Good afternoon, please welcome to one of the exciting session personally. I feel like this is the future of the medicine Digging deeper use of omics to improve therapy So we have three announcements to make first if you would like to ask a question without coming to Microphone you can scan the QR code and if you want to ask question in person Here is the microphone and the third announcement is like after this session We have a reception at B405 to celebrate today's session And Our first speaker is dr. Hemus Hemus from Cleveland Clinic And she will be talking about the convolution of the cell cycles impact on prostate cancer transcriptome to find new drug targets Great. Thank you for that introduction I would also like to thank the organizers for the opportunity present some of our work here today I have to start with a disclosure. This is a very broad title I'm actually going to talk about our work isolating a new regulator of prostate cancer cell proliferation and cell cycle progression And then show vignettes of data that illustrate how omics have helped us understand the biology of this particular protein in prostate cancer Its role in prostate cancer progression and its evaluation of its potential as a therapeutic target So I have no disclosures And I was asked to put up this slide. So let it sit for a few seconds so people can scan if they want to Know that I have so we work on prostate cancer, and I think for this audience. I don't have to introduce The concept that this is an in a significant health problem. It remains the second leading cause of cancer deaths in the u.s. And these deaths actually occur because the treatments for Advanced prostate cancer for metastatic prostate cancer stop working resistance develops to them now since prostate cancer is a hormonal cancer It's driven by androgen receptor Target therapies that target metastatic disease mostly rely on inhibiting androgen activation of the androgen receptor as you can see from this slide Consecutive treatments eventually initially work well the induced remission But they stop working until advanced disease occurs about 80% of the advanced cases are still relying on androgen receptor signaling But 20% are now viewed to be of the neuroendocrine phenotype which are also hard to treat So in view of these kind of limitations to treatments a lot of groups including ours have been trying to conceive New functional functionally diverse targets that could be inhibited to prevent The growth of cancers that have become resistant to these available treatments I'm going to show you the protein that we've become interested in this is a protein known as cytokines You may not know it. We actually didn't know it if you didn't have it on our radar It's a poorly characterized kinase that belongs to the AGC family of serine-threonine kinases and Was originally identified because of its interaction with the Rho and RAC proteins and viewed as being an effector of these proteins Which incidentally is why we started studying it We had this line of investigation that looked at AR activation of RhoA signaling and we viewed this as a potential effector Which should help us elucidate that pathway in our lab in a better way However, as we started this work, we noticed that this protein actually didn't really seem to be a very effective RhoA effector But instead it seems to be functioning as a major regulator of cell proliferation Which actually fits with how cytokine kinase is now viewed in the field It's really recognized as an important mitotic kinase that is essential to control cell division and cytokinesis steps Before I show you the data that got us interested in this protein I want to make two points. The gene that encodes cytokine kinase actually encodes two major isoforms One is cytokine kinase contains the kinase domain Another one is similar in structure, but lacks the kinase domain. Those two proteins are very different They are characterized by different tissue distribution. Cytokine N is exclusively expressed in the central nervous system Whereas the kinase domain containing version is ubiquitously expressed and particularly in highly proliferating cells So all of the cell lines and tissues that I'm going to show you of prostate cancer origin Actually only express the kinase domain containing version of this protein Second, and you probably wondered this already, if you're looking at mitotic kinase in the context of cancer treatments There certainly are challenges associated with this. I've seen from previous targets that have been moved forward in the clinic We were aware of this caveat. So before we started doing a lot of work on this, we carefully reviewed literature to see if the problems that occurred with other mitotic kinases might occur here as well And we thought it might not be because there's a number of transgenic animals in which the kinase has been deleted or inactivated There are also patients who show SNPs in this gene and the common theme is that these animal models or humans are viable. That's not a lethal event The main issue that occurs when cytokine kinase is inactivated or depleted is that the development of a central nervous system is delayed which we viewed as less of a problem in adult men who might be undergoing treatments targeting this particular protein So this is a data that got us excited in this particular protein If you look at the left-hand side again, we viewed we started looking at this particular target Because we thought it would function downstream of the androgen receptor in another signaling cascade And a long time ago I was trained as a molecular endocrinologist and my then promoter had this rule You always had to look at a couple of things when you started looking at a new gene involved in androgen signaling One of the things you always had to do was to perform a dose-response curve with androgens to see if or what not there would be an effect on its expression So we still do this and on the left-hand side You can see what the impact of a dose-response curve of the synthetic androgen RATD1 is on the expression of CIT We actually noticed what we thought it was a quite interesting biphasic response in which low doses of androgens induced expression of this protein and Higher doses tended to restrict this expression Those of you who work in prostate cancer know that these biphasic responses to the androgens are linked to proliferation Lower doses typically one nanomolar or below induce proliferation higher doses restricted So what we saw here was induction of cytokines expression that seemed to correlate with cell proliferation We did another independent experiment using an independent growth stimuli essentially serum starved prostate cancer cells for 24 hours and gave back different doses of FBS which induced cell proliferation and again the doses that induce cell proliferation were associated with increased expression of cytokines This doesn't seem to be a merely bystander effect because when we combined these growth stimulating effects With a knockdown of cytokines as you can see here for both conditions The cell proliferation that was induced Under the control transfected cells was completely lost when you lost also cytokines expression indicating there was causal involvement of cytokines feedback feed-forward loops that implicated cytokines in proliferation of these cells Since this was a new finding we tried to figure out what's going on What is the impact of cytokines on cell cycle progression and prostate cancer cells? What is the impact on cell division and prostate cancer cells? So not surprisingly on the right hand side when we knocked down CIT We noticed it and marked the increase in a number of multinucleated cells indicating a defect in cell division Consistent what literature has reported before for cytokines and other model systems What was my bigger surprise was with the results from our cell cycle analysis where we did see at later time points an increase in G2M phase which you could associate with defects and cell division But at earlier time points, we also saw restriction of the G1S phase, which has not been previously implicated for this protein We wanted to take a closer look at this So we performed a number of experiments in which we arrested the cells at different stages of the cell cycle using protocols that the Karen Hudson lab has previously Optimized we had we could confirm that our cell cycle nutrition works in fax assays and in by looking at markers indicative of the Arrest in certain and different stages of the cell cycle What importantly when you look at the top lane of that middle panel of Western bloods? We notice high cytokines expression in G2M, which is kind of where we would expect it but also expression in earlier stages of the cell cycle indicating that there were dual roles for cytokines both in the G2M transition and as well as G1S We're not going to bother you with the left hand side of the right hand side of the panel But together this kind of put forward the idea that cytokines was growth induced in prostate cancer cells that it seemed to be an important mediator of Cell cycle progression cells proliferation and it had roles beyond roles in mitosis Doesn't mean that's relevant to the clinical situation to start answering this question We performed a number of immunohistochemistry assays on localized prostate tissue treatment naive disease compared expression level of cytokines between those cases with matching new non Malignant tissues and we found that more cases of prostate cancer were positive for cytokines than benign tissue We also found that if you look at the average expression level of CRT and cancer versus benign it tended to be higher in cancer cases That doesn't tell us anything about treatment resistant disease to answer that question It's hard to get enough cases to look at protein expression for CRT. So we switched to omics data We started performing transcriptomics data to figure out what are these cytokines dependent gene signatures? So we identified about 100 genes each that either went up after loss of cytokines Or it's whose expression went down after loss of cytokines and then we performed gene GSA analysis using these signatures Comparing the enrichment in localized treatment naive disease versus castration resistant disease in which AR targeting therapies have already failed As you can appreciate from that middle panel If you perform a marked enrichment of genes that went down after say loss of cytokines So that actually depend on cytokines to maintain expression levels in the treatment resistant disease Indicating that this protein was active in that setting and seemed to be important in that setting as well This result was consistent with proteomics data from phosphoproteomics data from the Owen Whitty lab Which also indicated that serotonin kinase is active in this stage of the disease What are those genes what could you possibly be mediating so we looked before perform some pathway analysis on these genes that depended on cytokines Particularly those I went down after loss of cytokines and noticed in these functions that you could see fitting with more resistance and more proliferating Cells or more advanced disease e2f targets apoptosis g2m checkpoints Indicating again that this is a function that's that's that's relevant to advanced disease and treatment resistant disease Get a better handle of this and to really In clinical specimens value better the translational potential of our proteins of interest We started performing analysis in the prostate cancer transcriptome Atlas some of you may be aware of this this resource We set off by Mike Freeman And it's essentially a really user-friendly web tool that allows you to organ to access Well organized transcriptomic data derived from different stages of prostate cancer It also provides information on recent great and metastatic status of the disease and It has a function where you can correlate the expression of different gene signatures is called correlation view so essentially used or we analyzed our CIT dependent gene signatures against gene signature that we felt were either relevant to the function of cytokines and Particularly to the progression of prostate cancer, so we looked at things as e2f signaling Mitosis we had signatures that assess stemness both at the embryonic level at the adult level We also had a number of gene signature that would that were generated to by prostate cancer researchers That allowed us to assess the metastatic status of the disease and the treatment resistance of the disease So this is a busy slide But if you look at again genes that were down regulated after loss of cytokines Which is kind of the top plane in each of these panels you can see that they're positively correlated with e2f signaling mitosis and Embryonic stemness indicating that they are important for self-proliferation and for potentially tumor genesis Along the same lines if you look at the meta 16 signature. It's positively correlates with genes that depend on Prostate on CIT. This is a signature that was derived by chorea by the chin and really provides a prognosis for treatment resistance and metastatic disease Conversely if you look at association with a number of androgen receptor dependent signatures That is an inverse relation indicating that the function of site cytokines in advanced Prostate cancers differs from that significantly different from that of the androgen receptor All right now we wanted to see if these kind of bioinformatic prediction actually held true when we did cell biology assays So we started working with overexpression of cytokines In first and but benign models rwp1 cells are benign prostate epithelial cells. They are transformed by the non-molecule models and Interestingly when you knock down cytokines in those cells consistent with literature. It doesn't seem to impact proliferation at much conversely When you overexpress cytokines kind of mimicking what we see in clinical cases We see that the proliferation of these cells goes up markedly and importantly we also see that a pluripotency mark Markers such as SOX2, SOX4 go up in expression Same thing for EMT markers, vimentin and SLOG. All of these have been implicated in carcinogenesis and in tumorigenesis for this particular model We wanted to verify if this sort of observation in in vitro setting held true in an in vivo model So we switched to AR positive LN cap cells Which we engineered to inducibly overexpress cytokines. As you can see from that left top panel Overexpression also induces cell proliferation in this model LN cap cells for those of you who do prostate cancer work You know if you graft them in castrated animals They will not grow and that's exactly what we saw when we put those cells Into animals none of the animals developed cancers and they happily lived out the duration during the duration of the study However, if you induce cytokines at the time of grafting Tumors start to form and five of the ten animals showed signs of tumor formation they had to be sacrificed or died before the end of the Study indicating again that this overexpression of cytokines contributes to tumorigenesis Another question we wanted to ask ourselves What about overexpression of cytokines under conditions that you could view as mirroring AR targeting therapies in prostate cancer So you can see we tried here to knock out the androgen receptor which Decreased the proliferation of the cells in the middle panel in the right-hand panel We treated those cells with very high doses of androgens Which as I've just shown before restricts the proliferation of those cells Under both conditions when you overexpress cytokines you can overcome these crop restrictions Indicating that the overexpression of cytokines during disease progression Actually, it has a causal role in development of treatment resistance Does this mean that treatment resistance cells actually become more sensitive to inhibition of cytokines expression? We think it does we did a study using ten Cancer cell lines that represent the late-stage disease the top five are essentially castration Resistant prostate cancer cells that have lost their responsiveness to androgen targeting therapies But are still androgen receptor positive the bottom five lines have are either neuroendocrine Prostate cancer cells that also have become resistant to AR targeting therapies But no longer rely on an androgen receptor or the cells are entirely androgen receptor negative In each of those ten cell lines a knockdown cytokines prevented their proliferation We saw that not just in in vitro setting but only left on the right-hand side in an in vivo setting again using cells That were engineered to knock down in a doxycycline inducible manner cytokines We also saw that the knockdown of a site in established tumors decreased tumor volumes and delayed disease progression Indicating that this particular protein might be an interesting target to pursue for therapy However, probably a lot of proteins behave this way. It doesn't make it a viable Therapy, but we were dealing with a kinase and literature indicates that a single point mutation in cytokines Can render it kinase dead? So we did a couple of studies comparing the overexpression of wild-type cytokines But the kinase dead version and we found that overexpression of kinase that certain cytokines Not only prevented the growth induction that you see with the wild-type But it actually rendered the proliferation rate to look lower than that what you see with an empty vector control condition We tried many times to overexpress this kinase dead version stably in the cells. It simply wouldn't tolerate it Indicating that indeed this cell seemed to rely on this kinase activity of this particular protein Doesn't make it a good good drug target. We as you probably were wondering we is there a cytokine kinase inhibitor We were not aware of one. We searched literature. We couldn't find one However, we were aware that a lot of the clinically tested kinase inhibitors actually have off-target effects There's a lot of polypharmacology associated with these proteins and there's proteomics data available for instance one of the study provides a very nice database that you can mine for such proteomics data that allows you to query the ability of these Tested kinase inhibitors for kinases other than the ones that they were intended to to inhibit So essentially mind one of these data sets which compiled chemical proteomics data For inhibitors that could as an off-target effect inhibits cytokines activity Mining this data we found five of these for which the cytokines EC50 was less than 100 nanomoles Which was interesting and led us to collaborate with our colleagues in the Center for Therapeutic Discovery They helped us set up an in vitro cytokines Activity assay and Then we screened the same inhibitor that had popped up from Proteomics study to see if we could validate that effect on the kinase activity of cytokines Two of these inhibitors, the second one from the top and the one at the bottom actually did have IC50s at or around or below 100 nanomolars. The most bottom one is a drug known as OTS-167, which may ring a bell. It was originally developed to be a specific milk inhibitor, and then it turned out to be a specific inhibitor that also impacts the activity of a lot of other mitotic kinases. We've implicated this protein as an inhibitor, this drug as an inhibitor of cytokinase activity, thereby identifying cytokinase as a drug-able entity. By no means do we claim that it's a great drug. It's not something we would move forward and include into clinical trials, but it certainly provides a proof of principle and a potential pharmacological probe to develop more specific inhibitors. So when we use this drug on 12 prostate cancer cell lines, the two left ones are treatment-sensitive. The other 10 are the treatment-resistant cells I just mentioned before. Each of the cell lines is inhibited at fairly low, the probability of each of these cell lines is inhibited at fairly low doses of this particular drug. More importantly, when we turn to more clinically relevant model systems, in this case, an organized system that was derived from autotopic bone PDX by Chris Jamieson over at UCSD, we see that this drug also inhibits the normal progression of the growth of this model system. These organized form as spheroids, become bigger spheroids, eventually become cysts and then become bigger. And you can see that addition of this drug, fairly low doses of this drug, kind of arrests the growth of this particular organized model in the small spheroid stage, indicating that in the more clinically relevant model, inhibition of cytokines also inhibits the progression of disease. Okay, I'm getting to the last couple of slides. Knowing that the kinase activity of cytokines is relevant to this progression, one wonders what it might be doing in the cells. What is it phosphorylating? By what means does it make the cells more aggressive? So to start to answer this question, we're employing a number of approaches. Several of these approaches are proteomics approaches, essentially trying to combine different mass spectrometry approaches with phosphoenrichment steps. So we kind of have two major lines of studies that we have going on here. One of them is a simple IP mass spec for endogenous cytokines, combined with phosphoenrichment steps. The other is a biotin-based, proximity ligation-based mass spectrometry approach, in which we are comparing the phosphorylated interactors between a wild type version of cytokines and a kinase Z version of cytokines. So these studies yielded a number of phosphopeptides. Several of them are kind of overlapping between the two studies. What are these proteins doing? Several of them, I have to say, are linked to events in cell proliferation, as you would expect. At least a third of them had functions that we were not expecting. They were linked to RNA binding, metabolism of RNA, RNA splicing events, which is a function that's sort of new for cytokines, but has been linked to other mitotic kinases. So we were able to validate in the left panel on the right, and the left panel on the bottom, that the interaction of some of these new putative substrates with cytokines. But analysis on these peptides, on these phosphopeptides, also returned a significant number of RNA metabolism, RNA binding, RNA splicing-related functions, which prompted us to go back into our RNA-seq data. We didn't really have antibodies specific for the phosphocytes in this protein, so we couldn't directly query those. But we could go back and see if the functional consequences that were predicted by these proteomics analysis held true when you looked at RNA at the RNA level. So essentially, we went back to our cytokine-dependent RNA-seq data, and used an RMS algorithm to figure out if the loss of cytokines impacted alternative splicing events. As you can appreciate from this top lane, it does, we saw quite a few different events occurring. Most of them were skipped exome events. We were able to validate representative examples as shown in the bottom by first plotting sashimi plots and then doing RT-PCR, specific RT-PCR assays that target these specific splicing events. More importantly, the question really was, are these events relevant to clinical disease progression? So we tried to answer this question by overlapping the genes that were impacted by cytokine-dependent splicing events with the genes that are known to be impacted by splicing events in clinical progression. So clinical progression comparing localized treatment naïve disease with benign prostate, comparing either treatment naïve disease with CRPC, and then comparing NEPC versus CRPC. You can see here, by the bottom left panel, we found significant enrichments of several of these events during the progression of prostate cancer, specifically skipped exome events that became more significant as these progressed. So I think I'm gonna stop here. So let me just summarize the work that we've done. Using a number of omics approaches, we've been able to verify or isolate cytokinase as a new regulator of cell proliferation and progression in prostate cancer. We know that it's now a new draggable entity. It could be conceived as a new therapeutic target that might be viable. And we also identified new roles for cytokinase, splicing, for instance, that are relevant to disease progression. So with this, I'm going to stop, but not before thanking a lot of people. We work in a very nice team at Cleveland Clinic with the help of a lot of other investigators to obtain our results. I want to acknowledge my current lab member, previous lab members, and a number of collaborators, both within and outside of our institution, and acknowledge the number of funding agencies that have generally funded this work. So with this, I thank you for your attention. I hope I haven't gone over time. And I'm happy to answer any questions. Thank you. The phenotypes you're seeing are, you know, obvious. You know, do you know anything about what, and maybe I miss this, like, in terms of what regulates it? Because it seems like the patterns are strikingly similar to E2F. I mean, would it be enriched in like RB loss cells, or? Yeah, actually, no, to answer your last question, we've looked at RB losses. If you look at E2F2, it's kind of the opposite question. So if you lose RB, you don't impact the growth induction of cytokine expression. The presence or absence of RB doesn't also seem to affect the ability of loss of cytokines to decrease cell proliferation. In terms of a question of what might be regulating it, I didn't talk about this, but we know that cytokines expression is regulated at the protein level. And we implicated a new signaling cascade in which growth stimulation activates E2F2, not E2F1, but E2F2, which leads to skip two over expression, which leads to decreased P27 expression. And then it's the interaction between P27 and CRT that controls the level expression of CRT. And that does differ based on the stage of the cell cycle you're looking at. One other thing, I was really interested by the fact that the kinase dead version actually is functioning almost like a dominant negative. I mean, I've actually seen this, too. And I would scratch my head. I was like, why do we see this? Now that you actually have some substrates, have you tried to do anything where you think, is it basically like grabbing substrates and holding on to them for much longer or anything? Or how do we explain the dominant negative? So the moment I saw those data, I assumed that cytokines was working as a dimer or a multimer, because you're doing this in cells that still express the endogenous version. So we have a new postdoc we started about a couple of months ago trying to address this. Because again, cytokines is one of those poorly characterized things. We don't know how it is activated, how it's working. We've done a couple of studies comparing cytokines expression in native gels and denaturing gels. We've put two differently tagged versions of cytokines in the cell. And what you can see, there's certainly some polychromatization going on. I'm not sure it's a dimer or a higher-order mirror, but something is going on there. And I think that's causing the dominant negative effect that you're seeing. All right. Dan Juel, UVA, had a great talk. We had a hit with citron in our kinome screen that we published a decade ago. So it's wonderful to see this story. If you knock down citron kinase or with your compounds, do you see any sensitivity to ionizing radiation? We haven't tried ionizing radiation. We've tried things like docetaxel. And it doesn't seem to be a negative effect or anything like that, which surprised me a little bit. So far, we haven't seen it. We need to take a closer look at that. Zeynep Balakardian, University of Illinois. Great talk. I think I have a more general question about that U-shaped or induction of the kinase in a U-shaped method that you mentioned at the beginning of the talk. What are the challenges associated with it in terms of the identification of the inhibitors of these kinases? Is there any indication why you do those response, for example, with these kinase inhibitors that you showed that there is also such a response? Yeah, we haven't tried that. We should probably do that. We haven't done that. I think there might be something, but we have no data to support that. Thank you. We still have some time for questions. Now that you've got this new angle with possibly our co-controlling the RNA, it would be interesting to see, as you're doing your assays, like a next round of xenografts, it seems like one thing that somebody would want to eventually see is also if you compare this head to a CDK4-6 inhibitor or some other cell cycle inhibitor. Yeah, you're absolutely right. It's on a list of things to do, but we haven't gotten around to it. But that's exactly what I would like to compare it to as well. So when you knock it down, how are the cells dying? Or are they dying? You pointed out a G2 arrest. Is there any cell death? Yeah. I don't see overt signs of apoptosis, if that's what you're getting at. So that's not what you're getting at. Because you're RNA metabolism, I'm wondering if there's another type of cell death that's OK. Good question. We haven't, I don't know. So we've done crude experiments to look at, again, apoptosis generally, but nothing else so far. Thank you so much. Let's thank Dr. Hemmertz. Thank you. OK. In this session, our next speaker is Dr. Hisham Mohamed, who's an assistant professor at the Knight Cancer Institute in Portland. And his talk is entitled, Interrogating Hormone-Driven Cancers One at a Time. Thank you. I have no financial disclosures. Oh, sorry. This is not what works. Yep, thank you for the introduction. So like discussed, we're based in Portland, Oregon at the Knight Cancer Institutes. And yeah, so essentially, to give a very high-level overview of what we do, the lab's work is divided into two main sections. One is looking at interacting partners of nuclear receptors, specifically estrogen receptor, using this method that I developed a few years ago when I was in Jason Carroll's lab called RIME. It's basically an IP mass spec method. But more recently, we've been focusing a lot on understanding heterogeneity of hormone response. And key questions fall along the lines of how does underlying genetic and epigenetic heterogeneity impact hormone signaling? Do we expect changes as tumors progress? And how do we study these changes? We're trying to develop technologies and model systems to study these. And we'll try to address some of these things that we've been working on today. So when you talk about heterogeneity, what comes to my mind, the key concepts that come to my mind are three concepts. One is genetic heterogeneity. And what I mean by this is heterogeneity in terms of CNV pattern changes or DNA mutation changes between individual cells. And you can expect that this causes transcription or other advantages for the cell and the tumor. The other level of heterogeneity that people have been increasingly talking about is transcriptional heterogeneity. And the concept of cell states. And what I mean by this is in a relatively homogeneous genetic background, cells have in some cases the ability to fluctuate between transcriptional states. And this has been very well characterized in stem cells. And there is increasing evidence that this is the case in cancer cells as well. And the last bit that is less explored is the concept of epigenetic heterogeneity of epigenetic states. And this includes DNA methylation and chromatin accessibility for a start. And as you can imagine, the two of these in particular go hand in hand. And again, there's increasing need for multi-omics studies where you can assay all or a few of these omics at the same time, allowing you to understand how when one variable changes how it impacts the other one. So in terms of genetic heterogeneity, there's been tons of bulk sequencing and single cell sequencing based data sets where people have been tracking how tumors evolve in breast cancer and other cancers. In terms of transcriptional heterogeneity, there's been recently multiple single cell RNA sequencing studies, this one from Alex Swarbrick in Australia. They've sequenced single cells from multiple primary metastatic breast tissue. And one interesting concept here that I want to show is when you take tumor cells from an individual patient, so these are all individual patients, and when you assign PAM50 classification at a single cell level, you see that there's a spread of identity. So they don't all behave in the same manner, suggesting that there's this plasticity in cell type identity, either driven by genetic or transcriptional changes or epigenetic changes. And this is something that's been intriguing us and many others in the field. And trying to understand this is an important objective of ours. So one of the things that we've been doing, and I'm not going to detail on this side of things, is this multi-home assay where you can integrate ATAC-seq and RNA-seq from the same cell. So essentially, you get transcriptomics, and you get chromatin accessibility from the same cell at the same time. And we've performed this in more than 20 tumors. And these are mostly frozen tumors ranging from 20 to 15 years ago with prognosis data. And we find the expected subtypes, and we can find epigenetic differences between your main subtypes, et cetera. But an important point that I wanted to raise based on our previous slide was we can, using this data set, so this is just one tumor. This is an ER-positive IDC that I've picked. In this one tumor, we can identify at least two different clonal subtypes in terms of CNV patterns. And when we compare the transcriptomes and the ATAC profiles of these, we can see that from RNA signatures, we can see that we predict that some of these clones have a higher proliferation score than others. We also see very distinct RNA differences, and also motif differences in important genes, such as CTCF, FOXA1, and others, suggesting that not necessarily surprising, but we can see that genetic differences are creating very specific chromatin accessibility and transcriptomic changes in these patients. But our question for today that I'll be talking about is something a little, you know, it's going to be on questions based on non-genetic variability. So the question is, in a relatively homogenous genetic background, how variable is hormone response? You know, do every single cell which has a similar DNA identity respond in the same way? So the heterogeneity of hormone response is our first question. And do these cells have different transcription and epigenetic states? And what drives or regulates these states? And the motivation behind asking these questions is that we need to establish these questions in very controlled homogenous cell states and cell lines before we go into patients and start manipulating patient samples, given the heterogeneity that we have in patients. So to do this, we embarked on this large single-cell RNA-seq experiment, where we performed single-cell RNA-seq from multiple breast cancer cell lines and organoid models. For sake of simplicity, I'm just going to be focusing on our estrogen treatments. But we have essentially treated cells with estrogen, progesterone, androgens, fulvestrin, and tamoxifen, and different combinations of all of these for multiple time points. And I'm happy to share this data set with anybody that's interested in looking at this. And essentially, we did single-cell RNA-seq. And what you expect is, if you have two signaling patterns, let's say the blue and the purple pattern over here, you expect some cells to have different amounts of the blue or the purple. And using single-cell traditional analysis platforms, what this does is it looks at the transcriptomes of all of these cells, and it would identify these clusters of cells as cluster one, and these would be cluster two, based on the expression of the blue or the purple. But as you can imagine, you might have some amounts of the blue and the purple bleed over into other clusters, suggesting that, in cases like this, traditional clustering doesn't really function very well. The clustering methods and algorithms work really well when you're looking at individual lineages. Let's say, if you take a PBMC cell line, you have multiple lineages. It clusters them really nicely. But in places where you have gradients of signaling, gradients of transcriptional difference, it's really hard to evaluate this. So to address this, we developed a method called Titan. This is a machine learning method based on the concept called topic modeling, and topic modeling's been traditionally used for news classifications, and people have been increasingly using this for other purposes like this. And essentially what it does is, instead of classifying your data set into cluster one or two, what it does is it gives you scores for your blue network for your purple network for every single cell, and you can basically, the data set says you have 20 networks here, and this is the score for every single cell. And looking at an example, so this is MCF7 cells treated with different time points of estrogen. From a quick look, you can see that this is relatively boring, it's a big blob of cells. For those unfamiliar with single cell RNAseq, every single dot here represents a single cell, and the closer the transcriptomes of two cells are, the closer they cluster together. For example, these cells look very similar compared to these guys. And you know, the clusters are not distinct enough to be able to separate out our different treatments. You can see there's quite a bit of overlap between our different treatments. So we then applied our Titan method, and what we find is that the method predicted 20 different topics, or transcriptional networks to be present in our data set, and what we can see is that we can see several of these topics go up with estrogen treatment, like these guys over here, and several others go down with estrogen treatment, like these guys over here. But what was really interesting to us was, so for example, the case of estrogen upregulation, there's more than one topic that goes up, which was interesting. And you can see that some of these gene signatures have different time dynamics, like these guys come up earlier than these guys do. So we did pathway analysis of these signatures, and what we identified was that this was a classical ER-driven pathway, and consistently this other one came up as a FOXM1-driven pathway. And then we asked the question of, okay, are these in different set of cells? You know, what does it look like when we plot one of these signatures against the other? And that's what we did here, and what we see is that with increasing estrogen time, the amount of both of these signatures go up with time, but what is really striking is that they go up in a divergent manner. So if you are high for this FOXM1 topic, you basically cannot activate a lot of these genes in these topic, which include key ER targets, such as progesterone receptor, et cetera. And we validated this in multiple, in four cell lines and two organoid models, and consistently what we see is that all of these models have this ESR1 topic up-regulated, and they have this, the FOXM1 topic also up-regulated, but importantly, they're happening almost at the same time in different set of cells. We looked at this and published a single-cell RNA-seq in patient data sets, and what we see is, and this is a density plot where you plot the amount of the proportion of ESR1 topic or the FOXM1 topic, and interestingly, and somewhat not surprisingly, the ESR1 topic is enriched in ER-positive cells, and they have low of the FOXM1 topic, and vice versa with the ER-negative. But what's really cool is that if you plot them as single cells, one against each other, they almost recapitulate what we see in our cell line models, which is, you know, they are more or less mutually exclusive with some amount of the ER-positive cells expressing the FOXM1 topic. And this separation is, again, on a subtype-specific manner. Lastly, we've been looking at transcriptional and chromatin differences, and we did, again, the joint ATAC RNA experiment. We can see that, you know, estrogen treatment separates cells by RNA and ATAC, and essentially what we see is that, again, the FOXM1 and ESR1 topics happen to be in different set of cells, both at the epigenetic and the transcriptional level. And this is an example of a gene which is highly expressed in the FOXM1 high cells, lowly expressed in this cell, and we can see that chromatin accessibility patterns are different, and code accessibility patterns marked by these linkages over here are different between two set of cells, suggesting, again, that in our cells, estrogen treatment leads to at least two distinct transcriptional and chromatin states. And we validated, you know, knocking down FOXM1 leads to reduction of the key FOXM1-predicted topic genes. So, switching gears a little bit into something else that we think is also driving heterogeneity. So, with all of these data sets, we've been trying to identify more regulators of heterogeneity, which is a difficult task to do. Essentially, you know, coming back to the first slide on this concept of non-stable cell identity, we looked at our cell line RNA-seq models, and we find that, you know, there's gradients of luminalness and basalness, and again, I wouldn't call these our basal cells, but it's basically transcriptional gradients that the cells are fluctuating between. And what we did was we took these genes which are heterogeneously expressed, and we linked them to over 2,000 ChIP-seq data sets in these cell lines, and tried to identify what could be master regulators of this heterogeneity, and what unsurprisingly came up with estrogen receptor, but some of the other factors were p53 and KLFO. Now, KLFO was really interesting to us because previously we hadn't identified it as a ER-interacting factor with RIME, and it's a key pluripotency factor, and now our question is, does KLFO regulate pluripotency or heterogeneity in ER-positive breast cancer? So, again, coming back to a brief overview of KLFO, it's a pluripotency factor, one of the four Yamanaka factors used in reprogramming cells. It's a pioneer factor shown by Kinzarath to bind and open condensed chromatin, and there's multiple studies offered in breast cancer with this particular one showing that it's heterogeneously expressed in a proliferating tumor. We performed ChIP sequencing experiments of KLFO in ER, and there seems to be some amount of overlap. Interestingly, the overlapping regions have KLFO and ERE motifs. The non-overlapping regions seem to be driven by FOXA1 and some EREs, suggesting that perhaps KLFO and FOXA1 happen to be regulating distinct set of ER binding sites. When you look at this from a global view, plus minus estrogen doesn't really impact KLFO binding. However, loss of KLFO leads to reduced ER binding at some of these sites. And these are example sites. You can see that loss of KLFO in this condition here leads to reduced ER binding, obviously loss of KLFO again, and reduced H3K27 STO at these specific sites. We also did ATAC-seq experiments, and this is just a snapshot of our results, or representative snapshot. What we basically see is that when you knock out KLFO, independent of treatment, these sites close up, suggesting that KLFO plays a role in keeping these sites open or opening these sites in these cells. And again, unsurprisingly, the impact on RNA is that ER-driven target genes at these sites are impacted. They're not up-regulated or down-regulated as much. And we then went back to the question of heterogeneity. We did single-cell RNA-seq experiments where we knocked out KLFO, silenced KLFO with siRNA plus-minus hormones, estrogen in this case. And what we see is that specific sub-clusters of cells disappear, indicating that the loss of KLFO seems to be more detrimental to cells in particular cell states than the system as a whole. Again, we did the topic modeling analysis, and the key take-home is if you look at things that's different between the si-KLFO versus control, these two topics over here come up as very high, as different, and when you do pathway analysis and motif analysis of the promoters of this, they come up for KLFO motifs, and these signatures seem to be having a much poorer prognosis in patient data sets. We also did an analysis where you look at overall heterogeneity in the system. There's multiple different analysis that you can do for this. And in this analysis, what we see is that induction with estrogen leads to an increase in heterogeneity but silencing KLFO reduces heterogeneity in all conditions. And all of these results have been consistent in at least two cell lines, and we've performed this using the ATAC RNA, and we see chromatin-based differences as well. So this also further intrigued us. We were looking back into KLFO's role in pluripotent cells, and something that's been consistent and shown by quite a few papers is the concept that KLFO is involved in maintaining 3D loops and 3D chromatin structure. And also more recently, there's at least a few papers that show that KLFO plays a role in formation of these liquid nuclear condensates, and removal of KLFO leads to reduced condensate formation in these cells. So similarly, there's at least a handful of studies where estrogen receptor has been shown to form these nuclear condensates, particularly in an estrogen-stimulated manner. So we had naturally the question, does KLFO play a role in forming these condensate structures in breast cancer? And this is just very early data, but what we see and what we think is that it does indeed play a role in forming and maintaining these nuclear foci. So you have a number of ER foci per cell on this axis, and this is intensity of ER stains, and we compared control versus a knockout of KLFO, and you can see that when you knock out KLFO, you have some cells that maintain high ER intensity, but they don't form foci anymore, suggesting that KLFO plays a role in at least recruiting ER to this foci. Something that we're doing right now, and it's too preliminary to present here, is we're looking at the physical formation of these foci using EM, and we can see very clear, nice structures, and we're trying to quantify these changes, plus minus KLFO, and our key question there is, is disruption of KLFO impacting the formation of these foci, or is it just impacting the recruitment of ER to these foci? So this is our model, like I just discussed. We think that KLFO is probably bringing a lot of factors together, like it has been shown in the pluripotency setting, and loss of KLFO leads to at least loss of ER, potentially loss of this entire foci. And the key summary for my talk is, estrogen treatment leads to divergent transcriptional and epigenotic responses in cell line and organoid models, and that KLFO is required for ER function, and is likely a pioneer factor for estrogen receptor. Everything that I've shown today is unpublished, but if you want to use the computational package, feel free to take a picture of this, and it's available online right now. And this is the team, and most of the work has been done by Ye Hong Wen, and Aishagul Orse, and Aaron Doe. Thank you. Great talk Hisham. So my question is going back to the ER and FOCSAN1 topics. Do you think if we had the technology to follow exactly the same cells over time you know these cells would have the ER topic and FOCSAN1 topic you know so maybe the actual question is do we have transcriptionally asynchronous cells to begin with and we are just capturing after the you know hormone treatment we are just capturing those different you know sequences of topics. So so one thing I don't think that's your question but one thing that I want to point out is we did those experiments in single clone based cell lines where we've grown them from single cells to make sure that it's not a genetic effect and they show the same thing and the other thing is using unspliced RNA you can predict the trajectory that our cell is going to and what we see is when we do this analysis we think that they are basically cycling through these different states. So we do see that but we don't have you know enough we do have the data we haven't got got to a place where we can model the whole system we you know that great in modeling the things but yeah so so we don't think that they are stable states they are cycling through states. So I guess also with regards to the cellular heterogeneity in terms of response of estrogens you know getting back to Hannah's actually talk like is it how much of that could be attributable to changes like where the cell is and its cell cycle you know I mean I don't know if they're all if you're all locked in the same state when you do this or you know. So that's a key question we synchronize cells but we do see very strong cell cycle effects but at minimum what this tells us is that key estrogen driven genes cannot be activated in particular cell cycle states and and to speculate beyond this and this is like pure speculation what we think is that when you're in the Swaxman-Heist state you're not up regulating ER genes so what we think is this could be a state situation similar to a heterogeneous ER expression tumor where you have let's say 50% ER expression we are hypothesizing that the non-ER expressors are in this Foxamon state but they will transition to the other state at some point suggesting that this could be a therapeutic vulnerability it's in some sense. Great talk. I would like to ask about the proteomic heterogeneity. Have you have you thought about the protein itself the the sequence of protein if there is heterogeneity in different cells or something like that? The heterogeneity of protein expression? Yes. So protein expression is really hard to assay at the single cell level with the exception of you know using imaging or barcoding based technologies that is something that we're trying to do so we've just recently got some you know single cell data from nanostring where we looked at subcellular proteomics and transcriptomics we're trying to make sense of the data set that we have but you know ultimately that that's that's key to look at for sure. Yeah absolutely. David Shapiro, Illinois. Very impressive presentation. Do you have any indication of whether KLF4 has any role in this very dramatic change in the properties that you see in the ER mutant cells in breast cancer where they have different metastatic potential and so forth and they're much more flexible? That's a really good question. I don't know. We haven't looked at that. I mean you know it's worth looking at models so yeah thanks for the suggestion. I'm Jennifer Ricker from University of Colorado. There's been a couple of really high impact papers showing that in metastatic breast cancer well you can predict even in the primary tumor and then it's definitely true in primary compared to metastatic that you get more heterogeneity of ER and that it's a predictor of poor outcome and you often lose a lot of ER expression and then metastases compared to the primary tumor. So I'm wondering if you just do immunostaining for KLF4 and metastatic disease can you see higher expression like on a per cell basis of KLF4 where you're getting more ER heterogeneity? So that's a very good question. In primary disease we've done that. It is very heterogeneous and there's one publication that says that it's mostly centered in the middle of the tumor where you have less access to nutrients and oxygen and it enables a more you know aggressive phenotype in these regions. What we're also doing is so we can see heterogeneity we've not presented today. We're getting sections that we're staining for KLF4 and we can laser capture individual KLF4 positive cells and negative cells and we're comparing the transcriptomes and epigenomes of these cells. It's a bit tedious but you know it should answer you know exactly what you've been asking. Great. Yeah, thank you. Matt Sikora, Colorado. Great talkieshow. Kind of coming off a dance question. In your synced cells, the ones that are deprived at time zero, is there still substantial heterogeneity especially coming from the single cell clones and then can you predict based on maybe your topics the cells that are going to go down that ESR one path versus FOXET one path and then is that different across cell lines the extent of time zero? So that's a good question. There is some amount of heterogeneity. It's not as much as it is in estrogen treatment. Is it possible to predict? I'm sure somebody would be able to do it. We can't. There's the data there. So yes, theoretically you could because again you have the spliced and unspliced information. The unspliced telling you the trajectory that it wants to go in, right? So I think there could be you know modelings that you could do with this data set but yeah we can't do that. You can kind of imagine a setting where you could almost predict which cells are going to go down an anti-estrogen resistance. Yeah, so one way to do it is you know if we can identify like say top ten genes that can be used for prediction we go back and do imaging or live reporter-based assays then that could actually be possible, right? We could start tracking cells and predicting where cells go. We have time for another question. Hi Hisham, it's a nice talk. I was wondering like if we were to do the same analysis with or without physical resolution intact, do you see the changes? Physically dissociating? Physical resolution, because this is based off of single cell RNA-seq. Like in cancer tumors, the contact is very important versus normal cells with contact inhibition. So if you were to keep the physical resolution intact, do you see the same difference? So again, that's a great question. What we're hoping is spatial transcriptomics and spatial omics will tell us that. Because especially the thing that I've completely not talked about was a big set of genes that come up with KLFO inhibition as the immune pathway. We have no models to assess that in the ER positive context. But if we have patient samples where we can say KLFO high cells are next to such and such immune cells, that will give us great insight. And that's something that we're really looking forward to. Any other questions? If not, let's thank Dr. Mohamed again. Thank you. Our next speaker is Dr. Robert Clark. Currently he's the professor in the Department of Biochemistry and Molecular Biology and Biophysics at Hamel Institute at the University of Minnesota. And his title of the talk is Systems Biology to Understand and Predict Steroid Hormone Signaling. Please welcome Dr. Clark. Thank you, and thank you all for staying this late. And thank you for putting up with another speaker with a funny accent. What I'd like to do today is first thank the organizers for the opportunity to tell you what I'm going to tell you today. But I'm going to take you on a journey that explains how we think there's a framework for understanding the tremendous diversity of signaling that we see with endocrine resistance in breast cancer. These are my disclosures, and they're in the program. And this simply shows you the problem we're trying to address. As you can see, after about five years of treatment, ER positive breast cancers have a worse annual mortality rate than ER negative breast cancers. And while we have some very good therapies for these particular cancers, and again, tamoxifen as a monotherapy reduces the 10-year mortality rate by a third, many women still don't get that overall survival benefit. And what's been particularly frustrating is how difficult it has been to move the needle past where we are with tamoxifen. Almost all the meta-analyses that have compared tamoxifen with the aromatase inhibitors have failed to show any significant overall survival benefit beyond that of tamoxifen. It was only with the advent of 500 milligrams of fulvestrant did we have a monotherapy, monoendocrine therapy, that significantly improved overall survival for ER positive breast cancer. So why? Why is it so difficult to move the needle? Why do so many breast cancers recur? And why is recurrence fatal and so difficult to manage? I'm going to argue, for the sake of an argument, if you want one, that the way that we have, and we're as much at fault, if fault is the right word, at this as anyone else, is we tend to take the view that we can find a single mutation or differential expression of a single gene or a single canonical pathway that will explain this complex dynamic phenotype. And I think when we do that, we miss the complexity of this system. We know that some individual functions are critically important. ESR1 mutations are a great example. They explain probably 40% of all AI-resistant tumors. What's driving the other 60%? Mutations in PI3 canines and 8KT are fairly prevalent in ER positive breast cancer. We have drugs that target them. None of those have yet shown any improvement in overall survival. There's a modest clinical benefit, and there's a lot of toxicity that goes with it, but we have not moved the needle. And there are reasons why that's been so difficult, if you simply look at the regulation of signaling within that particular feature. And the bottom line really is, what you see depends on how you look and what you think you're looking for. Whether you're measuring it or not, the entire human interactome is operating in the system that you are measuring. And it is a phenomenally large functional system. When you start to look at all of the potential nodes When you start to look at all of the potential nodes and edges in that network, they are all functioning, and they're all being perturbed. And the property of data sets and systems that function that way are fundamentally different from the smaller systems that we tend to look at the way we conceptualize things. And so we often find ourselves in the situation like the blindfolded scientist and the elephant. Given the information that we have, we make rational decisions and interpretations of what we've got, but we're often wrong. And what we really need is a framework to interpret the data that we get that allows us to understand how the system, as an integrated, coordinated system, if you think of the ER-positive breast cancer cell as a system, how it responds to the stress of endocrine therapy or when you target the estrogen receptor. And that's what I'd like to propose to you today, a different paradigm, a different way of looking into the problem. For this, about 10 years ago, we published our initial framework for how we think this works. And unlike most others who approach this from a systems perspective, we looked at the data that we had and what was in the literature, and we identified five functions within a cell. Autophagy, cell survival death, metabolism, proliferation, and the unfolded protein response. All of these were known to be implicated up to 20 years ago in this phenotype. But how are they regulated and controlled, and how, when they work together, does the cell manage the stress of an endocrine therapy? And so the problem that we face is that we're looking at a fundamentally dynamical system. And you just got a wonderful example of that in the last couple of talks. So we need to be able to build models that identify the dynamical directional properties, the adaptive, reversible, irreversible, there is redundancy within the system, there's degeneracy within features within the system, there's non-linearity, and all of those together allow the system to do things differently in ways we hadn't done before, and that's the property of emergence. And all of this is true of all human cells. And so we're building input-output models that attempt to put this in a framework that's interpretable and allows us to understand where each of the pieces of data that you see in the literature may fall. So to do that, we need a series of questions that guide our way of filling out that framework. I don't have the time to show you all of the data, but I've been a little picky in what I've given to make a number of different points. The first one, and this is true probably for what we all do, is every time we ask a question, the question, the experimental system, the data type that we get, the analytical tools and workflows that we use, when they look at the system or reflect back some piece of the truth, but they don't always reflect it back the same piece, and they don't always reflect the same piece back the same way. And that we have to understand when we look at our own data, in my humble opinion. So you're gonna see many of the bigger picture things reflected back, but the details that underlie those may differ from time to time, depending on the way we look at the system. So here's the first question. What are the primary features of an endocrine-resistant phenotype? Are all recurrences the same? Do we have experimental models that will allow us at least to ask some questions of the human disease and give us something back that's reasonable or interpretable? Is the phenotype itself stable or reversible? And if the framework features those five modules that I described to you, are any of them represented in the human disease? And if so, can we learn something about how they're represented and how they may work? First question. Are all tamoxifen recurrences the same? We took a very simple approach to that. We said, let's look at naive patients and let's see if we can build a molecular classifier that will predict when that patient's tamoxifen, on tamoxifen, when that patient will recur. And we wouldn't be able to do that if they were all the same. So this simply shows you in the top panel that we can build a nice classifier. We're as good at classifying early recurrences as we are late recurrences. Of course, these are all recurrences. So we're working in a very narrow bandwidth here and the tool will work on independent datasets. So first question, are they all the same? No. Are our cell line models or our animal models useful? I've picked a simple example here. Here we're taking cells that are sensitive, LCC1, making them resistant, looking at the proteomes by two different techniques. And you can see that two different tools reflect by different things, different ways. And then asking, do those changes in the proteome, are they reflected in the transcriptome of primary tumors? And the answer is, yes, they are. And boy, that shouldn't come as any surprise because of the principle of evolutionary parsimony. We learned how mammalian cells go through the cell cycle by studying yeast. We learned about wind signaling by studying Drosophila. So of course we can take human breast cancer cells and learn something about human breast cancer. That shouldn't be a surprise. And when you ask the question this way, you see the same things. You see, in this case, in this particular experimental example, metabolism, the unfolded protein response, and cell death are what reflected back to you. And you can see some of the genes that are listed there that can explain why we see those as the dominant features when we look at the question this way. Is the resistant phenotype stable or reversible? Well, that kind of depends. If you've got an ESR1 mutation, it's driven one direction. If you don't, we took some data that Dick Santan had published, a lovely study he published back in 1995, where he just looked at the emergence of different phenotypes by giving trace, physiological and higher levels of estrogen. Over time, he did it in vitro and in vivo. We took a Waddington landscape model, built a mathematical model of how that dynamic changes went, the phenotype changes over time. And we were able to learn a couple of things about the system. First, the estrogen receptor functions as a bistable switch. It's probable all steroid hormone receptors function as bistable switches. The phenotypes themselves are reversible. But, and this is a reflection of bistability, when you acquire one phenotype, in this case, if you acquire an estrogen-independent phenotype, it's much harder to go back to an estrogen dependence than it was to become estrogen-independent in the first place. That's what we see clinically. We treat constantly for five years. We push the cells towards a resistance phenotype. And then we wonder why we can't get it to go back. That implies that if we give an intermittent treatment, we actually might be able to push out therapeutic responses for longer. And that's shown on the second part of the slide over here, that we were able to do that using a mathematical model where we simulated intermittent treatment and predicted what the outcomes would be. And now there are a couple of clinical trials ongoing looking at intermittent therapies. Next question, what are the framework features are present? You've already got a sense that some of them are present because I've shown them to you. But this is going back now to the two breast cancer transcriptome data sets and using a completely different way to look at the question. And I don't have time to explain all the tools, but what this particular tool does is it gives us a map of the signaling as based on the representation of protein-protein interactions as predicted across the differential expression of the RNAs in early versus late tamoxifen recurrences. No canonical signaling here, you're not gonna see a lot of it in this talk. And from this perspective, the modules that are most clearly defined are apoptosis, which is part of cell death and cell proliferation. The cell proliferation one makes perfectly good sense because if you look at the Key67 data, this is a study Mitch Dossett did, you can see that in tumors that are responding the Key67 levels fall dramatically within as little as two weeks. So now we're beginning to understand another feature of the system. You mess with the estrogen receptor, cell stop. And that allows them to do a bunch of other things as you'll see in a moment. But from this perspective, now we're seeing cell death and proliferation as being reflected back as the dominant features and some of the molecular signaling that underlies that. Here's another approach. Now we're gonna take matched biopsies. ER-positive breast cancers, we have data from prior to treatment and the time of first recurrence in this same patient. In this case we used, in this slide we used differential dependency network analysis, which simply gives us a series of small molecular features. Doesn't matter whether you read them or not because you can take a different approach to ask the same question of the same data. In this case, you just do a simple gene set enrichment analysis. You still get the same features reflected back. And that is in this case, cell death, metabolism and the UPR. And now we're beginning to see some other molecular signaling features that explain how those are differentially regulated in cells as they respond. Let's take a step even earlier then. Let's look in the neoadjuvant setting. In this case, we've used ultrasound as a measure of responsiveness and being able to biopsy the tumors as they go forward and look at the changes in gene expression. And again, just very quickly because it's easier to do it quickly this way is to show you an ingenuity analysis. Now you're seeing UPR, cell death, metabolism and proliferation reflected back. All of these are the dominant features. They keep coming back. And in many cases, it's the same genes. So let's flip the question. What mechanistically relevant features change and when do they arise? So this is now a cell line study. Time course, transcriptome over 72 hours. MCFs are T47Ds treated with estrogen, plus or minus estrogen, plus or minus fulvestrant. And what we see here are a series of other features, small signaling features that again reflect back, in this case, cell death, proliferation and the UPR. So we are seeing what we've known for a long time, but we're seeing it consistently represented back no matter how we ask the question. We see the same features and we see many of the same molecular reactions that underlie those features. Now, in that previous one, I'd highlighted XBP1 and BCL2. That was a non-canonical, unknown relationship before we predicted that from our DDN model. You'll see we went in and we showed that that's mechanistically true. But this is actually an earlier study. But it addresses the question of, are any of those early changes rewired permanently? So here, sensitive cells are made resistant and we simply are listing the top, the beginnings of the top genes that are up-regulated and down-regulated. And right at the top is XBP1 and BCL2. And so we had known and implicated the integrative stress response in UPR 20 years ago. We weren't the first to implicate it in this phenotype. So an NF-kappa B and a bunch of other things you've heard of and EGR1, which I'll come back to. So when you see those changes, that you can make cells resistant, you can take away the selective pressure and you still have that phenotype. It implies that you've got an epigenetic rewiring. And you've heard about that also today. And so we showed that a number of years ago too. When you think of epigenesis, remember that the substrates that allow that process to go forward are coming from the metabolome. And so there already is a direct relationship between changes in cell metabolism and the ability to rewire epigenetically. And we were able to show that if you come along with DNMTR8-stack inhibitors, you can partially reverse the phenotype. Now it's pretty clear that there are significant epigenetic remodels that goes on within this particular phenotype. And there's now a bunch of clinical trials that are ongoing to test this component of the framework with at least some interesting data coming out from the phase three ACE trial. But, you know, it keeps on going. So how do those signals flow? You've seen some examples. I wasn't asking you to pay attention or remember them because that wasn't the point of showing them to you. But now we have to begin to understand if these different modules are consistently changed and it doesn't matter whether we're looking in cell lines, whether we're looking in adjuvant or neoadjuvant, matched or unmatched cases in breast cancer tumors, and they're all correlated with clinical outcome, how do these functions coordinately work together? Because you've got autophagy, cell death, metabolism, proliferation, and the UPR. And I don't have time to explain to you all of the coordination events, so I want to get the principle across to you only. And for that, I'm just picking the unfolded protein response because it's one of our modules. And the UPR itself is a wonderfully integrated system. And so it's triggered by a whole bunch of stressors, changes in oxygen energy, changes in the metabolism module is gonna regulate the UPR. DNA damage will do it, a whole bunch of different things will do it. The upstream regulator is glucose regulated protein 78. So that's telling you that it's obviously related to cell metabolism. But that one upstream regulator regulates three different arms and those integrated arms do different things in different ways and in different timescales. So one arm reduces the rates of transcription and translation. That allows the cell to rebalance its metabolism to adjust its energy to the amount of proteins it's able to fold. And it will target the unfolded proteins and everything else for degradation if it can't do it. And so you're changing the rates of transcription, you're changing the rates of translation, and I've already given you a hint that it can regulate cell death through its ability to change BCL2. And jumping just into one little feature of that to make it a particular point, here it shows you canonically how BCL2 is thought to be regulated by the UPR. And it is degenerate in its regulation. So it can be regulated by junk or it can be regulated by CHOP. And I've just shown you it can be regulated by XBP1. So you've got tremendous degeneracy coming out of the UPR just to control BCL2. And I'm only using BCL2 as one example. I don't want you to think that BCL2 is the only explanation for the control of cell death. So if we were looking at this in the past, we would have said, well, let's target junk. Or let's target CHOP. Or let's target XBP1. None of those, as Mono approaches to targeting just this one feature of the control of cell death would work. You might get a nice response in the short term, but the cell's got two other ways to get rounded. And every cell's gonna take its own route to get past that one blockade. And this is the challenge that we face with dynamic integrated coordinated systems. And these panels at the top just show you that mechanistically that's the case. But this panel here, it tells you that, well, it's just not BCL2, is it? BCLW can do it. In fact, all the pro survival BCL2 family members can do exactly the same thing. But they do it in different contexts with different affinities and different functions. And if we look at XBP1, XBP1 doesn't just affect BCL2, it also regulates autophagy, the cell's recycle plant. So you take the cell's recycle plant and you block that, which you can do with chloroquine. And we've shown that it interacts in vivo synergistically with tamoxifen. Now there are clinical trials testing that. But that doesn't address the complexity of the system either because if you just look at BCL2, most of us initially think, I do anyway, of BCL2 as sitting in the mitochondrion and protecting the cell from dying. But that's not all it does. It has other protein partners, which should be no shocker to anybody, but one of them is Beclin-1. And it can sequester Beclin-1. And when it does that, it affects the activation of autophagy, the cell's recycle plant. So depending on how much BCL2 you have and where it is in the cell, it can be sitting in the mitochondrion protecting the cell from dying. And at the same time, if there's not enough of it left, it frees up Beclin-1 and activates autophagy. And that means the cell can feed its intracellular metabolism through autophagy to execute this cell survival signal. So it's another example of coordination among these functional modules and within the complexity of the human interactome. And it kind of just goes on. If you're looking at blocking cell death and you're seeing changes in autophagy, that's implying there are changes in metabolism, which you already knew because we'd implicated the UPR. So we can take the transcriptome, we can take the proteome, we can map one onto the other and get another view of how the system works. And in this particular case, there's an example of how that gene that's differentially rewired EGR1 functions. But what I'd like to focus on for the purposes of today are the high energy phosphate metabolism components down here. Because when you see that, you usually think it's carbohydrate metabolism, it's cancer, it must be glucose. And you wouldn't necessarily be wrong because as this shows you, you take sensitive cells, you give them an anti-estrogen, you take away estrogen, whatever way you want to do it, glucose uptake drops dramatically. Doesn't happen in resistant cells. And again, the clinical correlate of that is FDG-PET. So, and you'll see that as I've shown you before, everything links back to something that's clinically actionable or present in tumors. And that's partly a reflection of these changes in GLUT1, which is the primary transporter for glucose, which is also upregulated by XBP1 out of the UPR. So there's another connection. And if of course it's, there's 400 members of the SLC family. So there's more complexity in the system there also. But if you take a simple input-output view of the world, and you think glucose in, ATP out, what are the intercellular ATP concentrations look like? Well, they fall dramatically in estrogen responsive cells when you treat them with an endocrine therapy, and they don't in resistant cells. But look at the levels in resistant cells. They are exactly the same as the levels in cells that are dying from tamoxifen, which implies that resistant cells now have a more ATP efficient metabolism. And they may therefore also be more sensitive to further blockades of ATP production. And we've shown that, but I don't have time for that. But if you're gonna see GLUT1 upregulated, what other members of that family are upregulated? Well, over a hundred. And 55 of those are universally associated with per responses to endocrine therapy in breast cancer patients. 55 of them. So as you look at the, as you look at the system, as you look at it from 30,000 feet, every time you look into it, you see something different, but you also see something that's similar. So what are cells doing with these transporters? They're pulling nutrients in from outside the cell to feed intermediate metabolism. And that's going to work along with the activation of autophagy to support intermediate metabolism and drive a more ATP efficient metabolism. And therefore cell sensing, energy and nutrient sensing must be changed. That's also the case, I don't have time to go into it, but you can see here, GRP78 and AMPK are key sensors. And so there's degeneracy and redundancy within the system for how changes in nutrient flow is altered. So how on earth do you put all this together? So if you slept through everything, this is the only slide you need to see. This is the way we believe the framework functions and how cells respond to an endocrine therapy. And I'm going to take a very simple way of walking you through the system. It doesn't have to function this way in all cells. It'll always be perturbed when there are mutations in the system that change the probability that one area of the system will function versus an other. But I'm going to tell you everything that I've just told you, but in sequence. So you give an ER positive sensitive cell an endocrine therapy, you block function of the estrogen receptor. Cells take up less glucose. They actually take up less lots of other things. They take up less glutamine, bunch of other amino acids, whole series of things go down. Consequently, they don't have enough energy to fold their proteins. Now you've got a cell that was about to make a copy of itself, that's got all this material and it can't fold the proteins to finish what it needs to do. So it stops, stops proliferating. And I've shown you that's the case with Q67, for example. As a consequence, activating the UPR. Can't fold its proteins, drops glucose, two different ways in which the UPR is going to get regulated. The UPR is going to turn down transcription and translation to try and get out of that proteotoxic state back to proteostasis. To ensure that that has time to do that, it's going to do two other things. It's going to send a cell a signal to the mitochondria and it kind of says, don't die, I've got a plan and I'm going to activate that plan now. And one of those things is, I'm going to activate autophagy because I don't need all that stuff that I was making to make another copy. Degrade it, feed intermediate metabolism. And just to make sure that that's not the only issue, it's got the cells going to up-regulate where it can, nutrient transporters and pull in from outside and transport through the cell to where they need to be the nutrients of the system. And so you have this beautifully coordinated, integrated system that functions inherently to address the stress of endocrine therapy. Why do breast cancer cells do this? What do you think the breast is trying to do during lactation? It's making large amounts of proteins that it has to synthesize and fold properly and push out of the cell and not activate cell death or get its metabolism out of sync. This is what the breast is wired to do. And we've shown that that's exactly what happens during involution in mice. The same system functions and the cells start to die and they go back to sleep. So now we've got an integrated framework that allows us, when we see changes in different components of the system, to say this belongs over here or this belongs over there, or this is part of the coordination between these functions here and there. Where there are rate-limiting steps in the process, we have new places to intervene in the system to limit the ability to get past the degeneracy and redundancy in the signaling of the system. And that's basically what says here in this summary that we don't need to find new mutations. We don't need to find new genes. We have to look at how they are regulated and work together. It's the rewiring of a naturally programmed system that breast cancer cells are using. And it's probably not just breast cancer cells to survive. I'm gonna stop there. And I listed all the people who did the work at the bottom and they're listed again here. But I would like to thank the patients because as you know, I have no idea who they are. I'm not an MD, I'm a crazy basic scientist who can't wait to get to the beer next door. But the point I wanna make is, these people gave us a piece of themselves, not knowing that they would benefit. And they're not alive today because those survival curves that I showed you in the first slide all came to the X-axis. I mean, that's a tremendous gift and it's a wonderful thing that we all should be thankful for. And so I really will stop there and thank them for that I think I've hit my 30 minutes. We have room for a quick couple of questions. Angelo UVA, Bob, great talk. So if I understand it, then these five fundamental features of the system, if I wanna be an optimist, represent opportunities to perturb the system and get more durable responses by hitting more than one of those. The pessimistic approach or interpretation is that those are the five most robust aspects of a incredibly robust system. And that no matter how we intervene on those, the system will be robust to those perturbations. So in systems theory, I'm sure you know, highly optimized tolerance, robust systems pay for that robustness for fragile points. So are there underlying features well past those five that might represent those fragile nodes that are independent of those? So first I would say, we cure some patients. So the system is not so robust that it can't be perturbed adequately for long enough that enough cells die that patients live past their, they're still by date with breast cancer, if you will. So that's a positive thing. I think what we're looking for is those areas where there is coordination and integration only amongst modules, because that's the opportunity to hit two at once or three at once. And those are much more limited. So, you know, you could, for example, you could try to block cell death, which we do quite well with CDK4-6 inhibitors, but they have a bunch of other things that they do. But that doesn't address fundamentally the system. And so the benefit that we see, we don't see it with all the CDK4-6 inhibitors and the improvement in overall survival is maybe nine to 12 months at best. So you can hit a module and you can shut it down and you can get benefit just from doing that. But the real benefit's gonna come from shutting down more than one. And I think we have to learn how to do that. But now we know where to look. That's Cora, Colorado. So when you're tracking the activity of these different nodes, does it assume that we have the nodes figured out? So when you're building these, is this all based on annotation and our knowledge of how this works? So how do you? So it's a mix of both. We built a whole series of tools that try to teach us what the differential signaling would be. And that's why we focus on the edges less than the nodes. So, and it's another talk to explain how these tools work and what they do. And you couldn't have picked it up because you couldn't have seen it. But in some of the modules, the nodes are red or green because one node, one connection is present in one phenotype, the sensitive or responding, and it's gone in the other. And that's just how we're able to capture that information and represent it graphically. And so it's not that those lines that joined the little circles are the same in every phenotype. They're actually not because it's a dynamical system. So we have to learn some. And we never see canonical signaling in any of it. We see pieces of canonical signals. We never see a canonical signal in anything. Yeah, I think my question is a follow-up to what Dan asked, especially thinking PI3 kinase inhibitors, mTOR inhibitors, even the combination with ER targeting agents, you know, they fail. So based on what you showed, then would it be a better strategy to take out ER targeting and, you know, target all those, you know, other five nodes or... Yeah, if I had the answer to that, I'd probably be getting my Nobel Prize in Stockholm next year. And if anybody, if you have the answer, you might want to go home and write it down because then you can have it. I don't know. Well, the goal of what I wanted to do today was to say that when we look at the system, we see different features of it reflected back. And once we know where to put those, we can actually learn so much more about what we should be doing next than if we just think of it in isolation. And the PI3 kinase AQT is a great example because it sits up there, and it was on the slides. Even if that hadn't worked, and it's probably failing because of internal feedback within that signaling feature, there is tremendous redundancy in the system for the control, for example, of metabolism that gets driven through mTOR, which is the downstream component, the most downstream part of that feature. So you'll get a better head if you had mTOR, but you're not gonna get cures. And we've seen that you haven't, and there's tremendous toxicity and all the rest of it. But you can think of it like a balloon. You push your finger in, and that's what you're seeing with the mutations. You get this local, huge distortion of the network in that one area, but the rest of the network remains the same. And so signaling can recover from a whole series of different directions. And that's the challenge. You don't wanna be playing whack-a-mole. You wanna actually try and shut this system down so it can't recover. Hi, Dave Shapiro with Illinois. You've shown beautifully that the traditional strategy of inhibiting one part of these pathways at a time is unsuccessful. I just wanna just remind people of the alternative strategy, which is how our cancer drug works, and I'm not plugging it, but it's a different approach altogether of over-activating one of these modes, and that can be enough to overcome all of these other things because, as you say, the whack-a-mole strategy of one inhibitor at a time is not as effective. So it's important that folks begin to think of a turn-on strategy for some of these modules because that will also kill cells. All of these are bidirectional pathways. Absolutely, and again, the UPR is a great example of that. If you look at the CHOP arm, if you shut down transcription and translation too long, you actually activate cell death and the cells die. That simple, and the system is designed to do that because there's a problem if it can't fix it and it doesn't want the cell to survive. So the same system that can drive survival can drive cell death, and you've probably all seen autophagy expressed that way too. Is it a cell death or is it a cell survival mechanism? You know, if it goes on for too long, the cell eats up too much of itself, can't survive it, it activates cell death. So you can take different approaches to do it, but you have to know what you're doing and you have to understand where the compensatory functions are within the system. Let's thank Dr. Clark for the wonderful talk. Thank you.

heterogeneity

hormone response

breast cancer

Titan

transcriptional networks

estrogen response

KLFO

ER-positive breast cancer

ER binding

gene expression

endocrine resistance

systems biology approach

improve patient outcomes