Read the Full Video Transcript

Andrea Miyahira: Hi everyone. I'm Andrea Miyahira here at The Prostate Cancer Foundation. With me is Dr. Himisha Beltran, an associate professor at Dana-Farber Cancer Institute in Harvard Medical School, and Dr. Francesca Demichelis, a professor at the University of Trento. They'll discuss their recent collaborative paper, "Noninvasive Detection of Neuroendocrine Prostate Cancer through Targeted Cell-Free DNA Methylation," published in Cancer Discovery.

Dr. Beltran and Dr. Demichelis, thanks for joining me today.

Himisha Beltran: Thanks so much for having us.

Francesca Demichelis: Thank you, Andrea, for having me too.

Himisha Beltran: Thank you so much, Andrea. We are really thrilled to be here today to discuss our recent paper, which was truly a team effort, co-led by myself and Francesca Demichelis from the University of Trento and our groups, but with a number of collaborators from around the world with institutions represented here as co-authors. We'd really like to thank The Prostate Cancer Foundation, which was truly a catalyst for this work. I'd also like to acknowledge Gian Marco, who was the first author of this study, a PhD student in Francesca's lab.

So just a little background. Small cell neuroendocrine prostate cancer is an aggressive histologic variant of prostate cancer. It rarely arises de novo, but can develop in patients in later stages of prostate cancer progression as a mechanism of resistance and in its pure form can look and act like small cell lung cancer. The NCCN guidelines do recommend doing a biopsy when neuroendocrine prostate cancer is suspected as there are clinical implications to treat this patient like small cell carcinoma with platinum-based chemotherapy regimens.

But it's challenging as biopsies, first of all, are invasive and there can also be a lot of discrepancy or heterogeneity even within neuroendocrine prostate cancer within individual lesions or across metastasis in individual patients. And we know that there are some tumors that are truly AR-driven luminal adenocarcinomas, and those that are pure small cell carcinomas, but in between, as this is a process, we can often see heterogeneous or mixed features. So this was really the motivation for us to try to develop better ways to identify patients developing lineage plasticity and transforming to this neuroendocrine variant.

One of the most robust molecular features in neuroendocrine prostate cancer are its epigenetic features. It can be distinguished from prostate adenocarcinoma through DNA methylation in particular, and we've seen this now in multiple clinical cohorts and in model systems. And in prior work, we demonstrated the feasibility of being able to detect these DNA methylation changes in the plasma using cell-free DNA with good concordance between tissue biopsies and matched plasma samples, raising the exciting possibility that we may be able to use a blood test to identify patients developing lineage plasticity and neuroendocrine prostate cancer.

And this is truly the motivation for the current study, which is focused on developing a targeted methylation panel called NEMO, Neuroendocrine Monitoring, where we recognize that doing whole genome bisulfite sequencing, as we had done in our prior study, was not necessarily feasible or scalable or efficient or cost-effective when we think about the clinical cohorts and the patients that we see that we might want to test. So we focused on developing this targeted panel that's designed to distinguish cancer from white blood cells based on methylation patterns in the circulation and also distinguish adenocarcinoma from neuroendocrine prostate cancer.

We tested the panel in multiple model systems and patient cohorts, and there are a number of downstream applications embedded into the panel design, including inferring tumor content, which is really important when thinking about any cell-free DNA assay. There is a phenotypic evidence score, which is along the spectrum of adeno to neuroendocrine, and then the ability to infer transcriptional state using methylation patterns.

So the panel was designed by looking at multiple cohorts, including tissue cohorts and normal tissues. We took both an unbiased and a knowledge-informed approach towards developing this targeted panel that covered about 150 kb.

As mentioned, there were two specific modules within the NEMO platform, including tumor content estimation, which was based on looking at differential methylated regions that distinguish cancer from white blood cells, and then the phenotypic score, which is specifically geared towards the adeno to neuroendocrine. And again, this is not a yes or no, it's really a continuum to try to capture the heterogeneity that we see.

We applied the NEMO methylation panel to a methylation atlas of 39 cell types from healthy tissue samples, over 200 healthy tissue samples. And what we can see here is that there is a segregation between white blood cells and other tissues, which we expected and which is reassuring. And then, there was another segregation that distinguished endocrine and neuro lineage from prostate epithelial tissues, which I think really speaks to the biology of what we're picking up with these methylation calls speaking to the genes and pathways that we often see dysregulated in the neuroendocrine prostate cancer tissues.

Here on the left, you can see just a snapshot of certain areas of the genome where you can see differential methylation in cancer versus white blood cells and neuroendocrine prostate cancer versus adenocarcinoma, just to show the robustness of the calls. And on the right, you can see different clinical cohorts where we applied the NEMO cell-free DNA assay, including two CRPC cohorts where we didn't see a signal. These were sort of standard CRPC patient cohorts. But when we started to go to the more aggressive variant and the biopsy-confirmed neuroendocrine cases, you can see a higher phenotypic evidence score, again speaking to the intention of the panel.

Importantly, tumor content was independent from neuroendocrine fraction or phenotypic score, which is really important, meaning you could have a high tumor content and low score and vice versa, and it's important as we start thinking about applying these also to different tumor content situations and looking at the fraction that's not dependent only on tumor burden itself.

We applied the NEMO platform to different preclinical models, including PDX models developed at the University of Washington, where we see that the phenotypic score could distinguish the neuroendocrine prostate cancers from the adenos, but also we saw that there are other variants that were well-characterized within these models, including double-negative and amphicrine tumors that seem to show separation, including the double-negative tumors having a higher phenotypic evidence score, similar to neuroendocrine prostate cancer. And on the right, we also applied the score to organoid models that were recently described by Weill Cornell and Memorial Sloan Kettering that had specific transcriptional phenotype subtyping, as described in their paper. And we saw that the phenotypic score could distinguish the neuroendocrine from the AR-driven organoids, but the Wnt in the stem cell-like subclasses that they described had intermediary scores. This suggests that perhaps there is a continuum or a spectrum across other phenotypic variants that is picked up by DNA methylation patterns.

We looked at baseline plasma samples from patients enrolled on a phase 2 clinical trial of the aurora kinase inhibitor alisertib. This was a trial that was designed for patients that had either histology-confirmed neuroendocrine prostate cancer or aggressive clinical features defined by neuroendocrine marker expression or new liver metastasis without PSA progression. So not a typical CRPC population, but really enriched for what we think might have more of a neuroendocrine-like clinical spectrum. All patients in this trial had a baseline metastatic biopsy to confirm histology. And what we found is that the tumor content estimation by DNA methylation was not that different between the aggressive variant and the neuroendocrine-confirmed pathology cases, and tumor content was prognostic in this trial, similar to what's been seen in other contexts with DNA-based tumor content and tumor fraction estimation from ctDNA.

On the right, you can see the performance of the phenotypic score in the trial where you can see a high phenotypic score in those neuroendocrine cases. But in the aggressive adenocarcinoma cases, we saw this bimodal distribution with some cases having actually a high score, which could be due to heterogeneity of the tumor, could be due to pathologic differences, or even molecular similarities between aggressive variant adeno and neuroendocrine prostate cancer.

We also collaborated with Ana Aparicio at MD Anderson to look at patients enrolled on a phase 2 trial of carboplatin-docetaxel for aggressive variant or anaplastic prostate cancer. This trial enrolled patients based on aggressive clinical features shown at the bottom right for the eligibility, and it also included patients that had biopsy-confirmed small cell neuroendocrine prostate cancer. Again, tumor content was prognostic in this trial. When you look on the right, the phenotypic score was higher in the patients with confirmed small cell neuroendocrine prostate cancer, and again, we saw this bimodal distribution in the aggressive variant prostate cancer. These patients did not have a biopsy to confirm histology, but these high cases could represent either undiagnosed neuroendocrine prostate cancer or again, some epigenetic similarities that may exist across these variants.

Overall, when we look across all the cell-free DNA samples that we analyzed in this study, there was an iAUC of 0.93 for the phenotypic score in identifying histology-confirmed neuroendocrine prostate cancer, and this was even higher in higher tumor content plasma samples.

So we were also interested in understanding how we can use DNA methylation to understand biologic features and therapeutic targets beyond its role as a biomarker. For instance, we found hypermethylation of an EZH2 intronic region that correlated with EZH2 over-expression. Either gene expression or methylation of this region was prognostic in the West Coast Dream Team suggesting that perhaps we could use cell-free DNA methylation of EZH2 as a proxy for gene expression. And we know there are a number of clinical trials currently ongoing exploring EZH2 inhibitors for patients with prostate cancer.

We also explored the potential in applying NEMO to other clinical contexts. For instance, when we look at DNA methylation patterns in lung adenocarcinoma versus small cell lung cancer, we can see a segregation based on phenotypic score, also pointing to potential similarities between neuroendocrine prostate cancer and other small cell carcinomas.

So just to summarize, we showed how the neuroendocrine monitoring or NEMO platform could be a robust and scalable cell-free DNA platform to quantify tumor fraction and provide a phenotypic evidence score. This score can distinguish neuroendocrine prostate cancer from castration-resistant adenocarcinoma and could potentially identify other subtypes along the spectrum of lineage plasticity. I think there are a number of potential downstream clinical applications of NEMO that require further validation. I could envision a blood test like this if it could provide evidence that a patient might have undiagnosed neuroendocrine prostate cancer in the appropriate clinical context, which could help prompt a biopsy to look for neuroendocrine prostate cancer or maybe even in the future, replace the need for a biopsy. There may be additional prognostic value of the assay or it may be useful for disease monitoring, especially in low PSA situations.

Ultimately, we hope that NEMO or a test like this could help select patients that are most likely to benefit from traditional CRPC therapies and identify other patients that may need alternative approaches such as neuroendocrine therapies or clinical trials. We're really excited to understand how dynamic methylation changes occur in patients, and this is something that we're working on now to be able to understand how and when these occur in the context of treatment resistance as this could potentially identify patients for early intervention using strategies that are being developed to target lineage plasticity.

Thank you to the patients and their families for participating in this study, to all of our collaborators, and our funding sources.

Andrea Miyahira: Well, thank you so much for sharing this study with us today. So a large fraction of patients with NEPC also have adenocarcinoma. What fraction of the tumor must be NEPC for the NEMO test to accurately detect it and measure the degree of heterogeneity?

Francesca Demichelis: Maybe I can take this question, Himisha and Andrea. So as we know, in reality, in the circulation it is very hard to really say what is the lower limit because the cell-free DNA is not always one hundred percent tumor, right? Whereas when we tested that with a control experiment with in-vitro dilution by mixing adenos and neuroendocrine, what we saw is that as little as 5% of neuroendocrine fraction can be detected. So in principle, we can go really, really low. Now in practice, there are two components, one that I already mentioned, the fraction of the tumor in the circulation is a limiting factor and that's why we try to develop this phenotype score to be bound by the tumor content fraction. And the second component is that quite often, as you said, there is heterogeneity and it's hard to distinguish those two things, right?

So the framework that we developed should allow for that. And what we were able to test with a real sample is that we can get the spectrum at the same time. When we see a fraction of let's say a P score of 0.3, we cannot be fully sure if it is an admixture or it is a phenotype that it makes from the same metastasis or if it is a phenotype within the transition.

Andrea Miyahira: Thank you for that. So there has been some debate surrounding how exactly to define NEPC. How did you define NEPC for the development of this assay?

Himisha Beltran: I can take this one. You're right, I think that there's a lot of debate as to how to define neuroendocrine, whether you use morphology and what's the role of IHC markers. And for this study and a lot of our prior studies, we've used morphology-based definitions. And I would say that's probably the strictest definition. And we've had central pathology review for our cases, for our trials. In practice, there can be variability across pathologists as far as the morphology, IHC is commonly done. And I think that the quantification, getting to the first point of what percentage is neuroendocrine, especially in small biopsies, is really not necessarily reproducible and also the significance is really not that clear.

So I think what we try to do here is to take our strictest definitions and then see how a molecular definition or an epigenetic, the DNA methylation profile, can identify the spectrum. And what was interesting about, I think, the PDX models, for instance, is that the double-negative prostate cancer had a higher phenotypic score and those didn't have classical neuroendocrine markers. Of course, ultimately we need to show that this test improves our treatment selection, but I think it's pointing to biologic similarities that extend beyond morphology and IHC because those have been so challenging. So at the moment, I think it can complement, but I think that the way it was designed was really to try to understand the ends of the spectrum.

Andrea Miyahira: Okay, thank you. And how do you envision NEMO being used clinically, and what are your next steps towards translating NEMO into a clinical test for patients?

Himisha Beltran: Yeah, I think ultimately I would hope that as with an assay that we use in the clinic, that it can add to and help us sub-stratify patients and help us figure out the best treatment and ultimately improve clinical outcomes. And so there's a lot of steps that have to happen between now and then. And if it could even, I tried to outline some of them in the presentation as to possible clinical utilities and maybe it's really about just helping us figure out when to do a biopsy, I think that would also be very helpful.

We are currently testing NEMO in a number of clinical trials to try to further demonstrate clinical utility, looking at larger cohorts to figure out prognostic value, looking at patients with CRPC, seeing how often this emerges after AR-directed therapies or PSMA-directed therapies. We're looking at clinical trials that are specifically geared to the neuroendocrine prostate cancer variant to see how NEMO associates with target expression and outcomes in those trials. And I think all these validation studies are really needed now.

I think as far as next steps, we'd really like to continue to demonstrate clinical utility. We're really looking forward to any collaborations that people are listening and have specific cohorts that might be amenable to this. And then ultimately, looking to develop this clinically in a clinical-based setting. This is still a research assay, and that's something that we are really excited to think about.

Andrea Miyahira: Okay. Well, thank you both for joining me today and sharing this, Dr. Beltran and Dr. Demichelis.

Himisha Beltran: Thank you so much for having us.

Francesca Demichelis: Thank you.