Ashish Kamat: Welcome to UroToday's Bladder Cancer Center of Excellence. I'm Ashish Kamat from MD Anderson Cancer Center in Houston, Texas, and it's my pleasure to welcome a friend and a colleague, and a true expert in the field, Professor Lars Dyrskjøt, from Aarhus University Hospital in Denmark. As many of you know, Lars is an expert in the field of molecular medicine and also holds an appointment in clinical medicine, and a lot of the work that he does is essentially at the border and the transition between the hardcore, multi-omic analysis type work and the actual clinic. So, it's a real pleasure to have you today, Lars, to educate and take our audience not only through the field in general but then also update us on your latest publication, which is the molecular correlates of cisplatin-based chemotherapy in muscle-invasive bladder cancer. So, Lars, the stage is yours.

Lars Dyrskjøt: Thank you very much, Ashish, and thank you for the invitation to present our latest work here. So, I would like to start out by sharing a little bit of background of what we've been doing first. So, cisplatin-based chemotherapy is recommended in both the neoadjuvant and the first-line setting with response rates up to as high as 50%. Actually, no robust predictive biomarkers for chemotherapy response have entered into clinical use yet, and studies have identified some lacerations in the DNA damage response machinery to be associated with good response, but replication studies are really needed in this setting. And overall the inability to predict which patients would respond to chemotherapy, present a major clinical problem, as significant overtreatment of patients not responding is currently performed, and what we're doing now in the clinic.

First, a little background of cisplatin-based drugs and cellular effects. So when cisplatin enters the cells, it binds chromal [inaudible] and the DNA informs intra inside cross-links, and the cisplatin DNA binding causes various cellular responses, such as replication arrests, transcription, inhibition, cell cycle arrest, DNA repair, and apoptosis. Actually, the level of DNA damage and repair efficacy determines the cell fate, consequently, when multiple DNA damage is present since we'll undergo apoptosis. So if DNA damage repair pathways are impaired by, for instance, ERCC-2 mutations that we have seen reported heavily. In bladder cancer, cells were more frequently killed by cisplatin. So that's the overview of the cellular effects. There's also a huge effect from the immune system also when we treat patients with these cytotoxic drugs. So conventional drugs can activate anti-cancer immune responses through different mechanisms and that's, for instance, through direct stimulation of T or B-cell responses.

So I'm not going to go that into all of the different things we see on this slide here, but it's becoming more and more clear that the immune system plays an important role in chemotherapy response. And also recently, as some of the studies that have been showing some effect of the [inaudible] pathways is also very important to say. So it's very important in future studies that we include analysis of the immune system also when we are looking into chemotherapy response pathways.

So a little bit of background of what has been found so far in the bladder cancer field that I'm going to start with some gene expression, subtypes studies, and chemotherapy efficacy. So this is a study from the recent consensus, MIBC classification, where we actually showed a significant difference in survival between these six different molecular subtypes we have in muscle-invasive bladder cancer.

But what does it look like if we only look at patients treated with neoadjuvant chemotherapy? When we do that, we actually don't see any significant difference in survival for these subtypes. So they're not predicting an outcome that way. And also, if we look at the pathological response and pathological downstaging, as we see it to the right, we actually don't see any significant difference when we compare the different subtypes and the response to treatment. That's been investigated in other studies also where, for instance, in the P53-like subtypes reported some years ago now there was identified a significant difference between responders and non-responders, but it could not be replicated in further studies. There was another study by [inaudible] reported a couple of years ago also when they did find that the basal subtypes showed differential survival between different clinical cohorts, but there was no significant association between the different subtypes and pathological downstaging in that study either.

The different DDR mutations and chemotherapy efficacy has been intensely analyzed for an element calling they identified as ERCC-2 mutations to be associated with treatment response. As we see in this cohort here, and we have the respondents here and the ERCC-2 mutations here, and that was reflected in the overall survival for this cohort also. And it was actually validated in the follow-up study also in 2016 in an independent cohort. So in this study, they find, and in other studies that find that ERCC-2 mutations aren't independent or, I would predict, as of treatment response.

This is an overview of the different biomarkers and chemotherapy efficacy that has been reported in the literature. And what we see is that a few of these biomarkers have actually been replicated in independent series, and for instance, it's striking here. Also, we see that ERCC-2 is significantly correlated to a pathological downstaging response in two different cohorts, but in another cohort here, we don't see any significant correlation to pathological downstaging. So there's room for improvement and replication here.

So this leads me to our recent study, where we tried to take a deeper dive into this and the different mechanisms associated with cisplatin-based chemotherapy response. We investigated several and molecular layers in this study, DNA using exome sequencing and SNP array analysis, the methylation analysis using microarray analysis and RNA transcriptomic analysis using RNA sequencing, and then finally protein analysis. So I've looked at in immuno-profiling and spatial organization of the different immune cells using multiplex immunofluorescence analysis.

So we did a single-layer analysis and then tried to integrate some of these layers into some models of treatment response. Then we compared these different correlations to our detailed clinical information on response and outcome in this study. So this is an overview of what we did. We included the total of 300 patients in the study. About half of the patients receive neoadjuvant chemotherapy, and the response was pathological downstaging. And then the other half received first-line chemotherapy, and they were treated based on the detection of metastatic disease. So we had a complete response of progressive disease estimates for this part of the group. Also, one other thing that's important to note is that we didn't have a complete overlap of the different omics layers in this study. So it's 300 in total, but the layers did not overlap completely, so we could not perform a complete integrated analysis for all molecular layers simultaneously.

So first, we set out to look at the results from the DNA analysis, as we can see here. So this is an [inaudible] plot showing the most frequent immune significant mutated genes. And we actually see ERCC-2 among these genes. However, in this study, we actually did not find ERCC-2 mutations to be associated with treatment response. Here, we saw the tumors based on the mutational signature composition, as we can see on the top. And they could actually be grouped into a signature five mutational signature and an APOBEC associated mutational signature. So the SPS5 or a signature five mutations have been correlated associated with ERCC-2 mutations earlier. And what we found here was that when we coordinated these two subgroups to outcome and response, we could actually see that the signature five tumors were significantly associated with the treatment response here.

And we looked further into this to see what could actually drive this signature five mutation pattern. And here we found that tumors with the ERCC-2 and BRCA mutations actually had a higher number of signature five related tumors, as we see in this figure here to the right. Interestingly, the BRCA tumors mutation tumors showed an elevated response compared to wild-type in our data set, as we can see here. And we saw a similar trend in the TCGA data, but there were fewer patients where we could actually identify that they had received cisplatin-based chemotherapy. So there is a tendency.

In our data we also found that tumors with many indels and a high level of allelic imbalance showed a better response as we can see here. And when we looked into the gene expression subtypes from RNA sequencing in this cohort here, we could identify four of the six consensus subtypes in muscle-invasive bladder cancer, as you can see here. And when we compare these different subtypes to treatment response, we could actually see that the non-patient squamous tumors were responding better to chemotherapy compared to the patient squamous subtype.

This is reflected in this Kaplan-Meier plot here showing that the patients with squamous tumors have worse survival compared to the non-patient squamous patients, as we have seen in other reports also. Finally, we've performed some multiplex immunofluorescence analyses to investigate the presence and composition of immune cell infiltration in this cohort. So we looked into T-helper cells, toxic T-cells, T-regs, B-cells, different macrophages, and then we stain for PD-1, PDL-1, and MHC also.

So by digital image analysis, we separated tumors into desert subtypes, excluded and infiltrated immune subtypes as shown here, to the right. Here, you can see the desert subtype and this desert subtype we actually found to be associated with a specific poor survival compared to the excluded infiltrated subtypes.

We also found that day decreased expression of PD-1 was also associated with a poor outcome in this study here, as we can see here. So again, indicating that when PD-1 is expressed, we have immune cells in the tumors, and then we have a better response to treatment.

So to try to integrate these predictors, we looked at tumors where we had overlapping data on DNA and RNA information only. And so based on the DNA information, we assigned tumors to high and low genomic instability. And after that, we added the information on RNA subtypes, as you can see here. This way, we could actually demonstrate that the extreme groups with high and low response rates with tumors with high genomic instability and a non-patient squamous subtype that actually had the best responses. You can see here with the 80% response by the other extreme, low genomic instability, and patient squamous subtype had a 25% response rate. So now we're down to small numbers, and this is just an investigation into what is possible when we try to integrate several layers, but we see some significant difference here.

So to conclude, we found a number of predictors of response to chemotherapy. It points towards several of these molecular processes, and that there are similar molecular processes and a single biomarker to predict this may not be adequate for clinical use. So we need validation studies for other predictive biomarkers and biomarkers found in this study, optimally, where all of these molecular layers and biomarkers can be analyzed simultaneously to have a better view into this. We are actually doing a validation study here in Denmark right now, where we are investigating these predictors in an ongoing prospective observational study. As you can see here, it's being done in the framework of this TOMBOLA trial we have ongoing where we subsequently treat patients with immunotherapy upon detection of circulating tumor DNA, but we assessed the predictors of chemotherapy response here in the beginning of the disease causes of these patients going into this clinical trial to see if we can validate our biomarkers. So this work is ongoing. So finally I'd like to thank, especially, the first authors on this work, Ann Taber, Emil Christensen, and Phillippe Lamy who did a lot of this work. So thank you very much.

Ashish Kamat: Thank you so much, Lars. That was really very informative. And thank you for covering a lot of information in a relatively short time. Certain questions come up when people look at publications such as yours that are really rich with such data, but then, as you mentioned, there are disparate results from different publications. So, with your immense knowledge of the field, what would you advise a clinician today that is seeing a patient with muscle-invasive bladder cancer? What tests should they be looking at? What should be considered in some ways useful for clinical decision-making and what should be considered purely research tools?

Lars Dyrskjøt: I think what I've covered here now is purely research tools. So as we can see in different studies, we see some of these predictors pointing in different directions, and we're not always finding that, for instance, ERCC-2 mutations are associated with treatment response. So I think we need to look much more into this and do larger studies and more control studies because I think what our study here points towards is that there may be several mechanisms associated with response, and we need to look at different predictors in order to actually robustly predict a patient's response to chemotherapy.

And I think one problem may be, so we know from different studies and cell models that DNA damage response genes is important if patients should respond to treatment, but I think we have a little bit more shallow knowledge of what is going on when we look at the tumor biopsies also, because some of the inconsistencies we may see maybe because of different ways, we call the mutations. It usually needs to be deleterious on serrations receiving other protein as being the functional damaged. It could also be that these mutations are subclonal. So if they're subclonal, we would not expect to see a huge effect on the [inaudible] tumor. So I think we're still far away from clinical use here, but we need to investigate it in prospective clinical studies and trials also to get some robust data of this.

Ashish Kamat: Yeah, and I think the point that you raise is very important because, as our audience recognizes the PD-L1 status of tumors was supposed to be the holy grail for IO response, but it clearly is not, because now we have sufficient data to suggest that patients will respond and can respond despite their PD-1 status, and it depends upon which assay you're using to measure it. And that's similar to a lot of the subtyping data and RNAC and DNA mutation. So having a multi-omic analysis such as yours at least provides a level of comfort that you're interrogating multiple different pathways and different assays and providing possibly a more holistic approach to what's happening at the tumor biology level. With that in mind, Lars, and in the interest of time, I'll ask you one last question, maybe. When it comes to these sorts of data and these sorts of analyses that you've done, do you see these as being complementary to clinical data, or is the hope that one day we will be able to actually just profile the tumor and get the answer irrespective of the clinical data?

Lars Dyrskjøt: So I think it's probably going to be a combination, but of course the holy grail would be to make an assessment based on the primary tumor before you start treatment. If we can predict that this patient with high profitability will not respond to treatment, then, of course, that would be ideal. And then you could go to other treatments and you can spare the patient from unnecessary side effects from chemotherapy. The question is just if we'll reach some levels where it's adequate for actually making this decision. So I may right now be a little bit skeptical about this because of the high level of tumor heterogeneity we see also. So I think we are going to have some predictors that point in some directions, but as long as we don't have a multitude of different treatments to offer these patients, that may also be difficult.

I think another thing is that we've been working on in the field of circulating tumor DNA also. And I think that's a combination of tumor centric analysis that we've been performing here and then circulating tumor DNA handled. So this may actually be an optimal way of doing it because then we can track the CT DNA through the first cycles of chemotherapy. And if we can see that the tumor's not responding to the treatment, then maybe the treatment should be stopped and other therapeutic options could be considered. So I think it's going to be some clinical and tumor centric and longitudinal analysis that may be optimal for predicting this.

Ashish Kamat: Yeah, great points. And in fact, that's a perfect segue because I would love to invite you again to talk to our audience about the longitudinal studies that you've done and the work on CT DNA and those sorts of things. And we will certainly do that. In closing, I do want to thank you again, Lars, for taking the time from a busy schedule to share this data with our audience. Stay safe and stay well.

Lars Dyrskjøt: Thank you for the opportunity. Thank you.