivity of only three. But now you look at the spatial subclone two in green color. It's a much bigger spatial subclone. It has only single one volume, but with maximum connectivity of eight. So with the experimental tool and the computational tool, now we can study pre-cancer to cancer transition in breast and prostate. You must be wondering, why do we bring these two cancer types together? It's not because we studied breast cancer in phase one, H10, and prostate in phase two. It's because breast and prostate cancers share important biological features, including sex hormone regulation, germline predisposition, physiological function, and tissue architecture. Let me show you. So the loss of the basal layer marks pre-cancer to cancer transition in both cancer types. Now, the question would be, what molecules and features drive this particular transition? So we created a complementary stacked and native 3D data set for studying pre-cancer to cancer transition for these two cancer types. This work was accepted yesterday. Now it's online in Cancer Discovery. You can get access to this paper through this QR code. I'm going to stay here for a few seconds. Okay, so this is an example serial section workflow for a breast cancer sample with both DCIS, the pre-cancer, and IDC, the cancer. So we conducted 44 sections across 460 micron of distance, and then we assayed those 44 sections using a variety of technologies, including Xenium Codex, H&E, Cosmic, and Visium HD. So by expanding this approach to multiple samples for both cancer types, now we can map the pre-cancer to cancer 3D transitional path. On top, you are looking at the DCIS to IDC path for breast cancer. At the bottom, you are looking at the normal gland to high-grade PIN, then to the GP3 path for prostate cancer.So native 3D night sheet can help us highlight the complex structure and connectivity of DCIS that can only be inferred by stacked 3D. So the white color indicate CK5 staining, and the red color indicate pan-CK staining. As you can see, the white DCIS dots are embedded in the sea of the tumor. So by combining the 2D and the 3D approaches, now we can identify progression driver genes for this pre-cancer to cancer transition. We found some of the genes are enriched in the pre-cancer, some of the genes are enriched in the cancer. So those genes are involved in a variety of processes, including tissue integrity, ECM remodeling, metabolism, cell-cell interaction, and so on. So MGP, matrix gla protein, has been found to be highly enriched in DCIS, the pre-cancer, at both transcript and protein levels. MGP binds directly to calcium and the BMP2, preventing pathological calcification of breast tissue. So this function maintains the flexibility and the integrity of the ECM, therefore preventing transition to invasive morphologies. So not surprisingly, when we lock down on MGP using shRNA, using a DCIS cell line, we found enhanced proliferation and wound healing. So ERG has been overexpressed in a subset of prostate cancer, as we all know. ERG overexpression here has been found in morphologically benign luminal epithelium cells. In blue color, you can see the ERG expression here. It's not surprising to see high ERG expression in high-grade PIN and invasive cancer. What does this tell us? This suggests TMPRSS2-ERG fusion could happen way before we observe morphological changes under the microscope, as Hunter was indicating for skin cancer. So in summary, we have identified unique and common genes and the TME features associated with pre-cancer to cancer transition for these two cancers. So the common one includes EPHA6, GDF15, FASN, and MYC. So due to time constraint, I will only be able to discuss TME association with cancer transition, but you can get it from the QR code. So I want to use the rest of the time to talk about breast and prostate bone metastasis. So in addition of sharing the common biological features, these two diseases tend to metastasize to the bone. 90% of prostate metastasis are bone metastasis. To address this, we build a breast and prostate spatial cohort for studying bone metastasis, including both primary tumor and the metastatic tumors. So I want to draw your attention to this linear plot here. So, we can observe abundant tumor cells in pink color, and we can observe abundant fibroblast and macrophage in light purple and blue color, and the robust vasculature in the red color. So when we compared the breast bone metastasis to its matching primary tumors, we found a subset of bone metastasis show increased expression of this gene called ACKR3. Another name is CSR7. It is a regulator of metastasis by interacting with CXCL12 to facilitate the tumor cell homing to the bone. We have also found FGF receptor expression in both the primary tumor and the metastasis, but with enhanced expression in the bone metastasis. So how about prostate bone metastasis? We found enrichment of AR-driven program as expected. You can see highly expressed AR in the bone metastasis in comparison to the primary tumor. We also observed FASN highly expressed in bone metastasis, enzyme involved in metabolism. And not surprisingly, we found the canonical markers such as PCA3 got lost in the bone metastasis. So we are intrigued by the key steps of cancer metastasis in the bone marrow microenvironment, including homing, dormancy, colonization to expansion. So when we compared the breast and the prostate bone metastasis to the normal bone and to MPN, a hematological disease, we found the normal hematopoietic niche are replaced by a tumor-associated microenvironment. I want to draw your attention to the green box here. You can see GMP, megakaryocyte, MSC dramatically go down.This is replaced by this tumor-associated microenvironment, which I will discuss in detail. So we observed a significant increase in osteoclast in breast bone metastasis and osteoblast in prostate bone metastasis as expected. So then what's happening at the molecular level? So we found two very strong interactions between tumor and osteoclast may contribute to osteoclast activity in the breast bone metastasis. PDGF-beta and its receptor can promote osteoclast precursor recruitment, and TGF-beta I and ITGB1 can also stimulate osteoclast activity. So when we look at the prostate bone metastasis, we found SPP1 and the CD44 interaction and the BMP signaling likely can strengthen osteoblast niche in the prostate bone metastasis. So the SPP1 and the CD44 interaction can enhance matrix adhesion, and the BMP signaling can lead to osteogenic remodeling and bone-forming activities. So how about the vasculature and the fibroblast? We observed a significant increase of the vasculature in both bone metastasis, as indicated here by comparison to normal bone and the MPN. We also observed a tremendous increase of fibroblast in prostate and breast bone metastasis. Most of those fibroblasts are MYCAF. So we found in addition to the well-established VEGF-A and the KDR interaction for promoting angiogenesis, we found that FGF signaling is also very strong in breast bone metastasis, as indicated by the high expression of FGFR2 in the tumor. We believe this strong interaction might promote the expansion of the breast bone metastasis. So again, we observed this VEGF-A and KDR interaction in prostate bone metastasis. In addition to that, JAK1 and the Notch 3 and 4 interaction seems to be strong as well, and this particular interaction can promote angiogenic activation and vascular remodeling. So now let's switch to talk about immuno cells. We observed increased macrophage and limited T cell infiltration in breast and prostate bone metastasis. So when we look at the T cell stage, we found cytotoxic T cells, although with exhaustion markers, are more abundant in breast than prostate bone metastasis. We observed comparable level of Treg between these two. So to fine map the TME and the tumor relationship, we developed a density-based spatial analysis strategy to locate the immuno and the stroma cells from the peripheral of the tumor to the core of the tumor. So we found the increasing macrophage from the peripheral to the core and the decreasing T cell from the peripheral to the core. When we dive deeper, we found that breast bone metastasis were enriched with interfering gamma positive macrophage, suggesting a inflammatory microenvironment. On the other hand, prostate bone metastasis show the accumulation of SPP1 positive macrophage suggesting a matrix remodeling and bone associated immunosuppressive microenvironment. With that, I want to share with you a few thoughts. Our study suggests potential new metastatic tumor targets including FGFR2, fascin, and others. Our study might also indicated targeting some new interactions between tumor and the vasculature could be a possible options. In addition to the VEGF and the VEGF receptor interaction in breast and prostate bone metastasis, we found additional niche specific signals such as JAK1/Notch 3/4 interaction. A combined anti-angiogenic and a macrophage-directed strategy is also worth considering. Finally, immunoblocket combined with macrophage-directed strategy could represent another path forward. Breast bone metastasis harbor an inflammatory immune responsive niche while prostate bone metastasis offer a immunosuppressive microenvironment. Breast bone metastasis may be more responsive to immunotherapy, checkpoint blockade, and macrophage reprogramming, while prostate bone metastasis might require strategies for targeting SPP1 positive macrophage for improvingimmunotherapy efficacy. With that, I want to thank my close collaborators, Feng, Ben, Ryan, Russell, Cassie, for their close collaboration for many years for making this work possible. I want to thank my lab members for their contribution to this study and to many other work going on in the lab. Finally, I want to thank NCI Human Tumor Atlas Network, H10, and everyone here, Shami, Shannon, and my colleagues here, Ying Du. And I want to thank CPTAC, PECDS, and the Siteman Cancer Center at WashU. Thank you. Hi, Ding. I think the data are amazing. I'm a GI oncologist, as you know, but I wanted to share with you some data we have generated on CSF1 receptor as well as CD47. As you know, in colon cancer, immunotherapy does not work in liver metastasis, but we have shown that CD47, when you use that as well as CSF1 receptor, you may overcome immune suppression. So what you are seeing may be also applied for the tumor immune suppressive environment in liver metastasis. We have also data on FGF receptor 2 in colon cancer, which showed highly predictive and prognostic features for outcome. We have never looked for the association for bone mets because it's very rare in colon cancer, but it's certainly linked to the efficacy to each FGF receptor inhibitor. So there are certainly interventions which have direct implication for treatment, at least in colorectal cancer. Yeah, that's fabulous. Thank you for the comments. That's really tell us about the commonality of different cancers. Have you observed this interaction happening in the colorectal cancer liver metastasis, although it doesn't go to the bone? So there are data using a CD47 antibody that- Uh-huh ... actually works in liver metastasis in combination with immuno. That's fantastic. Thank you. Hello. A very interesting- Hi. Oh, sorry. No, go ahead. Oh, I can go. So thank you very much, Li. Wonderful talk. Thank you. I was wondering, for those of us that are designing spatial studies, should we do 44 different patients in 2D or the same section 44 time-- I mean the same sample 44 times and build a 3D model? What do you think for a single sample? Where do you think we're going to learn the most biology, through spreading into the across many patients or building a huge 3D model of a single sample? I sense there are pros and cons for both, and depending on what's your study design, if I'm trying to figure out what are the common drivers, either tumor intrinsic or TME drivers for a particular disease, I probably will do the 2D work spread out for multiple samples. But if you are trying to understand the cancer transitions, like what we are trying to do here, a 3D model will be extremely helpful. Thank you. No problem. Hi. Really interesting result on your comparison between the gene expression between the metastatic and the primary. Mm-hmm. You showed that-- I remember in the breast cancer, it seems it's quite heterogeneous. Some patients upregulated ACPR3, other patients upregulated FGFR2, and still there are other patterns. But for prostate cancer, it seems quite homogeneous, or like everyone upregulated the ER. And do you have explanation? That's a great observation. You have really good eyes. And so indeed, we see more heterogeneity in the breast cancer bone metastasis, although actually for the breast project, we had the matching primary tumor. And all of the breast tumors I'm talking about here, they are ER-positive breast tumor. So it's not like we are using different subtypes of breast tumor, but we did observe this tremendous heterogeneity. And for prostate bone metastasis, actually, we have included both hormone sensitive and the castration-resistant prostate bone metastasis. We actually were expecting higher heterogeneity from the prostate cohort, but we see uniform expression for the genes I'm showing here. But when you look deeply, though, when you look at the interactions I'm showing you, I mean, I'm just showing you a subset of the interactions, then you see tremendous heterogeneity in prostate bone metastasis. Thank you. No problem. Professor Ding, thank you for your wonderful talk. The bone-directed therapies are commonly used in these two type of cancer, such as Rank Ligand or RadioLigand therapy, radiotherapy plus. So how would you speculate the changes in the microenvironment TME or the tumor and these genomic signatures if the bone receive bone-directed therapies? How would you speculate the dynamic changes of these? Thank you. This is a tough question. There are always a lot of different possible combinations for treating bone metastasis because those patients have already developed a primary tumor and then develop metastasis. So they have already gone through a lot of therapies. At this point, I think a single strategy would not be... possible. And as the gentleman just indicated, how heterogeneous are breast tumor, breast cancer bone metastases are. So I think what you suggested, plus some of the options I'm proposing today, could all be used depending on patient-specific genetic alterations and the patient-specific microenvironment. Thank you. Two quick questions and two quick answers. Okay. I will try to be short. Thank you for your talk. I had a question about thinking also about the adipocytes within the bone marrow space. You show a lot of metabolic markers, like FASN, you show FGF2. So I was wondering if, looking at the interaction, you would also consider looking at the interaction between the tumor and the adipocyte, which is actually promoting and providing the lipids for the tumor. Yeah, that's a great point. Although we did see FASN, and to be enriched in invasive primary tumor as well, but I think, yes, the bone marrow microenvironment might further promote the overexpression of those genes. To be honest, we are still at the beginning phase of analyzing the data. What you suggested definitely will be the next step forward. Thank you. Yeah. Great talk. Thank you. Thank you. Matt Leventhal from Dana-Farber. Could you comment on the fact that, and I'm just curious what your perspective is, on prostate cancer being seen as a slow progressing disease, but AR is highly expressed and amplified throughout the life- Yeah ... even pre-cancer. So, what do you think is going on with AR going from the pre-cancer state to the metastatic state? I need to answer that quickly. So, yes, AR has been critical all the way, but enhanced AR pathway really allow the bone metastasis to keep going and that's just the nature of the disease. And I will stop here because I'm way over time. Thank you for the question. Thank you. All right. Thank you, Li. I'd like to invite our last speaker, Irene Ghobrial, who is PI of the Multiple Myeloma Precancer Atlas from Dana-Farber Cancer Institute. All right. Thank you. I'd like to thank, of course, AACR for this wonderful time and, above all, NCI and H10, because this has been really a wonderful collaboration between all of the H10 members, and we've been learning a lot from all of the different cancers. So I'll take you through our multiple myeloma atlas and really understanding the transition from precancerous MGUS and smoldering myeloma to overt myeloma and how we can prevent it. These are my conflicts of interest. So if you think about multiple myeloma, it has two well-known transitions of precursor conditions. One is called MGUS, monoclonal gammopathy of undetermined significance, and the other one is smoldering myeloma. And they're very different because MGUS is very common in the general population. Anyone over the age of 50, it's about 3% to 5% of us walking around with it, and it increases more and more as we get older. Smoldering myeloma, which is defined by having 10% plasma cells or cancer cells in your bone marrow, is the next transition of that. And in these patients, their chance of progression to overt myeloma to end organ damage is much higher. So instead of being only 1% per year, now it's 10% per year, so 10 times higher. And in some of those patients, when we see them, they have something called high-risk smoldering myeloma, meaning that they will go on to develop end organ damage or overt myeloma, 50% chance in two years. So when I sit down with a patient and I see them and I tell them, "You have high-risk smoldering myeloma," I'm basically telling them, "You have a 50/50 chance of getting fractures in your bones, kidney failure, anemia in the next two years." And yet we do nothing about it. So we had a question, and part of the H10 was this: Can we be more precise in our definition of who is high risk in the population, and who are those patients who we should be treating early, while others who may never progress in their lifetime, we should leave them alone? And currently, the markers that we use are basically very clinical markers. How much tumor burden you have, how many plasma cells you have in your bone marrow, are they 20%? How much M spike you have, which is the protein we secrete in the blood or light chains. And these are really only tumor burden markers. They don't tell you anything about the biology. They don't tell you anything about the immune system or genomic alterations. So we asked the question, should we really intervene in those patients early? And we're trying to look basically like the solid tumor people, like breast cancer. In breast cancer, we screen early and we use mammography or colonoscopy in colon cancer. Yet I can tell you in multiple myeloma, a blood disease, we can use a very simple blood test called serum protein electrophoresis, and it's much easier to screen by a blood test than doing mammography or colonoscopy. And even when we find those patients with high-risk smoldering myeloma, which is very much like a malignancy, like a Stage 1 or Stage 2 breast cancer, we tell our patients, "Watch and wait. Do nothing until you have metastasis everywhere, basically." Just like metastatic breast cancer. Wait until you have fractures, bone disease, or kidney failure, and then I treat you. And that's a problem. That's basically saying to a patient with multiple myeloma, "Wait until you have advanced disease, and then I treat you." So the way we were thinking of this is can we diagnose those patients early and potentially treat them early and shift the way of thinking into an early interception?To do that, you need to be more precise in your risk stratification. So I'll show you some of the work that we've been doing with our H10 collaboration here. And the first one was trying to understand the genomic alterations that can be predictive of progression in our patients. So here we took 218 patients with MGUS and smoldering myeloma and compared it to patients who have already gone to overt multiple myeloma, and we did whole genome sequencing. And this is hard to do from peripheral blood or bone marrow cells, especially when you have very small numbers of cells. But we were able to find those alterations, and then we put them on a scale. We wanted to make this something that's clinically relevant for our patients, so I can tomorrow go to the clinic and tell them, "Here is your genomic score." And we call that the MM-like score. And it's a simple score of how many you have that are more MGUS-like, which is only 14, 16, or 14, 20, and how many are myeloma-like, which, as you see here, NRAS, KRAS, 1q amplification, MYC alterations, and so on. And you balance them, and if you have more of the one, then you're more myeloma-like, and if you have more of the MGUS-like, then you're actually MGUS-like, even if you have a high tumor burden. And that's a critical one that we can now use in the clinic. But we said, "Well, that's not really using single-cell level information, and that's also not looking at the transcriptional changes that occur in the patients." We know that many patients may have a 14, 16, or a 1q amplification, and yet they don't progress to multiple myeloma. There is something more beyond the genomic alterations of those patients. So we decided to do single-cell sequencing with BCR sequencing so that we can truly look at the malignant plasma cells versus the normal plasma cells. And here we looked at 1.7 million cells from 226 patients. And this was a collaboration with Gary Guss at the Broad Institute, where we tried to ask the question transcriptionally, what's happening in those patients? But because those are the same patients, we also did whole-genome sequencing. Another question was, what can you define beyond the DNA level and beyond genomic alterations? Now, this was a hard thing to do because we're now comparing tumor to tumor, but also bone marrow plasma cells versus peripheral blood circulating tumor cells, and we're also looking at normal cells versus normal cells. And you can ask a lot of the questions here. The first one we wanted to do is, can we look at those transcriptional changes that occur during transition from a normal plasma cell to an MGUS cell to a more malignant transformed smoldering myeloma or myeloma cell. And you can see that we found 16 transcriptional signatures that were defined in those populations, including cell cycle regulation, as you probably expect, MHC Class II, checkpoint regulation, and so on, circulatory capacity. But then we took those and went and tried to see which ones truly predict for you progression to multiple myeloma. And we found that six of those signatures were very critical for that progression from MGUS to smoldering myeloma to overt myeloma. And this is very critical, especially beta-2M, which is in the MHC Class I. And you can tell that this is critical because we use beta-2M as one of our predictors of progression in our patients, and the ISS classification of multiple myeloma. Now, we further validated that in the COMPASS data. This is a larger data set that has overt multiple myeloma from MMRF, and we indeed found that in the same signatures were found even if you have overt multiple myeloma. So this is not just the precursor conditions, but can also be applied in patients with overt myeloma, even when they receive therapy. Now, the next question we said is, well, that's good enough to have signatures, but let's try to find a classifier that can predict for me, if I take a patient with smoldering myeloma and I looked at their cells, I can tell where they are in their transition from normal to MGUS to smoldering to myeloma because the way that we classify our patients clinically may actually be wrong. I may be classifying some patients who have MGUS that truly have more smoldering myeloma in their transcriptional biological changes and vice versa. A patient with 10% plasma cells, so I call them smoldering myeloma, but they will never progress. They're behaving more like MGUS. So can I take your cell and by single-cell signatures, tell you where you stand in that transition period, and can I put it in the axis of progression? So here what we did is we tried to do a malignancy score and a disease score. Malignancy score just tells me, are you normal or a malignant plasma cell? And with 100% accuracy, we can tell at the single-cell level if this is a normal plasma cell or a malignant plasma cell. Then the next question is, what is the malignancy score as a disease progression from MGUS to smoldering to myeloma? And here we actually classified it from the MGUS and myeloma samples and then came back and looked at it in the smoldering samples. And we indeed found that you can create a score, we called it the MST score, that can predict for you progression, and it correlates, of course, with disease burden, with the light chain, with the M spike, but it can actually predict for you progression, not only in the MGUS and smoldering myeloma and myeloma patients, but also again in the COMPASS database, which is the overt myeloma patients. And that classifier is critical for us because again, beyond the DNA data, we can use it now in the clinic as a classifier to tell us at the transcriptional level what's happening to the patient. Now, the next question we asked is, can it add anything more beyond the DNA level? I already know that my patients have genomic signatures, which is the MM-like score. Should I add on it the RNA data, or is just reflecting what happened at the DNA level? And here what we did is we actually removed the DNA information because we found very weak correlation between RNA and DNA from the same patients, and we went back and said, "What is the transcriptional change that is occurring above and beyond what you have at the genomic level from the DNA mutations?" And you can find that the circulatory capacity, the ability of the cells to circulate in the peripheral blood, as well as histone chaperone proteins and genes, were critical for that transition, which tells you that there is a lot of work to be done beyond the DNA level to really tell us why those cells become more malignant, even in the early MGUS stages in our patients.Because circulatory capacity and because the CTCs, or circulating tumor cells, are so prognostic in our patients, and we know that by flow cytometry, we went back and said, "Then let's do a larger cohort of single-cell sequencing in the peripheral blood so that we can really define what's happening in the bone marrow, but only in the peripheral blood." Can I just use a blood sample from you and tell you if you have the genomic alterations that are in the bone marrow? And potentially, we can also stop doing bone marrow biopsies in our patients because they're very hard to do, and patients don't like to do it. So here, this is work that we did from 74 patients and 29 healthy control, where we looked at both the bone marrow biopsy and the peripheral blood over time, again, from healthy MGUS, smoldering, and myeloma. And what we did is, again, with 1.2 million cells, looking at single-cell BCR sequencing and transcriptional changes, looked at the tumor versus normal and peripheral blood versus bone marrow. But we asked a different question here. We said, "Can I predict what happens in the bone marrow on the FISH or whole genome sequencing data in a peripheral blood sample?" And here you can see with this confusion matrix, we were able to predict everything that you see in a bone marrow biopsy by FISH or whole genome sequencing in a peripheral blood sample, even in some of those samples that had some alterations that are not detected in the bone marrow. It was because they had a couple of clones that were different in the peripheral blood. And in fact, you can do that with only 10 cells. And that was the SwiftSeq method that we developed to try and predict progression in our patients, but also to give you value in a blood sample with 10 cells in the peripheral blood. Now, it's critical for us to think, as you always do, that tumor cells don't live alone, and they have an ecosystem of immune cells, non-immune cells, and the whole microenvironment. So we asked the question, can we look at the immune cell signatures, again, across time from MGUS to smoldering to myeloma, and what happens with an early myeloma cell interaction with immune cells, and can we harness that in the future for developing therapeutics, especially immune therapeutics, for our patients with smoldering myeloma? So here we took 877 tumor samples as well as immune cells and tried to do a large atlas, again, of immune cell signatures across the spectrum, healthy, MGUS, smoldering, myeloma. But because we also have the single-cell sequencing of the tumor cells, now we can look at the interaction between both of those types of cells. So we don't have time to go over all the data, but I'll show you some of the compositional and transcriptional changes. And over time, you can see that there is a lot of compositional changes that occur, including CD4 naive T cells, central memory T cells, including Tregs increasing over time as you go to myeloma, CD16 monocytes, and you have less plasmacytoid dendritic cells and CD14 monocytes. But what was interesting, because we have the tumor cells, we found a strong correlation between the cycling of the tumor cells and the granzyme B CD8 TEMs. And that was showing us that indeed, patients who are progressing from smoldering myeloma to overt myeloma are the ones who have more of those granzyme B CD8 TEMs, sort of trying to respond to the tumor cells and trying to respond to them in a way that is basically exhausted despite being effector cells. If you look at the immune programs in those and actually try to look not only at the compositional changes but at the signatures across all of the immune cells, you find that interferon signaling was present not only in the tumor cells but across many of the immune cells, T cells, myeloid cells, and so on, as well as other signatures that were correlating with myeloid suppressive cells. And the other thing that we did was T cell receptor clonality, and we tried to look at the TCRs in those patients. And what we found is that as you progress to overt myeloma, you have less and less diversity because you have very specific sub-clones of T cells that are responding potentially to those tumor cells. And we're now trying to look at all of those TCR clones that are expanding and responding to the tumor cells. Now, what else are we doing with the spatial question with the H10? Here we're trying to take all of this information and take it at the spatial transcriptomic and proteomic level. And to understand that, as we know very well, the bone marrow biopsies that we do in our patients are very hard to actually look at them because we do decalcification, which actually messes up all of your DNA and RNA, and then we put them in paraffin-embedded, and then we try to say, "Let's try to do what the solid tumor people do." But it's very hard to do that in bone marrow biopsies. But bone marrow biopsies are very critical because all of the data I showed you in bone marrow aspirates, which is what we pull first from the first pull of our patients, does not reflect everything that's happening. It's not just the spatial localization of the cells and how they interact with each other. You're also missing a lot of the stromal cells, adipocytes, as we talked about, endothelial cells, and so many other architectural changes that occur in the bone marrow. So we spent a lot of time trying to develop a spatial proteomic method, and here we used IMC, which is imaging mass cytometry, and we developed 36 antibodies that would work on IMC that we validated by immunohistochemistry and immunofluorescence to make sure indeed they work on a decalcified bone marrow biopsy. And you can see here that we have CD3 and CD138, then you can classify it as T cells and myeloma and so on, and finally have enough signatures of all of your tumor cells and immune cells to ask those questions. So when you look at all of this data, we started seeing that adipocytes, actually, because you lose them a lot when you're doing all of those cuts and decalcification in your bone marrow, we couldn't even call the adipocytes, and we couldn't call megakaryocytes because they're too big. So we had to develop a new package, which is available, called the AdipoFinder, where you can actually recapitulate the data of the adipocytes, and now look at all the cell-to-cell interactions in those patients. And we decided to look in a study of 98 samples of CAR-T, either cell or silta cell, so patients with multiple myeloma who received the cellular therapy, CAR-T. If I take their sample before going on the CAR-T therapy, can I predict whether they will respond to therapy or not? And can I predict their toxicity? These are very critical questions for us clinically.So the first thing we did is we started to look at those cellular neighborhoods, cell-to-cell interaction between myeloma cells, immune cells, and others. And it's very important to show that indeed it's not just any change that occurs in the bone marrow. The cells actually do sit in very specific cellular neighborhoods, and it's very similar to what we see in solid tumors. So when we look at those cellular neighborhoods and put them together, we start seeing indeed that some of those neighborhoods are actually predicting not only a progression-free survival to CAR T, but also overall survival in our patients. And many of those signatures were not just about tumor burden, which is in red here, the plasma cells or the tumor cells, they were also about the mixture between the tumor cells and the immune cells. So here is an example of how a tumor and immune cell interaction occurs, and sometimes it gives you a very bad prognosis, and sometimes it gives you a very good prognosis, which are very different depending on the types of myeloid cells or tumors to immune cell interaction that are sitting in that cellular neighborhood. And finally, we said it's not just about response to therapy, but also the cytokine release syndrome and the prolonged cytopenias that we see with our patients. Many of them may not recover their counts for a long time. Can we look at that inflammation that occurs in the microenvironment, and can we predict with accuracy that those patients will have potentially that inflammation, CRS, or prolonged neutropenia because we can potentially prevent it? And indeed, you can find that specific interactions between myeloid cells, and dendritic cells can predict for us prolonged neutropenias in our patients. So finally, why do we do all of this? We do this because we want to treat our patients. And I can tell you finally, we have the first FDA-approved drug for smoldering myeloma, which is daratumumab. But that's the beginning, not the end. And in fact, we want to actually improve things by using immunotherapy in our patients with smoldering myeloma. And to do that, we've done two clinical trials as a proof of concept that indeed you can make a difference by treating early and by harnessing the immune system with cellular therapy or bispecific antibodies. So here with a bispecific antibody called teclistamab versus LenDex, which is our traditional way of treating smoldering myeloma, we compared those patients and treated them to see if truly our hypothesis is true that your immune system is better in the earlier stages, that your tumor burden and your clonal evolution is not as extensive as it is in overt myeloma, then you should probably have an amazing response if you use your best therapy, immunotherapy, as early as possible. So with that, we showed the first data of the first 12 patients, but the study is completed, so hopefully we will show the rest of the data soon. But this shows you the first proof of concept that a bispecific antibody, a T-cell engager, showed us 100% response rate, 100% MRD negative disease, which means there's no evidence of tumor cells in the first 12 patients, and with less infections and less cytokine release syndrome compared to the overt multiple myeloma. And here is one of our patients. This was patient number two, and you can see his light chains were significantly high. And after just receiving the first few doses of teclistamab, you can see complete resolution of his lambda light chain, and it's still in an MRD negative disease three and a half years later. And he sent me this picture when he was bungee jumping in New Zealand, so I'm using this as the best quality of life measure for patients who are on a bispecific antibody. Now, the other study, which was also the hardest thing to do, was can we convince the FDA and pharmaceutical companies to give us cellular therapy, CAR T therapy, not in the relapse refractory setting, not in the newly diagnosed, but in smoldering, in asymptomatic patients. And I can tell you this will be a plenary session Monday morning, so please come and visit and hear us talking about it, of 20 patients who Omar Nadeem will be presenting with CAR T. But we showed the first six patients a couple of years ago, or actually last year in the American Society of Hematology. And this is a teaser for what you're going to see Monday. But again, 100% MRD negative disease with one and done cellular therapy, which tells you again that the proof of concept matters if you use your best therapy or immunotherapy and you use your T-cell engagers early on. So with that, I hope I took you through a journey of precursor multiple myeloma. Can we diagnose them early? Can we risk stratify them with genomic, with immune, with single-cell sequencing, and spatial transcriptomics and proteomics to really define those patients who will benefit from therapy? And then can we apply our best therapeutics, especially immunotherapy, in those patients to potentially intercept and cure our patients with myeloma? With that, I'd love to thank, of course, our lab members, clinical members, all our funders, Gaddy Getz, who all our work we do it in collaboration with him. And above all, our patients. Thank you. Yeah, thank you for the talk. It was super nice. Well, you said that you had BCR data for the single-cell analysis, and as far as I know, this premalignant condition can be kind of oligoclonal from the BCR level. Have you looked at it? Do you observe this oligoclonality? And if you do so, do you see differences in these scores from progressing to myeloma? Yeah, great question. So yes, all of our samples have BCR sequencing. And in fact, the more you go towards malignancy, the more there is a major BCR clone. And this is true also when we do VDG sequencing for adaptive. You may have other subclones, smaller subclones, but most of the time it's a major one clone. But you're right. And we have other studies that I didn't show here called MGAP, which is the very early event of a change of monoclonality, and you do see those oligoclonal cells, and they compete with each other until one of them actually becomes the major clone. This was not a journey, this was a rollercoaster. It was incredible. I know that multiple myeloma is also classified by significant hypomethylation. So we heard about transcriptomics and DNA. What is methylation play a role in the characterization and outcome for multiple myeloma? Yeah, great question. I think we are deficient in a lot of the epigenetic regulation, methylation, and so on, and we do definitely need to do more. There has been many studies, but not enough and not at the single-cell level. So absolutely, we need to do that. Okay. If there are no more questions, I want to thank the speakers for keeping us on time and for these great talks, and thank you, audience, for all of your questions. If you have any questions that you were too shy to ask, you can feel free to come up and introduce yourself afterwards. That closes today's session. Thank you.