Read the Full Video Transcript

Andrea Miyahira: Hi, everyone. Thanks for joining us today. I'm Andrea Miyahira at the Prostate Cancer Foundation. Today, I'm joined by Dr. David Quigley, an assistant professor at UCSF, to discuss his group's recent paper, The Genomic and Epigenomic Landscape of Double-Negative Metastatic Prostate Cancer, that was published in Cancer Research. Thanks for joining me today, David.

David Quigley: Pleasure to be with you. Thank you for joining us, and it's great to be here with you and UroToday. So the paper is called The Genomic and Epigenomic Landscape of Double-Negative Metastatic Prostate Cancer, and I am David Quigley, I'm at UCSF. So our group has focused on prostate cancer for a number of years now, and we're really interested in how the disease adapts to targeted therapy. Before this disease is treated, it has relatively few mutations and generally is sensitive to androgen therapies, and these therapies are quite effective for many men when they're used for many years. However, unfortunately, they will inevitably eventually result in the selection of cells that are resistant for some reason to these therapies. And so the big question our lab is interested in is, how does this resistance arise, what are the mechanisms that are by it, and then how can we do something about it? And so these mechanisms are various, and we have taken an approach of looking primarily at patient samples that are donated by men who have prostate cancer.

So for a number of years there was a sort of understanding that prostate adenocarcinoma that becomes resistant to androgen receptor signaling inhibitors, or ARSI, can take one of two paths. Either it sort of keeps its adenocarcinoma nature, and that's the wing on the top there, and so those tumors often have somatic alterations at the androgen receptor itself, gains of copy of the AR or mutations, and they continue to evolve under therapy but remain fundamentally looking like the adenocarcinoma that they were before they were treated. And you can see on the right-hand side, I'm including some stains that were published by Colm Morrissey and Pete Nelson's groups at the Fred Hutch. And I'll use their stains again on the next slide, so thank you to them.

However, the other track that it can take is to do this really interesting thing where it kind of develops into a so-called neuroendocrine or small cell phenotype. And these tumors have a very different phenotype from the original. They are androgen indifferent, they rarely have AR gains, they characteristically have P53 and RB mutations that are inactivating, and they tend to have worse outcomes, and clinically that's how they're really recognized. And you can see on the figure on the right here, where this is an example of a NEPC tumor that doesn't express the androgen receptor but expresses this other marker, synaptophysin, which is associated with the neural lineage.

However, in the last couple of years, a more complicated model has emerged for this disease, and you can now see there are four different groups on the right-hand side here. We've got the prostate adenocarcinoma, but there are a few other possibilities here. Basically, they come down to whether you have either of those two markers, AR or synaptophysin, both, or neither. And so the ones that are new here, this so-called amphicrine group, which is double-positive for both neuroendocrine and AR lineage, and this double-negative group on the bottom that shows neither AR nor neuroendocrine phenotype. And I'm referencing here a paper that I encourage you to check out in the Journal of Clinical Investigation by Pete and Colm's groups from which these stains were taken, and I think they really helped the field look in this new orientation. And this was about 2019.

So the questions we were interested in, in investigating for our study is, when did these subtypes emerge? Are some transitional or not? In other words, are they the final state for these tumors or are we catching them between states? Are these subtypes linked to patient outcomes? And what are the molecular drivers of these subtypes? And so the people who I'm highlighting here in this slide, the person on the left here is Arian Lundberg, a postdoc in the lab who I co-supervised with Felix Feng. He did this work, and so I'm pleased to present it on his behalf. And then on to the right-hand side, my colleagues and mentors, Felix Feng and Eric Small, who are co-senior authors on this work and without whom this work would not have been possible. So I want to make sure to credit them as well as being really important to this overall project. It's not just my lab alone.

And in fact, we really want to emphasize that this work is coming out of a larger group, called the West Coast Dream Team, that was funded by the Prostate Cancer Foundation several years back that's borne a lot of fruit. And one of the things we've done with that is to generate this large cohort of men with metastatic disease and then look, using various genomic and epigenomic assays, to try to derive tumor subtyping information, biomarkers, and understanding about disease etiology. And so the first and the third are really the kind of focus of today's talk.

So briefly I'll summarize some of the work that came out of this paper. The first thing was we confirmed that there are, yep, indeed five transcriptional subtypes. We can see them both in the West Coast Dream Team data, that's a heat map of gene expression profiles shown on the top, as well as data published by Misha Beltran and her colleagues previously. I have a star right by that AR-low group there because the AR-low group is a little bit less clearly defined transcriptomically and it's a little bit more dependent on exactly which other tumors you're comparing to to say which tumors have that AR-low phenotype. So in the rest of this analysis we really focus on the other four categories, the double-negative, the double-positive, the adenocarcinoma, which kind of stayed the same, and the neuroendocrine, or NEPC, tumors.

So it had been known previously, like I mentioned, that patients who have this neuroendocrine phenotype tend to have worse outcomes. We stratified our patients in our study by the subtyping that was done transcriptomically and identified the fact that those double-negative patients, which are the ones illustrated on this Kaplan-Meier survival plot in blue, had the worst outcome of all of those subgroups. So that wasn't super surprising, but I don't think we'd really established that before, that there might be a sort of even worse category compared to the neuroendocrine patients as well.

We also did a couple of specific studies. We were looking at genes that were frequently amplified in different subtypes or frequently lost. And we looked in particular at this gene called CHD7. This is a chromosome remodeling gene. And if you're familiar with prostate literature, you might be familiar with CHD1, also in the same family, which is frequently inactivated in prostate tumors. So CHD7, we first looked at it because it's amplified in this double negative phenotype, but in looking more closely, we realized actually it's actually expressed at the highest levels in neuroendocrine prostate tumors.

We know from prior work in a lot of other contexts that CHD7 is linked to the neuro lineage, so it's important for neural crest development, and it is kind of a classic one of these transcription factors borrowed from neuroendocrine lineages that's active now in this subtype. And in work that you can read more about in the paper, we were able to identify a candidate enhancer within the gene itself that is differentially methylated, sort of more exposed, in the neuroendocrine subtype. And when it is more exposed, that's correlated with elevated expression of CHD7. So it's this interesting preliminary evidence that this gene actually may be activated by a differential methylation, and that may be relevant specifically in neuroendocrine disease.

More generally, we wanted to ask which transcription factors drive all these subtypes. And so this is kind of the heart of the study. We were really looking at a combined methylation and transcription assay analysis. And the idea is, first, you do differential methylation by subtype to find different parts of the genome where there's low levels of methylation. And that low level of methylation, for those of you who are less familiar with this, might be associated with activated expression if the methylation is impacting either some form of regulatory DNA, like a gene promoter. And then we asked which transcription factor binding motifs are enriched in those regions, and that can kind of give us a handle on, well, what are the candidate transcription factors that might be driving these subtypes?

So without going in a lot of detail, like I said, it's all in the paper, we were able to identify a lot of transcription factors that are probably enriched in each of these subtypes. And as a positive control, looking at adenocarcinoma, we knew that androgen receptor and FOXA1 and other genes that people are familiar with are likely to be really important, and in fact, those were the top hits in the AR-positive genes. The neuroendocrine samples came up with transcription factors like ASCL1, which has been extensively studied by Amina Zoubeidi and others who are really interested in these topics, but we kind of decided to focus on the double-negative tumors, the ones on the right-hand side there, because those are the ones we kind of knew the least about. And since they are not neuroendocrine and they're not adenocarcinoma, it's a little, "Well, what are they?" That was the question we were trying to get at.

So we focused in on this gene, KLF5, and nominated it as a driver of this disease. It has high expression in double-negative prostate cancer, it's actually low in primary disease, and this is something that had been observed previously by Scott Dehm and his colleagues in a paper in Nature Communications a year or two back, which I encourage you to check out if you're interested in this topic. And we found that KLF5 expression was inversely correlated with that of RB, but only in double-negative prostate cancer. That's what these plots here are showing on the bottom. That inverse correlation is only present in the double-negatives and it's actually linked to RB1 copy loss, so the tumors that have lost RB1 copies are the ones that have the highest expression of KLF5. And there's more details about this in the paper, but this is the kind of analysis we can do with these big genomic unbiased studies.

So basically what we found here is that double-negative patients have the worst outcomes, that this unbiased epigenetic analysis can identify subtype-specific tumor drivers, and we were going to need to do experimental work to establish if and what CHD7's role is in NEPC because that's a very reasonable question, but... These studies don't answer that question, but they open up something for us to look at. And more broadly, maybe something we can discuss later on is, are these sort of double states, like double-positive or double-negative, are they transitional or an end stage? What do they mean? And so we're really interested in engaging in longitudinal and single cell studies of heterogeneity that might help us to unpack those questions.

So I want to thank Felix and Eric for being really important parts of this study, the whole West Coast Dream Team, and the PCF for funding this work in a really important way, and Arian Lundberg, who did the work. So, thanks. And I'm happy to discuss all this work with you guys. Looking forward to that.

Andrea Miyahira: Thank you, David, for sharing that with us. So what are the major differences between AR-negative, NE-positive, and AR-negative, NE-negative tumors?

David Quigley: So there's a couple. The first is sort of the obvious one, that one of them has this neuroendocrine phenotype and the other doesn't, but we don't really know why that is. Several investigative groups have been trying to dig into this, but the neuroendocrine status itself is the fundamental thing. They also have a different genomic profile in terms of which alterations are present. So the neuroendocrine tumors tend to have a loss of RB and P53, whereas we observed in our cohort that the double-negative tumors had the most frequent loss of P10, which is a gene that's very frequently inactivated in prostate cancer overall.

The other thing that we really dug into is that the predicted transcription of factor activity profiles are really quite different. And this is linked hand in hand with the fact that the neuroendocrine, NEPC, tumors have this neuroendocrine phenotype, but really it's not the case that the double-negatives look sort of just like the adenocarcinoma but they lack AR. They really have a different profile, and so we're trying to figure out, does this represent some kind of other lineage or is it sort of a transition between two states?

Andrea Miyahira: Okay, thanks. Did you identify any therapeutic weaknesses in the AR-negative, NE-negative tumors? I don't know if the KLF family is a candidate or not. And if so, what are your next steps for translation?

David Quigley: So that's a great question. So it wasn't the first focus of this study in that we really tried to understand the wiring diagram. And the focusing on transcription factors other than the hormone responsive ones, like AR or estrogen receptor or glucocorticoid receptor, means that immediately targeting those individual proteins is challenging, because most of the time targeting a transcription factor directly is hard. But our next steps are really to understand, so where do these proteins bind? What do they activate? What does the tumor get out of them being active? And does that activity produce any dependencies that we could then potentially exploit? And so the idea then is to understand, is there a subtype-specific dependency that is revealed by this transcriptional factor activity?

Andrea Miyahira: So thank you so much. We've learned so much from these sort of bulk tumor studies, but now we're starting to get a lot more single cell data coming out, I know with your group also, and also advances in our ability to understand this data with your computational advancements. So as we go forward, how do you see our understanding of inter- and intrapatient heterogeneity progressing, particularly in clinical utility?

David Quigley: So I think the question is only going to get more and more complicated because we know that this disease, like probably all tumor types, continues to evolve as patients are exposed to therapy. And I think the understanding is beginning to shift from the idea that all the tumors in a person's body are kind of basically the same to the fact that, as they get treated, and as we do a better and better job of developing therapies, and patients live longer on those therapies, those tumors are going to continue to kind of differentiate.

And there's two levels of that kind of heterogeneity. One is differences between tumors in different parts of a person's body. Maybe the tumor that's in somebody's lymph node has a different phenotype from the tumor that's in somebody's spine, and maybe we're going to need to take approaches that can be complementary to each other and hit both of those tumors if they have differing vulnerabilities. The other is intratumoral heterogeneity, so heterogeneity within one individual tumor. Oftentimes, when we've looked sort of broadly using relatively low throughput methods, metastatic lesions tend to look relatively uniform, but work from a Michael Hoffman and other folks who are looking at this, pathologists who have really paid attention to this kind of material, have understood that actually there is considerable potential for differences within the same tumor. So you can have a neuroendocrine and an adenocarcinoma lineage within the same tumor.

And so those kinds of differences are also things we're going to have to be grappling with. So I think there's a lot of scope for understanding with spatial technology, using both spatial RNA sequencing but also spatial proteomics, which is a really interesting emerging area, where people can now multiplex eight or even 30 or 40 different protein markers in the same slide and do that with archival tissue, like FFPE even, and be able to understand, what is the diversity of different phenotypes within one tumor?

And, as you say, for therapeutic application... Because as a genomic scientist, part of me is always just interested in knowing what's going on, but as a translational scientist, I want to know, well, so what are we going to do? What use is this information? And so I think both in understanding, are we driving tumors in a certain direction with our therapy and is there a way to stop that, but also there's a lot of emerging therapeutic effort going into targeting surface markers that are unique to a given type of tumor. DLL3, among many other targets, are really emerging as potential ways to target tumor cells without having to know exactly what those proteins are doing. If the tumor just expresses that protein, that then you can then attach a toxic payload to something that will stick to that protein and cause tumor cell death.

So understanding the diversity of phenotypes within and between tumors I think is really going to be an important part of our next sort of wave of studies now that we've done a lot of these bulk tumor sequencing studies and I think learned quite a lot, but there's going to be a lot that will be opaque until we can look at that level of detail.

Andrea Miyahira: Well, fantastic. Thank you so much for all your work and thanks for joining me today, Dr. Quigley.

David Quigley: It's a real pleasure, and I want to thank you for having me on. It's great to be on this production. And again, like I said, I want to thank the University of California, the Prostate Cancer Foundation, and all of my colleagues with whom I stand together in trying to do this work.