Andrea Miyahira: Hi everyone, I'm Andrea Miyahira here with the Prostate Cancer Foundation. With me today is Dr. Abhijit Parolia of the University of Michigan. He will share his group's new paper, Divergent FOXA1 Mutations Drive Prostate Tumorigenesis and Therapy-resistant Cellular Plasticity, published in Science. Dr. Parolia, thanks for joining us.

Abhijit Parolia: Thank you so much, Andrea, for that introduction and particularly the invitation to share this new publication from our group that is primarily focused on understanding the central oncogene in prostatic malignancies called FOXA1. In this study, we were the first to actually make some of these knock-in transgenic mouse models. So I'll share some of the salient observations from these first-in-field transgenic mice carrying some of the human oncovariants of FOXA1.

In one of the earlier publications from Arul's lab, which I was fortunate enough to lead as a graduate student, we were the first ones to classify FOXA1 mutations into three distinct structural classes where Class 1 mutations predominantly comprised of missense and in-frame mutations clustering towards the C-terminal end of this forkhead domain, which is also the domain through which FOXA1 binds to the DNA. Class 2 mutations in sharp contrast comprised of predominantly frameshift mutations immediately after this forkhead domain.

As a novel classification for the first time, we described that even the centenic locus, which is the chromosome region where the FOXA1 gene is encoded, you see a host of rearrangements including duplications of the gene itself, which we annotated as Class 3 alterations. And notably, if you, in this particular classification schema now catalog the frequency of FOXA1 mutations across distinct clinical stages of the tumor from both European Caucasian men or Asian men, it's pretty obvious that FOXA1 mutations are one of the primary principal oncogenes in these tumors reaching about 35% frequency in the CRPC tumors in Caucasian men. And most notably in a recent publication, it was reported they actually replaced ETS fusion in the Asian population to drive prostate cancer formation in upwards of 40% of the cases. So really it is a principal oncogene in prostatic cancers. If you now go back and in this classification, just look at the clinical incidence of FOXA1, it's pretty interesting because you actually see in about 9 to 10% of the cases in the primary tumor, you're able to detect a Class 1 mutation of prostate cancer.

Now again, these are estimates from the Caucasian population. We don't have as rich data from the Asian population as yet, but it would be much higher in that population. These mutations stay around the same frequency in the metastatic tumor, which is usually castration-resistant disease. If you now in contrast to Class 1 mutations, look at Class 2 mutations, you see a dramatic or significant enrichment of this class of FOXA1 mutation in metastatic tumors. What's even more impressive is if you now go and look at the clonality of these mutations, in the rare cases that you actually do detect Class 2 mutations in a primary tumor, they're actually subclonal and they attain full clonality only in CRPC tumor cells, which really suggests that Class 1 and Class 2 mutations are two independent mutually exclusive events and likely define distinct molecular trajectories that prostate cancer really takes in the clinic going from a primary tumor to metastatic disease where Class 1 mutations would be clonally found in the primary tumor, meaning that they were likely the initiating Class 1 events.

And this tumor when it goes on to adopt some sort of co-alteration to get metastatically competent, these mutations persist even at similar frequencies in the metastatic disease. In sharp contrast, Class 2 mutations are rarely if at all detected in the primary tumor. This is the organ-confined disease, so it's likely initiated through some other genetic event, say perhaps ETS fusions. But in the course of progression, these tumors evolve Class 2 mutations, which are then clonally selected to comprise and make up the metastatic disease burden. So Class 1 and Class 2 mutations highlight or identify distinct etiologies of prostate cancer in terms of its progression. Given all of this and given such high clinical relevance, we were surprised at the time, and this is way back in 2018 to find that there were absolutely no mouse models for the FOXA1 oncogene. People had even knocked in KRAS mutations which are rare and barely found in prostatic tumors, but no one had knocked in an oncogenic variant of FOXA1.

And this was primarily our motivation to make some of these first-in-field mouse models. And what we did in these models is we took several representative transgenes of the human FOXA1 gene carrying the FOXA1 mutations either belonging to Class 1 or Class 2, and we knocked in that cassette or that transgene in the ROSA26 locus of the mouse genome, which is flanked upstream by this puroR STOP cassette that has loxP sites on either side. When you now take this transgenic mouse and cross it with a Cre recombinase, in this case, the Cre is driven by probasin, which is a promoter that is exclusively expressed in the mouse prostate luminal epithelium, you're able to actually release this puroR cassette, which is again this translational block which then relieves the FOXA1 transgene to get translated. So again, it's a release of the translational block and you can appreciate that in the absence of probasin Cre, there is no expression of the FOXA1 variant, which we can uniquely track by doing IHC for this exogenous epitope that we have added onto the transgene, the V5 tag.

But as soon as you have the probasin Cre, you're able to detect FOXA1 expression only in the luminal epithelium of the prostate tissue, not in any other cell including the stroma. We've also interrogated other tissues from this animal. It's a conditional and specific knock-in of the FOXA1 variant in the mouse prostate epithelium. For the Class 1 disease we primarily chose, again, these are mutations that comprise, that really affect structurally the Wing 2 region of the forkhead domain and to model, we selected these particular mutations to knock in into the mouse, the R265-71 in-frame indel being the one that shows a dramatic effect. So when we take these mice and we just inspect the normal tissue, you can actually appreciate that the prostate comprises this monolayer epithelium. It is a single layer of epithelial cuboidal cells that line the lumens of the prostatic gland, the lumen being filled with the prostatic fluid.

But as soon as you express the mutant form of the Class 1 FOXA1 oncogene and in this case, in the context of biallelic loss of Trp53, which is another co-alteration that frequently co-occurs with FOXA1, you're actually able to see this hyperstratification of that epithelial lining emergence of these atypical cells with increased nuclear size. And if you now stain with Ki-67, which really highlights proliferative capacity, almost 40 to 50% of these epithelial cells stain strongly positive for the proliferative index, suggesting that these are highly proliferative atypical lesions that start appearing in these prostatic glands when you overexpress the Class 1 mutant. Of note, it's important to note that Trp53 deletion alone has no effect in this context to drive transformation. Now if you age these animals, it's interesting that you actually see a larger span of the prostatic tissue showing these cancerous lesions, but you also have appearance of these higher grade lesions and in some cases starting at around 60 weeks of age, you even start seeing these tumor cell islands that invade into the adjacent stroma, which we have labeled in this particular study as grade IV lesions.

So you indeed through overexpression of Class 1 mutations in the mouse prostate epithelium drive invasive forms of adenocarcinomas in these mouse models. What's really impressive is now if you come in and you interrogate through multiplex IF, what is the status in terms of AR expression of these tumor lesions? You can actually see pretty nicely in this case tissue shown in the bottom line here, the adjacent benign tissue has AR expression, but even the tumor lesion that is right beneath it retains the expression of the androgen receptor protein. It, as you would expect, expresses the exogenous V5 tag transgene and the benign areas do not. So that's probably suggestive that it is actually the expression of the transgene that's driving these oncogenic transformations. It stains positively for CK8, which is a cytokeratin associated with luminal identity, all of which suggests that these truly are AR positive luminal lesions that Class 1 mutations trigger in these mouse epithelium.

Now, consistent with this idea, if we now do a mini pilot trial and we take some of these tumor bearing mice and we castrate them or do not castrate them, followed by inspection of the tissue using IHC or 10X multiomics, you're able to actually see that in the intact case, meaning that they have surplus of androgen, we are able to detect robust amounts of cancerous lesions in these animals. But as soon as you castrate, we were unable to find histologically any presence of these cancerous cells. We further went ahead and performed for a select number of these animals single cell RNA-seq analysis. Even here, we were able to actually see an increase in the prostate cancer score in an intact scenario. But as soon as we castrate these animals, these tumor cells somewhat disappear suggesting that indeed consistent with the expression of AR, these lesions continue to be reliant on constant supply of androgen, which is a key feature of human primary adenocarcinoma.

So we were particularly excited to observe that our model recapitulates the salient features of the human disease. Now moving on swiftly to Class 2, as a reminder, again, these are frameshift alterations that immediately follow the forkhead domain and this often leads to truncation of the C-terminal regulatory domain that has been shown in gold in this particular schematic. So this results in formation or rather production of a truncated form of the FOXA1 oncogene. And right off the bat, we aged these animals all the way up until 145 weeks even in the backdrop of PTEN deletion, which is known to be a co-alteration that frequently happens in humans carrying Class 2 mutations.

Class 2 mutations do not drive formation of tumors in the prostate epithelium, nowhere near the kind of lesion that we saw for Class 1 tumors. Instead, what we saw is we were able to reproduce this idea of cistromic dominance that again, we had suggested that these mutations display in the 2019 study wherein you can clearly see in these prostate organoids, when we overexpressed through Cre recombinase addition, the Class 2 variant, it's able to replace on the chromatin the wild type FOXA1.

Again, when we overexpressed the truncated form of the Class 2 mutation, it displaces the wild type FOXA1, the endogenous wild type FOXA1 from the chromatin, and it's the only variant or the only FOXA1 type that occupies chromatin in these particular cell types. This concept is perhaps better seen in these read density tracks shown on the right where you can again appreciate that the wild type protein is replaced in presence of the Class 2 mutant from the chromatin and Class 2 mutations seem to be the only protein that binds to the chromatin in these cells. This phenomenon we refer generally to as cistromic dominance. And to my knowledge, this is the only transcription factor that has been described to have such a dominant effect in terms of its DNA binding capacity.

And now in these tumors we wanted to really understand what happens to the transcriptome and if there is a dramatic consequence, if not histologically, at least transcriptional identities are altered in presence of a Class 2 mutation. So we took these animals, we took out their prostate tissues, which was followed by single cell multiomic analyses, and in this case you're looking at RNA expression of luminal cells and if you now just cluster them based on their transcriptional similarity, there were about seven unique clusters that emerged and I'd just like you to focus on this particular schema or rather the stacked bar plot on the right where we were able to in the control animals detect distinct clusters. But in the case of animals where the Class 2 mutations were expressed, there were two unique clusters that dramatically swelled up, clusters three and clusters four. When you actually now look at what is the identity of these cell types, we were able to see that both of these clusters, cluster three and four, strongly expressed the luminal cells or rather markers associated with the luminal cells.

But in sharp contrast to these normal luminal cells, we actually saw a dramatic increase in stemness potential. So these are cells that are inherently stem-like in the normal epithelium, which is cluster five in normal tissue. They strongly express the luminal stem-like cell markers. But what we found is that the clusters three and four, in addition to luminal markers partially gained expression of the luminal progenitor or the stem-like cell associated markers as well. So this was the reason why we went ahead and relabeled these cell populations as Class 2 induced luminal stem cells. So again, in presence of the Class 2 mutations, we see this dramatic rewiring in terms of the transcriptional program within these cells. If you now just take all of the mouse atlases and there are about three major atlases that were published at the time, we went ahead and defined a consensus stemness signature comprising of about 42 genes and just simply did a differential expression analysis comparing Class 2 expressing mutant luminal cells to the wild type luminal cells, we were able to see the signature showing dramatic overexpression.

And one of the primary markers that even others have used in the field to follow these stem-like populations called Trop2, was among the most highly upregulated genes in these induced luminal stem cells. These again, are interesting cell types that have been described by others as the predominant cell type that regenerates the prostate epithelium post-castration. These are proliferative, they survive better upon androgen withdrawal. So we were particularly interested in profiling in situ too, what happens to Trop2 expression in Class 2 expressing prostatic tissue. Now as shown in this normal control tissue, Trop2 positive epithelium is about 1 to 2% of the tissue architecture. That's what's been consistent with prior observations and we in our tissues were able to detect about 1 to 2% of the cells staining positive for Trop2. Notably, these sit at the proximal buds of the prostatic tissue, so that again, was consistent in our cell types.

But as soon as we interrogate Class 2 mutant tissues, we were in for a surprise. When we overexpressed the Class 2 mutant, we saw a dramatic increase in Trop2 staining in these cells. Again, these are androgen intact tissues, normal tissues, but when you now come in and look at the expression of Trop2 in contrast to the normal tissues, we saw about 10 to 20 fold increase in Trop2 positive luminal epithelial cells. Consistent with this stemness character, when you now come in and castrate these tissues, we were able to use V5 as a marker of cells that express the Class 2 mutant, in intact tissue, about 50% of the epithelium stained positive for V5. Now this is a leakiness of the system. We always get about 50% overexpression of the transgene, but when you castrate about 90 to 95% of the remnant cells were positive for V5, really suggesting that these V5 positive, Trop2 positive epithelial cells are better able to survive castration.

Moreover if you now take and stain these residual cells with Ki67 only in the case of Class 2 mutant tissue, we were able to see sustained expression of Ki67. So not only do they survive, but in presence of the Class 2 mutants, these epithelial cells instead of dying or atrophying, they actually remain proliferative and continue to divide. So this really is interesting and suggestive that Class 2 mutations perhaps reprogram the luminal epithelium to impart this resistance character.

So this brings me to the summary slide, which is quite exciting. It really suggests that there is a divergence in terms of the oncogenic traits that FOXA1 mutations provide. I did not show you a lot of the mechanistic data that we have in this study and I encourage you to perhaps review this paper. It's now published in Science. It's no longer under review, but we found that Class 1 mutations, if you overexpress in the backdrop of p53 deletion, you drive rampant formation of these AR-positive luminal adenocarcinomas that notably gain NSD2 expression, rewire the AR cistrome and remain strongly reliant on androgen supply for survival, which again is very consistent with the human disease and notably has been extremely hard to model in mouse prostatic tumors.

In sharp contrast, when we overexpress truncated Class 2 mutations in the prostate epithelium, we do not form tumors, but instead we reprogram these luminal cells to adopt some sort of stem-like progenitor-like character wherein they're better able to withstand androgen withdrawal and they continue to proliferate even in absence of androgen. And this is rather achieved through activation of latent enhancers belonging to KLF5 and AP-1.

So with that, I'll perhaps leave you with this acknowledgment slide. It really is massive teamwork lasting and spanning over six to seven years of active work in this project from my lab, particularly Sanjana Eyunni who is a graduate student and the first author in this study led the predominant majority of this effort and this was done in collaboration with Arul and several individuals in his group. He continues to be an important member in my independent phase. And I'm grateful to the Prostate Cancer Foundation. I must acknowledge that this is the project that they funded me for as a young investigator and I'm extremely grateful because it's extremely hard to raise money for such basic science projects, but it really is extremely gratifying to see some of the findings and be able to share it with the rest of the community. So with that, I'm happy to take any questions.

Andrea Miyahira: Thank you so much Dr. Parolia and congratulations on this study. So what are the differences in disease development progression and outcomes in patients with different classes of FOXA1 alterations?

Abhijit Parolia: Right. I think a lot of that data still needs to be evaluated, Andrea. It's been extremely hard to gather long-term survival studies or data to do these survival studies, although I am aware of some efforts from the University of Minnesota. But they have suggested that certain classes of FOXA1 alterations do associate with poor prognosis. In-house we have looked at some of these cases, notably Class 3 alterations, which are duplications of the FOXA1 gene itself are associated with poor prognosis. Interestingly, Class 2 would certainly associate with poor prognosis. Sadly, their frequency is extremely low, so we are not able to do a statistically significant analysis yet. But I really speculate that Class 2 mutations that again are associated with metastatic disease seemingly impart resistance to castration, they would be prognostically among the worst tumors if I were to speculate.

Andrea Miyahira: Okay, thank you. And based on these data, did these studies provide any new insights into how to treat different FOXA1 altered subclasses of prostate cancer?

Abhijit Parolia: I think they certainly do, although these are conceptual at this point. There's a lot of prior evidence suggesting a lot of these therapies may or may not work. But nonetheless, that said, we did find... I didn't get an opportunity to share the mechanistic component of this study with you in this talk today, but we found Class 1 mutations actually in parallel co-activate AR as well as mTORC1. And the activation of mTORC1 is completely independent of AKT and PI3K. So it's really interesting that could speculate as to why maybe inhibition of some of these upstream effectors of this pathway may not have potency therapeutically speaking in a Class 1 driven prostatic tumor, but I would speculate that one should likely re-evaluate some of the PI3K inhibitors, that class of drugs, by blocking effectors downstream of AKT because in some forms like Class 1 mutant driven tumors, it seems to bypass some of these upstream effectors and drive the disease formation.

For Class 2 mutations, I think in our prior study we had shown that they de-repress Wnt. That's one of their salient phenotypes. I don't see them to be responsive to therapy, in particular androgen deprivation therapy. I think they would evolve even in a tumor that is naturally evolving. And there is some data showing that even hormone naive metastatic prostate cancer actually has Class 2 mutations emerge. So I really think that one could speculate Wnt or some of these similar pathways being vulnerabilities in this tumor type. But all of that said, these ideas have been tested but have not shown efficacy in prostate cancer in the clinic. So one wonders if selection of these tumors... patients who would better respond needs to be evaluated and maybe some of these observations might contribute to that objective.

Andrea Miyahira: Okay, thanks. And what are your take-home messages for modeling prostate cancer using transgenic mice?

Abhijit Parolia: Right off the bat, our group is extremely young in this area. This is our very first study trying to describe some of this. Arul's group is much more experienced. They've done this quite extensively. That said, I really think one thing that the field was perhaps struggling with is generation of models that truly respond to castration. PTEN biallelic loss has been widely adopted in the field because of the lower latency. But I do believe that the tumor that results from PTEN deletion isn't necessarily androgen reliant. It may be still AR dependent, but not androgen dependent. And that was one thing that we really tried to communicate in this study, that these tumors seemingly rely on androgen and androgen supply in addition to AR itself. So I would speculate that these perhaps Class 1 models mimic an earlier form of the disease, rather than a primary form of the disease.

That said, we also have p53 deletion in there, so it does complicate some of those scenarios, but that would be I think one of the salient takeaway points from this study that Class 1 mutants really drive an androgen sensitive form of prostatic adenocarcinoma. In terms of Class 2 mutations, I think it really suggests that these two mutations are independent in their biology. They do not have overlaps and Class 2 mutations certainly cannot drive primary prostate cancer formation. It really is a mutation associated with progression and perhaps it could be used as such for diagnostics to really detect or catch these early clones that might be in the blood to detect this secondary tumor in a distant site. But all of that of course needs a lot more work. This is just us laying some of those ideas out there for the field to evaluate.

Andrea Miyahira: Wonderful. And what are your next steps in these studies?

Abhijit Parolia: I think I'm particularly motivated and I would be really excited if we could raise some money around this. I think one of the primary issues in prostate cancer in general has been the lack of models. So we really, right now my lab is committing a lot of effort in taking some of these organoid models. We are right now making syngeneic lines. We are re-injecting some of these organoids back in immune competent mice. We are trying to make tumor xenografts. We are trying to make derivative 2D lines.

All of this is something which will not necessarily lead to a high-impact study, but I do believe it's extremely impactful in our ability to interrogate prostatic biology. Having these syngeneic normal tumor models would be extremely exciting. So that's something that we are headed towards next. We are trying to really write a smaller study focused on some of the derivative organoid as well as syngeneic tumor xenograft, allograft models that we are able to make and we are now molecularly characterizing to see what aspect of the donor tumor or the donor biology they still retain and what is their genomic profile, what mutations might have somatically appeared. So that is where we are headed next.

Andrea Miyahira: Okay, wonderful. Congratulations again on this study and thanks for sharing it with us.

Abhijit Parolia: Thank you so much, Andrea. I really appreciate PCF's continued support and thank you again for this invitation.