Read the Full Video Transcript

Ashish Kamat: Hello, and welcome to UroToday's Bladder Cancer Center of Excellence. I'm Ashish Kamat, Professor of Urologic Oncology and Cancer Research at MD Anderson Cancer Center in Houston. And it is my pleasure to welcome Dr. Dan Theodorescu, who is a friend, a colleague, and for the purpose of this talk, most importantly, a true trendsetter and ingenious in the field of not just urologic oncology and bladder cancer, but cancer in general, Dr. Theodorescu has had multiple, multiple high impact publications and just impact on the field in general. And he is now Director of the Comprehensive Cancer Center at Cedars-Sinai. He is a PHASE ONE Foundation Distinguished Chair, and really needs no introduction, so I'll cut it short and essentially hand the stage over to him today to talk to us about this new, exciting epithelial cell subpopulation that he has discovered and his group has discovered, that predicts response to bladder cancer therapy and potentially provides a new subtype of bladder cancer. So Dan, with that, let me hand the stage over to you.

Dan Theodorescu: Well, thank you Ashish for that very kind introduction, overly kind introduction. It's a privilege to be here and share this information with the group. So my laboratory has been interested in bladder cancer, understanding its biology, and predicting its behavior for the benefit of patients for the last 25 years. And one of the things that have driven us is shown on this slide, which is whether it's first-line therapy or second-line therapy, we have a number of tools, including cisplatin-based chemotherapy, and then more recently immune checkpoint inhibitors that are effective in our patient.

But however, as is shown in this slide, we also know that these therapies are not perfect and could be improved. So the two options are, really to discover new therapies or to discover ways to enhance current therapies, be it chemotherapy or immune checkpoint inhibitors. And our laboratory has been involved in both of those aspects, but fundamental to really developing these two approaches is really to understand what is really happening in the tumor during its genesis and evolution.

So let me then explain the basics of how these therapies can be developed or improved in therapies. It all hinges on the concept of tumor evolution. Tumor evolution is a very interesting idea because it starts with the genesis of the tumor. And in this case, we have tumors that are composed of multiple cells, and the important thing in tumors and the reason why cancers are difficult to treat is that these tumors, while they are clonal in many regards, they also have multiple heterogeneities of these clones. So they may start from one cell, as the transformed cell, but as the tumor grows, they acquire genetic, epigenetic, and other changes to cause this clonal heterogeneity. And as you select tumor variants, either by growing in other places, such as metastasis, or treating them with various treatments, you have a progressive evolution and selection of clones for metastatically competent clones to various organs.  And then when you treat them, you have obliteration of certain clones, but the persistence of others. So what this slide really emphasizes is that cancer is difficult to treat because we have tumor heterogeneity. And unless we really come to grips with this tumor heterogeneity, we are not going to be able to either treat cancer effectively or predict cancer behavior effectively.

So, based on that, and based on the desire to improve therapy, we had a hypothesis that had three parts. Number one, that there are specific cells in tumors whose presence can determine the risk of progression of metastasis and therapy resistance. Number two, that we can discover these cell types using modern single-cell technology. Number three, that we can use this information to eventually target these therapeutically and prevent either the emergence of metastatic disease or resistance to therapy.

So here is what we have done to address and test this hypothesis. We took 25 muscle-invasive, non-metastatic tumors from patients that had undergone cystectomy and carried out an analysis called, single cell sequencing. What this analysis does is provide a very nice pictorial of, including this picture here, what this shows is a pictorial of the various populations of cells in the tumor. And without going into a lot of detail, what that does is, the computers analyze the transcriptome of every single cell, which is amazing. It's magic, magic, it is science fiction from, what science reality is now is science fiction 10 years ago. And what this does is look at the sequencing of every individual cell and then takes that information and then does a nearest neighbor analysis, and then clusters cells according to their expression, transcriptomic RNA signatures. When you do that, it generates these beautiful pictures that show you what populations are present in your sample.

And as you can see here, the computers are completely unbiased and identified some critical populations. Now, when you look at those populations, you see that there are quite a bit of epithelial cells, and we will talk about that in a minute, but these epithelial cells and other components, fibroblasts, et cetera, were defined by a variety of genes that are shown in panel C, and you could see these various genes that defined the various populations. These are sort of the emblematic genes of the various populations.

But then when you look at panel D, it's pretty clear that there are five critical epithelial populations, which are really the cancer cells. The other cells that are nonepithelial are basically the stromal supporting cells within the tumor, fibroblast, lymphoid, et cetera. So when you look at these epithelial cells, you have classical cells that have been defined previously, but we also defined a new population that we picked one gene to sort of use as sort of our name, if you will, a temporary name for this population, CDH12, and it is one of the genes that is highly expressed in this population. And when we look at this population, in terms of the canonical signatures that have been published by many groups in the field, we see that this population does not really track any single subtype of bladder cancer.

So the next question was, how do these populations track with prognosis? And when we took these populations in the genetic signatures and the RNA signatures of this population called CDH12, we looked at them and analyzed them in the TCGA and realized that only this population was predictive of outcome in the TCGA. And so that's in panel A and then we looked in panel B, we looked at this population in terms of its projection, again on the single-cell pictures, to see if there are any other, if you will, cancer generic signatures that project and cluster in this area. And as you can see in panel B, we have the TeratoScore and Pluryi tests, which speak to stem cell signatures that tend to project on where the CDH12 population is. And interestingly enough, neuroendocrine.

So that made us look at the consensus signatures in regard to that, and you could see in panel C when we do heat maps, the CDH signature lights up very nicely with neuroendocrine signatures, as does another population called the cycling population. And in the TCGA signatures, it also shows up in the neuronal. So both in the consensusMIBC signatures and the TCGA signatures, the CDH12 shows up in the neuronal as well as in the completely non-bladder specific neuroendocrine generic cancer signatures. So obviously that is a clue to where it's intrinsic biology.

So then what we wanted to look at was to see how this population behaves in chemotherapy-treated patients. And we took a population that had been previously published, pre-and post-chemotherapy, with platinum-based agents. And as panel A would show you, these are the levels of the various populations, as a response in the pre-treatment and post-treatment. And if you look at the various epithelial populations, the CDH population, when it's starting at low levels pre-treatment, increases the levels. There is no change if you start with high levels and there is no other population that increases its levels in a similar way when the intrinsic levels are low, so that was pretty interesting.

The other interesting thing was shown in panel B, which is when we looked at various gene families or ontology analyses that we like to call them, based on the change of the score. So in other words, whether the tumor started low and ended up high or vice versa, or ended up high and stayed high, as a function of chemotherapy, you see that the genes that are changed as a function of chemotherapy in cells that stayed high with CDH12, the red line, versus the dotted blue line, are different. And that actually says something, because what that means is that when tumors have high levels of CDH12 populations, the response to chemotherapy is different than in cells that actually have low levels of CDH12s. What it also tells you is that killing cells that are not CDH12 to allow the CDH12 cells to grow, generate something, and you could see that a lot of the pathways are cytokine or immune response pathways, whereas the other ones have to do with fibroblasts and other things that would prevent an immune response.  Again, very strong clues as to how this population does what it's supposed to do, in terms of the immune response, which is the next slide.

So of course, given the popularity of immune checkpoint inhibitors, we wanted to look at this, and we looked at a population that was treated with atezolizumab, IMvigor210, and we basically looked at patients that had biopsies before chemotherapy and biopsies after chemotherapy.  Given what we have shown before in the previous slide, the chemotherapy does make a difference, and therefore we wanted to be mindful of that potential effect. So if you look at panel A in samples that were taken before chemotherapy versus after chemotherapy, before chemotherapy, there are very few markers, including CDH12, CD8T cells, PD-L1, and PD-L2, nothing is significant.

Conversely, post-chemotherapy, everything is significant, which is really interesting, including CDH12. Now the other populations were not as robustly significant. And therefore, again, highlighting the importance of the CDH12 population. But it's important to note that the pre-chemo and post-chemo biopsy, even in patients that are treated the same, does not predict. It is a really important point in this paper that when you do the biopsy, in order to stratify patients to various therapies, is really critical. So going by the tissues in the cystectomy, when patients have had chemo, and then you're putting them on checkpoint inhibitors, going by the cystectomy specimen is probably not a good idea. This is a cautionary tale to be mindful of that. And B shows basically that the stratification by RECIST criteria in the post-chemo, showing that the response also is significant for the lowest quartile.

Interestingly, in panel C, we see that the CDH population is very strongly expressing PD-L2, as opposed to PD-L1, compared to the other four epithelial populations that I've described before, which kind of have a similar expression of PD-L1 and PD-L2. Finally, in panel D, we compare the CDH12 signature with other signatures that have been published of bladder cancer subtypes, and we see that the CDH signature is the one that has a very, very strong prediction compared to the other signatures here in green.

So almost towards the end here, one other important thing that I would like to leave you with, but I alluded to in the chemotherapy slide, is that a lot of the really interesting biology that is driven by the CDH12 cells has to do with their very, very solid and strong, compelling interactions with other cells. They do interact with fibroblasts in a very significant way and I haven't had time to show you this, and also with T-cells. Because of the strong interactions and the implications that has, and because of the gene signature of this population that helped us define it in the original description, in the first few slides, we are naming this population, the C3 population, which stands for Cell-Cell Communication.

So finally, the conclusions. I think it's pretty clear that single-cell biology and single-cell transcriptomics are just very, very powerful to really identify new populations of cancer cells. I think this is going to revolutionize our understanding of cancer and really bring to fruition, the description of tumor heterogeneity, that started in the eighties, and now we are going to be able to actionably intercept and understand these populations and intercept them therapeutically. We use this technology to identify a new bladder subtype and other subtypes that I show in the first slide, and the hope is that this will create novel tools to stratify patients for various therapies. So what we are doing now is analyzing this population with other therapeutic modalities. We're also doing a lot of mechanistic studies to understand the genesis of this population, which is now clear that it is, and we have this in the paper, it is present in normal cells in a nontransformed way, so as a transient population during differentiation.

So obviously the hypothesis potentially, those oncogenic genes that transform the cell at the phase of development when this cell is in play, can lead to a different type of bladder cancer, which is sort of the high working hypothesis of this whole project. We are working on identifying various ways to intercept it. And in the meantime, we are also thinking about how to integrate this signature into a clinical test that could be provided when we molecularly profile patients. And on the right side here, we have a hypothetical way that we could use the C3 signature score that has been used in the Kaplan-Meier curves before, to potentially guide therapy going forward for patients that have locally advanced urothelial carcinoma, going through chemotherapy and immunotherapy.

So with that, I would like to close. I would like to thank Ashish and UroToday for the opportunity to present this data.

Ashish Kamat: Thank you so much, Dan. Anytime I listen to you talk, it's amazing how you take a complex subject and make it so simple, and you did exactly that, even though you didn't have much time and I'm sure everyone listening and watching will really appreciate that. So thank you for that. You actually addressed a lot of the questions that I was going to ask you in order to sort of clarifying the manuscript, in your presentation, again, not surprising at all, but let me just ask you a few questions. So first off, cadherin 12, the way you looked at it, would you also be able to identify this on circulating tumor cells? I'm sure you are well aware of Tom Powles's data, suggesting with atezo, how that can really predict who may or may not require adjuvant atezo. Are you able to do that in your lab at this time?

Dan Theodorescu: So, I'm glad you asked this question very much. It turns out what we have in the gene signature, we have actually refined the gene signature to only look at genes that are expressed on the surface of cells, exactly, we're thinking very similarly. And what it turns out is with 14 genes, all on the surface, we have the same predictive ability, this is not published, the same predictive ability of the C3 signature as we do with the entire signature, which is 200 plus genes.

So I would be welcoming any collaborations for anybody out there listening. I would love to do a circulating analysis either at the molecular level.  We also have tools that can look at this population at the DNA level and at the RNA level, and now at the surface level. We would love to collaborate with any individuals that are interested to address exactly what you say because I obviously have a hunch that it could be useful, so thank you for that question.

Ashish Kamat: Yeah, absolutely. It's almost like I read your mind.

Dan Theodorescu: You read my mind perfectly.

Ashish Kamat: No. That's great. And again, just one question, because you've always been a strong supporter of the fact that you have to have a hypothesis and then test it out. And clearly, you have this discovery, I'm sure you have already planned and chocked out a prospective evaluation of this. Is that something you can share with us? What do you have cooking? What do you have ongoing as far as evaluating this in a prospective manner?

Dan Theodorescu: Well, so again, great question. Right now, again, for anybody that is listening out there that would be interested in collaborating, right now, what I would really love is I would love to run just very simple Kaplan-Meier curves to validate the signature on other data sets with a variety of therapies. Be they surgical, I have the TCGA obviously, but it would be nice to do other data sets. I would love to look at more immunotherapy data sets. We're in the process of looking at other cancers, but really it's whatever is available that we are gravitating to. If there are any other data sets that could be evaluated outside of IMvigor210, that are available, I would love to evaluate them because the more information we get, the more refinement we could do on the signature and the better it would be to reduce the number of genes that we could potentially evaluate in a prospective way, because the ideal perspective evaluation that I'm thinking, and Ashish, I love your comment on this as well, your perspective, is I would love to take the signature, refine it through the available data sets that are published, then come up with a really, really smallish signature that we could evaluate on tumors and the smallest signature that we can evaluate in the blood, whether it's surface or otherwise, and then do a prospective trial. That would be the dream scenario, but I would like to really put it through the wringer and multiple data sets out there, to really refine it and craft it, so the prospective trial has a chance of being definitive, and then we could use it clinically.

Ashish Kamat: Yeah, no, I totally agree with that. And especially since the cell-cell interaction that you've shown, and we can go back and look at the slides for those that are viewing this online, I think this would actually be something worth looking at in patients treated with trimodal therapy in bladder sparing conservative therapy as well. Any thoughts about that? Do you have any-

Dan Theodorescu: Great idea. Great idea. I did not have an available dataset for trimodal therapy. And honestly, we had so much data coming at us, it was like drinking from the fire hose. Which was great, I love it. I love it. I'm just having a blast here, okay? But having said that, now that you put this bug and you're reminding me, it's an important issue. I'm going to try to get a hold of some radiobiology data sets and see if this population is available and if it's there and if it does anything, but thank you for that.

Ashish Kamat: Absolutely. And again the audience of UroToday is so vast and all over the globe. So, anyone that is listening to this and has a data set in this manner, I'm sure you would welcome them, reaching out to you and collaborating.

Dan Theodorescu: Absolutely. And we have various mechanisms to do this in a blinded way. So I have different tools that it does not compromise confidentiality and all these things, and we do it in a blind way, which is actually good for validation. I have a paper now that is in review at another journal, where we did validate a biomarker completely blinded, with another entity. And neither of us knew who had the biomarker algorithm and who the other entity was that had all the patient data, and we did it in such a way that neither knew the other's data set. So it was really great. We have it all figured out. So I welcome any contact.

Ashish Kamat: Professor Theodorescu, always a pleasure chatting with you. Thank you so much for taking the time.

Dan Theodorescu: Thank you very much. I appreciate the opportunity. It was nice to discuss this.