Ashish Kamat: Hello everybody, and welcome to UroToday's Bladder Cancer Center of Excellence. I'm Ashish Kamat, urologic oncologist in Houston, Texas, and it's a distinct pleasure to welcome Joep de Jong, who has joined us before in person at several conferences. But today, he's joining us to talk about the latest publication in Clinical Cancer Research, talking about molecular characterization of residual muscle-invasive bladder cancer that identifies a scar-like transcriptomic profile with favorable prognosis after neoadjuvant therapy. So Joep, welcome.

Joep de Jong: Thank you very much, Professor Kamat, for the introduction, and thank you, UroToday, for the opportunity to discuss our latest research. First of all, as a quick background, I'd like to acknowledge that multiple groups have been interrogating bladder cancer biology, and there's a specific interest in the molecular alterations before and after applying neoadjuvant systemic therapy for muscle-invasive bladder cancer. So the Choi and colleagues have focused on molecular alterations after neoadjuvant chemotherapy, comparing the TURBT with the cystectomy tumor samples. The Seiler group has focused on neoadjuvant chemotherapy as well. And Necchi and colleagues, where I was part of their collaboration, we focused on molecular changes after neoadjuvant pembrolizumab, which is like a PD-1 inhibitor.

And the group from Dr. Powles with the ABACUS trial, they also focused on molecular alterations before and after neoadjuvant atezolizumab. And this is of particular interest because we are always seeking the mechanisms of resistance from muscle-invasive bladder cancer because these tumors are residual at radical cystectomy. And generally, having residual cancer after systemic therapy is associated with worse prognosis. And these two charts highlight some of the findings from these biologic characterization studies. So, both of them are heatmaps. And I'll start on the left. In the heatmap, we should generally acknowledge that every vertical line is one patient. And what we can do with gene expression analysis, that we can tell the computer to say, "If you look at all the biologic patterns of these tumors, which kind of tumors are on the biologic spectrum resembling each other?" And then the computer basically groups these tumors from patients into separate biologic subgroups.

And this is all looking into the tumors from one particular cohort study. So the chart on the left is from patients that were treated with neoadjuvant systemic chemotherapy. And mind you, these are the tumors that are residual at radical cystectomy, so not the TURBT. And there's a general acknowledgement of basal and luminal subtypes, but also more immune-infiltrated tumors. And also, one group on the right, which is here indicated in orange. That is characterized by more of the stroma infiltration. And these are all components of the tumor microenvironment that we found not only in chemotherapy-treated tumors, but when we looked at the study from Dr. Necchi, the PURE-01 study, we also compared pre- and post-tumors that were treated with neoadjuvant immunotherapy. And even though the tumors were either treated with chemotherapy or immunotherapy, we could highlight the luminal group, which was typically expressing more of the luminal marker genes; a basal group, which is typically expressing more of the basal marker genes; and then again, a third scar-like group.

This group was named scar-like because they typically express genes associated with wound healing, or scarring, or the stromal microenvironment. And this was a separate finding in two independent cohorts, basically resembling different subgrouping of the tumors, of which one was called scar-like. And because we wanted to know if it was actually resembling scar tissue, we had previously also profiled actual scar tissue. So, this was bladder tissue from patients that had a complete response. And we compared the actual scar tissue on the biology levels with these tumors that were classified as scar-like. And these charts are... They seem not easy to interpret, but basically, what we showed by principal component analysis or differential gene expression analysis is that the scar-like residual cancers were, on the biologic spectrum, not that different from actual scar tissue. And therefore, we were actually curious to see if it could also correlate to prognosis because, of course, scar tissue in complete responders relates to good prognosis because they have complete response to the systemic therapy.

But the residual tumors that, on the biologic aspect, looked as if they were scar tumors, they were also correlated to good prognosis, but only in the cohorts that were treated with neoadjuvant systemic therapy. So again, PURE-01, which is chart B. The scar-like group was associated with favorable outcomes after neoadjuvant immunotherapy. We had the same observation for patients that were treated with neoadjuvant cisplatin-based chemotherapy here in chart C. And we also acknowledged a scar-like group in a cohort that was treated with radical cystectomy only at the University of Texas Southwestern. And importantly, while we did find a scar-like biologic subgroup, this did not correlate with better outcomes if they were not treated with neoadjuvant systemic therapy.

We thought this was of particular interest because we could find this subgrouping in three independent cohorts. And in the meantime, the consensus molecular classification, which is a big collaboration of almost all of the groups that were involved in this biologic type of research, developed a classification model that could be transported onto future research data. They also acknowledged a stroma-rich group, which was basically resembling the scar-like group. So, we were interested if this stroma-rich subtype could identify the scar-like tumors. And we saw this was possible for the PURE-01, which is chart B and chart A. So the scar-like group was almost similar to the stroma-rich group, but when you look at the chart for the post-NAC cohort, we saw that the stroma-rich group was not identifying the scar-like cases that were correlated to good prognosis. And this led us to investigate the study by Professor Powles, the ABACUS study, where we actually again looked at the biologic profiling of these patients that were now treated with neoadjuvant atezolizumab.

So, PD-1 inhibition. And we again found a scar-like cluster or class that was correlated to good prognosis. And to actually translate this clustering analysis onto future clinical application, one step is required. And that is that the clustering, which happens within the cohorts, needs to be translated to a single-sample classification model. And that basically means that if we have one RNA expression profile from one patient, can we say if this patient belongs to the scar-like subgroup, yes or no? So we trained a machine learning model, and this model accurately identified the scar-like cases in all of the four cohorts. And you can see in these Kaplan-Meier plots that patients that harbored a scar-like residual muscle-invasive bladder cancer after therapy were actually related to good outcomes in the chemotherapy, pembrolizumab, and atezolizumab-treated cohorts, whereas this was not the case in a cystectomy-only cohort. So, we basically think that the TURBT can already induce a wound-healing response that can resemble biology at the radical cystectomy specimen, but it has to synergize with the neoadjuvant systemic therapy to basically correlate to the good prognosis.

And therefore, we do think that scar-like tumor biology, importantly at the radical cystectomy specimen, is correlated to favorable outcomes. And we do think that the single-sample classification model can accurately identify the subgroup. And we are now trying to collaborate with future systemic therapeutic studies to see if we can validate this with different systemic therapies, and whether we could also use this classification model to accurately select the right candidates for further adjuvant or maintenance therapies. These are my acknowledgements. Importantly, the PIs that have made this data available for further research and also for developing the single-sample classification model. I also would like to acknowledge that Elai from Veracyte is adding this classification model to the Decipher Bladder Grid, which is basically the research-use-only report of the Decipher Bladder test. And in that way, we can further evaluate this model across multiple research studies to see if it actually holds up and could be used for selecting the right candidates for further therapies, yes or no. So again, thank you for the opportunity to quickly highlight the research, and I'll be very happy to answer more questions or simplify anything wherever you see possibilities, Dr. Kamat.

Ashish Kamat: So a couple of questions for you because this is really what we are moving towards, personalized therapy for patients, right? And we're using molecular markers. A lot of work that you've done, because you've really led the field... And for a long time now. It's just not something you're passionate about overnight. But tell us a little bit about how you think this would factor in with circulating tumor DNA and other parameters that we know are being used to de-escalate therapy for patients. Because clearly, we want to escalate therapy where it's needed, and de-escalate where it's harmful and not as needed. So, share with us your insights and some of the work that you've got planned.

Joep de Jong: Yeah. So first of all, I think it's an exciting question. Thank you very much. In detail, in the ABACUS trial that we evaluated here, we found the scar-like group where some of those patients actually had circulating tumor DNA. And we know from literature that having circulating tumor DNA presence is basically associated with worse prognosis because it's an indication of micrometastatic disease. But we saw that the patients, and there were only a handful of them that had circulating tumor DNA, but this favorable biologic subgroup still had the good prognosis line in the ABACUS trial. I think what we need to do in the future is actually combine tissue-based gene expression biomarkers, that we know from literature have been correlated to outcomes but also with therapeutic response, together with the presence, yes or no, of micrometastatic disease. So for example, in the neoadjuvant setting, before applying any of the novel therapies that are emerging, we can say basically, if this patient is luminal or basal, we know what kind of therapeutics in the future we need to apply.

Whether it's like ctDNA positive, yes or no, could basically be used to tell us how intensely we need to apply the systemic therapeutic regimen. So for example, my future wish would be if you're luminal, you have favorable biology. But if you're circulating-tumor-DNA positive, we need emerging therapeutics that actually inhibit the luminal tumor biology. For example, EV. If you're basically non-luminal, you're more immune-infiltrated and likely more to respond to pembrolizumab or maybe even one of the other PD-L1 inhibitors. Then based on the ctDNA presence, we could basically guide how much of the cycles we need to give and monitor response. So I do think that the benefit of these biomarkers becoming more studied, more evaluated by multiple groups, will guide us into how we could use them for therapeutic decision-making. But I do think they go hand in hand because they basically give different information from different aspects of the tumor. I hope it answers your question because I could philosophize about this with you for hours, but that's basically where I hope to see the field going.

Ashish Kamat: I'm glad you said that because I was thinking the same thing too. We could philosophize on this for hours, but I think the key thing is to recognize that these are all parts of the puzzle. It's not one that's going to inform the field completely. And of course, then there's always going to be the outliers where despite the tumor biology, despite the circulating tumor DNA, patients have their own outcomes and prognosis. So really, I think this is where not only the science, but also, I think the machine learning and AI, that again, you're doing a lot of work in, is really going to help us because there's going to be a lot of data, a lot of factors, a lot of predictive prognostic factors that we have to blend in. Thank you again for taking the time. Congratulations on the work, and see you soon.

Joep de Jong: Thank you very much. See you in a few days in Munich.