Skip to content

2026-04-21-pm-dissecting-hematologic-malignancies-single-cell-spatial

Date: 2026-04-21 Primary topic: single-cell-spatial-omics Cross-referenced from: bioinfo-tools, single-cell-spatial-omics, virtual-cells Word count: ~11,126

Auto-generated captions

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

Let's start the session. Thank you for coming to the session, Dissecting Hematologic Malignancies at Single-Cell Resolution in Spatial Omics. This session will be chaired by me. My name is Mamiko Sakata-Yanagimoto, University of Tsukuba, Japan. Hematologic malignancies, the tumor growth is supported by a complex interaction between tumor cells and the microenvironmental immune cells and stromal cells. So in this session, we will discuss how spatial analysis will visualize and dissect the microenvironments of hematologic malignancies, especially focusing on lymphoma. So today, we'll have three speakers. The first speaker will be Dr. Rong Fang, Yale University. So Rong, please start your presentation. All right. Okay. Thank you for kind of staying around for the last day of the AACR. And so what I'm going to talk to you today is, I call lymphoma transformers, because of my close collaborator, literally, the kind of co-PI driving the entire project, Mina Sue, and she loves Transformers so much. And so, okay, now, the story is about the transformation of lymphoma from low-grade or indolent to higher grade or often large B-cell lymphoma. But they're kind of like transformers. We try to catch them in a space. All right. And so this is my disclaimer. And so again, Mina, and I was really inspired by her vision, also sometime her demanding questions. One is, "You can do the spatial transcriptome and protein and the spatial epigenomics, and can you really look at the same tissue I'm examining every day for diagnosis of patients, right? And so often, those tissue specimens are in paraffin blocks, and can you really make your spatial omics sequencing fully compatible with FFPE tissue blocks?" And so I think that's a kind of good question for us to think about as technology innovator, but that also kind of reflects the desire by our pathologists to learn much better molecular information and really help our patients. So over the past ten years, we have witnessed the emergence of the whole field called spatial omics, and twinned the spatial RNA profiling called spatial transcriptomics, which was chosen as the Method of the Year by Nature Methods. And two very nice review articles written by Professor Xiaowei Zha and Professor Joakim Lönnerberg, pointing out two distinct approaches towards spatial transcriptomics. One is by imaging, the other is high-throughput NGO sequencing. And we were probably among the first groups thinking, "Ah, can we do beyond just the RNA transcriptomics?" And really venturing into multi-omics across the central dogma of molecular biology. And so in recent years, they've become even more exciting. For example, Professor Matthias Mann, Max Planck Institute, developed the Divap, which is one of the few examples nowadays you can do kind of unbiased proteomics using mass specs. And now you have other tools, not just the imaging, NGO sequencing, also have mass spec in play for spatial omics. And also, can you connect the omics to functions, for example, metabolic activities? We have been interested in this question for years, and our approach is the collaboration with my colleague, my friend here, Lin Ye, UC San Diego, utilizing optical chemical imaging such as stimulated Raman scattering microscopy to map out different metabolic activities. In the same tissue, we can do protein RNA or epigenome, and you can truly connect the omics to what the cells are actually doing. And so this is kind of a roadmap of the sequencing-based spatial omics. I don't want to go into the details, but the beginning of this journey is really the landmark paper published by Professor Joakim Lönnerberg back 2016, using DNA microarray to capture transcripts, and now that's 10X Genomics. And so we came up with a somewhat kind of opposite approach. Instead of capturing within the DNA barcodes into the tissue, such that we can barcode transcripts, proteins, but also epigenetic features. So we demonstrate the pretty kind of diverse applications of this spatial deterministic barcoding to look at, in particular, spatial epigenomics and also spatial multi-omics, including chromatin accessibility, histone modifications. And recently, my former post-doc now a professor at Penn, further demonstrated you can also map out DNA methylome spot by spot across the entire tissue as well. Okay, back to RNA. And as I mentioned, I think that's the beginning.Of this revolution of spatial omics and RNA or transcriptome remains to some degree, still in the mainstay of the whole technology landscape. But if you Google, or now you have better tools, ChatGPT or Gemini, to ask, "Okay, what is spatial transcriptomics?" Basically, two keywords. One is RNA, and the other, of course, spatial, but not much about what type of RNA, just RNA expression. But if you're a molecular biologist, you know that's not a full story at all. And actually, even the same messenger RNA molecule has such a dynamic life cycle, starting from transcription to the further maturation to different isoforms, and some going to give you the proteins, and then the degradation. This process is tightly regulated by many non-coding RNAs, actually, that constitutes 95 or even more than 95% of the RNA molecules. And unfortunately, most of the spatial transcriptomics technologies do not tell you anything about that. I really think that they are the most important part of RNA biology, but the expression of different messenger RNA transcripts are largely just the outcomes. Then we started to ask, can we look at all kinds of RNA molecules and understand this RNA regulation, RNA regulatory network, but in a tissue Mina looking at every day? The FFPE tissue block for clinical diagnosis. And so the little trick we came up with was actually originally developed by Professor Steve Quake. Instead of you capture messenger RNAs based on their own endogenous polytail, you can use an enzyme to add a polytail to whatever RNA molecules, RNA species, but also to the fragment of RNA. In that context, turns out that's a good thing, even if in FFPE, the RNA molecules are very fragmented, and you can add many polytail to those short fragments. As a result, you'll actually get an even better detection sensitivity compared to fresh frozen tissue spatial transcriptome. But what's more exciting for the study of RNA biology is now you can see many different RNA species, not limited to protein coding messenger RNAs anymore. So now you see link RNA, microRNA, and tRNA, and snoRNA, sRNA, and so on and so forth. And for example, tRNA can further quantify the different amino acids associated with tRNA that very specific amino acid associated tRNA. Now you can see spatially they are different. And probably no one ever noticed this because no tool existed in the past to allow you to map out the tRNA distribution at this resolution. It turns out tRNA is the first small non-coding RNA discovered and recognized by Nobel Prize. And so back to the key topic and the core topic I want to discuss today, lymphoma transformers. So I've taken the clinical tissue specimens, as I alluded before, and performed this type of spatial total RNA sequencing. This is a patient who suffer from stomach for two months and no clear reason came to Yale, and a couple of biopsies taken from the stomach, and Mina look at the tissue specimens. And her diagnosis by looking at the histology is, okay, the one on top probably already lymphoma, but relatively benign, and the one on the bottom has a lot of large B cells, probably already diffused large B cells, but from the same patient. And that might really indicate a transformation from this low-grade lymphoma to diffused large B-cell lymphoma. And so it actually took a lot more than just looking at that histology image to do diagnosis. Mina order a lot of molecular pathology tests, and up to 10 different IHC and a gene arrangement test. Altogether, actually, it's pretty expensive, and also take quite a bit of time to get the results back to give definitive diagnosis. It's a relatively easy case. If that's a complicated, tricky diagnosis, it's probably taking even much longer time, and still without sufficient molecular information, you cannot make a diagnosis at all. That does happen pretty often. So in this patient, if you need to treat, there's still a standard R-CHOP, but some patients might respond to PI3K inhibitor, but we don't know who's going to respond, who cannot. And we took this sample and performed this spatial total RNA sequencing called a PathoDB, and very nice spatial clusters you can see. And by annotating a gene expression profile associated with the clusters, you can already see the dominant cell types in each of those tissue regions. However, what is very powerful nowadays, given the latest advances in computational biology and AI modeling, now we're able to integrate seamlessly with the high-resolution H&E, such that you canTurn this H&E into this colorful spatial molecular profile and almost single-cell resolution, not exactly but close, using the computational method developed by Professor Mingiao Li at Penn called iSTAR. And so now I get much more granular cell types by combining the information from the molecular and also from the tissue and the cell morphology or even subnuclear morphology. And we're doing the same thing for the diffuse large B-cell lymphoma biopsy from the same patient. Now we can ask a lot of questions. And for example, the tumor cells themselves, but you can also ask the question about other cell types in a, since this is in situ spatial, for example, we were asking a macrophage in the low grade versus on the high grade, what's the difference? And we notice indeed in diffuse large B-cell lymphoma is a macrophage polarized more or less to M2. And also you see a lot of interesting metabolic pathways you would never suspect and notice. And then now we got all those mapped out in this unbiased genomic spatial total RNA sequencing. And so recently we further developed even more powerful AI-enabled tool to integrate them together to give you a truly single-cell resolve, as you appreciate from this image, in particular, in the enlarged view, it's a truly cell-by-cell, single-cell spatial total RNA transcript, and you can derive by combining the images and the spatial transcript on total RNA sequencing. So for cancer, and you'll get to understand or at least look into the genomic alterations, including the copy number variation, the single nucleotide variants. And the latter has been very difficult to decode de novo at a genome scale. And largely because in the past people sequenced this, a very short fragment from the three prime end, you don't have sufficient coverage to do unbiased genomic calling of SNV. But in our cases, even though this is now a long read, but we do get a full coverage across the whole length of RNA. And then we ask, "Hey, can you do de novo calling of SNV?" It turns out that worked out pretty well, but I don't want to say that trivial. There's a lot of computational, we need to come up, we need to develop. And now you see the SNV count high rating corresponds in the tumor, and this relatively easy case further confirmed by Mina using the standard IHC staining. But if you don't know, I would say the genomic changes should be your gold standard, should be your ground truth to call the malignant versus non-malignant. Also, since there's a whole genome scale, SNV profile can take the profile to do a further clustering, and then you can see where they are. So that's what's super cool in this type of data. And you can further look into sort of this tumor evolutionary phylogenetic tree, but mapped out where the different clones in a spatial context. So now I can see the most sort of high SNV count clones actually have multiple different clones spatially segregated in different regions. You can also look into the specific mutations, and we have no kind of prior knowledge which matters, which not, and interestingly, we found one. And then we look back to the literature and we found, wow, Ari Melnick, a couple years ago, demonstrate this mutation actually can drive the extranodal lymphoma proliferation by inducing more kind of memory-like phenotype. So that's cool, and now we see that happens, that occur in this patient. All right. And so another important advantage of this technology can look at not just protein coding transcripts. As I mentioned, more than 95% of the transcripts are non-coding. And this is from our data, you can see the coding rates still 60% to 70%, but because they are in general longer. However, we do get a sufficient number of rates corresponding to various non-coding RNA species. What is most difficult to detect, that's a microRNA, because they are extremely short and alsoOften kind of embedding the protein. But it turns out this approach worked out reasonably well. And if you study microRNA, in human cells, we have a little over 2,000 microRNA. From this data, we detect 1,800 approaching the whole pool of the microRNA. For microRNA, each microRNA has been kind of annotated and also computational, that you can predict which ones are their target genes and also upstream of each microRNA, which genes might regulate the biogenesis and the function of that microRNA. So now we can look at, first of all, different microRNA differential expression since this spatial resolved data. You can ask, "Oh, which ones are upregulated in the tumor, downregulated in the tumor?" Or as I mentioned, in different tumor clones, right, which ones are upregulated in different tumor clones? You can ask all kinds of questions you are interested in. The data extremely informative. But afterwards, for example, we saw miR-21, which is one of the well-characterized oncomirra. That makes perfect sense, and the other is miR-155. We don't exactly know what the function of that might be. But as I mentioned, you have the target genes you can predict. You have the genes upstream you can predict, but now we have real data. Okay, you can see upstream which genes are actually detected and downstream which ones are actually detected, and you take those ones you actually detected to perform gene ontology analysis. You'll figure out, "Oh, wow, that's exactly what happened." And upstream, that's sort of PI3K pathway activation based on most of the genes enriched for that pathway. And that corresponds to persistent chronic inflammation and then activates miR-155, and then miR-155 further upregulate the activation. For upregulating the activation of PI3K/AKT pathway. So, as I mentioned before, I think some patients might respond to PI3K inhibitor but it turns out that's a very toxic drug and often you don't want to use without knowing this patient might benefit. And from this data, we suspect this patient actually might benefit, and this mirror, this microRNA, potentially can serve as a biomarker to stratify the patient for the targeted therapy. That's how I imagine this type of data can actually inform the precision medicine in individual patients. So this work has been published and also chosen on the cover on "Molecular Biology." I wasn't very clear why, but later I said, "Oh, actually first time you can see such rich molecular biology in clinical FLT tissue specimens." A little bit, kind of couple of more small examples to show the transformers. And so recently, as I mentioned, there are two distinct approach towards spatial omics. One by imaging, the other by sequencing. We're asking, "Can we combine both and you get a best of both worlds?" Because I think if you really want to get a single-cell or subcellular mapping, you've got to go to imaging, that's the best. But if you want to have unbiased base-by-base sequencing to do de novo genome-wide discovery. So there are two definitions of resolution. One is spatial resolution and the other genome resolution, right? So sequencing gives you the ultimate genome resolution at a single-base level. But if you combine both even though the different modality, but if we can do both on the same tissue, computationally integrate them together, you may get the best out of both worlds. So that was published a couple of months ago in "Nature Methods," but even though that's "Nature Methods" paper, we apply this to various human lymphoma transformations. This is the marginal zone lymphoma. So actually in the same tissue you can see some region already manifest as diffuse large B-cell lymphoma. After this DIBER+ protocol, we can turn every spot into a single-cell transcriptome. And now we can perform this kind of U map and also the pseudo-time trajectory analysis at a single-cell level. So on the right, the dot by dot, they are from single-cell transcriptome now. So another is Richter transformation. Basically, the chronic lymphoblastic leukemia can transform into DLBCL. And so that transformation can be seen, I think, in this interesting sample. You see particular region that's still a cell, and majority of the tissue is DLBCL, but you also have a lot of cells in the DLBCL regions requiring this DIBER+ protocol to deconvolve and to give rise to truly single-cell resolution, spatial by single-cell transcriptome. So we did that and got on a single-cell transcriptome, but each cell we know where they are. And now you perform the clustering, you see very distinctive separation between large and the small B-cell lymphomas, which is very different from the marginal zone... lymphoma transformation. However, when you're looking at small cells in this case, so it turns out they are not homogeneous, and the small cells in a different region already manifest some phenotypic change. And then you take them to perform in real time, you can actually see how they gradually change themselves step by step towards finally transforming into diffuse large B-cell lymphoma. As I mentioned, we're able to look at not just the protein coding, but also non-coding RNAs. And we show in this lymphoma, large versus small cell, and also different regions of the small cell, now we can identify the microRNAs that might be associated with the lymphoma transformation. All right. So probably I'm going to stop here. I have a few slides to show. You can also look at this at epigenetic, but I'm going to skip that for maybe ACR next year. All right. I just want to use this slide to summarize. I think the spatial omics now is much more advanced with both the resolutions, okay. We talk about resolution a lot, but I would say there are two resolutions. Okay. Now, by combining multi-modality, multi-omics approaches, and also empowered by AI, now we can generate the same tissue Amina looking at every day. Now I get a single-cell resolve, all kinds of molecular information, at least across all different RNA species, to really empower our clinical pathology practice in the future. I would like to conclude by thanking my former and current lab members and all amazing collaborator funding support, in particular for this grant, and part of the NCI H10 consortium and also the Lymphoma FL, supported by Mark Foundation and the Follicular Lymphoma Foundation. Thank you very much for your attention. Thank you, Rong. So are there any questions from the audience? So please identify yourself. Hi. Ian Market from Barcelona. Thank you for your talk. It's amazing. I saw in the last slide that you mentioned that you can capture B-cell receptor and T-cell receptor data, but how's your experience with this kind of data in this technology? Yeah. So, I think Professor Joakim Lundeberg published a paper about spatial VDJ. And so I think it has this unique advantage, even without doing long reads, technically no way to do long reads since that's paraffin tissue specimens. But we do see a pretty decent amount of TCR/BCRs we can detect. But now, for given sequencing depth, we want to detect everything, right? And then the number of TCR/BCR reads are really low, still pretty low. However, we have been talking to Joakim, actually we can use the same approach they developed, published in their science paper, to enrich the TCR/BCR sequence before you do the sequencing. But you don't have to change everything upstream. Just once you get a PCR amplicon, you can do another round of enrichment. I believe that should give you a much better coverage for immune repertoire. We haven't done that, but that should be durable. Go ahead. Hi, I'm Eric. I'm from BC Cancer in Vancouver. So, excellent talk. I really enjoyed it. Mm-hmm. There was one particular slide that kind of interested me, was the tRNA stuff that you can detect, and there you were able to detect spatial distributions of tRNA, non-coding RNA, et cetera. I just want to ask, so did you find anything in regards of differential spatial distributions between, let's say, lymphomas that transform and don't transform? I was interested in, did you find any biological results that reflect or are predictive or prognostic of transformation? Yeah. Thank you. Yeah. I think from the same tissue or just a couple of those, we already identified some of those, but I think that requires further validation across a very large number of specimens, large number of patients. And so, I have a few there. I just keep those slides. If you look at all those kind of differential volcano plots, the top rank, those are the ones we identify. And so a lot of interesting program, including the metabolism I didn't talk about. I think associated with the transformation, you'd also see a significant alteration in the metabolic program. And often, I don't know why, that was kind of always among the top-rank programs. And we also use the metabolic chemical imaging to not just verify, actually discover what happens in terms of iNL. And very interestingly, we noticed some either nuclear or perinuclear lipid droplets that give you a dramatic change during the transformation to diffuse large B-cell lymphoma, whether or not it's targetable and how to target. I think no one ever thought about this kind of lipids in lymphoma. Yeah. Dr. C, yeah. Can I ask you one question right here at the table? Okay. Okay, thanks. It was a fantastic talk. Thank you. In the case of where you had the Richter transformation and the CLL- Mm-hmm ... in the same tissue, were you able to look at proliferation centers in the low-grade part of it and see were they related... more closely related just to the small cells, so that you could see where your cells in the proliferation center and the small center would be cycling back and forth, and there's something just drastically different about the Richter. Or could you see things similar in the proliferation center that you could point to that that's where the Richter came from? Oh, you are totally right. So that's when we take all the small cells. And they are in a rich cell region in that tissue, but also in a DLBCL region, you also see a lot of those kind of small cells. And then since now that becomes single-cell resolved, and we take those cells and compare to the ones still in the, that you call proliferating center, they are different. But there's not a distinctive difference. That's actually a continuum. That's when we take all those cells and plot as a kind of shorter time. Now you see the whole continuum within the small cells. Yeah, that's exactly that got our attention. We just look at all the small cells, and we saw the gradual shift. By a certain point, that's still what I suspect, because the separation between the small and the large cells is dramatic, and at a certain point, they're going to make that big jump into large cell phenotype. So may I have one question to you? Sure. Dr. Wei, it is quite impressive that the spatial biology analysis can be applicable to clinics. So my question is, if I give back the patients their sequencing data, we do a deep sequencing, more than sometimes thousands of sequencing to ensure the accuracy. So how about that? It is quite similar to the TCR B cell sequencing question. Yeah. I think it's slightly different from TCR B cell, because we can do better, we just haven't done better. We haven't taken that approach to get better data. But back to just the RNA gene expression, I think now, so data analysis and from data to insights. Okay, first of all, that might take, nowadays, there's a huge amount of time and effort by our computational guys. And how to really change that in a clinical practice, maybe going to be done by agent in the future. That's one. Number two, I think the assay itself, nowadays, still academic prototype and how to really not just scale, but really industrialize everything. Just in a path we have to go through, but we need to have a good case to show it's necessary. It provides critical additional information really needed for accurate diagnosis as compared to the standard, for example, FoundationOne. I've been talking about FoundationOne a lot. And so you got to show this is critical additional information beyond the FoundationOne. Yeah, I do have a few cases I'm thinking we can really pursue of spatial genome testing in a clinical. Yeah. Yeah. It's interesting. So are there any questions? No? So thank you, Rong. Thank you. Oh. This is different. Hello. So I'm talking. My name is Mamiko Sakata-Yanagimoto, University of Tsukuba. My title today is Spatial Multi-omics Dissects Immune and Stromal Cell Heterogeneity in T-cell Lymphoma. And it is a great honor for me to have been invited to talk here today. Here is my COI disclosure. So already, Rong greatly have a nice talk. Lymphoma tissues are consisted of not only tumor cells, but also a rich variety of stromal and immune cells. Sorry. So we have elucidated lymphoma pathophysiology by clarifying the lymphoma microenvironments. And now by using spatial multi-omics, it is a powerful tool for delineating the microenvironments of lymphoma. So by using single-cell analysis, we constructed single-cell analysis atlas of stromal diversity in human lymph nodes in 2022. And now, by using spatial transcriptomics, we could identify spatial distribution of diverse stromal cells in human lymph nodes at single-cell resolution. And this approach is, of course, applicable to understand the lymphoma microenvironments. We captured whole T-cell architectures in follicular lymphoma in this study. Firstly, we identified novel T-cell subsets using single-cell RNA sequencing, TCR sequencing, and then we identified these spatial distribution by spatial transcriptomicsAnd we further identified their prognostic impact by using multiplex digital spatial profiling for more than 240 follicular lymphoma cases. So today, my role is covering the spatial transcriptomics in T and NK cell neoplasms. So firstly, I will provide a brief overview, and then I will talk about our original data regarding vascular niches in angioimmunoblastic T-cell lymphoma by spatial transcriptomics. So I will start from this topic. T and NK cell neoplasms are consisted of up to 34 subtypes, and there are many epigenetic risks are known. For example, chronic infection by EBV virus for NK/TTL, HTLV-1 for ATLL, and germline variants, and clonal hematopoiesis are known to be the risk for T-cell lymphoma development. And most subtypes are known to have poorer prognosis like this. And until today, there is a series of published spatial analysis regarding T and T cell lymphomas. These cover three major subtypes: extranodal NK/TTL, angioimmunoblastic T-cell lymphoma, AITL, and mycosis fungoides and cutaneous T-cell lymphoma, CTCL. Most published studies are published in these two years, 2024 or '25, using multiple platforms including Xenium, GeoMX, Visium, CODEX, CosMX, and so forth. And across these major subtypes, shared features and distinct features are discovered. For example, shared TMA features include tumor-associated macrophages, cancer-associated fibroblast remodeling, and CD8 T-cell exhaustion, as well as Treg. And as an example of a distinct feature, TLS formation has been known to be found during the progression of CTCL development. Then I will talk about our recent data about vascular niches in AITL by spatial transcriptomics. I will introduce you about AITL. AITL is an intractable lymphoma with a five-year overall survival of 35%. It is the most common T-cell lymphoma worldwide, predominantly affecting elderly individuals. AITL patients display lymphadenopathy, high fever, and skin rash, as well as autoimmune-like manifestations. Pathologically, AITL tumor cells mimic T follicular helper cells. And as the name angioimmunoblastic T-cell lymphoma, the proliferation of vascular vessels and EBV sometimes infected immune cells, especially immunoblasts, are the hallmark of the disease. So firstly, I will talk about immunoblasts in AITL. AITL genome has three characteristic features. Epigenetic mutations, including TET2 and DMT3A, RHOA mutation, and T-cell receptor signaling pathway mutations. And notably, most of TET2 and some DMT3A mutations come from clonal hematopoiesis. So AITL is a representative cancer derived from clonal hematopoiesis. Based on the gene mutation distribution pattern, we have raised the concept multi-step tumor genesis in AITL development. Firstly, hematopoietic stem cells acquire TET2 mutation during the aging process and evolve into premalignant cells. The tumor-specific RHOA mutations instruct these cells into AITL tumor cells, and simultaneously, AITL premalignant cells can be differentiated into various inflammatory cells and infiltrated into tumor tissues. So in the AITL tissues, the tumor cells have TET2 and RHOA mutations, while inflammatory cells have TET2 mutations alone. So we hypothesize that TET2-mutated inflammatory cells function as a niche to support tumor cells. By using a single-cell approach, combining with mouse data, we propose that TET2-mutated germinal center B cells have both light zone and dark zone properties simultaneously, and it function as a niche to support AITL tumor cells in both human and model mouse tumor through CD40 ligand/CD40 axis. Then I will focus on proliferation of vascular system in AITL. As I have mentioned, hyperproliferation of blood cells is the origin of the disease name, and tumor cells are known to exist alongside blood vessels.And the blood vessels have been thought to be high endothelial venules, HEV. So we have research questions. Are there any heterogeneities of blood vessels in AITL and do they function as microenvironmental cells to support tumor cells? Before we address this question, we performed single-cell analysis to clarify the heterogeneity of blood endothelial cells in human lymph nodes. Before the single-cell RNA sequencing era, only three subtypes of BEC are known: artery, vein, and high endothelial venule. However, by using single-cell RNA sequencing, we identified 10 subclusters including TIP cells, essential for angiogenesis, and HEV, high endothelial venules, can be sub-classified into four subtypes. So to capture the AITL BEC heterogeneity, we performed single-cell spatial multi-omics analysis on BEC in human AITL. We analyzed 189 AIT samples in total by using single-cell RNA sequencing, spatial multi-omics including spatial transcriptomics, and multiplex digital spatial profiling. Firstly, I have to explain about single-cell RNA sequencing data. By using single-cell RNA sequencing, we profiled more than 50,000 cells like this for normal and AIT samples. And consistent with pathological findings, BECs were increased in AIT samples. Unsupervised clustering with single-cell RNA sequencing data identified 10 BEC subclusters. And notably, transitional capillary activated HEV comprised two subclusters, chHEV1 and chHEV2. And by using pseudo-time analysis, both chHEV subtypes were post-capillary states differentiated from tBEC from here like this. And by checking the subclusters of BEC of AIT and normal samples, we found that proportion of chHEV2 and TIP cell, tBEC, was significantly in AITL comparing to control. And by using differential gene expression analysis, we found that both chHEV in AITL exhibited elevated expression of integrin-related genes. Then to characterize the spatial distribution of these subclusters of BEC in AITL, we performed single-cell transcriptomics using Xenium in situ. We constructed two panels, custom panels containing 289 genes designed by top markers of subclusters of single-cell RNA sequencing data, and the prefixed 5K panel, including 500 additional gene panels. We used custom panel to analyze discovery cohort samples and prefixed 5K panel to analyze validation cohort samples. We finally found that spatial transcriptomics by Xenium in situ identified lumen-like vasculature structure in AIT samples. By using a clustering of cells based on Xenium in situ data, all 10 BEC subclusters and multiple immune cell clusters found by single-cell RNA sequencing were successfully identified like this. Then we mapped these subclusters onto the H&E specimen like this. So we found that a spatial transcriptomics analysis could identify individual blood cells composed of multiple BEC subclusters. In the left figure shows that the tumor density high area is colored by yellow, the tumor low density area is colored by blue, and each BEC subcluster is color-coded as follows. Artery subcluster with a red color, HEV in yellow, and veinous subcluster in green or blue. So we could find that in the tumor density high region, HEV yellow color are very enriched like this. And then we calculated the distance between tumor dense region and each BEC subcluster, and we found that chHEV1, was a capillary activated HEV, was closest to regions with high tumor density area. Then we performed neighborhood analysis to identify tumor supportive niches. Here, we show the proportion of tumor cells as follows.We could identify three niches: GCB niche, T cell-enriched niche, and cHEV FDC niche. The GCB-enriched niche was consistent with our previous finding that AITL tumor cells having T follicle helper features are supported by clonal hematopoiesis-derived TET2-mutated GCB cells. GCB niche are shown on the specimen like this. And the cHEV FDC niche is a novel finding. We identified a novel niche containing cHEV1, FDC, and of course, tumor cells. And this niche was characterized by selective enrichment of exhausted T cells. And in this figure, black dots showing exhausted T cells are enriched in cHEV FDC niche comparing to artery or HEV niche. And in this graph, shows that exhausted T cell density was higher in cHEV FDC niche comparing to artery or HEV niche. So to validate these findings by protein expression, we performed multiplex digital spatial profiling using 40 antibodies, again designed by single-cell data, top markers of single-cell RNA sequencing subclusters for more than 120 samples. Combining these markers successfully delineated multiple BEC subclusters, including artery, capillary, cHEV, HEV, and vein. And this is the data showing the clustering, and then we mapped this cluster onto the specimen like this. We successfully identified the spatial distribution of 10 immune and six VE subclusters onto the specimen of a phenocyclofusion multiplex digital spatial profiling data like this. And in this left figure shows that white arrows indicate cHEV expressing PEVAP and MECA-79 simultaneously. And the purple cells indicate TIM-3 expressing exhausted T cells. So as shown in this figure, exhausted T cells preferentially accumulated around regions enriched with cHEV. And then we calculated peri-vascular index, indicating the enrichment of specific immune cell populations around each BEC subtype. We found that the PVI of exhausted T cells was highest around cHEV1. cHEV1 is this. Then we checked the cHEV signature and the clinical outcome. In the left graph shows that PVI of exhausted T cells around cHEV1 was significantly higher in non-responder groups comparing to responder groups, while PVI of cytotoxic T cells was comparable between non-responder and responder. And by using the analysis of RNA sequencing published data, we found that patients with high expression cHEV and HEV signature had a worse clinical outcome. Here are the summary of my talk today. Single-cell RNA sequencing revealed an expansion of BEC subclusters, particularly capillary activated HEV in AITL. The second, spatial transcriptomics demonstrated that among BEC subclusters, cHEV1 was located closest to tumor cells and formed a cHEV FDC tumor niche together with exhausted T cells. Multiplex digital spatial profiling validated the spatial transcriptomics findings and further highlighted vascular heterogeneity in AITL. And we further found that BEC signatures were associated with worse clinical outcomes. These are the collaborators. Manabu Fujisawa, he is a postdoc in BC cancer and simultaneously staff of our department, and Yuki Shimizu, a talented PhD student, did most of the works. This is my final slide. Thank you for your attention, and I'm happy to take any questions. Hello. It's Antonio from Ohio State University. You mentioned briefly about integrins. You have a result with integrins. Yeah. Can you tell me a little bit more about the relevance of that finding, please? Thank you so much. Because now we analyze the functional analysis, but we are thinking that cHEV interacting with, I didn't mention about the stromal cell today, but specific stromal cells interacting and making the vasculature structureSo maybeintegrin or some collagen-related gene and the protein expressed by these cells may be important, is what we are thinking. But this is just our speculation. We need some functional validation. Thank you so much for question. Hello. Yes. Great talk. Thank you so much. I was wondering, you using TIM-3 as a marker for exhausted T cells, is there a particular reason why you took that particular marker? Or do you have corroborating results with other exhaustion markers like LAG-3 or any other like TIGIT, for example, stuff like that? Thank you so much. Yes. We included this marker into the 40 antibodies of the subsequent multiplex digital spatial profiling panel, and of course, in the single-cell transcriptomics data. Just we show to show just see the, because the eye, we can see one color. So just we pick up TIM-3 today, but of course, these cells express LAG-3, TIGIT, or something. Thank you so much. Thank you. Dr. Suren. Yeah, that's a fantastic talk. So in that niche with the follicular dendritic cells and the tumor cells and the HEV cells, in tumors, you see these big clusters of FDCs there, and you usually don't see individual FDCs scattered around the tumor. So can you get at that niche just by looking at the shapes or the collections of FDCs without having to do 50 markers to kind of figure that out? Thank you so much. Because I am doing many research on AITL, and I am thinking that FDC in AITL distributed into two parts. One is colocalized with B cells, and the other part is separated from B cell cluster. And this part of FDC is colocalized with so-called cHEV in forming the vascular niche together with FDC. So by using spatial analysis, honestly, it is difficult to isolate the B cell coexisting FDC because of the technology, by dissecting technology. So now we only capture the separated FDC in AITL, and this is important for exhausted T cell accumulation. Thank you so much for question. Are there any question? Amazing. All right, amazing talk. Quick question. So you mentioned those niches in particular, for example, exhausted T cell in the close to tumor. Do you know how they got exhausted? How tumor talk to them? Do you know the communication between them? Thank you so much. That is very important fundamental question. Now we are analyzing the functional data to enhance the why, answer the question why. Thank you so much. We are thinking that some kind of cytokines are very important, which is known to promote the T cell exhaustion. Thank you so much. Thank you so much. So thank you so much. The next speaker is Dr. Tse. His title today would be Spatial Biology Concepts in B Cell Lymphomas. Yeah. Thanks a lot for the invitation to come and speak here. So I'm a hematopathologist, so some of the things I'm going to say are going to come from a hematopathology perspective. Just for disclosures, I didn't make a slide. I have research support from Eli Lilly and equity in Abcon Therapeutics, and none of that is relevant to this talk. So advances in our understanding of the molecular underpinnings of lymphomas has really largely been driven by bulk analysis, and we have the concepts, for example, from gene expression profiling of the cell of origin or the stromal signature or the importance of a dark zone signature to recognize highly aggressive diffuse large B-cell lymphomas. We've done next-generation sequencing across multiple lymphoma types to define the mutational landscape, and even molecular subtypes that we can kind of agree upon, for example, in diffuse large B-cell lymphomas. And the application of single-cell technologies really provide direct data at the individual cell level that enable detailed descriptions of tumor cell states, the tumor microenvironment constituents, and even clonal trajectories. But it's still kind of early days. If you do a search in PubMed, I just did this last week, just wondering. If I did single-cell RNA-seq in lymphoma, how many hits would I get? And you get about 800. But we're here today to talk about spatial omics and spatial technologies, and if you search spatial omics in lymphoma, you only get about 30 hits. So we're still really trying to figure out this technology and get reproducible results. So what I thought I would do is just kind of review some recent progress and some papers that I found interesting in both in follicular lymphoma and then quickly in diffuse large B-cell lymphoma, and then really think about how are we ever going to apply these learnings actually in patients in the clinic.So I'm going to start with this paper that caught my eye back when it came out, a study out of the NCI. It doesn't use a spatial transcriptomics or anything, but it really uses very detailed image data to get a lot of spatial insights. And so they came at the problem that so in follicular lymphoma, while most patients follow an indolent clinical course, a subset really suffer early progression and have a short survival. And perhaps a greater understanding of both tumor cell intrinsic and extrinsic factors governing progression and outcomes would be really helpful. And so a multimodal approach then could potentially identify B-cell programs, cellular composition of tumor microenvironment, and even topological factors like follicular and stromal features that could predict outcome. And so they used bulk whole exome and RNA sequencing, single-cell RNA sequencing, and then multiplex immunofluorescence platforms to try and identify this. And so the bulk analysis in the figure there and the single-cell analysis is really good at getting very detailed description of what's going on at the genomic level, even telling us what's going on in clonotypes. But you have to really apply these advanced imaging techniques to really get a granular idea of what the cell phenotypes are, what the composition is, and then really only the imaging technologies are going to give you that kind of detail about the relative abundance of rare cells that by bulk sequencing or single-cell sequencing are very underrepresented, and you really don't understand where they are and how many of them there even are. And so you get that from the imaging data. If you apply that to a series of cases of both normal lymph nodes and follicular lymphoma with known outcomes, you should be able to begin to get some insights. So what they found was that in B cells in the early progressors, upregulated B-cell receptor signaling and extracellular matrix remodeling pathways, and that progressors were associated with proliferative type B cells and a tumor microenvironment that was much more capable of remodeling its stroma. The progressors also had non-class-switched lymphoma cells and that these B-cell intrinsic factors, along with tumor microenvironment factors, could help identify high-risk patients. The single-cell data characterized the lymphocytes in rare cell populations in great detail, so things like plasmacytoid dendritic cells that were in there, macrophage subtypes, and follicular dendritic cells. And we noted that follicular lymphoma tissues were much more enriched in exhausted CD8 positive T cells compared to normal lymph nodes with molecules that were just mentioned by the person asking the previous question. And if you apply then this relatively high plex immunofluorescence assays, you could really begin to get a handle and detail on diverse subsets of B cells, T cells, myeloid cells, stromal cells, and then really see that there are many more myeloid and stromal cells in those tissues than you thought, rather than if you're just looking at the single-cell RNA sequencing data. And then what resonated as a pathologist is that when you start to look at some of these pathways and molecules and do immunofluorescence staining, the early progressors really had expansion of desmin positive fibroblasts, and those are those yellow streaks in the image. So in the progressors, you can see a lot of desmin positive fibroblasts within the follicles. And then DC sign positive macrophages also within follicles, and they were seen in contact with IRF4-positive lymphoma cells giving evidence for antigen-independent B-cell receptor signaling in follicular lymphoma that we knew was going on, but now you can see correlates with progression. While the spatial data then allows you to also do community analysis, and you can then look and when they did that, they kind of found 15 different cellular ecosystems or neighborhoods that they named for the dominant cell types. And they found that the cellular communities that are seen in normal lymph nodes are actually altered in malignancy, not surprising. And cellular communities that were distributed heterogeneously across entities and some were only seen in follicular lymphoma or expanded in the progressors. And given that the extracellular matrix remodeling might be important in relapsing patients, gene signatures associated with that fibrosis were looked for, and the bulk and single-cell RNA sequencing data identified things like TGF-beta and chemokines like CCL4, 5, and CCL13 that were involved in recruitment of many different cell types into that tumor microenvironment. And then if you apply fancy shape analysis of the follicles, you could identify five different follicle types, and applying them in paraffin tissues using sort of a reduced plex immunofluorescence assay, you could actually reproduce almost all of those neighborhoods that you could get with a 40 plex assay. And then really look with more detail in outcomes and show that the high-risk patients had actually different kind of follicles that you can see there. So the high-risk progressors had much more kind of disrupted follicles that were set far apart. Again, really resonating with hematopathologists because we've seen these for years. Most people do a follicular dendritic cell stain to evaluate a lymph node in a follicular lymphoma patient, and we know that there are cases certainly that have very large well-formed FDC networks that are back-to-back. But then there are these other cases where there's almost remnant follicles, but you know it's a follicular lymphoma. And so those pictures really show what we've been seeing for years, but really didn't know what the significance of that was. And then one of the particular shapes, follicle type IV, was seen and associated with early relapse.Now, community analysis also showed differences. There's a progressive decline in communities that were combined tumor B cells and FDC networks, and then in the progressors in the early relapsers and then an increase in communities that were enriched for B cells, fibroblasts, and extracellular matrix together were seen as well And actually, you could then make a score, a linear predictor model of some of the most highest risk communities, which are these stromal communities that were enriched for desmin and vimentin positive fibroblasts and extracellular matrix proteins, and show that those patients that had that high linear predictor model score were associated with early relapse. And you could then interrogate the tissues and see those things like the lumican that you see there in yellow and other extracellular matrix proteins in these early relapse patients. So the conclusions from this paper were really specific architectural features like follicles and distribution of extracellular matrix and cellular communities were lymphoma specific and are present actually before treatment and really do predict for outcome. Now, Dr. Sakata and Dr. Abe in her lab actually focused on the T cells, and I wanted to look at this paper because this paper really looked at spatial biology and with the idea that we've been looking for a long time at T cells in follicular lymphoma, but with old style tools, immunistic chemistry and things like that. And there were papers that would come out and say certain subsets were associated with outcome in follicular lymphoma, but they weren't really reproducible, and it wasn't really until we could approach it from a multi-omic approach that we could really get a handle on that. And so they show that the follicular T cell ecosystem really plays a major role in follicular lymphoma biology, and that follicular lymphoma contains very distinct non-TFH follicular T cell subsets that are actually developmentally linked, they're spatially organized, and they're functionally anti-lymphoma and also prognostic. So they applied single cell RNA sequencing, single cell T cell receptor sequencing, spatial transcriptomics using a Xenium platform, and then multiplex spatial proteomics platforms in formalin-fixed paraffin-embedded tissues. And they supported those observations with functional assays that really supported the conclusions of that transcriptomic and spatial data. So the single cell RNA-seq and T cell receptor sequencing data show that there indeed were distinct non-TFH follicular T cell subsets, namely T follicular regulatory cells and CD4 and CD8 follicular cytotoxic cells. And these T follicular regulatory cells, as you might expect, expressed Treg markers like FoxP3 and CD25 and TFH markers like PD-1 and 6, IR5. And then the cytotoxic CD4 and CD8 follicular cells express, as you would expect, cytotoxic markers, a stem-like marker which was actually identified in a previous paper, and then these cells were really less aligned and terminally exhausted cytotoxic exhausted cells. And the spatial data and the Codex analysis really showed the strength of the spatial data was that the TFRs and the TFHs obviously were localized to the follicles. And then the cytotoxic CD4 and CD8 positives cells really were localized to the very edges of the follicles at the borders between the T cell and B cell receptor areas. And then communication analysis really shed light on what was responsible for that, and that was the CXCL13, CXCR5 path, and then the CCL19, 21, and CCR7 axis. And so this interplay and location of these cells is very important in the biology of a follicular lymphoma follicle. Now, there are functional studies then that supported these concepts. Isolated T cell subsets were shown to have migratory competence for T follicular reg cells towards CXCL13 gradients. The follicular cytotoxic cells were able to produce things like interferon gamma and TNF alpha and were able to degranulate. And then the ability of the cytotoxic follicular T cells to inhibit malignant B cells was also shown in co-culture experiments. Interestingly, the T cell receptor sequencing and single cell RNA velocity analysis actually showed that there was clonotype sharing between all these cells. So they were related and could differentiate into one another. And so the T follicular regulatory cells and the cytotoxic CD4 positive T follicular cells were really mainly derived from Tregs and TFH cells respectively, and the cytotoxic T follicular cells were derived from resting cytotoxic cells that could then further differentiate it finally into the exhausted cytotoxic population. And then upstream regulatory analysis with differentially expressed genes identified key players like IL-21 as a key inducer that was supported by cellular interaction analysis and also functional evidence to show that you could actually differentiate Tregs into T follicular regs and TFH cells into the cytotoxic CD4 cells, and then the naive T cell and resting cytotoxic cells into those follicular CD8 positive T cells with IL-21 and TGF stimulation.So now you can think about this model and see how this kind of all works together with a TFH cell in the kind of center being stimulated by the malignant cells and the FDC cells to actually secrete an IL-21 and force differentiation of Treg cells into these T follicular reg cells that then would modulate the TFH cells and the help to malignant cells or drive differentiation into T follicular CD4+ cytotoxic cells or differentiating those naive CD8+ cells into these cytotoxic T cells that would then modulate malignant B cells. And the proof of that pudding would be, well, then if you can quantitate these cells, which you now can with these high-plex immunofluorescence assays and look for those cases that had those high TFRs and TFCs, which should be modulating the malignant cells across multiple cohorts, you could see in the upper curves that those cases with high populations of those cells did better than ones that had lower numbers of those inhibitory cells. So the concepts from these kinds of papers, you can start to see now that there is outcome information in follicular lymphoma that's encoded in the TMA, the tumor microenvironment, and its organization. It's not really only dependent on lymphoma genetic driver genes. And early relapse is associated both with baseline architectural and stromal features and then proportions of lymphoma-specific follicular T cell subsets. And that follicular lymphoma niche is really shaped by chemokines and its stromal infrastructure, and that the stromal TGF-beta biology is important in follicular lymphoma. So let's spend a couple of minutes talking about diffuse large B-cell lymphoma. There have been multiple papers over the years using these high-plex immunofluorescent platforms looking at cell types. They've all been relatively smaller studies. And I wanted to focus on the last one, which is the really, at least to my knowledge, one of the early ones that used true spatial tools and used CosMx platform along with the Codex spatial imaging to look at a relatively large number of cases and then include some other kinds of cases and really develop a kind of a rich understanding of cell types within diffuse large B-cell lymphoma. So the question, this came from out of the MD Anderson group in Michael Green's lab with the question of, can combined spatial transcriptomic, proteomic, and genomic analysis provide a prototype understanding of the canonical patterns of spatial organization and immunologic interactions in large cell lymphoma? So they had a TMA with about almost 80 diffuse large B-cell lymphomas, but they included some EBV-positive cases and also some large cell lymphomas of immune privileged sites, which is getting its own play and that are going to be recognized in the new classification because they are a little different. And so those were included in the kind of cases that were looked at. And so they applied the CosMx platform to do the spatial analysis along with Codex and then had a whole exome sequencing data to kind of give the whole genetic background and show now with that kind of data, you can really look at there's many, many cell types within diffuse large B-cell lymphoma. The tumor cells are green just for thing because you can reference it to these other things. But there's many other kinds. There's a couple of different kinds of plasma cell types and many kinds of macrophage and tumor-associated macrophage cell types, other stromal cells that are now defined. And then once you understand those cell states, you can organize them into cellular niches, and about seven cellular niches were organized, and those are shown here. And you can see that although tumor cells are present in all of them, there is a couple that are highly enriched. This is the so-called CN5 niche that's very a dense tumor B-cell niche, and then one that's more diffuse, and it has more mixed cell types in. As you would expect, red is the actual T-cell, C5, in that there's a niche that's really rich in T-cells. And if you look at the genes that are expressed there, they make sense for the kinds of niches that you're defining. So that detailed spatial analysis really showed that DLBCLs have these cellular niches that do co-associate into communities, and they have distinct patterns of cell-cell communication. So for example, the T-cells in that T-cell-rich niche, they actually showed increased expression of memory and naive markers and cytotoxic markers and didn't really have a lot of exhaustion markers compared to other niches like the myeloid and B-cell-rich niche. So the T-cell niche was a niche where maybe they were more kind of healthy and functional type T-cells. There's differences in T-cell proportion and function that can be linked to differences in the chemoattracting and co-inhibitory signals that were seen in those niches, and that the EBV-positive large cell lymphomas and these unusual diffuse large B-cell lymphomas in these immune-privileged site, like the CNS and the testis, have high degrees of T-cell infiltration that are really cytotoxic, but they're also highly expressing multiple co-inhibitory signals. So there are these cells that are really cytotoxic but unable to function, and that might represent then a therapeutic strategy to deal with these kinds of lymphomas. So when you start to think about these kinds of findings and what we're seeing from these papers, I think it brings some questions that I have to mind. I don't have the answers, but some observations that there's actually sort of a kind of a not great correlation between what we're finding with spatial data and the bulk analysis. So I think that's telling us that that spatial data is really bringing in a bunch of neighborhood information that was really not considered from the bulk studies. There's kind of not also a great correlation between the niches and some of the genomic classifiers that we're seeing. So again, I think the spatial transcriptomic data puts the genomics in the context of a local tumor microenvironment, and that that tumor microenvironment is really not entirely shaped or entirely driven by the tumor genetics, and that there's either host factors or given the fact that the immune system is plastic, that there is room for variation in case to case, and it's not all preordained by the driver genes that are present in the tumor cells.And then, are these cellular niches lymphoma type specific that we're defining? Or are there canonical ones that really apply across all cell types? I don't think we've done enough studies to really understand whether we can just think of generic cellular niches and just start plugging them into different lymphoma types, or are we going to have to study every lymphoma type differently, separately, because they are that different. And then as a pathologist, I think a lot about how are we going to scale this? How are we going to actually bring this into the clinic? I think that is a problem. Gene expression profiling is common in lymphoma for over 25 years now, and we're still not great at bringing that into the clinic. And so I think we have some work to do. This whole meeting, I think, as you peruse different titles and sessions, has been a meeting where there's been a ton of talk about AI, and I think here is a question that's really quite relevant to this problem is can we use AI tools and foundation models to scale findings that we're learning from spatial transcriptomic data? I think we, along with other people, are trying to develop pathology foundation models, and we have developed a lymphoma-specific foundation model that we're calling LymphoVision that does better than the general pathology models. The only reason I bring that up is that I think if we're going to do this, we're going to have to use these models because they outperform general pathology foundation models. And so if we use specific models, we can then start to think about how do we bring them into the spatial transcriptomic and spatial biology area, and that's by pairing the data with those images. And so I think there's two approaches. These are all papers that are out in preprint, and basically there's two approaches, right? There's one where you can basically start from scratch, where you have paired transcriptomic data on all your cases in the paired image data, and you train it from scratch with image and spatial data. The other way, which is that last paper, the SEAL method that came out of Dr. Mahmoud's lab at Harvard, takes a little different approach. It starts with a well-trained foundation model and then fine-tunes it with spatial data to make it a spatially aware foundation model. And they have shown that you can, whatever task, it doesn't have to be a spatial transcriptomic task, but once you have a spatially aware foundation model, it does better in almost every task that you ask it to do. And so then you could think about taking that and then training or fine-tuning it on a true spatial prediction task. And so these are the kind of things I think we're going to have to think about and how we bring things like this into the clinic and democratize the findings into something that is answering a clinically relevant question. And with that, I will invite you to Rochester to see our amazing Corn Tower, if you ever want to go there. And I'll stop there. Yes, take any questions. Thanks. Thank you, Eric. Great talk. So do you have any questions from the audience? Thank you so much for the great conclusion summaries. So I'm wondering how to improve the statistical sound of the spatial technology since the tumor can be much bigger and each slide's only about five to 10 micrometer, and it is relatively really expensive, and normally the lab could just get a few slides. And even though we say from the second paper, the "Nature Genetics," they say the T cells have less exhaustion marker than the B cells. Is it possible just from that one slide or they conclude from many slides? Yeah. So those were done on tissue microarray cores across all the things. So I think that was an inference from across the entire set. But your point is well taken, right? Especially in an example like follicular lymphoma or AITL, where there's a lot of heterogeneity across a tissue, and now if you're basing all your analysis on two TMA cores per case or whatever, I think you could get into some issues. And then, of course, in the clinic, if you're doing things-- we are doing things on just tiny little needle core biopsies, and trying to infer some spatial data from that, I think we can also start to run into problems. Yes. So regarding that, is there a certain workflow we can follow? Like from Professor Yana Kikmoto, and she started from the transcriptomics data, like bulk data, and then followed a final signature. Then we look at specific signature from the spatial transcriptomics and validate with the spatial proteomics. So, if we already have certain marker that we can focus on, maybe can reduce a bias. Is there a certain workflow you could recommend? Thank you for your question. I think the bias and the pathological or insights are quite very difficult to discriminate is what I'm thinking. Because I focus on-- I would like to ask Ron and Eric because I am focusing on angioinvasive T-cell lymphoma, because classically, angio is very important because it is the origin name. So there should be something that angio-- Because when we collect the angio proliferative sample, then it is AITL, so it should be important, is my pathological insight. But how we can discriminate this pathological insight from a bias? Sorry, I think my question was not clear. So, I think you made a good example, the workflow from spatial started from the bulk RNA-Seq. Yeah. So we can be more focused. My question is, do we recommend the spatial transcriptomics, the technique always followed after the bulk all single-cell RNA-Seq, so we can be more- Ah, yeah ... focused, but not just from one slice? Thank you. For that issue, myself, I'm confident that we should do single-cell RNA sequencing before we do spatial transcriptomics, because each specimen has the thickness, so we cannot avoid the segmentation issue completely. So we are pretty sure that we should do single-cell RNA sequencing before, and then we identify new cells, and we determine the mapping. Is my stance. How Rong is doing? I'm not a pathologist. I'm afraid I can not make any kind of expert recommendation, but I want to share what Mina told me. So often, if you get a fine needle biopsy, often she needs to order 10 IHC slides. You pretty much would exhaust the entire tissue, and then you put a patient at risk, right? If you don't get a good result back for those IHC and you cannot make a diagnosis and the patient cannot be treated, that's a real problem. And so that's why Mina demands, "Okay, can you get all the information from exactly the same tissue section or adjacent, now 10 of those, and then I can make sure I don't put a patient at risk, and I can get a sufficient amount of information to make a proper diagnosis." I think that no matter bulk single-cell sequencing or bulk genome sequencing, I think FoundationOne is widely used, but still it got to consume a lot of tissue specimen. I would say it depends. So my goal is you don't need to do that, and you get a sufficient amount of info. But the sampling variability, that's a problem I think pathologists, oncologists, they know for decades. You cannot sample entire tumor. But I know for prostate cancer, they know how many cores you need to take to get a relatively good representative of the tumor. That has been investigated. I don't think the very informative spatial omics technology is going to address that question that has been intended to be addressed during the past couple of decades. But at least that single piece of tissue, if you can get much more information, that should help, yeah, in making the whole thing compatible with pathology. Don't try to ask them to take a different tissue processing protocol. That's not going to happen. That's not going to happen very soon, maybe. Thank you. Thank you so much. So it's time, so we would like to wrap up the session. Thank you for coming. Thank you. Yeah. Thank you. Fabulous talk, you guys. Thank you.