Andrea Miyahira: Hi, I'm Andrea Miyahira at the Prostate Cancer Foundation. Here with me is Dr. Michael Haffner of Fred Hutchinson Cancer Center. He will discuss one of his team's latest papers, Patterns of intra- and intertumor phenotypic heterogeneity in lethal prostate cancer. This was published in JCI. Dr. Haffner, thanks for joining us.

Michael Haffner: Thanks so much for the invitation, Andrea. Well, thank you again for the invitation to present this paper on behalf of my co-authors.
What I really wanted to do today is present some of the new findings that are presented in this manuscript. But start out by defining a very simple molecular classification framework for castration-resistant prostate cancer first. Then, delineate some of the patterns of molecular heterogeneities that we observed in advanced metastatic prostate cancer, and correlate them with genomic associations.

But first off, we know that prostate cancer is a highly heterogeneous disease with clinical presentations vary greatly between different patients. And this really calls for a refined way of molecularly classifying these different disease manifestations. And one of the classification frameworks that was presented initially by my colleagues Colin Morrissey and Peter Nelson, makes use of a very, very simple way of markers that stratify tumors essentially by their expression of luminal epithelial or androgen receptor-related markers, and neuroendocrine markers.

And in doing so, you can classify now castration-resistant prostate cancer into four different flavors. Those that are very much in tune about classical prostatic adenocarcinoma expressing the androgen receptor and luminal epithelial markers, and showing no expression of neuroendocrine markers. Those that are commonly known as neuroendocrine prostate cancers, they lack the expression of luminal markers and the androgen receptor that express high levels of neuroendocrine markers. Those that have expression of both, either in admixtures or co-expression patterns of AR and neuroendocrine markers, and those that lose both, the so-called double negative prostate cancers that are negative for AR and neuroendocrine markers. So this framework is very helpful to contextualize some of our findings.

And of course, we know from a lot of anecdotal work from our clinical experience that often patients showed very discordant responses to certain therapies across different metastatic sites. And this really now begs the question, what is the level and pattern of heterogeneity of these molecular subtypes in patients with advanced metastatic prostate cancer?

Now, in order to do this and to ask this question in a very formal way, we leveraged a very unique sample cohort from the University of Washington Autopsy Project. And this is an ongoing effort more than 20 years and counting now for patients that die of metastatic prostate cancer can undergo a very quick sampling of their entire tumor burden, allowing us to procure specimens that represent the entire spectrum of the disease that the patients died of.

So, in applying these strategies, we looked very closely at 52 cases from this rapid autopsy cohort, and used first an in situ approach applying validated immunohistochemical markers to subtype these individual tumors. These are more than 630 individual tumor samples studied here. And what we found is that there are different patterns. We see mostly still that in this patient cohort tumors showed expression of luminal, epithelial and AR related markers. Some patients showed exclusively expression of neuroendocrine markers in the absence of AR markers. But there were about 19%, so 10 of these 52 patients. They showed a high level of intra metastatic heterogeneity.

So maybe it's helpful to show you how a case with such high level of intra metastatic heterogeneity looked like. This is a patient had a very large liver metastatic tumor burden, but also metastasis to the lung, to the bone, and a very bulky tumor in his prostate. Just histomorphologic profiling of these tumors showed that there are two different types of tumors in the prostate, a clear adenocarcinoma, and then also morphologically defined a hybrid neuroendocrine carcinoma. The lung showed sarcomatoid carcinoma features, a very, very rare manifestation of prostate cancer. The liver was uniformly neuroendocrine carcinoma, and then in the bone we saw adenocarcinoma morphology, but the tumor cells did not express any androgen receptor or neuroendocrine markers, consistent with a double negative subtype.

So despite this high level of morphologic and molecular heterogeneity, we saw that on the genomic level these individual tumor lesions within a patient, they're very similar, all of them characterized by key driver gene alterations, in this particular case, loss of PTEN and RB.

So that got us really thinking about the genomic underpinnings of this level of heterogeneity. So first we asked the question, are there particular genomic alterations that are associated with these distinct subtypes? What we found was that loss of PTEN was very strongly associated with double negative disease. Gain of androgen receptor copy number, or AR ligand binding domain mutations was very commonly found in AR positive tumors. And loss of RB, or correlation of RB and PTEN was most commonly found in AR negative neuroendocrine marker positive neuroendocrine carcinomas.

Then the other question that was very important for us is, were any of these alterations associated with the diversity that we found across the patient's tumor burden? So we calculated based on our molecular assessments a heterogeneity index. And to a great surprise, we started to see that tumors that had androgen receptor gene locus alterations, and particularly those with high level amplification, had a much, much lower heterogeneity index overall. Whereas tumors that had PTEN alterations showed a higher level of heterogeneity.

So this is for molecular subtype assessment, but how about some other very basic characteristics of a tumor cell such as cellular proliferation? So, in order to look for cellular proliferation differences, we use a validated immunohistochemical assay for Ki-67, stained this entire cohort. And the data is summarized in this graph here where every column represents an individual patient. Every dot is an individual metastasis. And you can appreciate from this graph that there's, in some cases, a very high level of diversity in cellular proliferation index across different metastatic sites. And this is really something that is previously underappreciated.

So in summary, what I wanted to show you is that castration resistant prostate cancer shows a very high level of intra and interpatient morphologic and molecular heterogeneity. In this probably largest study to date, we assessed more than 630 tumor samples. And this showed phenotypic diversity at the patient level, at the single metastasis level, but then also at the individual cell level. We've seen that certain genomic alterations, in particular PTEN and androgen receptor changes, are somewhat associated with the patterns of heterogeneity that we see.

And the other thing that's also important to note is that core tumor features such as cellular proliferation can vary dramatically across different metastatic sites in the given patient.

So, what are the big take home messages here? I think we can conclude that tumor heterogeneity is really a major challenge, both from a diagnostic, but then also from a therapeutic perspective. What our study implies is that in some patients a single tissue biopsy may not be sufficient to capture the entire diversity of the patient's molecular tumor burden, and might not be enough for accurate tumor classification. That's why I need to think moving forward more comprehensively about integrated approaches that take both molecular imaging data, liquid biopsy approaches, and tissue-based biopsies in consideration. And of course, future work is needed to define what the therapeutic implication of this heterogeneity that we observed in this study is. And also to think about strategies that would allow us to potentially reverse some of these heterogeneous cell states, and overcome barriers that we currently would envision that this heterogeneity poses. Thank you very much.

Andrea Miyahira: So, thank you so much, Dr. Haffner, for presenting this really interesting study. So, what additional insights does this study give into the patterns and pathways of tumor evolution to these molecular subtypes and clinical features?

Michael Haffner: Yeah, thanks so much, Andrea. This is a very important question. We're I think starting to scratch on the surface a little bit. So, some of our data suggests that certain genomic alterations, say AR copy number gain, are associated with a higher level of homogeneity in the patient's tumor burden, whereas others, PTEN loss, for instance, are associated with a higher level of heterogeneity.

But to be quite honest, I think there's much more to this study. And while this manuscript focuses on the genomic associations with heterogeneity, I think there's much more to be learned about the epigenome as well. And we're currently conducting more in-depth epigenome profiling studies within this cohort to link DNA methylation, as well as chromatin accessibility changes, with these patterns of phenotypic heterogeneity.

Andrea Miyahira: Thanks. And in these patients, can an exact cause of death be determined? And if so, did those final lethal lesions have anything in common?
Michael Haffner: Yeah. So often it's very difficult to identify a cause of death in these patients. And the tumor burden also varies dramatically of some patients that present with relatively limited tumor burden, whereas others have extensive metastasis, for instance to the liver, where you can envision that just the amount of the metastatic tumor burden could result in end organ failure.

So, cause of death still I think very, very challenging to delineate. And this is also I think a broader question in the field, maybe we would need to try to think about addressing both in this postmortem examination, but also then in patients that are still in clinical care, is what triggers these dramatic switches in disease presentation, in particular in the very, very latest stage of the patient's disease journey.

Andrea Miyahira: Thanks. And you note that because of this heterogeneity within patients that single site biopsies are unlikely to capture all diversity. So, could that be captured with ctDNA? Or alternatively, are there shared or truncal features that can be therapeutically targeted?

Michael Haffner: Right, yeah. So, I think in our group, we've been spending a lot of effort in trying to develop ctDNA based assays that can detect some of these molecular subtypes that we had discussed at their incipient phases. So this would allow for an additional way of measuring potential patterns of heterogeneity in a less invasive manner. Of course, some of the genomic alterations, and there's been a lot of studies by our group, as well as many others, we know that the vast majority of genomic changes are actually truncal in nature, in effect most, if not all, of the metastases. But I think there's more to be said about the plasticity that relates to epigenome changes coming back to the first question. And I mean, those are the type of molecular changes that we just need to profile and understand much better.

Andrea Miyahira: Thanks. And based on these studies, what are the biggest next questions to ask? And what are your next research steps? And are there plans to look at any tumor microenvironment, metabolic programs, immune features or correlations with other clinical factors such as host factors or prior treatments in this cohort?

Michael Haffner: Right, yeah. Excellent question, and I think very timely for us to think about this. I see this study as a starting point, laying out a framework for how we can use this particular type of biospecimen. But then also allowing for a rough categorization of these individual samples. Now, and this is a wonderful study supported by the PCF TACTICAL Award. We are very interested in diving much, much deeper into the molecular differences, not just at the single site level, but at the single cell level, applying both spatial transcriptomic approaches, as well as single cell sequencing approaches. And those will hopefully help us to, A, characterize more of the molecular changes, more of the molecular diversities that we see in these patient samples, but also get us to think a little bit more about the tumor microenvironment. Are there particular features of the tumor microenvironment, stromal cell changes, immune microenvironment, changes that predispose or maybe prevent for some of these patterns of heterogeneity that we've described in this paper to emerge?

Andrea Miyahira: Thank you so much for sharing this wonderful study with us.

Michael Haffner: Thanks so much.