Read the Full Video Transcript

Ashish Kamat: Hello, everyone, and welcome to UroToday’s Bladder Cancer Center of Excellence. I'm Ashish Kamat, Professor of Urologic Oncology at MD Anderson Cancer Center. And welcoming us today is someone who really needs no introduction.

Michael O'Donnell has done so much for the field of bladder cancer and is constantly innovating and pushing the envelope. And he's here to share with us his data and insights, beyond just the abstract, on the impact of gene expression signatures of immune infiltration in response to sequential intravesical gemcitabine and docetaxel versus BCG in patients with high-risk, non-muscle-invasive bladder cancer. So, Michael, welcome, and take it away.

Michael O'Donnell: Thanks, Ashish. Well, thank you. This is the collaborative work of four different institutions, including the University of Iowa, where I am, but also Rutgers University, Erasmus Medical Center, and the Veracyte group. And as a background, we know that BCG is still the most recommended treatment for high-risk, non-muscle-invasive bladder cancer. But we're still facing global shortages, and alternative strategies are needed.

We pioneered the doublet of gemcitabine and docetaxel, intravesically, several years ago, and it's now become a potential alternative to BCG for BCG-naïve, high-risk disease, for which the availability of BCG may not be there. And I'm just showing, on the right, our early data looking at our patients that received either BCG—172 of them—or 138 who received Gem/Doce—very similar data sets. They were done not as a randomized trial but just to get a sense of how this treatment is performing. And as you can see, Gem/Doce does pretty well compared to BCG. All the curves begin to approach each other at about three years.

There is a trial underway that should settle the question, however, of which one is superior, or if they are, in fact, near equivalent. And that's the BRIDGE trial, being led by Max Kates at Johns Hopkins. But it may very well be the case that, for particular patients, one treatment might be better than another. And for this, we need biomarkers in order to make this kind of decision.

So in our study here, we took those patients—a subset, about 40% to 50% from each of the groups, the BCG-alone group or the Gem/Doce group—and we were able to obtain paraffin blocks for genomic interrogation. We were able to use a commercial assay, the Decipher Bladder Genomic Subtyping Classifier (GSC), which provides a clinical-grade, transcriptome-wide assay to tell us about the expression of the area of interest.

This allowed us to give a classification by the GSC and also assess consensus subtyping models, such as UROMOL, which is useful for classifying four different types of non-muscle-invasive bladder cancer. And then we used a particular feature of this called ESTIMATE, which stands for Estimation of Stromal and Immune Cells in Malignant Tumor Tissue Using Expression Data. What it basically does is give us a bird’s-eye view of what’s happening in the immune microenvironment in these slides from these patients.

To simplify, we established the median ESTIMATE score and then subdivided patients into those above or below the median. We then looked for an association between both the genomic subtypes and the ESTIMATE scores with the high-grade recurrence-free rate in these two populations of patients receiving either BCG or Gem/Doce. They were all high-risk NMIBC.

The summary of the results is given here. Both cohorts were very well balanced: a third of them were Ta, high-grade; two-thirds were T1 high-grade; and a small proportion had pure CIS. But among the papillary groups, 30% had some form of CIS. We observed the high-grade recurrence rate in the BCG-treated patients at 24 months was 36% versus 16%, which is similar to what we saw from the unselected groups. Again, not randomized, so I don't think you should put too much weight into that.

Interestingly, as expected, most of the GSC subtypes were luminal (85%), and 71% reflected the consensus luminal papillary subtypes. Neither of these were associated with clinical outcome. However, it was the ESTIMATE score that gave us some interesting new observations. As you can see on the far right, we have the two groups separated into the lower-than-median group, and the bottom panel is the higher-than-median group.

Let’s go to the higher part. What we see here in blue is the BCG group underperforming the Gem/Doce group. And even with these small numbers, it was statistically significant at p ≤ 0.02, with 90% of patients in the Gem/Doce group at two years being disease-free versus only 63% in the BCG group. However, at the lower median immune scores (ESTIMATE scores), there was no statistical significance. And, in fact, the Gem/Doce performed almost identically, with 86% at 24 months, versus 90% when it was in the higher median group. Importantly, this gives us a potential immune biomarker of response.

At least from this high-level view, patients with a higher immune score—likely indicating subsets of immunosuppressive cells—seem to have a suboptimal response to BCG, whereas Gem/Doce appears agnostic to the immune subtype score. This is probably not surprising, considering that Gem/Doce is more of a chemotherapy-based approach rather than an immunotherapy-based one. But the real take-home message isn’t just about these particular results—it’s the idea that, although these two treatments may be similar in overall clinical outcomes, there may be identifiable subsets of patients who respond better to one therapy over another.

Ashish Kamat: Thanks so much, Michael. As always, you distill a complex topic into very understandable, bite-sized chunks of information. So, again, as you said, the take-home message is clearly that it helps us identify subsets of patients that might do better with one treatment versus the other. And that’s the Holy Grail, right? We have so many different treatments. We want to be able to personalize therapy for patients.

For so many years, we've been relatively empirical, using clinical data, grade, stage, etc. And this gives us a little bit of a clue about separation, or, in some ways, predictive ability. But that’s what I wanted to ask you—do you think this ESTIMATE score is prognostic or predictive, based on the nuances of what you've just shown?

Michael O'Donnell: I think it's more likely to be predictive because Gem/Doce almost acts as an internal control, showing that, with similar sub-characteristics, these groups are almost identical. Yet, if it were prognostic, we would have seen the two groups behaving the same. But Gem/Doce remains relatively constant. It's only the BCG group that seems to be affected.

Ashish Kamat: And why is that? I know you can't make much based on immune scores, et cetera. But why do you think that the immune score is predictive of one treatment and relatively prognostic for the other?

Michael O'Donnell: Well, although it's tangentially related, we know from studies of BCG-unresponsive patients that they seem to have a signature expression demonstrating an immunosuppressive signature. That is, there are more Tregs, MDSCs, and M2 macrophages. Suppressive molecules are subtypes of cells that can turn off what should be a productive cellular, immune, anti-cancer response.

And so, it's not surprising that some patients, even in the naïve form, may, in fact, already have an unfavorable signature that predisposes them to be less responsive to BCG. As we know from years and decades of BCG clinical trials, there seems to be about a 30% drop in response rate in the first six months. And we're seeing an absolute 30% difference in this group as well.

So we may very well be looking at the same thing. There’s also some data, for instance, in the BCG-unresponsive or BCG-exposed space, from Eugene Pietzak at Memorial Sloan Kettering, showing that when they alternate BCG and gemcitabine in patients exposed to BCG, they're getting a very high early complete response rate—close to 94% at six months. This indicates that there may be a way to use this knowledge to change the immune microenvironment to make it more favorable, to prevent BCG unresponsiveness or to rescue BCG-unresponsive patients.

That, I think, is going to be the real Holy Grail. We’re going to need to do more sophisticated analyses, such as spatial transcriptomics and immune phenotyping of these cells, to figure out exactly how this is happening. But this gives us a little clue that we might be closer to scratching the mechanistic surface than we have been before.

Ashish Kamat: Yeah, absolutely. I think spatial transcriptomics and whole-exome sequencing, as well as genomics—I mean, the whole proteomics field and metabolomics as well—are likely going to give us clues. Mike, a couple of questions. So you had recurrence rates, right? I'm sure you looked at it. Was there any effect on progression rates? Because you had a high number of T1 patients?

Michael O'Donnell: Right. Well, our progression rates in this trial, even at three years, were under 5%. They were better in the Gem/Doce group—98% progression-free—than in the BCG group, which was, I think, 94%–95%. But the numbers were too close to really make any sense of it. We would need longer follow-up with much larger patient populations.

Ashish Kamat: Yeah, and I think that’s key, right? I mean, responses to all therapies have improved because we’re better at detecting tumors and recognizing them. You do a whole extensive work-up each time, which we’ve talked about as well. So, luckily, our patients aren’t progressing like they used to, 15 or 20 years ago, right?

Michael O'Donnell: Right. No, it’s allowed us to begin to look at bladder-sparing therapy in a safer environment without fearing that the patients are literally going to be dropping out with bad outcomes early on.

Ashish Kamat: Right, right. You and I could talk about bladder cancer forever. And we do—every time we meet. But, in the interest of time, let me just ask you one last question. Taking a step back and looking at these sorts of studies that you’re doing, and also considering AI-based pathology, clinical nomograms, et cetera—where do you think we are heading? What’s your most favored horse to win the race for being predictive or allowing us to tailor therapy for patients?

Michael O'Donnell: Well, it comes down to a pretty familiar catchword, which is AI. We’re going to have sophisticated enough programs that can integrate the clinical data, digital pathology, immune phenotyping, and other aspects of it that we don’t even know about yet—things we can’t see or feel. But the computer can ascertain them from a raw data set. And I think we’ll get closer and closer to achieving a much more predictive capability.

Right now, we’re probably, at best, 70%–75% predictive. It’d be nice to get that into the 90% range. And then it would change all of the outcomes in our favor. We would know who we shouldn’t treat with a certain agent, and who would be much more likely to respond favorably. And I think that’s the excitement for the future. That will allow for the personalization that we’re all looking for. And maybe it will also allow cost-effective treatments. Everything is a plus, plus, when we know more about it.

Ashish Kamat: Exactly. Michael, always a pleasure. Thank you for taking the time. See you soon.

Michael O'Donnell: Thanks, Ashish. Thank you. And thank UroToday for their assistance in putting on this program.