Drugging the Epigenome to Target Lethal Disease

Session 3: Drugging the Epigenome to Target Lethal Disease
Moderators: Amina Zoubeidi (Vancouver Prostate Centre) Michael Shen (Columbia University)

The Landscape of N6-Metyladenosine in Localized Primary Prostate Cancer and how these Modifications can Drive Disease Aggressiveness
Housheng Hansen He (Princess Margaret Cancer Centre, University Health Network)\

Characterizing Lineage Plasticity using Circulating Chromatin
Sylvan Baca (Harvard: Dana-Farber Cancer Institute)

Utility of Epigenetic Profiling to Inform Resistance to ARPI
Wilbert Zwart (Netherland Cancer Institute)

Targeting the Epigenome to Alter Lineage Plasticity
Michael Shen (Columbia University)

View the Transcript Below:

Session 3 Drugging the Epigenome to Target Lethal Disease 

Amina Zoubeidi, BSc, MSc, PhD [00:00:09] Good afternoon, everybody, and welcome to the session on epigenetics. And I welcome you on behalf of Dr. Michael Shen from University of Columbia. My name is Amina Zoubeidi from the Vancouver Prostate Center. And here this session is very interesting, that it will take us to explore how diverse layers of the epigenome from RNA methylation to histone modification, and also the chromatin landscape, contribute to shaping our treatment resistance or treatment response, but also the evolution or the resistance or developments of aggressive disease. And also, we will hear at the end about how we can target the epigenome and how the epigenomes, targeting the epigenome still have an important place in prostate cancer. And to start with, we’re gonna have Dr.  Hansen He from University of Toronto that he will tell us about the RNA methylation landscape. In localized prostate cancer and how does methylation contribute to aggressive disease. Thank you.  

Housheng Hansen He, PhD [00:01:21] Okay, so thanks for the opportunity to share some of our recent work on what’s so called RNA epigenetics. So, there’s some disclosures. So, we know when RNA got transcribed, there are many kinds of modifications can happen and there are actually over 150 types of RNA modifications has been identified for now and N6 adenosine methylation, which we call the m6A modification, is the most abundant internal modification in human RNA, and that’s where we have been focusing on. So, the m6A modification has been characterized in the 1970s, and the writer, major writer, METTL3, has been identified in the 1990s. But the field really started in 2011 where the first eraser of FTO being identified. So, this really determines that m6A can be added and can also be removed. So that’s a feature, a typical feature of epigenetic. And then through the work in the last decade or so, now we have a very comprehensive understanding of the regulators of m6A, including large writer complexes, and then now we know another eraser called ALKBH5, and there are also many readers being identified, and then when the m6A being deposited, its function actually really dependent on which readers recognize it. So, with YTHDC1 mostly in the nucleus and helping the RNA splicing and then the modified RNA to export to the cytoplasm. And in the cytoplasm, there are also many kinds of readers and if it binds to YTHDF2 reader it’s mostly related to the RNA decay. If it’s a bind to IGFBP families, they’re mostly. Important for its RNA stabilization, and if it’s binding to YTHDF1 and DF3, it’s mostly promoting the translation initiation. So, there’s still some debate in the field, but I think the knowledge of the readers still holds true. So, in prostate cancer, actually, it has become quite relevant, as you will see from the reader’s main writers, Sorry. METTL3 is actually expression keeping going up along the disease progression from the CRPC to NEPC while the erasers FTO keeping decrease and it maintains very low in CRPC and NEPC. And then this consequently resulted in which we call the hyper-methylated phenotype of prostate tumors that have been, as shown by multiple groups, including ourselves. And then you will see that there is higher level of global level of methylation of m6A in patient tumors compared to the benign tissues. And in addition to the writer and erasers, the readers also change quite a lot in patient tumors and mostly related to copy number amplification. And the most obvious one is the major reader, YTHDF1, being amplified in almost 25 percent of the primary prostate cancer, and that percentage goes up even further in metastatic tumors. And then based on this, there are an understanding of the major regulators. There are many efforts in industry try to develop small molecule inhibitor drugs targeting many of the regulators, and with the writers being the most advanced ones, with the STORM therapeutics right now has inhibitor targeting METTL3 in a number of phase one clinical trials, mostly in leukemia. So, we’re hoping to see this moving to solid tumor, including prostate cancer very soon. So, in addition to the understanding of the regulators, reader, writer, erasers, there are also different. Another layer is the understanding of where the modification happens, and which transcript has the modification. And the technology to map that has been developed in 2022, one Nature paper, one Cell paper, use the antibody against m6A and then pull down followed by sequencing. And through this effort, we know that most of the m6A modifications happens near the stop codon, and it also has a specific sequence recognition, which we call them GGACU, with the A being modified. But this technology has a limitation because it needs a large number of starting materials typically. At least 30 micrograms of total RNA. So, this is impossible for us to analyze m6A in patient tumors. So, the first thing we did in 2017 is to develop a low-input m6A profiling approach and we can reduce the starting material from 30 micro gram to 500 nanogram and then with this it allows us to profile patient tumors with m6A modification. So, in prostate cancer, we leverage an intermediate risk prostate cancer cohort that established locally in Toronto, which is called the Canadian Prostate Cancer Genome Network, CPCG in short. So, we generated the m6A modification for 148 primary patient tumors from this cohort and identified over 32,000 of m6A modifications with almost 10,000 genes on average in the patient tumors being m6A modified. And the one thing we noticed that typically in the oncogenes, for example, we show here, including androgen receptor and MYC are heavily m6A modified while the tumor suppressors including TP53 and PTEN are usually alone in m6A modified or with no m6A modification. And with the m6A profiling data available, we actually can group the patients based on their m6A patterns. We can group them into a five-patient group. And remember, this is an intermediate risk group. And even in the intermediate risk group, we can still see a very clear separation of survival. And actually, this survival is strongly associated with tumor hypoxia with the most aggressive subtype P1 has the highest level of tumor hypoxia. And we know that the tumor hypoxia is very relevant to prostate cancer progression. There are a lot of studies that have been published in previous years, and in 2019 we have a study together with YC-1 showing that the tumor hypoxia actually goes up to the roof during the disease progression, in particular at the late stage of a CRPC and NEPC, and relevantly, if they have higher hypaxial level, the response much better to the hypoxia targeted therapies. We tested a drug called evofosfamide that is a pro-drug that’s specific and activated in low oxygen conditions. And based on this result, we’re launching a phase two, small phase two clinical trial, targeting hypoxia in patients who failed a first-line abi or enza treatment. And we also showed that using Amina Zoubeidi’s model of A16A(?), that in the hypoxia conditions that we saw a drastic increase of m6A modification. So, with this, if you remember that the readers, that METTL3 constantly goes up during the disease progression and then the hypoxia also goes up in the disease’s progression. And we think there’s, we call them m6A symbiosis, where you would expect a super hypermethylated phenotype in the late stage of prostate cancer, including CRPC and NEPC, and the question is that whether we can target both hypoxia and m6A modification by combining the hypoxic drug with the METTL3 inhibitors. So, this is work that is ongoing in the lab. In addition to the global patterns that we show, they’re very clinically relevant, individual modifications also can be clinically relevant. So, this is analysis looking at the m6A association with disease survival, and the top three genes are actually very closely relevant, all relevant to VCAN function. As you will see here, the high methylation at these genes are associated with worse clonal outcome. And we can include and encode a protein gene called Versican. It’s a marker for fibroblast cells. If you do a single cell analysis, you almost for sure will use this marker for cell type separation. And we can also find that the MR level and the protein level, on VCAN. Is also associated with the clinical outcome, but the odds ratio is much higher when we look at the m6A level. So how does this function? Why the m6A modification on VCAN is so strongly associated with clinical outcome? And to analyze this, we use the approach that is using the catalytic DAT version of CasRx, which target RNA, and you can use guide RNA. To specifically target any regions that you want to modify a modification by either fuses with the right or METTL3. You can deposit methylation there and you can fuse with the eraser, and you remove the modification there. And with this tool, we’re able to very efficiently add m6A modification on VCAN and you will see that the deposition of the m6A modification increase the VCAN MRI, but more importantly, increase the protein translation. And then we show later on is that the increased translation or protein production of VCAN can be secreted to outside of the cell and then form a major part of the extracellular matrix which promote tumor cell progression. And we know that the VCAN is a marker from fibroblast cells is usually generated by fibroblast cells, but in the cases, we found here is in tumor cells, if they have high m6A modification, the tumor cells itself will generate loss of VCAN cell itself, and it does not really need the fibroblast cells to promote tumor progression. And we recently also developed a screening platform because we identified so many m6A modification sites. So, using this CasRx writing system, we’re able to target 2,000 m6A modification sites in prostate cancer, and we did functional screening. We’re able identify a number of modification sites that are very important in prostate cancer and we show one of the top hits being a gene called CHD9 and the modification on CHD9 recruits reader YTHDF1 and DF3 that promote its protein translation and that we show later on being a new tumor suppressor genes and then there is a poster by my post-doc, Xin Xu, is in the audience so you can learn more if you’re interested from the poster session. So, you know, we showed in the cases when the modification happens, you know, it eventually will affect the protein. And my mass spec colleagues will say, Hansen, we don’t need your assays. We just do a mass spec, and then we got all the answer. And I always show this example where we found in lung cancer, we did the same m6A modification. And we did the association with the clinical outcome. And you will see the top gene called the BLVRA. So m6A pattern has a very clear separation of a clinical outcome, but if you look at the mRNA, you don’t see any separation, you look at the protein, you don’t see any separation, and we also don’t see any correlation between the m6A level versus MRNA or m6A level with the protein. And this phenotype can be very nicely validated in an independent cohort or another 80 patient tumors, m6A very nicely separate the clinical outcome. mRNA has no separation, we don’t have the protein for this cohort, but it seems that the clinical association only happens at the m6A level. But this is go against our textbook understanding of the function of m6A which eventually either affect the RNA abundance or affect the protein abundance. So, we are right now doing quite a lot of work, so we’re thinking that it’s probably in many cases that m6A modification has actually functioned through affecting the structure of the RNA, and that may not necessarily related to its RNA stability or translation. Okay, so to summarize, we showed that in prostate cancer, about 30 to 40% of the genes are methylated, and the global m6A methylation patterns are associated with clinical outcome, in particular related to hypoxia. And we showed that in prostate cancer, it’s typically hypermethylated, and this pattern goes up even higher in late stage CRPC and NEPC. And we showed the site specific m6A modification can also be clinically relevant either as a biomarker or therapeutic targets. And with that, I will thank you for your attention.  

Amina Zoubeidi, BSc, MSc, PhD [00:17:50] The floor is open for questions.  

Daniel Spratt, MD [00:17:53] Very nice Hansen, so I was really interested with the link between the hypoxia and in the RNA modifications too. So, could you show like, if you take like a reductionist approach, are you able to show this hypoxia if you would just put in a hypoxia chamber? Does that actually cause the modifications?  

Housheng Hansen He, PhD [00:18:11] Yes.  

Daniel Spratt, MD [00:18:11] So this is like a direct?  

Housheng Hansen He, PhD [00:18:13] Yeah, it’s a, you put the cells under hypoxia chamber.  

Daniel Spratt, MD [00:18:18] Do you have any idea what the mechanism is? Is it like the oxidation, or is it affecting something else?  

Housheng Hansen He, PhD [00:18:25] Yeah, we don’t know the exact mechanism, but I think it’s through the metabolism effect.  

Daniel Spratt, MD [00:18:31] So I mean your thought is that basically oxygen is modifying some oxygen-sensing protein and that’s causing the modification?  

Housheng Hansen He, PhD [00:18:39] Right.  

Daniel Spratt, MD [00:18:39] Okay.  

Unknown [00:18:43] Great talk. I’m just curious that as you go from ARPC to NEPC, now just looking at the m6 modification, is there a way you can now tell the genes that could be driving based on the m6 modification rather than the expression? And my second question is that, is it like those one or two genes that you’re talking about or is that a general readout that we can get from m6 being high has a higher induction of effect on the protein expression or is it just one or two targets that have that kind of effect?  

Housheng Hansen He, PhD [00:19:23] Yeah, that’s a great question. So first, I think it’s a global increase of m6A modification. We’ll call them symbiosis of hyper-m6A methylation. So, we’re now doing m6A profiling in CRPC and NEPC patients. So, we don’t know the answer yet. But my guess is that there is a global increase.  

Amina Zoubeidi, BSc, MSc, PhD [00:19:53] If you don’t have any other question, I will leave the floor to my co-chair, Dr. Michael Shen.  

Michael Shen, PhD [00:20:04] OK, it is now my distinct pleasure to introduce Sylvan Baca, who is at the Dana-Farber Cancer Institute and Harvard Medical School. The title of his talk is Characterizing Lineage Plasticity Using Circulating Chromatin.  

Sylvan Baca, MD, PhD [00:20:22] All right, thank you, good afternoon. So yeah, thanks to PCF for the opportunity to tell you about our work on characterizing lineage plasticity from circulating chromatin. These are my disclosures. And I figured I’d start since this is a session on epigenetics by just framing how I view the epigenome in cancer. So, I think of the genome as the hardware of the nucleus and the epigenome as the software. And the hardware is basically fixed in that cells in a multicellular organism tend to share the same DNA sequence, but it’s because of the epigenome and the programming that that allows that cells can take on very specialized functions as they developed and fit the needs of a multicellular organism. And we know through work of people in this room, as well as others, that this process can be sort of corrupted in cancer and developmental programs can be reactivated inappropriately to allow things like metastasis, invasion, and treatment resistance. I think of the major function of the epigenome as controlling gene expression. And it does this through a large array of chemical modifications to DNA and to the histone proteins that DNA is wrapped around, some of which are shown here. And because of recent technological advances in liquid biopsy, we can increasingly study these features and how they’re present in cancer using liquid biopsy, and along, you know, that brings some advantages that allow us to look at changes longitudinally in a way that we couldn’t if we were just studying this from tissues or in model organisms. So, this is a focus of my lab, and I’m going to talk about lineage plasticity and prostate cancer today. So, as you know, this is a process where prostate adenocarcinoma, which usually is dependent on androgen receptor, can become resistant to AR targeting therapies by shedding its dependency on AR and taking on other histologies. The most common of which is neuroendocrine prostate cancer. Which is insensitive to AR inhibitors. And our work and others have shown that this process is enacted by major changes in the epigenome. So, when we looked at this with ChIP-seq to measure histone modifications that mark active regulatory elements, you can see that there are tens of thousands of enhancers and promoters that get turned on or off across the genome when you compare prostate adenocarcinoma to treatment emergent NEPC. And this gives us an opportunity to build signatures that can detect in a very specific way the emergence of these neuroendocrine cells. And this has actually been done in a couple of different ways using several different epigenome profiling technologies from cell for DNA. And what these all have in common; they were basically all done by investigators supported by PCF. So, this is a field where PCF has really led. And whether you look through DNA methylation profiling, histone modifications or these signals of nucleosome depletion that can be read out from cell-free DNA, it’s pretty clear that we can, you know, detect neuroendocrine prostate cancer from cell-free DNA, you know if there’s sufficient tumor fraction. And this is just one example of work led by Jacob Berchuck that I got to work with him on when he was at Dana-Farber. So, this is looking at a DNA methylation assay where we’ve designed a signature of neuroendocrine differentiation by comparing patient-derived xenografts that are neuroendocrine prostate cancer versus prostate adenocarcinoma. Finding the most differentially methylated regions in the genome and then looking for those signals in cell-free DNA by measuring DNA methylation profiling. And you can see in this test cohort and also in a validation cohort that came from a separate institution, these patients who had greater than 3% tumor-derived cell- free DNA, this test did a pretty good job of separating patients with biopsy-confirmed NEPC from patients with prostate adenocarcinoma. And as you would expect, the patients who had elevated NEPC scores in their cell-free DNA had worse survival. So, I think at this point, it’s pretty clear we can detect this, and I’m really interested in thinking about where do we take this next in a way that it could maximally benefit patients. Prospective validation, I think, is going to be very important. I won’t talk about that today, but I’m going to tell you a little bit about our efforts on characterizing NEPC from circulating chromatin, as well as monitoring and trying to detect lineage plasticity early as it emerges. Recently, in work with Matthew Friedman’s lab when I was finishing my post-doc there, we described this assay called cell-free ChIP-seq where you can profile histone modifications on circulating chromatin that comes from cancer and is shed into the plasma. And what’s really interesting to me about this is that it gives you a very deep and broad view of gene regulation in cancer. So, we can look at a number of different histone modifications. They all annotate some function of the genome. We have a mark, H3K4 trimethyl that looks at active gene promoters. So, where genes are being transcribed, that mark tends to be high. And we can also look at this H3K27 acetylation, which captures activated enhancers as well as promoters. So regulatory elements where transcription factors bind and can increase expression of nearby genes. And then we also get DNA methylation as a readout. What I like about this is that you can get single gene resolution views of gene regulation and expression. So, this is an example of what the data look like. These are plasma samples from three patients with prostate adenocarcinoma in blue, and then neuroendocrine prostate cancer in green. And you can see if you look at the androgen receptor locus, there’s a lot of this, in this case, this is showing H3K27 acetylation. There’s a lot of this system modification along the promoter and end of the gene body of the androgen-receptor. When neuroendocrine differentiation happens this signal gets turned off, and you don’t see it in NEPC, but you see an increase in this histone modification along the gene body of ASCL1, which we heard about earlier and is a master lineage transcription factor that mediates this switch to neuroendocrine prostate cancer in many cases. And looking across a variety of different cancer types, you can pick out lineage-specific examples of genes that are highly expressed in these different cancers. So here, each row is a patient with the indicated type of cancer. We’re running this assay on their plasma, looking at this H3K4 trimethyl mark that tracks with active gene promoters. And you can see that in each cancer; you can see high signals of a diagnostic marker gene that you expect to be highly expressed in that cancer. So, the first row is prostate adenocarcinoma. In that patient’s plasma, we see high levels of this mark at the KLK3 promoter. In the second row, you see in a breast cancer that was ER positive, high levels of this marker at ESR1. And so, we can do this not just to look at diagnostic marker genes but more interestingly we can look at therapeutic targets as well where expression levels you know matter in terms of predicting response. And so, here’s looking at the ERBB2 locus which encodes HER2 we looked across a number of different cancer types where HER2 can be expressed including breast colorectal and esophageal cancer. And just by quantifying the level of this histone modification at the gene promoter we can distinguish between patients who do or do not have HER2 expression on IHC of matched biopsy tissues. And there was actually one patient with colorectal cancer who hadn’t been assessed for HER2 expression since it’s only present in about 2% of cases. When we saw high signal in the plasma, we went back and looked and found that this patient’s archival biopsy tissue did indeed highly express HER2. Now, getting to prostate cancer, we can look at things like expression of PSMA. So, the full H1 gene that encodes PSMA tends to have high levels of these histone modifications when PSMA is highly expressed. And so, this is not my work, but this is a really nice study by Praful Ravi, Heather Jacene, and Jacob Berchuck that was done in collaboration with Precede Biosciences where they looked at a cohort of patients who had PSMA PET measurements and they observed a correlation of this level of histone modification in the plasma with the SUV mean measured on these scans. And they saw this in a training cohort and then also in a validation cohort it held up. So, this could be a kind of blood-based way of evaluating the overall expression of targets like PSMA. And I think this is a timely thing to think about because as we’ve seen in earlier talks, there’s a lot of development of therapies that basically have this common principle that they’re targeting cell surface proteins that can be aberrantly expressed or specifically expressed in cancer. And these include antibody drug conjugates, radioligand therapies, CAR T. cells, T Cell Engagers. And there’s been a real acceleration of development of these agents. This is pretty dated now. This is from 2023. But already, you can see that the growth of targeted biologics was pretty fast. And this has only accelerated since then. This is showing FDA approvals. But there are many, many trials of these agents, and I suspect that as more of them make it to clinic, it’ll become helpful to have a way of assessing what targets are present in a given patient’s cancer. You know, enabling the frame of precision oncology, but rather than looking at mutations, looking at expression levels of a target. Obviously, expressing the target doesn’t mean that the cancer will respond, but I think it’s a first step, and in many cases, probably necessary for a robust response. So, you know in some of these cases, you know we had neuroendocrine prostate cancers where we had matched biopsy tissue. You know, we saw high signals of DLL3 promoter activity in essentially all of these, and in a subset, we were able to confirm this on IHC of matched tissue, that there’s membranous expression of DLL3 as we predict from these signals in plasma. And so, this is a genome-wide assay, so we can look not just at one target at a time, but we can ask across the whole genome, depending on how you count, there could be dozens to hundreds of tractable cell surface targets where there’s high expression in cancer. And these are just a couple examples, but we can see examples where there’s probably high expression of targets where we already have drugs. For example, NECTIN4, this is these neuroendocrine prostate cancer samples, we see high levels of NECTIN4 expression, you know, which we can target with Enforcement Abodotin. So, the nice thing about doing this in blood is that, as I mentioned, you can monitor this longitudinally in a way that would be very difficult with, you know serial biopsies. So, these are three patients where we had multiple time points sampled and we can kind of track, you know how is the predicted expression of these targets changing over time. As they develop neuroendocrine differentiation and respond to treatment, and as the cancer evolves in response to treatment. And so, this is a sort of vision for the future. This is obviously a little way off. There’s a lot of work to do here still in validating these ideas, but I think we could reach a place, especially as we get more of these cell surface targeting agents online, where a patient could be followed longitudinally, their expression of these targets could be inferred from cell-free DNA, and you know, you could change therapy and sort of have real-time precision guidance for these cell surface targeting agents, and, you know I think an emerging mechanism of resistance to these agents that makes sense is that the target can be downregulated. Along with that, there can be other changes in gene regulation that happen, and we’ve seen in some unpublished data, this is just a couple of cases, but anecdotally sometimes when a target is downregulated, you’ll see upregulation of another cell surface protein target. And that may enable new therapeutic opportunities that you wouldn’t necessarily know about if you’re just sort of going off assessment of archival biopsy for looking at expression of these targets. So next I’ll talk a little bit about monitoring and early detection of lineage plasticity. So, we’ve assembled a cohort of patients who underwent neuroendocrine differentiation after being treated with AR targeted agents. And this is kind of an overview of part of the cohort here and the X axis shows time centered on day zero when the biopsy showed neuroendocrine differentiation. And so, we have a number of different blood draws that were all samples of convenience where patients donated to our biobank. But we can follow how some of these signatures and plasma change over time. And what’s nice about looking at neuroendocrine prostate cancer is it’s so distinct epigenetically from prostate adenocarcinoma that it’s pretty easy to quantify these differences in a robust way and actually track the relative portion of prostate adenocarcinoma versus NEPC-derived chromatin in circulation. These are two examples here. You can see how these kinds of change in response to treatment. You know, it’s pretty early looking at this cohort, but one observation, first of all, is that, you know, many of these patients either have a lot of NEPC signature or prostate adenocarcinoma signature. Interestingly though, some patients have a mixture of both. I’ve highlighted with these red arrows four examples where these are blood draws at a time after neuroendocrine prostate cancer had been diagnosed on a tissue biopsy, and yet these patients have a large amount of this prostate adenocarcinoma signal in their circulating chromatin, which could have therapeutic implications. This, to me, obviously there’s a lot to be done to validate this, but this may mean that these patients have regression, not just with the neuroendocrine component, but also with the prostate adenocarcinoma component of their disease. We think these are separate cells contributing to these signatures, reasons I can go into later. But this may mean that if you have a target that’s really specifically targeting something or treatment targeting something just in the neuroendocrine prostate cancer component of their disease, this may not be sufficient for treating the prostate adenocarcinoma. So just to make a brief comment on detecting lineage plasticity early. So, we’ve grouped these patients by just time intervals leading up to and after the biopsy that showed neuroendocrine differentiation. And for a couple of these patients, we had blood draws that happened to be done up to three months before neuroendocrine prostate cancer was diagnosed on biopsy. In five of these patients, there were elevated signals that we could see up to a couple months prior to the biopsy. And there’s a lot more to be done in terms of validating this prospectively, but I imagine this could enable early treatment switches or therapy addition trials. For anyone who’s treated these patients you probably have examples where patients got sick very quickly and knowing about this ahead of time. You know, could be helpful. And I’ll just end by mentioning that you know this ability to look at lineage plasticity and circulating chromatin is not limited just to neuroendocrine differentiation. This is a nice example of a collaboration with Jacob Berchuck that was led by two talented post-docs in my lab, Karl Semaan and Rashad Nawfal, where they had these serial blood draws in a patient who developed resistance to lutetium and this patient after treatment had growth of a thoracic lymph node, it was biopsied and it was found to have histology and cell markers of squamous differentiation. And looking back, these cells have the TMPRSS2-ERG fusion, so they did originate from the original prostate cancer. And you can see, because, you know, squamous differentiations have a similar kind of epigenetic signature that can be detected in the same way that the neuroendocrine one can, you can sort of follow the level of the signature over time as resistance developed. So, I’ll summarize now, just wrapping up, that, you know, lineage plasticity can be detected from epigenetic signatures using blood biopsy. I think there’s a lot of exciting work doing this, including in circulating tumor cells, which I didn’t even get into. You know, epigenetic signals in circulating chromatin, they reflect gene expression, and because of this, I think, there are opportunities for looking at expression of targets and potentially guiding therapy with new cell surface targeting agents. Finally, early detection of lineage plasticity may be possible. Requires prospective validation in large cohorts, and this would be a great place to collaborate for people who are interested. And this may open opportunities for early intervention trials, particularly as we learn about ways to maybe slow or reverse lineage plasticity, which I think there’s some talks describing. So, with that, I’ll just wrap up and say thank you to the many people who enabled this work. In my lab, Gary Lee and Rashad and Karl Semaan were the main kind of champions of much of this work, none of this would possible without Matt Freedman, his lab basically developed this assay, and then Ji-Heui Seo is a senior post-doc in the lab, who spent a lot of time optimizing kind of the technical aspects of the assays that works robustly. Also, I’d like to thank Jacob Berchuck, who’s been a long-time collaborator, as well as my clinical and research mentors, Toni Choueri, Misha Beltran, and Mary-Ellen Taplin. And of course, also want to thank PCF, as this would not be possible without their support. So, I’ll stop there and happy to take questions if there are any.  

Unknown [00:36:49] Over here, hi.  

Sylvan Baca, MD, PhD [00:36:50] Oh, hi.  

Unknown [00:36:51] Hi beautiful talk. So, I was particularly intrigued by the fact that you are potentially seeing differences in blood in the squamous and the neuroendocrine. Can you talk a little bit more about that?  

Sylvan Baca, MD, PhD [00:37:04] Yeah, so, you know, just like neuroendocrine differentiation, there are specific transcription factors that seem to sort of drive the development of squamous lineages, and then also in this lineage plasticity setting, it’s the same ones we think that are probably mediating that and the same kind of sets of enhancers and gene promoters that are upregulated, so including promoters of, you now, some of the cytokeratin marker genes you would use to look at, you, know, squamous lineage, those ones are upregulated, and it’s maybe not quite as stark as with neuroendocrine prostate cancer, but still many hundreds of thousands of kind of distinguishing epigenetic features there.  

Unknown [00:37:38] And do you see, so what would be really cool is if you could see maybe a ratio between them and things like that, if you can see both the neuroendocrine and the non-neuroendocrine.  

Sylvan Baca, MD, PhD [00:37:50] Yeah, exactly. No, we’re really interested in quantifying the relative contribution of these different kinds of epigenetic subtypes and benchmarking that, which is an important step to do still. Thanks for the question. Go ahead.  

Unknown [00:38:03] I was wondering, if you were able to profile, like, in a temporal fashion, during the process of lineage transition, maybe there are differences in which modifications or DNA methylation appear first, or if there’s a certain order, and does that sort of tell you about causality of gene regulation?  

Sylvan Baca, MD, PhD [00:38:23] Yeah, that’s really interesting. I don’t think we’ve looked at that yet. You know, just in the cases that I can think of, you know, where I might have observed this, I think often there’s, you know concordant changes in DNA methylation and histone modifications at the same time, but you know definitely we haven’t sort of sampled densely enough to sort of say that for sure.  

Joshua Lang, MD, MS [00:38:46] Hi Sylvan, Josh Lang. I’m a huge fan of your work. This is really exciting. And of the many questions that are spinning through my head, one of them is thinking about the different subtypes, you know, neuroendocrine is not neuroendocrine is not small cell, you know what I mean, the list goes on. Can we start to better define with this, your approach to epigenomic profiling the different types of neuroendocrine transformation or transitions that can occur? And how can we quantify that and bring it back to the different mechanisms driving that. 

Sylvan Baca, MD, PhD [00:39:10] Yeah, I totally agree. I would love to get deeply into this. I mean, in small cell lung cancers, you know there’s these different transcriptional subtypes even within something that looks histologically very uniform. And here, there’s sort of different levels of neuroendocrine differentiation. I think there’s a lot we’d like to study. I think a lot of it will require rapid autopsy series where we can actually benchmark what’s going on across a patient’s tumor burden to the signals that we see in plasma. But I think that’s a really important sort of next step as well, and we haven’t done much there, but that’s the direction we’d love to go in.  

Joshua Lang, MD, MS [00:39:39] Thanks!  

[00:39:42] Sylvan, over here?  

Sylvan Baca, MD, PhD [00:39:43] Yes.  

Brian [00:39:43] Hi, Sylvan. Brian, great to see you. Enjoyed thinking about various collaborations with you. Love your work. Great talk. Off the wall question here, but the multi-cancer early detection tests that are out there using methylation patterns, do they ever pick up a lineage plasticity signal or any of these signals? And if so, is that something that we can use to think about the various causative processes behind prostate cancer initiation?  

Sylvan Baca, MD, PhD [00:40:07] Yeah, that’s really interesting. I don’t know of any data on this. I imagine they probably could. The one that’s furthest ahead in clinical development probably is GRAIL’s assay. And that one does include some neuroendocrine small-cell cancers. And just given how different the methylation profiles are, I wouldn’t be surprised if they can pick that up. I mean, I think this early detection of early-stage cancers has gotten a lot of investment. I think, this idea of trying to detect lineage plasticity early is kind of an area that I don’t think has been explored much, but I’d love to see more done there.  

Unknown [00:40:36] Can I answer that question about GRAIL? Because I’ve worked a lot with GRAIL. I think GRAIL is extremely interesting, but they’re definitely not ready to tell you about lineage plasticity. They’ll tell you prostate cancer or they’ll tell maybe lung cancer. They’re not there yet unfortunately.  

Sylvan Baca, MD, PhD [00:40:53] Yeah, that’s good. Thank you.  

Michael Shen, PhD [00:40:56] Okay, thank you very much. So now it’s my distinct privilege to introduce Wilbert Zwart. Wilbert is a group leader at the NKI, the Netherlands Cancer Institute, and he’ll be speaking on the utility of epigenetic profiling to inform resistance to androgen receptor pathway inhibition.  

Wilbert Zwart, PhD [00:41:21] Great, thank you. Thank you, Michael. It’s an absolute pleasure to be here. In my team, we’re, obviously, it’s my disclosure, all right, in my team we’re intrigued on how this very simple model on how androgen receptor drives its transcription networks, how that translates to the clinical reality. And in doing so, we have been investing over the last decade very heavily in trying to set up technologies, how to perform high quality ChIP-seq analyzes in clinical specimens, either directly from the surgery theater or from pathology. And now we’re capable to work with very, very small amounts of material. Here you can see some imaging data of an enlarged lymph node of a patient with metastatic prostate cancer. A H&G coordinated biopsy was taken, and from that we take three slices of a couple of microns, and that already enables us actually to have a genome-wide assessment of AR as well as K27 acetylation for active promoters and enhancers. Using this technology, we learned really a great deal on how androgen receptor works, and more specifically, that it shows a very, very high level of plasticity in tumor genesis and progression. So, we found that an AR is reprogrammed in tumor genesis. It actually also occupies very different regions throughout the genome upon the acquisition of protein coding somatic mutations. But also, in disease progression and adaptation to therapy, we see a very, very distinct AR cistrome occurring. In samples from autopsy programs, we could identify that the AR cistrome is actually very strongly conserved between different metastatic lesions from within the same patient, indicating also that biopsy bias when it comes to different metastatic lesions doesn’t really play a major role here. And another thing which is relevant to mention is that these reprogrammed AR sites, that they are indicative of context dependent patient prognostication. And what this means is that for those sites which are reprogrammed in tumor genesis, they regulate genes which are prognostic in the primary disease setting, while for those sites which are reprogrammed upon metastasis formation, they regulate genes which are also associated with prognosis in the metastatic disease setting. Now, these were tremendously fantastic collaborations with a lot of colleagues who were over the world. And they learned us a lot, but actually these were all cohort studies. So, we wanted to make the next big step in trying to implement these technologies now in prospective clinical trials and trying to actually see if those profiles would be indicative of response to AR targeted therapy in those settings. So, I’ll tell you two small stories, one study in the neoadjuvant setting and one in mCRPC. So, for the neoadjuvant study, it’s published a couple of years ago, so I’ll keep it brief, but just to give you sort of a feel on the directionality there. So, this was a neoadjuvant study where patients received three months of monotherapy enzalutamide. We had a biopsy before treatment and a surgical section after treatment on which multi-omics profiling was performed on the paired samples. Well, we obviously observed that AR activity was going down. But at the same time, we could see that there is this massive epigenetic reprogramming that FOXA1, the pioneer factor for AR, was relocated to new regions on the genome, and on those new regions AR was occupying those new regions, but AR was inactive. Now this other factor comes into play, ARNTL, I’m going to explain you later what that is, and that ARNTL, as you can see, the top figure, when FOXA1 was relocating ARNTL would follow. And it also got higher in its chromatin interaction capacities post-treatment in our patients, but also in our cell lines. That ARNTL protein was also upregulated in our patient samples, not only on protein level, as you can appreciate here, but also on transcript level, where the levels by themselves were associated with poor outcome. And that was only observed in post-treatment samples. While the samples isolated before treatment, there was no difference between patients who did or did not respond to the treatment. We also found an acquired dependency of the ARNTL. So, when we performed a knockdown of that ARNTL in our parental LNCaP cells, there was no effect whatsoever on the proliferation capacity. But when we do the same thing in our enzalutamide-resistant cells, we could actually see there was a decrease in proliferation capacity when we got rid of ARNTL. In vivo, the effects were even more pronounced, that taking resistant cells and then treating those with enza, obviously there’s not that much happening, but when we get rid of ARNTL, we could restore sensitivity to enzalutamide by just knocking out ARNTL. Now, that ARNTL, it’s also named BMAL1. It’s a classical circadian rhythm regulator. Classically got nothing to do with prostate cancer. And now we found it actually now to play a role in a non-circadian fashion to now become a driver of acquired resistance in prostate cancer. It’s a concept we position now as biological repurposing. Now, over the last years, we’ve been trying to find proper inhibitors of this process and ARNTL, that’s a hard one to target. But CLOCK, there are actually inhibitors for that. And in preliminary data, we actually see that while with monotherapy, in LNCaP cells, we don’t really see anything happening when we’re using a CLOCK inhibitor in increasing concentration, but when we combine it with enzalutamide, we can actually see a synergism with decrease in proliferation capacity by combining CLOCK inhibition and enza. Now, for the second story, I’d like to move, then, to the mCRPC setting. And this was also a phase two trial, 60 patients were enrolled in the study, and we have biopsies before treatment and also subsets of patients, also biopsies after treatment. We have long-term follow-up of those patients, and on the samples, K27 acetylation, ChIP-seq analysis was performed. On those samples, we did a supervised analysis, directly comparing those patients who did respond versus those who did not respond well to enzalutamide treatment. And we could identify a bit over 600 K27 acetylation regions that would hallmark resistance to enzalutamide before any treatments. This is what these peaks look like, and even though in the responder-enriched sites which was only 25 of them, so that’s a very, very concise list. But for the regions enriched in the non-responders, 657, those were highly robust, highly reproducible and observed in every single patient where we analyze these data and what you can also see quantified in this slide. Now as a next step we wanted to validate these results because this is at the end a supervised urological analysis. And here we run into trouble because there was not a validation cohort available. Well, luckily, we could team up with Eva Corey and Pete Nelson, who performed a generated lookup series on which response to castration data is available, and a couple of years ago, we performed ChIP-seq analyses for K27 acetylations over this very same histone mark together with Matt Freedman on the very same lookups. So, all we had to do is to couple the response to the castration with the K27 acetylation ChIP data, and that provided us with a validation cohort. And in doing so, we could see a very clear distinction between PDX samples which had signal at these regions versus those which didn’t, directly translating what we observed in our patients to these PDXs. And again, as you can see in the quantifiable plots on the right. Now, when we’re looking to the response orchestration in the animals, we see a clear distinction for those animals where there was weak K27 acetylation at these resistance-associated regions. We can see that there’s a clear response to castration. While if there was a strong signal at these regions, there was no response to any degree in the PDX samples. And of course, I’m showing you here one PDX. Here’s all the data. And you can see very consistently that for all of the PDXs analyzed, if there were a weak signal for the K27 acetylation, there was response to the castration in contrast to the other animals. And also, what you can say quantified here. Y-to-Y-axis shows you the doubling time of these tumors. So, the larger doubling time, the slower the tumor growth. And you can see that for the weak K27 acetylation containing tumors, we can actually see there’s an increase in the doubling-time response to castration, which is not observed if there’s a strong signal there. Directly confirming our patients observed predictive epigenetic classifier towards a PDX-type observation. So, the genes which are on the control of these specific regions, we could validate that those to be elevated and specifically elevated in the subpopulation of enzalutamide resistant cells. This is single cell RNA-seq data where the cells, and in fact in cluster three, which were most specifically enriched for this specific classification. So, indicating that the subpopulations of those cells are indeed positive for these The next key question is what proteins are now driving this specific phenotype? And in order to identify that, we made use of an InSilico database which contained over 14,000 individual ChIP-seq data sets. We projected those on top of our resistance associated K27 acetylation epigenetic regions. And we could then identify a couple of known drivers of resistance including GR and F3C1. I also identified FOXA1. On which we also perform ChIP-seq in our patient samples. And I hope you can appreciate that we, when we now look into those resistance-associated sites, indeed we observe a stronger signal at these specific regions. Now, to functionally interrogate direct biological contribution of this, we performed a focused siRNA screen, looked at three therapy-resistant cell lines, performed siRNA for each and every single one of those individual hits, and then looked for proliferation capacity when we got rid of those hits. As one does, then normally you then deconvolve the hits by individual siRNAs, and we can validate a couple of very well-known and classical markers. GATA2, FOXA1, NKX3 as essential for proliferation capacity. And that also includes HDAC3, which is actually something we can easily drug and easily target. So, we decided to do a drug synergy assay in parenteral LNCaP cells but also in castrate-resistant 16D cells between our vorinostat, which was an HDAC inhibitor on the x-axis and enzalutamide which is on the y-axis. If there’s a clear red indication, it shows there’s drug synergy between them, which you observed in vitro. Also, ex vivo, when we take the PDX samples into culture and then perform the drug synergy assays, again, confirming the synergy. But also in vivo, in PDX models, where we can see the enzalutamide by itself didn’t do that much. This was an enzalutamide-resistant PDX model. But then, vorinostat or the combination of the two effectively blocked tumor cell proliferation. And now, as a next step, we’re trying to translate this back again towards a new phase two trial where patients who are progressive to enzalutamide would then be eligible to continue on enza together with an HDAC inhibition. And I realize that the HDAC inhibitors have been tested before in such type of context, but not in this specific patient population. And also, we believe we can lower the dose, limiting toxicity. Now, I just gave you two examples on context-dependent acquisition of specific drivers, or more specifically of transcriptional regulators, dependent on the types of intervention you give. And what we’re now doing is trying to identify specific rules of acquisition of these dependencies dependent on the types of therapies. So, what we do is… We re-analyzed all the K27 acetylation ChIP-seq data from all the patient samples we’ve been analyzing. So that’s over 300 individual patients, including three naive metastatic lesions. And for all of those, we performed ChIP-seq analyses for our favorite histone mark, the K27 acetylation. What we now aim to identify is then see if there are specific epigenetic signatures which are demarcating specific faces of the disease. For metastasis formation, castrate resistance, or neuroendocrine disease. And for those, we performed computational InSilico analyses, trying to identify genomic regions which are typically behaving the same way between all the patients we analyzed. And that gave us the following analyses, there we go, where we could identify three distinct classes of epigenetic changes. Those regions on the genome which are specifically gained in neuroendocrine differentiation, those which you lose in neuroendocrine differentiation, and also those which we specifically gain in metastasis formation. Again, doing the same InSilico analyses now gave us a concise list of every single transcription factor which was enriched for these specific phases of the disease, which we’re now trying to interrogate with functional perturbation assays. Trying to restore the original epigenetic state of that specific lesion but also test whether or not that would then lead to a disease state-specific therapeutic intervention targeting exactly that phase of the disease at that time for those specific cells. Lastly, and this couples back to Sylvan’s presentation, is that we’re trying to develop now a minimally invasive assay for identification of the epigenetics states. And for the trial I’ll just show you, what we’ve also collected is plasma samples from the very same patients. So together with the lab of Matt Freedman, we’re now trying to perform on the same patients where we have the ChIP-seq profiles on the mCRPC samples, also to perform that on the plasma samples. Trying to actually then to translate this all back again towards a new therapeutic discovery based on our findings. So, with that. I’d like to end and I’d like to acknowledge everyone in the team, everyone who contributes to our work and I’m happy to answer any questions. Thank you.  

Christina Jamieson, PhD [00:57:16] Hi, Chris Jamieson, UCSD. So, I’m just wondering how this amazing epigenomic landscape fits in with copy number variation. So, have you looked at, you know, how the epigenetic changes compare between AR that’s amplified versus not amplified?  

Wilbert Zwart, PhD [00:57:34] Yeah, that’s a very interesting point. So, at the end, the copper number does have an effect on the strength of the K27 acetylation signal, that is for sure. So, it is something that we definitely see as a block, if you will, that those regions are also then further occurring. We do normalize for that. So, you typically also sequence an input sample along, but you then sort of correct for that. In the mCRPC patients, we also included AR ChIP-seq. And for all the patients we analyzed, only two of those AR ChIPs actually worked, and those were the only patients where we had a confirmed AR copy number gain. So, I do think that it directly contributes to the success rate of these analyses, but we didn’t really see any clear distinction for the K27 acetylation signal between those patients who did or did not have the AR amplification. Thank you. 

Sushant Kachhap, PhD [00:58:32] I’m Sushant Kachhap from Johns Hopkins. My question is, do you know whether or have you investigated whether which histone acetyl transferase makes these marks, because you can also use an inhibitor against this. I’m saying this because there’s a recent publication regarding H2B N-terminal acetylation by p300, and inhibitors of that is used in castration resistance, which synergizes. So, I’m just wondering whether the same acetyl transfer is made.  

Wilbert Zwart, PhD [00:59:05] Yeah, so I think the issue is not necessarily the histone acetyl transferase, which has the effect. It’s more the transcription factors which are active in that specific context. So, one thing I didn’t show you is that for those 657 regions which were specifically gained in the therapy-resistant patients, we had at least 50,000 regions which were not changed. So, it’s a very, very distinct subset of regions which change in their behavior. It also comes down to therapeutically targeting any of these enzymes was probably not necessarily gonna have a major effect, at least when it comes to the therapeutic window, because normal physiology will surely be affected as well. So, it is really about the transcription factors which are then changing their behavior, thank you.  

Joshua Lang, MD, MS [00:59:54] Well, congratulations on your technical successes in performing these in the context of prospective clinical trials, which is, you know, really remarkable work. My question actually comes back to this almost epigenetically unstable group of patients which you’re identifying. Are you identifying any tissue site specificity, meaning are there different metastatic lesions or sites of disease that contribute to the instability that you’re seeing, and is that giving us some clue in the tumor microenvironment contribution to these lethal prostate cancers?  

Wilbert Zwart, PhD [01:00:23] Yeah, that’s an excellent question. So thus far, our observations on this are anecdotal. So, we looked into one patient, was a collaboration with Michael Haffner, where we had access to autopsy samples, where we compared different metastatic lesions, and we didn’t really see massive changes. Also, here for this trial, we had biopsies from a number of different regions, and also there, both supervised and unsupervised analyses, we didn’t really see clear distinction between them. I mean, it’s absolutely true that fibroblasts, macrophages, T cells, they also, of course, express AR, and they also have their distinct AR cistroms. But when we would start to normalize for that, it doesn’t seem to be a major driver in observations we have.  

Joshua Lang, MD, MS [01:01:10] Thank you.  

Wilbert Zwart, PhD [01:01:11] Yeah, thank you.  

Amina Zoubeidi, BSc, MSc, PhD [01:01:23] Thank you so much. It is now my great pleasure to introduce you, Dr. Michael Shen from Columbia University. This is very something, very announcement, very interesting that, you know, Dr. Micheal Shen gonna tell us about his recently accepted Nature paper on targeting the epigenome to reverse. This is gonna be a proof of principle that you can use and target the epigenome to reverse the lineage plasticity. Congratulations for your Nature paper, Michael.  

Michael Shen, PhD [01:01:56] Thank you very much, Amina. It’s been a real pleasure to organize this session with you. Today, I’d like to talk about recent work from my laboratory. We’re very interested in understanding lineage fidelity and development, as well as lineage plasticity and tumorigenesis. And of course, we’re all familiar with sort of a canonical form of lineage plasticity in prostate cancer. Oh, this is my disclosure. As we all know, under selective pressure from androgen receptor pathway inhibitors, we develop a spectrum of CRPC, multiple subtypes, as defined here by Ekta Khurana’s group, ranging from CRPC-AR to neuroendocrine prostate cancer, which I’ll refer to as CRPCNE. And then there are other subtypes that are a little bit less well-defined. The relationship between these subtypes is not really understood, but it is believed that this increasing linear plasticity is being driven by epigenomic reprogramming. And what I’m going to tell you is that there is a key role for a histone methyltransferase, known as NSD2, in maintaining the neuroendocrine state. And this work has been largely driven by a very talented postdoc in my lab, Jia Li, who is also a PCF Young Investigator. So, this work started in collaboration with Cory Abate-Shen’s group. Who had developed a genetically engineered mouse model that can result in neuroendocrine differentiation. So, what Jia did was to develop organoid lines from these mouse tumors. There are two examples shown here, parental tumor as well as organoids derived from them. The NPPO-1 line is interestingly very heterogeneous. It consists of a mixture of neuroendocrine cells as well non-neuroendocrine cells that express androgen receptor. The NPPO-2 line is much more homogeneous. It contains neuroendocrine cells that express chromogranin A and synaptophysin as well as FOXA2. In collaboration with Andrea Califano’s group, in particular Alessandro Vasciaveo, we performed single-cell analyzes of these organoid lines and found that collectively we can discern three distinct clusters of cells within these organoid lines corresponding to cluster 1 which is an AR positive cluster. And cluster three, which is neuroendocrine, and cluster two, which seems to have more of a transitional state. One of the first questions that we sought to address was whether we could actually observe neuroendocrine trans differentiation in organoid culture. To tackle this question, Jia sorted neuroendocrine and non-neuroendocrine cells from the NPPO-1 organoid line and marked the non-neuroendocrine cells with RFP. So, this is a form of lineage tracing that we looked at. And then if we culture the non-neuroendocrine and neuroendocrine cells separately. They’ll form non-neuroendocrine and neuroendocrine organoids, but if we combine these, we can now start to see RFP-marked neuroendocrine cells appearing in the co-cultured organoids. Now, to definitively establish this conclusion, we perform single-cell analyses, and here in the top row, we show control. These are the non-neuroendocrine cells alone, non-neuroendocrine organoid alone, and then in the co-culture, the key point to make here is that you can see that there are RFP marked cells. That are now in clusters two and three, providing evidence for this trans-differentiation at the single-cell level. Okay, so how do we then get to identifying epigenetic regulation? So, to do this, we collaborated with Chao Lu’s laboratory at Columbia, and we performed an immunofluorescence screen for differential expression of epigenetic marks, and emerging from this screen, we found something very interesting, which was we saw a higher-level expression of histone H3 lysine 36 dimethyl marks in the neuroendocrine cells versus the non-neuroendocrine cells. So H3K36 dimethylation is actually generated by a small family of histones methyltransferases, the NSDs, and in looking at various published data sets such as this one from Samir Zaidi and Charles Sawyers’ lab, we can see that NSD2 in particular is expressed at high levels in CRPC-NE. So, what then can we say about NSD2? In collaboration with Johan de Bono’s group, we looked whether NSD2 is associated with survival outcomes and indeed in two independent data sets, Johan’s Royal Marsden Hospital data set as well as the PCF Stand Up to Cancer data set, we can see that high levels of NSD2 expression are associated with poor survival. So if we actually look now at these neuroendocrine organoid lines and ask, what is the level of expression of NSD2, we can see that in four different neuroendocrine organoid lines, we have high levels of expression of NSD2 as opposed to non-neuroendocrine organoid lines, and we also looked at EZH2, which of course has been widely studied as a potential epigenetic regulator implicated in neuroendocrine differentiation. So, in these organoid lines, NSD2 and EZH2 are both expressed at high levels. And this is commonly a feature in prostate cancer, as well as other tumor types. But if you look more carefully at the marks themselves, you can see that in the neuroendocrine lines, H3K36 dimethyl marks are at high levels, whereas H3K27 trimethyl marks, which are the enzymatic product of EZH2 or PRC2, are relatively low, still high overall, but relatively low. In the non-neuroendocrine organoid lines. This is switched. So, this is reflective of a general antagonistic relationship between NS2 and EZH2 function. So further extending this, we could look at the genomic landscape of HVK36 dimethylation using cut and tag, again, in collaboration with Chao Lu’s laboratory. HVK36 dimethyl marks are usually found in fairly broad domains, and we can see that they’re associated with the enhancer and promoter proximal regions of various neuroendocrine genes and regulators, such as chromogranin A, FOXA2, ASCL1, et cetera. And there’s a converse relationship with H3K27 trimethyl marks, which are instead at high levels in the non-neuroendocrine organoid lines. So, what is the function of NSD2? So, if we now use CRISPR to target NSD2, what we can see is that we see a loss of neuroendocrine phenotypes, so. Here is NPPO-1NE, the neuroendocrine line, as an example. We lose synaptophysin and chromogranin A expression, but now, interestingly, we gain expression of androgen receptor, which is not expressed in the control organoids. So, Jia made the very interesting observation now that with the gain of AR, there is also now response to enzalutamide, as shown here. So, we can now generate a very nice dose response curve to enzalutamide. Of course, this is all done by CRISPR targeting, but in order to really push this story further, you need a small molecule inhibitor. And NSD2 was considered to be largely undruggable for many years, but over the past two or three years, that has really changed, and now there are several small molecule inhibitors available. So, we’ve synthesized one of these, and you can see here that administration of a small-molecule inhibitor of NSD1 can actually really push all the cells away from clusters two and three towards cluster one. And if we look at, this is a principal components analysis of H3K36 dimethyl marks across the genome was by cut and tag, what you can see is that this segregates the neuroendocrine and non-neuroendocrine organoid lines, regardless of genotype. So, the NSD2 inhibitor treated neuroendocrine organoids grouped together with the non-neuroendocrine phenotypes. And finally, very interestingly, if you now ask what happens to these organoid lines after NSD2 knockout or treatment, you see that there’s now a response, and this is using a human organoid line, now a response to androgen, so to DHT. So, this suggests there’s a restoration of canonical AR signaling, and this consistent with recent work from the Chinnaiyan and Asangani labs suggesting that NSD2 is a cofactor for AR that directs AR to non-canonical binding sites. So, we further characterized our NSD2 small molecule inhibitor. This is in collaboration with Or Gozani’s lab at Stanford. And we can show that it’s very specific for NSD2. It has strong selectivity against other histone methyl transferases, but there is some response of NSD1. So, in order to use the inhibitor in organoid experiments or in vivo, it’s necessary to actually pre-treat with the inhibitor and then treat in combination with enzalutamide. So, if you do that, you get these very nice responses in the mouse organoids to the combined treatment. Of course, what we’re really interested in is responses in human organoid lines. So together with Yu Chen and Misha Beltran, we looked at five different human CRPC organoid lines. And three of these are CRPC-NE. They have different RB mutational statuses, interestingly. And then we have a fourth line, WCM1262, was originally characterized as neuroendocrine, but it was subtyped as CRPC-WNT. And then finally, MSKPCa2 is a classical CRPC-AR line. Now, treatment with the NSD2 inhibitor abolishes H3K36 dimethyl marks in all of these lines. And interesting, there is sort of a reciprocal upregulation of H3K27 trimethyl marks. And what you can see with treatment of the inhibitor alone is that in WCM1262, we now see increase in AR expression. And in treatment of MSKPCa2, which already expresses AR, we see now the appearance of cleaved caspase-3 indicating apoptosis. So, I think we’ve already seen synergy finder analyses in Wilbert’s talk, but you can treat with different levels of NSD2 inhibitor as well as enzalutamide and generate the synergy finder plots. You see evidence of strong synergy between the NSD2 inhibitor and enzalutamide. This is captured by the Bliss score. A Bliss score of over 10 is usually indicative of synergy. Our Bliss scores range much higher than that. And interestingly, in contrast to this, If you look at treatment with an EZH2 inhibitor. This is the Pfizer compound, also known as Mevrometostat. We don’t see any evidence of synergy here in treatment of this CRPC-NE line, MSKPCa10, and this is entirely consistent with recently published work from Himisha Beltran’s group. So finally, we wanted to ask, well, is there synergy of the NSD2 inhibitors with Enzalutamide in vivo? So, in order to address this, Jia grafted these organoid lines into immunodeficient mice, allowed the tumors to grow somewhat, and what you can see here clearly is that the combined treatment actually not only suppresses tumor growth, it may even cause the tumors to shrink a bit. So, this is true in the MSKPCa10, and 14 lines, as well as WCM1262. The interesting case is MSKPCa2. Which is a CRPC-AR line. It responds to the NSD2 inhibitor all by itself, although it displays a better response in combination. There is a slight response to enzalutamide all by itself. So, if we look at sections of these graphs, in this case from MSKPCa10, you can see that the graphs treated in combination have essentially stopped proliferating and now are instead are starting to undergo apoptosis. What do we make of all of this, all right? So, I’d like to offer sort of a conceptual working model for what we think is going on. And this is of interest, of course, because as many of you know, Mevrometostat, the EZH2 inhibitor, is currently in phase three trials for treatment of CRPC in combination with ARPIs. And so, it appears to be effective thus far. So, the question is, what do think is really going on? So, we would hypothesize that during progression from adenocarcinoma to neuroendocrine states, there is a lot of epigenetic reprogramming going on. And so initially, you have a rise in the levels of EZH2, which is well-documented. And then later on, we see an increase in NSD2 activities. But remember, there is this antagonism between EZH2 and NSD2 activities. So, the increase in NSD2 levels actually causes a decrease in overall levels of H3K27 trimethyl whereas H3K36 dimethylation increases. So H3K36 dimethylation is generally considered to be an activating mark, H3K27 trimethyl generally a repressing mark. And so, we think that there’s sort of a transitional stage here that might be associated with increased stemness. Some people might say higher plasticity potential. And this may be associated with interesting chromatin states that I think will require further investigation. So, I’d like to stop there, just summarize, tell you that we have organoid lines that recapitulate the phenotypic heterogeneity of human CRPC, including CRPC-NE. We can observe neuroendocrine trans differentiation in organoid culture. We see this interesting relationship of CRPC-NE with these histone marks. And if we knock out or inhibit NSD2, we can revert neuroendocrine differentiation, restore canonical AR signaling and enzalutamide response, and therefore NSD2 inhibition may represent a new therapeutic strategy for treatment of CRPC-NE. Along those lines, I believe that K36 Therapeutics is about to initiate a phase one dose escalation trial for their NSD2 inhibitor in combination with darolutamide in CRPC, so it will be very exciting to see the results of that trial. So, this work was only possible with a great number of collaborators, but it was led by Jia Li in my lab, and I’d also like to thank, in particular, Alessandro Vasciaveo in the Califano group, and Chao Lu at Columbia University. And we’ve been funded by many sources, but in particular of course, the Prostate Cancer Foundation. Thank you very much for your attention. I’ll be happy to take questions. Thanks.  

Unknown [01:17:31] Michael, brilliant as usual and congratulations on the paper. I can’t wait to read every detail, but the question is for K36, right? So, when you go into the clinic, do you need to select any patients? And if you do, how are you going to, how do you plan on doing that?  

Michael Shen, PhD [01:17:46] Well, I don’t know what they’re going to do, but this, I know that they were involved in extensive discussions about patient selection, so I’m not really sure what’s going to happen in their trial, but it’s possible that it’s just going to be all comers. Now, you note from the data I showed, it’s possibly that the NSD2 inhibitor would have a broader activity than just on CRPC-NE, so, I think, you know, all comers might be a wise choice to avoid this issue entirely.  

Daniel Spratt, MD [01:18:15] So actually, let me ask a follow-up question to that. So, I guess with that regard, though, because it’s not just NEPC, you know, you start with a genetic mouse model, do you know what’s upregulating it? Is it loss of p53 or loss of PTEN or something?  

Michael Shen, PhD [01:18:30] So, okay, let me clarify. That’s a great question. I didn’t really say what the mouse model was. So, as you know, Cory’s lab has generated a wide series of mouse models over the year. This one is called NPP53. It has induced loss of p53 and PTEN driven by an inducible NKX-3.1 driver. So, there is also heterozygous loss of NKX-3.1. This mouse model is distinct among all of the mouse models in Cory’s collection. And it has a very broad heterogeneity, perhaps recapitulates much of the heterogeneities of human CRPC, including neuroendocrine differentiation, but it’s not limited to that. So, it’s a particularly interesting mouse model. Now, I didn’t really go into it, but the organoid lines that we developed in the neuroendocrine cells, there is loss of RB activity, but RB is not actually mutated.  

Unknown [01:19:28] At the beginning of your talk where you show any versus non-any, one of the marks was H3K36 because you started. I’m wondering what about H3K27 trimethylation, do they not show significant difference? And my second question is, even among any, when we say it’s a broad group, there are scenarios of AR positive, any positive. So. Now, among those any, is there a gradient of H3K36 dimethylation or the distribution across the genome that kind of gives you an idea of what other different phenotypes fall within that?  

Michael Shen, PhD [01:20:04] Okay, great question. You’re really getting into the nitty-gritty details here, which I love. So H3K36 trimethylation is something that we looked at. It doesn’t seem to be particularly relevant to anything I talked about, but the gory details are mostly present in the paper. And the other question concerned, you know, whether there was AR expression or not. So, a couple of the organoid lines that Jia developed do express AR. They sort of have this amphicrine phenotype, and they’re a little bit different from the AR negative line. So, in those cases, NSD2 knockout by itself was not sufficient to revert, it didn’t have that much effect. So there, we actually used an oncohistone, which is a dominant negative H3K36, and that was able to abolish the neuroendocrine phenotypes.  

Varadha Balaji, PhD [01:21:05] Balaji from Misha’s lab. Wonderful work, congratulations. And actually, you partially answered my question, but can you talk a little bit about K36 dimethylation versus trimethylation and why K36 dimethylation is quite specific to neuroendocrine phenotype and any role for de novo DNA methylation?  

Michael Shen, PhD [01:21:28] Wow, those are great questions. I’m not sure I can really answer them. You know, H3K36 trimethylation is governed by a different enzyme, set D2, as you know. And as I said, we don’t seem to see much of a role for that here, to the extent that we’ve looked. But I think that if I had to speculate, I think that H3K36 dimethylation is particularly important here because of the antagonistic relationship to H3K27 trimethylation. There seems to be something very interesting and special about that that is driving plasticity. That’s a speculation, but that’s where I’ll put my money.  

David Quigley, PhD [01:22:12] Hi, David Quigley, UCSF. I love this work. I’m really interested in your thoughts about what are we talking about when we say neuroendocrine disease or neuroendocrine phenotype because there are, especially in the model literature but also in the clinical genomics literature, a lot of different ways of saying that something has a neuroendocrine phenotype or has neuroendocrine differentiation. We could be talking about protein markers, expression, chromatin accessibility changes or morphology. So, in your model, what is the marker or the constellation of markers that you think are the most valuable and are those generally the markers that we should be using as a field or is there sort of a not quite a consensus there?  

Michael Shen, PhD [01:22:59] Okay, so that’s a great question, David. I mean, obviously the definition of neuroendocrine is central to all of this. We tried to sidestep this issue as much as possible. We’re not relying upon histopathological definitions, although I should mention that everything that we’ve looked at in the paper has been evaluated by Mark Rubin at the histopathology level. So, we have phenotypes ranging from small-cell phenotypes to something more like large =-cell. We’re primarily relying on marker expression. To define neuroendocrine, but that’s also supported by the single-cell analyzes, you know, so at the RNA and also at the protein inference level where we see enrichment of neuroendocrine signatures. So, we try to have reinforcing levels of evidence for neuroendocrine states.  

Irfan Asangani, PhD [01:23:49] Hi, Michael, congratulations. It’s fantastic to see the paper isn’t going to be in nature. And the question I have was, you see the MSKPCa3 model, which was quite exclusively sensitive to NSD2 inhibitor, right? Which is the AR-driven model. So, you think this, eventually if you are going to start a clinical trial, those has to be a CRPC and the refractory patients, not the neuroendocrine positive patients. What do you think about?  

Michael Shen, PhD [01:24:17] So, I can’t really comment on the clinical trial, but I think it’s the idea that an NSD2 inhibitor might be effective in CRPC-AR is certainly supported by the work from the Chinnaiyan and Asangani groups.  

Irfan Asangani, PhD [01:24:31] That’s me.  

Michael Shen, PhD [01:24:32] Oh, that’s you, Irfan. Oh, yeah. I can barely see you in the lights, sorry Irfan. So, you know, your work, God, this is embarrassing. Your work is extremely consistent with what we found. I mean, the problem is that we don’t have a lot of models to work with here. So, further analyses are going to require, I’m not sure.  

Irfan Asangani, PhD [01:25:03] Yes, that’s why I was thinking, did you try any of these PDX models which is available for you?  

Michael Shen, PhD [01:25:08] Not yet.  

Unknown [01:25:11] This is really exciting, thank you for the presentation. To turn AR back on, so neuroendocrine, the gene is heavily methylated, the chromatin is closed, so presumably NSD1 knockdown is allowing it to become demethylated. I’m wondering, is there any connection between K36 methylation and DNA methyltransferase maintenance and could it be going through K27 and then K9 methylation would be interesting if you looked at the effects on K9 methylation.  

Michael Shen, PhD [01:25:46] Okay, so here I have to confess that my understanding of epigenetics is not as solid as it should be, but certainly there is a relationship between H3K36 dimethylation and methyltransferase activity, this is work from Chao Lu’s group. I should say that in the non-neuroendocrine models that we look at, that it’s not necessarily the case that AR is in a closed chromatin configuration. So, there is something about the non-neuroendocrine states that are most adjacent, let’s say to the neuroendocrine states in which perhaps AR is more poised for re-expression.  

Unknown [01:26:33] Fantastic work. You shared a little bit about how maybe the heterogeneity and the genetics of the model are very important to find these kinds of results. Historically, it’s also been that we’ve seen this happening in patients and then in mouse models and now in organoids. But it’s hard to get it to happen in cell lines and et cetera. With this project, have you learned a little more about those nuances?  

Michael Shen, PhD [01:26:59] Wow, that’s almost metaphysical. I mean, I think that organoid models have tremendous utility here in that they seem to recapitulate physiological states perhaps a little bit better in many cases, although not necessarily always. And certainly, one important attribute of organoid models is they can capture a lot of the heterogeneity and the spectrum of heterogeneity that’s observed in CRPC. And I think it’s apparent to all of us that CRPC is an extremely heterogeneous disease. So, you know, the only shortcoming is that we don’t have that many organoid models. And the ones that exist were developed after great amounts of effort by folks like Yu Chen. So, we’re very grateful to them, but we need more. Thank you very much.  

Amina Zoubeidi, BSc, MSc, PhD [01:27:54] Thank you. Thank you so much for all, for my co-chair and also the speaker and all the speakers in this session but also a big thank for all the engaging audience. So, we had fantastic session with a lot of questions. Thank you very much.  

Back to Video Hub

Back to Video Hub