Skip to content

2026-04-19-disseminated-tumor-cells-breast-cancer

Date: 2026-04-19 Primary topic: single-cell-spatial-omics Cross-referenced from: single-cell-spatial-omics Word count: ~12,317

Auto-generated captions

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

Okay, in an effort to keep us on time, I'm Louis Chodosh. I have the honor of chairing today's session, Advances in Understanding Disseminated Tumor Cells: An Emerging Opportunity to Prevent Lethal Breast Cancer? With a question mark and the emphasis, I suppose, on the question mark. I'm from the University of Pennsylvania. I will introduce our speakers momentarily. So by way of introduction, the focus of this session is on this problem, breast cancer recurrence. So we all know breast cancer is the most common cancer diagnosed amongst men or women. And fortunately, five-year survival is quite good, approaching 90%. And as a consequence, breast cancer is by far the most prevalent cancer in the world, with more than 12 million breast cancer survivors globally. The majority of women are in what we would call a no evidence of disease period, or the surveillance period, in which there is nothing. There is no disease evident on a scan. However, in following patients, up to 30% of women will ultimately recur following a variable latency period, and that recurrence, typically metastatic recurrence, is responsible for nearly all mortality from this disease. Now, breast cancer recurrence has the interesting property shared with prostate cancer in which recurrences can happen after many years. So in some cases, 25 years or more after definitive treatment of the primary tumor. And while these sorts of kinetics are typically associated with estrogen receptor positive breast cancers, it is also seen to some degree with estrogen receptor-negative tumors as well. And this very long stochastic latency period has brought up the question of whether these cancers may pass through a dormant phase. Now, arguably, the current clinical approach to breast cancer is front-end loaded. That is, there is definitive treatment of the primary tumor. Women enter the surveillance period, and there are really very few therapeutic interventions during that surveillance period to prevent recurrence other than anti-estrogen therapies and CDK4/6 inhibitors for ER positive breast cancers. Now, during this period, patients are observed. There's no active monitoring. There's no clinical test that will tell you whether a woman is part of the 30% who might ultimately recur. And this causes a lot of anxiety, understandably, and decreased quality of life for patients because they can never know whether they're cured or not. So I would argue that the failure to intervene during the surveillance period is a missed opportunity, that simply watchful waiting is not enough. So why do some breast cancers recur? Well, as one of our speakers, Christoph Klein, and others have shown, over the period of years in which a breast cancer grows to reach a clinically detectable size, there are cancer cells that make their way into the bloodstream, disseminate to distant organs, and can basically sit there as single cells or micro metastases as DTCs or minimal residual disease. And at some point, in some women, DTCs can reactivate growth and give rise to recurrent metastatic cancers. So DTCs are the reservoir from which recurrent tumors arise. And again, the fact that they can survive in their host for decades raises the question of whether they pass through a dormant phase. So if one looks at bone marrow aspirates in early-stage breast cancer patients, up to 40% will have detectable DTCs in their bone marrow, shown here by pancytokeratin immunohistochemistry. And that's important because the presence of DTCs in early-stage patients is strongly associated with an increased risk of recurrence and decreased overall survival, as you might expect for cells in this reservoir that gives rise to recurrence. So DTCs are really both a biomarker to identify early-stage patients who have an increased risk for recurrence, and we can also think of them as a potential therapeutic target, as we'll talk about in today's session. And I would say the main questions that we have to answer to get to that type of clinical approach is, first, what are the biological properties of DTCs? What are the mechanisms by which they persist and recur? And can we translate that information to prevent recurrence? And today, we have three speakers who will address each of those particular questions. So I will speak first about approaches to clinical translation targeting DTCs. Then we will have Dr. Amoya Sreekumar, who is one of the AACR NextGen Stars, who will talk about a particularly interesting mechanism by which dormancy is enforced in tumor cells. And then we will have Dr. Christoph Klein from the University of Regensburg in Germany, who will talk to us about the biology and molecular characteristics of patient-derived disseminated tumor cells. So with that, I will start my talk. These are my disclosures. Okay, so if really all breast cancer mortality or the vast majority is due to lethal recurrent breast cancers, the goal of preventing breast cancer mortality could presumably be achieved by preventing recurrence, if that's the lethal event. And the strategy that I'm going to talk about is depleting this reservoir of DTCs or at least preventing them from reactivating. And the approach, as I mentioned earlier, is really to say, look at DTCs as bothThe prognostic marker of risk and the target for therapy. And the steps are: understand the mechanisms of dormancy and reactivation in these cells, develop better methods to detect them and to characterize them at a molecular level, and then to use that information to develop clinical approaches to target these cells to prevent recurrence. So one of the approaches that my laboratory has taken over the years to understanding dormancy biology is developing genetically engineered mouse models for human breast cancer that make use of the doxycycline-inducible, mammary-specific expression of oncogenes of interest in the mammary epithelium. And as shown here, when mice are put on doxycycline in their drinking water, in this case, they turn on the HER2 oncogene, and that drives the development of hyperplasias, and then ultimately invasive mammary adenocarcinomas that will metastasize. And as an example of oncogene addiction, with these mice with extant primary tumors, if doxycycline is removed from the drinking water, HER2 is then turned off in a tumor which has multiple mutations. These tumors rapidly regress to a non-palpable state over the course of several weeks, essentially a genetic approximation of targeted therapy. But if you follow these animals over time off doxycycline, just as in patients, tumors will recur stochastically and via a variety of escape pathways, and they really do it after a variable latency period. If you look at the site of a regressed primary tumor and genetically marked tumor cells, you can see that there is a collection of residual tumor cells that persists in these animals, which enables us to isolate them, query their biology, and try to understand the mechanisms of survival and recurrence. One important element if one follows the proliferation of these tumor cells, within three to five days after instituting this genetic targeted therapy, the proliferation rates crash, cells exit the cell cycle, and remain in this persistent quiescent state for many weeks, as shown here as well by BrdU labeling of these dormant residual lesions. But if you put animals back on doxycycline, turn the oncogene back on in those residual tumor cells, they rapidly reestablish proliferation rates and reenter the cell cycle. So they are genuinely reversibly quiescent or serially dormant. If we isolate these dormant RTCs, what we find is that they exist in a distinct cellular state. So dormant residual tumor cells are in a state that is different, both from the primary tumor cells from which they arose and the recurrent tumor cells to which they give rise. And this is true in multiple different animal models for dormancy and recurrence. And interestingly, across these different models, dormant RTCs share a variety of gene expression alterations. So what makes us think that this reflects anything relevant about human disease? Well, Amalya Sreekumar, who's in my lab and here speaking today, did very interesting work where she developed a core mouse dormancy signature from in vitro and in vivo mouse models of dormancy, and then applied it to neoadjuvant therapy breast cancer patient data sets in which there were initial biopsies of primary tumors pretreatment, and then surgical excision of the remaining residual lesions. And what she found in the majority of data sets is that there is an enrichment across multiple patient cohorts of this mouse core dormancy signature in the post-treatment samples from patients. And interestingly, this was independent of therapy, so seen both for endocrine therapy or chemotherapy, short or long. And this argues that therapy-refractory residual tumor cells that persist in patients following neoadjuvant therapy exhibit changes in gene expression that are similar to those associated with cellular dormancy in mice, arguing for at least some relevance of these models to the clinic. So using models such as this, we ask the question: can we identify dormancy-specific responses to therapy that we could leverage? Well, the inverse of that is, of course, chemotherapy, which effectively kills proliferating tumor cells but has very little effect on dormant tumor cells. So can we find the inverse of that? I'll show you one example of this to begin with. So this is Notch activation, looking at the Notch intracellular domain in dormant residual lesions in mice, showing that Notch is activated in maybe 40% of residual tumor cells in these mice, and we know this is Notch activation because it's blocked by the gamma secretase inhibitor. So we basically ask the question, if we inhibit Notch with a gamma secretase inhibitor at different stages of tumor progression, what happens? Well, if we inhibit Notch in animals as they're developing primary tumors, it has no effect on primary tumor genesis at all. If once animals have developed primary tumors, we treat them with Notch inhibitors, again, it has no effect at all. And if we treat animals with recurrent tumors, again, Notch inhibition has no effect, and if anything, accelerates recurrent tumor growth at the highest doses of GSI. But in contrast, if we treat animals bearing dormant residual disease with gamma secretase inhibitor, what we see is a pronounced improvement in recurrence-free survival in animals that is dose-dependent, where at the highest dose, about 50% of animals never recur at all. So this is clearly a dormancy-specific effect. So using approaches such as this, we've been attempting to catalog different pathways in which dormant residual tumor cells show a much greater dependency than proliferating tumor cells.As one example of this, as we thought ahead towards translation, was autophagy. So my lab and others have demonstrated that autophagy is activated in dormant tumor cells. And if one does an experiment in orthotopic tumor model in mice, in which we inactivate autophagy by knocking down either ATG5 or ATG7, which are required for autophagy, form primary tumors, allow them to regress to a dormant state, and simply ask: Is there a difference in rates of recurrence? And what we see is that in tumor cells in which autophagy has been disabled, there is a pronounced improvement in recurrence-free survival, arguing that autophagy is required for recurrence. If we do an isogenic competition assay between cells that have disabled autophagy or their isogenic controls, inject them at a one-to-one ratio into mice, form primary tumors, allow them to regress and then spontaneously recur. If we look at the ratio of those two cell types, what we see is that there is no proliferative-- there is no advantage or disadvantage for autophagy-deficient cells during primary tumor formation or initial tumor regression. But during the dormancy period, a difference appears in which cells in which autophagy has been impaired are at a selective disadvantage in animals. So autophagy appears to be required for the survival of dormant tumor cells. And consistent with this, if we pharmacologically block autophagy by administering chloroquine to mice bearing dormant residual disease and treat them for two weeks and simply count the number of residual tumor cells, we see a decrement in the number of surviving residual tumor cells in mice, arguing that chloroquine in the dormancy period will kill residual tumor cells. While administering chloroquine to animals bearing dormant disease chronically and then following them to a recurrence endpoint shows that autophagy inhibition by chloroquine results in an improvement in recurrence-free survival. And so the important element here is not just that chloroquine has an effect when given to animals bearing dormant residual disease, but that it really has no effect on primary tumor genesis or recurrent tumor growth. So these are dormancy-specific vulnerabilities. In a similar manner, mTOR inhibition, in this case by rapamycin, a brief period of rapamycin treatment to inhibit mTOR in mice bearing dormant disease reduces the number of residual tumor cells, and chronic treatment dramatically improves recurrence-free survival. And again, rapamycin inhibition of primary recurrent tumors has, at best, transient and modest effects, so dormancy-specific effects. So the lesson of 15 or 20 years of studies in mice has been essentially that any drug that will kill or reduce the number of dormant residual tumor cells ultimately reads out as an improvement in recurrence-free survival and suggests that dormancy, which is in some ways akin potentially to a surveillance period, is a window of therapeutic vulnerability that may not exist in other phases. And so we sought to take advantage of these observations to translate this. We did this through the CLEVER therapeutic trials supported by a joint R01 with my close collaborator, Angie DeMichele. And essentially, these were concurrent parallel mouse and human trials targeting residual tumor cells in the surveillance phase. The targeted pathways were autophagy, mTOR signaling, or their combination, and there was a mouse preclinical trial and then a parallel patient trial. So the mouse CLEVER preclinical trial tested transient therapy and combination therapy. So we started with mice that harbored dormant residual tumor cells, and we treated them for either three weeks, nine weeks, or chronically with either rapamycin or everolimus to block mTOR, chloroquine or hydroxychloroquine to block autophagy, or the combination of everolimus and hydroxychloroquine. And we used a recurrence-free survival as an endpoint. And what we saw is shown here for everolimus is that this is the recurrence rate of untreated controls. This is chronically treated animals with everolimus showing this dramatic prolongation in recurrence-free survival. But animals treated transiently for three or nine weeks with everolimus also showed a marked improvement in recurrence-free survival. We saw a similar pattern, albeit more modest, with treatment for hydroxychloroquine or chloroquine. And I will say the combination of everolimus and hydroxychloroquine was at most modestly better than everolimus alone. So that was the recurrence-free survival endpoint. If we digitally counted the number of surviving residual tumor cells in these animals at the three-week and nine-week periods, what we found was that, again, mTOR inhibition or autophagy inhibition led to a decrement in the number of residual tumor cells, with mTOR inhibition showing the greatest effects. And interestingly, as we looked at the association between RTC burden and recurrence-free survival, what we found is thatThe greater the extent to which we depleted residual tumor cells, the greater the improvement in recurrence-free survival, and this regression actually has quite an impressive R squared value. So the extent of residual tumor cell depletion is highly correlated with the extent of improvement in recurrence-free survival. So that was the mouse trial. The phase two CLEVER patient trial for recurrence prevention had the major goals, first of feasibility and safety, since I can't tell you how many times we were told this is an impossible trial to do. No one will do it, no one will fund it. And this is really can one do a DTC-targeted trial in the surveillance window? The ancillary goal was, well, can a targeted intervention actually measurably decrease DTC burden? As an exploratory endpoint, are there any alterations of recurrence-free survival? And then long-term follow-up and repeated testing of these patients, and this trial was supported by this joint R01, major contributions from the Abramson Cancer Center through the To Prevent Translational Center of Excellence. So in the CLEVER trial, again, a phase two trial, eligible patients would've been diagnosed within five years of the study entry and had completed definitive treatment except for endocrine therapy, and had to have one of the following, either triple-negative disease, HER2 or ER-positive disease with positive lymph nodes or residual disease after neoadjuvant therapy, or ER positive disease with an oncotype score greater than 25 or a high risk of MammaPrint test. So these are all, of course, standard markers for high risk of recurrence. And then in addition, a patient had to be willing to undergo a bone marrow aspirate. For those patients who were, there was a screening bone marrow aspirate on the SURMOUNT screening study run by Dr. Demichelle. These aspirates were screened by immunohistochemistry using a pan cytokeratin antibody, so a well-worked-out biomarker for DTCs. And those DTC positive patients were randomized to one of four treatment arms. Six cycles of hydroxychloroquine alone, everolimus alone, or their combination, and then a fourth arm who were observed for three months, and then at the end of that three-month period, they were given six cycles of hydroxychloroquine and everolimus. Patients were monitored by bone marrow aspirates at three-month intervals, and after six cycles, patients who were still DTC positive were offered an additional six cycles of combination therapy. In all, 51 patients received study treatment. All arms met the criteria for feasibility. There were no unexpected adverse events, and there was no grade four or five toxicity. So it was clearly a feasible trial. So what were the other results? If we looked at treatment effects on DTCs as assessed by this immunohistochemistry endpoint, after three cycles of either hydroxychloroquine alone, everolimus alone or their combination, the DTC number was significantly reduced compared to the observation arm, with reductions in DTC burden of about 80% and a corresponding increase in the fraction of patients who did not have detectable DTCs using the IHC assay. For recurrence-free survival, which is of course an exploratory endpoint since we're not powered for that, after a median follow-up of 77 months, five-year recurrence-free survival across all groups was about 96%, which is substantially better than would be expected based on the high-risk nature of this population, and there's at least the suggestion that patients who cleared their DTCs, air quotes around cleared, they may have better outcomes, recurrence outcomes, than those who do not. But too early to tell. So the lessons of the CLEVER trials was first, we see concordant effects in patients than in mice, which gives us greater confidence in the predictive ability of these models. Transient inhibition of autophagy or mTOR during this surveillance period can both reduce RTC burden and, at least in mice and perhaps in patients, improve recurrence-free survival. Effects on RTC number and recurrence-free survival are correlated, suggesting that RTC burden may be a useful surrogate endpoint biomarker for recurrence. But I think most importantly, this is proof of concept, that a transient mechanism-based pharmacological intervention that's aimed at residual tumor cells may be a feasible approach to prevent recurrence, and there are follow-on trials being conducted by Dr. Demichelle and other investigators at our institution as well. Well, one of the things that became evident during the CLEVER trial was really a tremendous need for improved DTC biomarker development. And this really arises from a number of limitations of the immunohistochemistry test. So it's low throughput, it's low sensitivity. If you look at something that is a DTC versus something that the pathologist says is not a DTC, it's clear you need a pathologist who's trained in order to read these studies. There's also, I would say, a general dearth of information on the test characteristics of this, particularly sensitivity and specificity.And because it lacks sensitivity, there's a high false negative rate from studying just one bone marrow aspirate, since the most common result of this assay is either you detect zero DTCs or one DTC. And then, of course, since all you're getting here is a brown cell or not, you have no information on the biological properties of these cells. So it was apparent that there's a need for other tests, and there are many approaches to this. I'll tell you about one. But the need for increased sensitivity and dynamic range of a test, the hope of finding a valid surrogate biomarker to first accurately identify patients who have DTCs, and are therefore at elevated risk, to quantify their burden, to molecularly characterize their cells, to be able to assess their risk of reactivation and recurrence, and to be able to measure the effects of therapy targeting DTCs in real time. And perhaps the ultimate hope of being able to phenotype DTCs to be able to match the right therapy to the right patient based on the particular flavor of DTCs they might have. So we turn to flow cytometry as an approach for this, using markers EpCAM and HER2, as shown here in tumor cell spiking studies into normal donor bone marrows. We first depleted bone marrows of CD45 positive cells, and then, using HER2 and EpCAM, as well as CD45, stained them where we can detect DTCs compared to what are largely CD45 negative normal bone marrow cells in the shaded region. We can detect DTCs here shown down to a one in one million spiking frequency, but looked at over a wide range of dilutions, we can detect down to about one tumor cell in 10 million normal bone marrow cells, and the assay is highly linear. We can use this to detect DTC populations, here shown in metastatic patients who have substantial numbers of DTCs per mil of bone marrow aspirate. And then in early stage breast cancer patients as well, as shown by the much smaller number of cells shown here, here, and here, where we can see a range of phenotypes from cells that are HER2 positive, and I'll just say this is HER2 expressing, not HER2 amplified, EpCAM expressing, but not HER2, or some double positive cells. So we can detect these cells both in metastatic and early stage patients. If we visualize these cells using, in this case, ImageStream, we can see here, sort of using human breast cancer cell lines as an example, that claudin-low breast cancer cell lines tend to become EpCAM negative, but they still express some HER2, much lower than the amounts expressed in HER2 amplified cells. But garden variety DTCs frequently are EpCAM positive, HER2 positive at this lower endogenous level, and of course pancytokeratin positive. In some patients, we see that bone marrow DTCs become cytokeratin negative, EpCAM negative, but will still express HER2 above background levels, even in patients whose metastases have a mixture of cells expressing or not expressing cytokeratins or EpCAM. So staining for HER2 enables the detection of these claudin-low human breast cancer cells. If we do Smart-seq2 expression profiling of patient DTCs, and I'll just say we're at the very beginning of these sorts of studies, what we see is clusters of different human breast cancer cell lines, normal. These are CD45 positive normal human bone marrow cells. And then three distinct clusters of DTCs isolated from early stage breast cancer patients that have distinct expression profiles. In terms of physical properties, compared to human breast cancer cell lines in terms of size, estimated by forward scatter, that DTCs are smaller than human breast cancer cell lines growing in culture, but somewhat larger than different white blood cell types present in bone marrow. And similarly, their complexity is intermediate between breast cancer cell lines and at least monocytes and lymphocytes. So distinct physical properties. As you might predict based on levels of proliferation markers compared to human breast cancer cell lines, DTCs tend to express relatively low levels of proliferation markers. And then looked at more broadly, if we do hierarchical clustering, we see our breast cancer cell lines isolated at a single cell level, separate from our disseminated tumor cell clusters and normal bone marrow controls. Here are the three DTC clusters. Here are the different human breast cancer cell lines. We can detect mutations and copy number alterations, really across the DTCs and the human cell lines. Mutations somewhat less robustly, I will say, but this is in contrast, of course, to normal human bone marrow controls. And if we look at our DTC clusters, we see two of the clusters are luminal with estrogen response signatures, as well as some expression of ERBB2 that is lower than that in amplified cells, proliferation low, as I said. And then our third clusterAppears to be more mesenchymal, to have a basal greater than luminal set of markers, and a lower estrogen response. So clearly different flavors of patient DTCs. And then finally, I will just say that if one analyzes the genomics of these cells, that DTCs commonly share driver mutations that were present in the primary tumor with some additions, perhaps, at least the suggestion of that. And bone marrow DTCs share, of course, driver mutations with the metastases to which they gave rise, but may be missing key driver mutations that arose later on. So in this case, a patient treated with an aromatase inhibitor who developed an estrogen receptor mutation in the metastatic tumor that we did not detect in the DTCs that we analyzed. So on at least the suggestion of ongoing genomic evolution. So in conclusion, dormancy is a unique window of susceptibility to therapy. Treatments that reduce DTC burden when given during this surveillance period appear to prevent recurrence or at least have that potential. That intervention in this surveillance period does appear to be feasible and may have clinical utility. Having a better test is essential, and then really the ultimate goal is to change the paradigm. Who is it that needs therapy? That is, who are the people who actually have residual disease, and can you spare the people further therapy who do not have detectable residual disease? What kind of therapy should be given? Well, I think the direction should be analyzing unique dependencies of dormant tumor cells. Which therapy? Well, the best thing would be to select therapies based on the properties of those cells. And then finally, did the therapy work? The ability to do real-time pharmacodynamic monitoring of patients treated with DTC targeting therapies. And ultimately putting these things together, I think raises the hope for the ability to prevent breast cancer recurrence by targeting DTCs as well as its associated mortality. And with that, I'll thank the many people who contributed to this, but a particular shout-out to Angie DeMichele, who's been a close collaborator and research partner for decades now, and without whom this work would never have been done. And then also to our patients and our Translational Center of Excellence. So with that, I'll thank you. And I think we have time for just a couple of questions since I went a little bit over. Hi. If you can go to the microphones. Yeah. Thank you, Lewis, for such a great talk. I'm wondering, yesterday we heard a lot about persister cells. And I'm wondering what is the overlap or the difference between DTCs and persister cells? Is it a working definition or are there clearly differences? Yeah. So persister cells versus cellular dormancy, and I'd say there's clearly overlap in those Venn diagrams. The definition of cellular dormancy and reversible quiescence is a very straightforward definition. Persister cells undoubtedly share properties with cells that survive therapy in a state of cellular dormancy, in terms of the adaptive sorts of mechanisms. There are many studies of persister cells which are perhaps less focused on the dormant state per se, as well as the ability to persist. So I would say they are distinct, but undoubtedly overlapping. Yeah. Hi. SJ Shi from Cellensia. We are working on this functional CTC organoid. So I'm very interested and curious. Does that mean that this DTC is the CTC already find a new home and decide to stay there dormant for years? So during that time, there's no CTC or cell-free DNA to be functional, so you're focusing on this stage. So they are the, call it CTC already adapt to a new environment, a totally new entity. And my second question is that- Well, why don't we leave it for one question since we're kind of running out of time, and we have a couple other people. Oh, okay. Sure. Yeah, go ahead. So the question is CTCs, DTCs, what's the relationship? Yes. I'd say the easy answer is we don't know, but my nickel guess, based on the data that we see, and there are other studies I didn't have time to talk about, is DTCs are very long-lived. They sit in tissues for decades. My suspicion is many of these early-stage patients are CTC negative, and my suspicion would be in order to have CTCs in the circulation, or perhaps even detectable levels of ctDNA, there probably must be some proliferation in cell death that's happening. So I think if there's a strict phase of quiescence, it may well be the case that there are not biomarkers in the circulation. Once someone reactivates, and you detect it in the circulation, my best guess is that you're essentially talking about a subclinical recurrence that will come to clinical detection in some number of months, but that's a guess. Time for two quick questions. The question is, can you detect similar to DTC's cells already in tumor biopsies, maybe in metastatic breast cancer? I'm sorry. I didn't catch you. Can you detect the DTC-like molecules in the biopsies of patients, at the time when they are treated and then predicted or not? Now, maybe with the- Yeah. If you're talking about in- ... single cells. Yeah. If you're talking about in bone marrow aspirates, one can detect these cells both before therapy or in some patients after therapy. I'm thinking in the biopsy of the tumor itself. Yeah. So since there's no label in a primary tumor for an extended label for proliferation, it's unclear to me at the moment how one would detect it.But if we had better markers, perhaps that would be a possibility. And then last question. Yes. ctDNA as a surrogate for MRD, what do you think? So related to the prior question, at least in our studies, and I think in many other studies, once ctDNA appears in the circulation, typically recurrences are clinically evident certainly within a year. So my suspicion is that's marking the reactivation of one or more quiescent clones. But we'll see as the ctDNA technologies get better and better and better, maybe it will be detectable. So with that, I am honored to introduce our next speaker, Dr Amulya Sreekumar, who was one of the 11 senior trainees who was selected by AACR as a Next Gen Star, and Amulya will be leaving my laboratory and starting her own independent laboratory in Berlin in a week. So please welcome Dr Sreekumar. Thank you, Lewis, for that introduction. I also wanted to thank the organizers of the AACR for the Next Gen Star Award. I'm really excited to talk to you about this study that really, when we first saw it, took us by surprise, and challenged our notion of mammary lineage fidelity during breast cancer progression. The title of my talk is Residual Breast Cancer Cells Co-opt SOX5-driven Endochondral Ossification to Maintain Dormancy. As Lewis already introduced, the key question driving the clinical finding that's driving the research that we do in the lab is that breast cancer is the leading cause of cancer-related deaths in women worldwide. So advanced diagnostics as well as targeted therapies have rendered the primary tumors highly curable. But in about 30% of patients, these tumors come back after several years or decades and are incurable, and account for the vast majority of mortality that's associated with this disease. So this is plotted here as tumor burden over time, where you can see this long clinical latency window following definitive treatment of the primary tumor and prior to relapse, which necessitates a better understanding of the biology underlying minimal residual disease, as seen here. More specifically, what I mean by that is the residual tumor cells, or RTCs, as I'll refer to them, that make up MRD. If we could understand the biology by which they persist, survive, and ultimately give rise to recurrences, we could devise ways to either eliminate them or prevent their outgrowth in order to improve patient outcomes. So one of the emerging hypotheses that's attractive and is showing up from existing data is that the selective pressures such as therapy and metastasis might induce a bottleneck that results in convergent phenotypes on these RTCs, as you can see here. And what I mean by that is that maybe there is the possibility that RTCs converge upon phenotypes that are independent of the specific stressor that caused them to achieve that state. One such adaptive mechanism that's been observed is the acquisition of a reversible state of non-proliferation by these RTCs that we term dormancy. So this has actually been observed in patients where RTCs are observed following both therapy as well as metastasis, and are thought to be reversibly quiescent such that they can give rise to recurrences after a long period of time. So this also constitutes an attractive model for why these cells persist for long periods of time in the latency window prior to recurrence. So Lewis already introduced the genetically engineered mouse models that were developed in his lab that I was using to identify mechanisms by which RTCs selectively survive and give rise to recurrence. Just as a reminder, the model that I'll be focusing on is this MtbTan model, where you can use doxycycline to toggle oncogenes on or off in the mammary epithelium, in this case, of the HER2 oncogene. So in the presence of dox, HER2 drives primary tumor formation, and withdrawal of dox causes its regression to a non-palpable state. And after an extended period of time, these tumors recur in a manner that's independent of that HER2 driver oncogene. So using the cells derived from this MtbTan model, I created a Cas9 constitutive expressing version of these cell lines and used that for a targeted CRISPR-Cas9 screen to identify mechanisms that drive RTC survival as well as their subsequent recurrence. To do so, we designed a guide RNA library targeting genes that are selectively enriched in dormant time points relative to their proliferative counterparts. And this resulted in a guide RNA library of about 500 guide RNAs that we transduced into this Cas9 expressing cell line at a low multiplicity of infection to ensure a single guide RNA per cell. What we can do with that is that the presence or the abundance of guide RNAs can serve as a proxy for cell numbers. So after injection into mice, we harvested primary tumors, multiple residual lesion time points at both early and late stages, as well as recurrent tumors, and performed next gen sequencing to identify guide RNA abundance at each of these time points. So the first project that resulted from this was actually published a few years ago and identified the enrichment of a sugar-modified molecule called proteoglycan as being selectively enriched and required for dormant RTC survival early during this process.For the study that I'll describe today, we were interested in asking what are the mechanisms enabling these RTCs to persist in a dormant state, for which we focused on the latter half of the screen and looked for clonal enrichment at these later stage time points. So I mentioned that we had a library of about 500 guide RNAs, so the expected guide RNA abundance per guide in this pool is 0.2%. To call a clonally enriching guide RNA, we used a threshold of 5%, which is about 25-fold higher than this existing baseline abundance. And we looked for abundance that's increasing at these late stage time points in late residual lesions as well as recurrences. These data are depicted here in stack bar plots, and what you're looking at is the abundance of the top 10 guide RNAs in each of these samples boxed by time points. And looking across this, you can already tell that there is a stochastic enrichment of guides as you go through this time course. To sort of strengthen our confidence in the guide RNAs that we identify using this approach, we additionally looked for guides that were enriched in two or more samples, which is represented by the colored bars. And of these hits, SOX5, which is indicated here in samples that had enrichment by the red arrows, demonstrated enrichment in multiple samples as well as with multiple guide RNAs. So we proceeded with validating SOX5 as one of our top hits from the screen. So to do so, we first performed a recurrence-free survival assay where we introduced cells containing either the guide RNA targeting the inner Trosa locus or one targeting SOX5 that we identified from the screen orthotopically injected into the inguinal fat pad in the mammary gland here, and waited for spontaneous recurrences after Her2 downregulation. So the clonal enrichment we observed from the screen predicted to us that likely loss of SOX5 would result in an acceleration of recurrence, which is actually what we see here with two independent guide RNAs targeting SOX5 that we identified in the screen. You see this acceleration of recurrence indicated with these hazard ratios of about 2.5 to three. We next wanted to ask if this was also true in a non-orthotopic site. So for this, we performed an experimental metastasis assay where we introduced the same cells into the lung, allowed them to grow to where they demonstrated saturating luminescence, and then withdrew doxycycline to turn off the oncogene. So you can see here at day zero where we withdrew dox, you can see an acute diminishment in the luminescence signal that sustained over several months. In the SOX5 knockout arm, we see a similar diminishment in the signal of luminescence signal. But interestingly, you start to see the signal come back at later time points in this representative mouse. And this was also seen across the experiment, as indicated by the Kaplan-Meier curves, that loss of SOX5, even at a non-orthotopic site, was accelerating time to recurrence. So finally, we wanted to ask if this is related to the ability of tumor cells to proliferate in this latent phase. To do so, we used cells from the MTB-TAN mice containing either a scrambled hairpin or a hairpin targeting SOX5 to knock down SOX5. And what you see here is there were no differences in the Ki-67 prevalence in proliferative time points. But when you look at the residual lesions, but most profoundly at the late residual lesions, you see this increase when you lose SOX5 in the number of tumor cells that are proliferating. So taken together, these data suggest to us that loss of SOX5 is promoting recurrence by accelerating proliferation in this latent window. Our next question was: how might SOX5 be functioning to do so? And SOX5 is a transcription factor with very defined developmental roles, where it has functions in both neural development as well as in specifying cartilage. Interestingly, when we looked at gene expression profiles from both in vivo and in vitro experiments using these MTB-TAN derived cells, we observed an enrichment of the gene ontology term ossification in both settings. Additionally, we also observed enrichment of terms such as endochondral bone growth, bone mineralization, chondrocyte, and osteoblast differentiation. And this was really striking to us because we were looking at pure mammary epithelial cell populations in vitro as well as in vivo, where the cells were sorted based on a GFP label they had. So it suggested that mammary epithelial cells were undergoing a differentiation event that might render them more mesodermal in nature. But we wanted to confirm the prediction from the gene expression here. And to do this, we decided to use alcian blue and Von Kossa staining, which is normally used to look at endochondral ossification in embryonic development. So what is this process? Endochondral ossification is a process wherein you require a cartilage template to form bone, and this is visualized histologically using alcian blue and Von Kossa staining.Alcian blue is a stain that marks proteoglycans that are abundantly made by chondrocytes. And if you remember, it's also the sugar modified molecules that we had previously identified in RTCs as being essential for their survival. Von Kossa staining marks the calcified areas in the bone and is basically silver nitrate that replaces calcium phosphate. So we first wanted to use this histological approach to confirm the gene expression findings that we saw. So as a positive control here, I used an embryonic day 16.5 embryo and looked at Von Kossa staining and Alcian blue staining and confirmed that this is indeed marking the calcified areas as well as cartilage. We then looked at the staining pattern in primary tumors where we observed negligible staining. However, when you look at residual lesions, you see an abundance of proteoglycans being expressed, as well as Von Kossa staining begin to appear in a temporal manner, really confirming what the gene expression data were telling us, that there is this ossification-like process occurring. To further confirm this, we wanted to use small animal imaging, for which we chose to use a PET/CT approach. For PET, we used 18F sodium fluoride as our positron emitting tracer, and 18F sodium fluoride marks hydroxyapatite in the bone and is a very specific marker for the skeleton. With CT, we used X-rays to look at radiodense areas in the skeleton. And I hope you can see here that those two signals are almost indistinguishable. So the heat map here shows you PET tracer uptake overlaid on top of the CT signal. So what you're looking at here is actually from a tumor-bearing mouse. In these regions marked by the red dotted lines, that's the orthotopic mammary glands where we can see some PET tracer uptake. But because that was really hard for us to resolve from the knee joint, we decided to excise the glands from tracer injected animals and perform PET/CT ex vivo, which is what you're looking at here. So in our negative control, which is just medium injected without cells into the orthotopic site, you don't see PET tracer uptake. We see minimal PET tracer or CT signal in two primary tumor time points. However, when you look at residual lesions, you start to see a progressive increase in tracer uptake that is coincident with CT signal, seen here in white and marked with the arrowheads. Suggesting that this is a phenomenon that's occurring and it's getting aggravated during late residual lesion time points. And this is perhaps best exemplified by sections that we got from recurrences where we could still see a remnant residual lesion. So what you're looking at here in the H&E is a more eosinophilic area marked with the black dotted line, where we saw Alcian blue and Von Kossa staining. And if we looked at serial sections from this tumor, you can see that that eosinophilic area has very minimal Ki67 staining, as well as 24-hour EdU uptake, suggesting that the tumor cells in that area are actually low proliferating compared to the mass around it. Again, driving home that inverse correlation between ossification and proliferation. And then our burning question really was, is this actually regulated by SOX5 in the residual lesions? To do this, we used the knockdown cells that I previously described with the hairpins, and we found that loss of SOX5 massively decreased the calcification and PET tracer uptake that we normally see in the residual lesions in the control. And this is quantified here on the right where you see a substantial decrease in signal. And then finally we asked, is this relevant to patients? So in the neoadjuvant therapy data sets that Lewis previously described, we asked if we see an adaptive increase in chondrogenesis and osteogenesis gene expression signatures, which when significant is highlighted here in pink. And we did indeed see this in both estrogen receptor negative as well as estrogen receptor positive patient data sets. And in a subset of these patients for whom we had both post neoadjuvant therapy gene expression data as well as long-term recurrence outcomes, we found that enrichment of these signatures was associated with a delay in time to recurrence. Again, speaking to the fact that these signatures are associated with indolent disease progression. To summarize what I showed you, SOX5 loss promotes RTC proliferation and accelerates recurrence. It also impairs chondrogenic and osteogenic differentiation. We found a surprising dormant residual lesion phenomenon where you see SOX5 regulated endochondral ossification, and that this osteochondrogenic gene expression is associated with improved recurrence-free survival in patients. So there are several details in this story that I didn't have time to get to, but here's a QR code that would take you to the paper that just came out in April's issue of Cancer Discovery. And with that, I'd like to thank Lewis, the members of the Chodosh Lab, our collaborators in radiology, the Department of Cancer Biology for always being so supportive, and my former funding sources. Thank you. We have time for some questions We see microcalcifications in breast tissue very commonly associated with ductal carcinoma in situ. The calcifications are different than benign calcifications, and radiologists can identify them. Do you think that that phenomenon could be occurring in pre-neoplasia as well as the inflammatory recurrence disease? I think that's a great question. This has come up before, especially because mammograms can quite routinely detect microcalcifications. Whether that is the same process or, as you mentioned, it's very dependent on the type of calcification and the properties associated with it as to whether it gets called relevant to disease or requiring further management. We would propose that this is part of an active program of differentiation compared to calcification, which can also occur in an independent manner, for example, if you have necrosis or even from milk in the ducts. So it would require additional study to ask if there is maybe an association with a chondrogenic phenotype as well that might lead to more of a hypothesis related to what we're seeing in indolence. Thank you. Great talk. I have a question that is related. Did you try to see if there is a correlation between the chondrogenic signature and H&E images from the samples collected from those tumors? From patient samples, you mean? Yeah, from patient samples. That's something we're actually in the process of doing. We have access to patient samples, and we're trying to see if we can identify, as predicted by the gene expression signatures, are we also able to look for these features there. Thank you. Hi there. Thanks for the great talk and very interesting work. I'm wondering if you know what the exact link is between loss of SOX5 and enhancement of proliferation. What part of the cell cycle machinery does loss of SOX5 ultimately drive? Mm-hmm. And then if there's time, I'll have a second question. Okay. That's a good question, actually. There are many parts of the study that require further elucidation mechanistically, including how SOX5 is being temporally regulated, that we don't have a good handle on. It's interesting you mention that. There are contexts in which SOX5 has been studied predominantly for its oncogenic functions. One that is more similar to what we see with quiescence is actually in the context of neurogenesis, where loss of SOX5 can promote a proliferative burst and deplete neural progenitor cells. Why that context matters and what it's actually doing to the cell cycle remains to be studied, but I think constitute important questions here. Thanks. I think I'll let the other person go. One last question, which will be the person behind you. Great talk. Thanks. So one question for the preclinical translation aspect is, beyond the genetic SOX5 knockdown or knockout, is there anything on market for the osteogenesis blocking at the residual state that you could do in vivo or preclinically? Yeah. So I think the proposal or the best approach would be the opposite. If we could try and actually enrich this program and maintain cells in a dormant state, that would be great. We don't know how to go about that yet, but actually, as I mentioned in the previous response, having a better handle on how SOX5 is being temporally activated would get us there. So that's future directions and things we're working on. And with that, we'll thank Dr. Srikumar for an excellent talk. Our next speaker, we're honored, came all the way from Germany, across the ocean for this. We are honored to have Dr. Christoph Klein, who will bring things home by telling us things that probably most of us here don't know about the molecular characteristics of human DTCs. Yeah. Thank you very much for this invitation here. You learned yesterday evening, MRD is a hot topic since three years. I was surprised to learn this, because actually the study of the seed of metastasis is actually much older. So in 1987, Gertrud Müller had a wonderful idea, which is still used today, to use cytokeratin and EpCAM as markers to detect disseminated tumor cells in bone marrow. And as it has been explained, and I will not go deeper in this, this is an opportunity, after surgery or local treatment, to prevent relapses that may arise from disseminated cancer cells. This research has gone forward. Many patients have been screened. Also, this data has been shown before. And what you can see here is, yes, disseminated tumor cells have an impact, but counting is certainly not enough to achieve the goals. And therefore, more than 20 years ago, we started to isolate these cells by micromanipulation because these cells are super rare. You find one or two cells in one million bone marrow cells, and we use two markers, cytokeratin to analyze cells for DNA analysis, and EpCAM for RNA. And for the rest of my talk, I will make a very important differentiation. I classify the cells by the clinical stage of the patient, where we isolated these cells. So if a patient has no metastasis, I will call the cellsM0 disseminated cancer cells for M0 stage, and if the patient has metastases, M1 disseminated cancer cells. I will summarize a few results very quickly. So what we learned in mice and men or women was that dissemination actually starts very early. So here you can see this is a spontaneous mouse model where we followed the growth of the primary tumor, and we asked when can we find cells in bone marrow or other organs in transgenic mice? And the surprising finding was long before we have a tumor. That is actually why I call the cells mostly disseminated cancer cells and not tumor cells, because we have no tumor when dissemination actually starts. Now, this is similar also in patients. Here you can see the sizes, the diameters of the primary tumors, and then from patients, many patients, and you can see there's not much change. In most cases, you have no disseminated cancer cells, and if you find them, it's basically independent of the size of the primary tumor. So seeding starts early, and there is no increase as the primary tumor grows. And you can plot the same data in a different way, where you determine the risk of de novo seeding at a given size, and you can see that at the diagnostic size, seeding risk has already tenfold decreased. So seeding is an early process, something that we do not really understand yet, but it goes with the molecular biology that we found or the genomic profiles. Here, for example, you can see a typical histogram of copy number alterations. You see the gains here on chromosome one or eight or 17 in primary tumors. You can find very similar patterns in the metastasis. But when you look before metastasis in disseminated tumor cells, cancer cells, there are certain genetic changes that are not present yet, which raises a question, what happens if we cut out this tumor, the source of cells that have this amplification? It probably means that evolution needs to reacquire these changes. And therefore, we started to address the phylogenetics of individual histories by going for a single cell analysis of the phylogeny of disseminated tumor cells. And we teamed up with the Weizmann Institute in Israel, with Udi Shapiro and Ofir Raz, and tried to isolate, generate the history of disseminated cancer cells. So what did we do? We isolated the cells. We did a whole genome amplification. We sequenced 12,000 microsatellite sites and therefore performed a huge paternity test, basically. The problem is you do not have parents; you have just the descendants of the cells. But we had a very simple question. We wanted to understand where does the disseminated cancer cell leave the primary tumor, at what stage? Now I'm showing only one patient in detail, but what you get when you do this with several patients is first you get a very nice separation here between the cancer cells and the normal cells in most of the cases. I'm showing you one breast cancer patient where we took single tumor cells. At surgery, bone marrow aspiration was performed. We isolated the cells from the bone marrow. We used control cells of different types. And the patient, seven years later, relapsed with metastasis, and at that time point, we could isolate circulating tumor cells from the bloodstream that bona fide were derived from the metastasis. And we had the simple question: Are the disseminated cancer cells derived from the most advanced clone of the primary tumor, or does it develop somehow in parallel? So this is the disease course of the patient. Primary tumor was taken 2009, and the bone marrow cells as well. Seven years later, we got the circulating tumor cells, and in between, the patient received all types of therapies. And one therapy I just want to point out here, it was tamoxifen, so anti-hormonal therapy. And this is the tree that we get. What you can see here is the primary tumor cells, the metastasis-derived circulating tumor cells, and the non-tumor cells. And you can see here these triangles. These are boot signals of significance in a way that the tree at this position is very stable. So these are in bootstrap experiments. Here is a clear separation. The arrow points to the head. So this area is completely different from this area. This area here is completely different from this area. And the first thing that you notice, the primary tumor and the met-derived CTCs, they are very different. They separated early on. Now, the disseminated cancer cells, seven years before, one is here, one is here, and then surprisingly, we find circulating tumor cells very well separated from all the rest of this area. And then you look for estrogen receptor mutations. Here in the primary tumor, not treated, all Y type. Here, treated, you find the mutations.Now here you've the disseminated cancer cells. They don't have the mutation, but one of the CTCs actually has it, surprisingly, completely unrelated to this cell here. So the clustering seems to show that the circulating tumor cells left very long time ago, and because some of them cluster closer to the normal cells, an early distant. The circulating tumor cells here on this side, clustering with the normal cells may have been dormant for quite some time because they did not evolve into the killing clone here. And also it may reflect that important mutations may actually acquired independently. So very briefly to summarize this, I set these points. So it became very clear we need phenotypic information on these disseminated cancer cells, and that's why we isolated the cells and performed whole transcriptome amplification and RNAseq of these cells. And I'm showing you now data that are derived from more than 1,000 patients where we search for these individual cells from different cancer types, breast cancer, prostate, lung, and esophageal cancer. Now EpCAM is not the best marker for cancer cells because it's not so specific as cytokeratin is, because you will find in normal healthy donors also EpCAM positive cells, which is a big problem. So we noted that cancer-derived cells are stronger stained. They're brighter. They're a little bit larger. It's no problem if the patient has metastasis to identify morphologically clear cancer cells in the bone marrow. They are very, very different. So these were our criteria. And so we isolated 1,000 cells, and the data were long time unclear until Huijin Kirkulu, our bioinformatician, brought some sense into it. So these are the cells that we isolated from the patient, and we quickly identified that by EpCAM staining, we had isolated different cell types, which expressed EpCAM, plasma cells, erythroid progenitor cells, some rare immune cells. But then we had a big cluster here with the M0 derived disseminated cancer cells and the metastasis derived cells. And when we projected these cells into the bone marrow atlas that had been published, we saw that the erythroblasts integrated here, the plasma cells integrated here into the bone marrow atlas. But we find two new continents in the bone marrow atlas, one for the M0 DCCs, and here the M1 DCCs clustering closely to the stromal cells of bone marrow, rare cells, similar to the fibroblastic cells here, the M1 DCCs. So that was strange that they clustered so far apart from each other, and we wondered what is the differences here. So we looked into the gene expression profiles with unsupervised clustering. And since we had different cancer types, we used the M1 DCCs and did unsupervised clustering. What you can see here, the algorithm can perfectly separate these cells. It's clear what is a breast cancer, what a prostate, what a lung cancer cell. When we did the same with the M0 DCCs, the surprise was big. We couldn't differentiate these cells. What looks like kind of a cluster here for breast cancer versus the others is nothing but the sex of the patient. If we remove the sex-associated genes, we will have a complete mix of these cells. It looked like the cells had completely forgotten where they are coming from. So we compared these two cells here, and we found for the M1 DCCs, we found expected terms like luminal epithelial cells, epithelial cells, epithelium mesenchymal translation. The surprise came when we looked into the M0 DCCs. You find here stem, stem, embryonic, placenta, pluripotent stem cells, terms related to complete differentiation loss of these cells. So we looked, we used algorithms for stemness and high potency scores and found that actually the M0 DCCs had very, very high stemness and potency scores, even higher than ERPs. And this is surprising because ERPs are actually the cells in the adult with the highest stemness and potency scores because they produce 2.5 million red blood cells a second. So here you can see this, and this is data from fetal cells versus adult cells, and the fetal cells always have the higher score except here for the erythroid progenitor cells. Now we use the data, the stemness scores, and applied it to the lung cancer patients where we had the survivals quickly. So we took from each patient where we had isolated a cell, the highest stemness score, and if the patient had a high stemness score, it was in this category, or it had a low stemness score. And what you can see, the cells with the highest stemness scores, the EpCAM positive cells with the highest stemness score, can perfectly separate the survival of the patients.So in this case here, that in about two or three, two and a half years, all patients were dead that had a single cell with such a high stemness score. Now, this was striking, so we moved forward, not relying on the bone marrow atlas, but taking an atlas of all human cells, and we projected the cells that we had isolated from the patients into the human atlas, which also contained embryonal and fetus-derived cells. And what you can see here, that the M0 DCCs actually cluster very closely to human embryonic stem cells. Now, this was really surprising, and we looked further when in embryonic development do we find the highest similarity. And we started with the zygote, the fertilized egg, and went on to the 16 cell stage. And we find in these cells incredibly high enrichment of genes that are linked to these very early stages. Whereas the standard classical metastatic cells that we can isolate does not enrich for these cells, because they undergo a later stage gastrulation, which comes later in embryology, and express these typical genes. So maybe we have a very early stage here of embryonic development somehow activated in disseminated cancer cells, and so this is our current focus in our research to elucidate this a little longer. But then was obviously the question for breast cancer, do we also find a correlation with this? And this is very early data that we just got two weeks ago, the follow-up of the breast cancer patients. And what you can see here, this is just the finding, the separation of group patients where we did not find EpCAM positive cells and patients where we find EpCAM positive cells. And you can already see that EpCAM is actually a good marker to find these cells, because most of the events occurred in the EpCAM positive breast cancer population. So most patients we will identify. And now we use this group here with the EpCAM positive cells and separated those into those that had these embryo-like cells and those that didn't. And this is very promising, I think, although it is not significant yet because very early stage here in follow-up analysis. But so far, all events occurred in the group with the embryo-like disseminated cancer cells. So there's one thing that I would like to point out, which I find striking. I mentioned that we could not separate the cells from breast, lung, and prostate cancer. But in lung cancer, when we find these embryo-like cells, most patients die here in an early time point. In breast cancer, most of the patients die later. So the cells are very similar, but the clinical course is very different in these two cancers. And this is very puzzling, and we started to look into this. And with the help of David Blumenthal, I learned that you can have similar gene expression profiles, but different gene regulatory networks underlying them. And here in this regard, you can see here the comparison between lung cancer and breast cancer are very different. So the gene regulatory networks that regulate the speed of regrowth, if this is linked to it, are very different here. So now we have this phenotype. The metastatic seed appears to be very different from manifest metastasis in primary tumors. The embryo-like DCCs form a subgroup that is associated with risk of metastasis. And although these embryo-like DCC phenotypes appear to be similar, there is some underlying differences that needs to be explored, and maybe these differences are linked to the dormancy periods that we see more frequently in breast cancer. So the last few minutes, I just want to point out the direction of the research that we are taking now, and I just want to introduce you an approach to get a little bit more ideas about the biology in patients. And here in the lab Ati Aliozin and Speaking of Cheech develop a new way to analyze the phospho proteome signaling pathways in single cells. I just want to mention this. This is a technology which has been used before. You label antibodies with a DNA tag. The specificity here for us is that we add as a tag exactly the same adapters that we use to amplify the genome of a single cell. So when we isolate a single cell, we cut the genome into pieces, link adapters, and do PCR. And now we have the same PCR primers here on the identifier that tells us which antibody has bound. So when we isolate the cell together with the antibody, we can sequence the antibodies bound and the genomic amplification... changes. So basically, the experiment looks like you detect the cells with cytokeratin and EpCAM, you add the DNA tagged antibodies, you isolate the cells, and you can read out the whole genome and the profile of the antibody. And you can do this also with patient samples. And the method is very quantitative, which I want to show you here. So we use different cell lines that express different levels of HER2, and then you have the antibody here. You detect it with a secondary antibody, so the tagged antibody is detected by a fluorescent antibody, and then you can correlate the PCR value with the intensity of the staining. And this is really quantitatively doable here for even single cells. So with that, we wanted to look as a proof of concept into the activation pathway here, HER2, phosphor-HER2, phosphor, phosphor-Akt, phosphor-Stat. We used this, and I just want to show you one patient because I found this interesting and a bit provocative. It's a neoadjuvant setting. We take bone marrow sample when the port is given to the patient, we get the first bone marrow sample, then the patient is treated, and three months later, surgery is performed, and the second bone marrow sample is taken. So here we had a HER2 amplified breast cancer patient. We isolated one cell, and the patient got treatment against HER2, and we could identify in this patient six cells after treatment and isolated these cells. And here is the result. You can see here, I apologize for the poor quality of the-- this is a Mac problem. So you have here the normal cells for the readout of the antibody, and you have the putative DCCs, and you have a positive control. And you can see that in all cells, these pathways are all activated. And one cell has a very high increase in HER2 staining, and it also has an amplification here of the locus. So it goes together, and it tells you that even after three months treatment with this antibody, you can detect resistant cells against this treatment in the bone marrow of a patient. So these are the questions that we are currently working on. So which pathways actually regulate this phenotype of the DCCs? Has this pathway maybe something to do with the therapy resistance? Is there a functional link? And then when there is such a specific phenotype, and I think probably we do not have a two cell embryo like cell normally in the body, this could hint that there are additional targets to be identified, which may be specific for these disseminated cancer cells. I showed the people who did the work with some pictures, but there are many more people involved who isolated these cells, who did various aspects in this work. I want to thank the patients that donated the bone marrow for these studies over many, many years. And if you are interested in the topic, I just want to use the opportunity to inform you that there will be a conference next year also on dormancy and minimal residual disease. Thank you very much for your attention. Questions for Dr. Klein? That was fascinating. Congratulations. I'm curious in your perspective about whether the founder, the truncal founder mutation for those M zero cells is in a pluripotent, more stem-like cell to begin with. And also, is there an earlier onset of disease? Actually, this is the weakest point in my story that you're asking. We were looking obviously also for truncal mutations so far, but we have done this with a panel sequencing, and we find mutations in these cells. And actually in lung cancer, we also find shared mutations between the primary tumor. But in breast cancer, it's very difficult. Breast cancer in general is not a typical cancer for mutations, that is driven by mutations. I can at least say so far we have not identified PI3 kinase mutations or P53 mutations shared between these cells and the primary tumor. So maybe the beginning of cancer and dissemination is something different than our canonical mutations. Thank you. Hi, Christoph. So the cell that you described, the embryonic-like disseminated cancer, sounds really like a cancer stem cell. So there's studies that show that that EMT embryo-like state is kind of like a cancer stem cell. I wonder then, can you actually find them now that you have a signature in the primary tumors? And I'm wondering if even if they are there, whether there's an acquired dormancy phenotype on top of that embryonic stem-like signatureDoes that make them actually different when they do get to the bone as they enter dormancy? So EMT is not the prevalent term that we call enriched in these disseminated cancer cells. So it seems to be before gastrulation. That was actually why we looked into this zygote two-cell early embryos, because we did not identify EMT as this characteristic. Mm-hmm. Do we find cells of this type in the primary tumor? So Huijun has looked in published data of lung cancer, and actually there is a few studies where lots of cancer cells were sequenced, and she finds a few cells resembling this phenotype, this gene expression pattern. This needs to be extended, the studies in the primary tumors that we looked in of our patients, we couldn't do this because all this tumor went into paraffin. Hi, Christoph. Patrick from Genentech. I just have a question related to the profiling of the M0, M1. Do you see the interleukin 6 trans signaling in those cells? Yeah. Sort of to ask if that embryo-like signature contains some of the interleukin 6 signaling based on what you've previously- Yes ... mentioned. Yes, absolutely. Yeah. So that is the- Yeah. IL6 signaling is activated, yeah. All right. In these cells. Thank you. There you go. Hi, thank you for your presentation. I really appreciate it. I have a question regarding the embryoid DCC population get resistant, and you show that we have the up-regulation of HER2 in associated with the perforation of ERK, AKT, and STAT3. So I just wonder, do you have any plan to target those kinase? Because you say that they have no specific drug or there are no therapies to target the embryonic stem cell right now. So do you have any plan to target those kinase based on your pre-data that you already found? I'm sorry. I had difficulties to understand your question. So you were asking about these pathways that we found activated in these four cells that we analyzed? Yes. So you show that embryoid DCC have the in screen expansion of the phosphor ERK, phosphor AKT, and phosphor STAT3. So I just question, did you have plan to try some targeting those kinase to figure out how we can prevent embryonic stem cell? Well, at the moment, this is just a proof of concept. So we were just establishing the method, and we just wondered whether-- I actually expected all these cells completely dormant, right? So no phosphor HER2, no phosphor AKT, no phosphor STAT. Maybe STAT, yeah, activated by some cytokine signals. How this goes further, we will probably focus on-- So basically you can use many antibodies. You can scale this up. It's just a matter of money, 20, 50, 100 proteins that you label your antibodies. But these cells are so rare, so you need to get as much information out of a single cell. This is always the problem, because this nightmare of thousand patients investigating it, it's ongoing thing since 15 years. It's really a lot of work. And so we will scale this up and look for activation pathways, embryonic genes, and candidate dormancy pathways that may be activated. Thank you so much. One last question. Okay. Thank you so much for your talk. I was just wondering, since you do see IL6 upregulation, you mentioned, and then also the phospho-ERK, phospho-AKT signaling, how are you differentiating these cells from senescent cells that are potentially having upregulation of these same pathways and then that's driving the SAS production? Yeah. So you're completely right. There's more questions than answers at the moment. And, no, I can't tell you exactly what these cells are, what the biology is at the moment. The activation of the IL6 pathway, we have seen before in the transcriptomic data. Mm-hmm. So I think they are not senescent cells at the moment, so by other points from these profiles. Okay. Thank you. Yeah. And with that, that brings this session to a close. Please join me in thanking our speakers.