Molecular Subtyping of Bladder Cancer: Ready for Prime Time? - Eva Comperat & Hikmat Al-Ahmadie

September 8, 2020

Eva Comperat and Hikmat Al-Ahmadie, MD, join Ashish Kamat, MD, MBBS to challenge the use of immunohistochemistry, or traditional pathology, versus molecular subtyping in bladder cancer. Dr. Comperat discusses immunohistochemistry, explaining it is more precise, and because it is standardized and more widely used than molecular subtyping, it remains the gold standard. Dr. Al-Ahmadie continues by detailing the shortcomings of relying on pathology, taking the position for molecular subtyping. He notes that molecular classification can provide additional, in-depth information on tumors grouped by their morphology, and argues that while molecular subtyping is currently more expensive, technological advancements may make it more widely available in the near future. To end their discussion, Drs. Kamat, Comperat, and Al-Ahmadie consider the evolving staging and treatment landscape, and the future of molecular subtyping.


Eva Comperat, Head of the Department of Pathology at the L'Assistance Publique-Hôpitaux de Paris, Hôpital Tenon, in Paris, France

Hikmat A. Al-Ahmadie, MD, Pathologist, Department of Pathology, Memorial Sloan Kettering Cancer Center, New York, NY

Ashish Kamat, MD, MBBS, President, International Bladder Cancer Group (IBCG), Professor of Urology & Cancer Research, MD Anderson Cancer Center, Houston, Texas

Read the Full Video Transcript

Ashish Kamat: Welcome everybody to Uro Today's Bladder Cancer Center of Excellence. I'm Ashish Kamat from Houston MD Anderson Cancer Center. It's my pleasure today to welcome two giants in the field of pathology, Dr. Eva Comperat, and Hikmat Al-Ahmadie. Eva's calling us right now from Austria, but she does most of her day job at the Sorbonne University in Paris. And Hikmat's calling us from New York City where he is GI pathologist at Memorial Sloan Kettering.

But the purpose of this, today's debate, each one of them have been given one particular side to take. Obviously they will present the pros and cons of each. But for the purpose of the debate, Eva is going to tell us why she is believing that immunohistochemistry, or traditional pathology, trumps molecular subtyping, and Hikmat will then take the other side, and then we'll have a nice discussion at the end. With that, Eva, take it away.

Eva Comperat: Thank you so much for inviting me. It's a real pleasure to be with you. I will tell you as a real pathologist, why I really believe in real pathology. When we talk about subtyping, actually we should really all talk about the same things. Because we can talk about subtyping and going on the clinician side, who will say, "Well, there's non muscle-invasive bladder cancer, and there's muscle-invasive bladder cancer." You can also see it from a pathology point of view, and you say, "Well, there are types of course of these." You will see real carcinomas, but they're also variants and they really start since, I would say 10 years nearly now, to play a really very important role with all these kind of nested carcinoma, microcystic, micropapillary, as you know.

Of course, you can talk about molecular, but molecular from my point of view, is very, very complicated. As you can see, only on this very small slide, it is something which can be quite confusing. If you go to the real pure pathology, you see that we have a very good standardized, extremely well-standardized system with the WHO, but also with the CAP, we're reporting, ICCR. So these are standardized reporting, which tells you exactly what you have to do as a pathologist. And they also tell you about sub-staging. This tells you about how to report, if you have, for example, variant histology, and so on. So if you keep reading very closely to these kinds of recommendations, international recommendations, you cannot really get it wrong.

Of course, there's one very, very important point. This is that very many of these tumors, as you can see here, you have squamous, you have micropapillary, you have microcystic, you have nested. These tumors can look very much the same if you're not very experienced. Probably, of course, everybody should be more or less experienced in what he's doing, especially in the bladder. There have been some studies showing that if you have just a resection, a TURP, on the bladder, counts, and afterward you have the cystectomy, there are very often kind of discordance. Some of the clinicians of course have asked whether it is really enough to do only resection.

Then as you can see, there are some variants which are easier to recognize than others. So for example small cell, it's pretty easy to recognize. The squamous is pretty easy to recognize. On the other hand, micropapillary is something which is quite difficult probably to recognize. We'll come back to this a little bit later.

We also did a very small study on this, and we could show that if you do resection, very often you have the same subtype variant also in the cystectomy. Nevertheless, of course, there are about 20% of cases where you have discordant findings. So this is also one of the problems in bladder cancer because bladder cancer is a very heterogeneous lesion. So of course, as much, if you get more material, the more you will be precise.

As I already told you, there's a problem of course, with this interobserver variability, which is a huge problem. There has been a very nice paper in 2014, where 14 pathologists, really very well known people, had to type micropapillary carcinomas. And as you can see, as long as they are very typical cases, you have an agreement of 93%, but on the left side, you can see two pictures. And I'm not sure whether I would say micropapillary to these carcinogens.

You can see if they are a little bit atypical while the agreement really decreases very, very quickly. This is problematic because there has been a very nice paper from Ashish, which could show that if you have a micropapillary urothelial carcinoma, and you have to see, so clinical T1 tumor, and you make an upfront cystectomy, you will save all of these patients. On the other hand, if you give some BCG and the other treatments that will get progression that will have lymph nodes with metastasis and so on. So really of course, on the shoulders of the pathologist, there's a very heavy responsibility, because it would get the subtyping, the variant wrong. We do not recognize it at the very moment. Then of course the clinician will not be able to treat the patient in the best way.

The second thing is, which is in my opinion very important from the pathological point of view, is the sub-staging, which is really very useful in T1 tumors especially, but also it's very important to do the right staging, which is not always so easy. I mean, everybody thinks that staging is an easy issue. You can have very important issues between TA and T1. You can have very heavy issues between T1 and T2 too.

So if I just go a little bit to the molecular subtypes, that one of the very first papers on this was from Sjödahl, and he said you have five major molecular subtypes, and he described them very lovely, and he said it is completely independent from pathology. And then you go to the results and you can see it's not true, it's not independent at all from pathology. As you can see, you have very many urobasal A tumors, which really did very well, but they were mostly all low-grade pTa tumors. So staging is a very important issue if you do pathology and it can give you very, very important information.

And if you go a little bit further, the further you go your model, you see that all these kind of molecular sub-stagings become more and more complicated, and they're even more complicated that what we can do in normal pathology. And it's really getting more and more complex as a matter. So I'm not sure whether these complexity is easy to treat in the daily practice.

And there has been some papers once again, from students who said, well, if you have muscle-invasive bladder cancer, you need 13 markers, immunohistochemistry markers. And with these 13 markers, you can say, what are this? Is it urobasal, or genomically unstable, or whatever tumor. But the problem, and he says it himself in the paper, is you have accuracy of 75%, which is probably less than you have with a very good pathologist. And the second problem is, where do you put your cut-offs? Because very often you have a tumor which will express a little bit GATA3, but which will also express a little bit p63, where will you say, well, this is the cut-off? There's no cutoff, which is really officially fixed by any pathology organism in the world.

There have been, of course, other attempts from the molecular side to say, well, we try to compare well-defined pathology variants or types, then we look at the molecular side. And you can see, of course, a conventional Espinosa, very heterogeneous lesion, squamous, as you can see, we really recognize very well in pathology. Then of course there are tumors like the micropapillary or the plasmacytoid tumors, which seem to be pretty luminal tumors. So of course, then you can say, what was the state that if you recognize the micro popularity or the pesticides, which tumor anyway, you don't have to do probably any molecular code and molecular trials because already these are luminal tumors and they're not basal tumors.

And this is the same paper. You can see each line is one patient. When you have a closer look, you can see that very many patients have different tumor subtypes, variants. This underlines, once again, the problem of the heterogenous bladder tumor. And I think we can pretty well sort it out with good sampling in a tumor saying, well, you have so many percent of squamous tumor, we have so many percent of micropapillary tumor. And then the clinician who has the patient in front, of course, he will be able to say, well, I will treat him like this, or I will more or less treat him like that. Because this is an old patient, this is a young patient, he will do better, and so on. Furthermore, there was this paper from Kamoun and coworkers, and I was not very convinced of this paper because it was supposed to give a statement and say, well, these tumors, basal and luminal, and you have to treat them like this and like that.

But nevertheless, it is not as robust as it seems to be. It was quite problematic also to publish this paper after what I have understood. Because once again, the problem was the cut-offs — because these tumors, even on the molecular side are not always black or white. Sometimes they're mixed features, and you don't really know in which group you should put these tumors. I think from a pathologic point of view, you're much more precise when you say, well, you have 10% of nested carcinoma and you have 20% of microcystic tumors, and so on. So I think we can do as well, at least as well, as molecular. And so the more the molecular cannot tell you whether you have a lymphovascular invasion, can not tell you where the carcinoma inside you is uni or multifocal, is extensive or not. It cannot tell you all the variant histologies, and it cannot tell you the stage of the tumor. I think these are really issues which are recognized, and everybody knows that these are very important issues.

So in conclusion, I would like to say that subtyping of the bladder by pathology only is still the gold standard. It is pretty quick. It is pretty cheap. It's very precise, and it's extremely standardized. On the other hand, we have the molecular, we miss probably some important information like the lymphovascular invasion. It's pretty expensive. Not everybody can do it. And I think pathology should be something which can be done, by everybody in the whole world on the same level. It's not always available to do some molecular, it's not easy to achieve. The cut-offs are a little bit the problem, and there's no 100% agreement on how to handle. Thank you.

Ashish Kamat: That was excellent, Eva. Thank you so much for providing so much information in a very short period of time. Hikmat, if you can share your slides, and present to the audience your perspective on this topic, that would be great.

Hikmat Al-Ahmadie: 
Okay, great. Well, thanks Eva for the beautiful presentation. And first of all, thank you for the opportunity to participate in this interesting topic that's very relevant and timely. Again, I'm pretty partial to pathology because that is how I trained, and I still see the value in applying it. On the other hand, I share some of the shortcomings that we know exist. And from here, I wanted to highlight some of these shortcomings and how a molecular classification approach can help overcome or enhance what we can do in our pathology material.

So we know urothelial carcinoma is a heterogeneous disease at the clinical or the morphology and the genomic level. And we know at the morphology level, as you can see in this list, we recognize many histology variants that rely on unique and specific morphology creature, that was mentioned by Eva earlier, that can help us to classify these tumors.
But if you take these histology variance aside and you look at the most common subtype of urothelial carcinoma, which is a tumor that doesn't have any recognizable features to place it in one of these established entities, we call it urothelial carcinoma, not otherwise specified, or NOS.

And you can see from these three examples, these are three different tumors that are urothelial carcinoma NOS, that have different morphology appearance. So in a way you're forcing three tumors that have morphologic differences into one entity that basically goes against the basic principle of morphologic classification, which is relying on the morphology. This tumor, something like this category, may benefit the most from, actually, introduction of molecular classification that can provide more in-depth information, way beyond what a view pathology can provide. Going back to the variant histology, despite our best efforts and the morphology criteria that have been established for quite some time for these entities, there is still a significant lack of agreement and concordance, in identifying these entities.

As you can see in these two examples here that I've been sharing with you, either by having cases submitted to another center pathology review, or by re-reviewing cases within the same institution. Anytime you do a re-review, you're bound to identify more tumors with a divergent differentiation that have not been called initially. Which in a way, in essence, reduces the significance of these variants, if they're not always called every single time, and necessitates expert re-review, which may not be always available in every part of the world.

Now, if you look at this heat map from the TCGA cohort of the 412 muscle-invasive bladder cancers, that is pretty much familiar to everyone, one can see how powerful this analysis can be. You can see this analysis identified five distinct groups, separated by different differentially expressed genes, and major pathways that are related to different levels of differentiation at the urothelium, or squamous or neuronal subtypes, but also they provide insights into the components of the microenvironment, such as smooth muscle cells, fibroblasts, and immune microenvironment.

If you look back at the morphology, you can realize that these subtypes, for the most part, recapitulate many important morphology findings. For example, you can see high prevalence of the papillary Babylonian morphology in tumors that have enrichment with FGFR3 alterations. And you can see near-exclusive squamous differentiation in tumors that have been classified into the basal squamous subtypes.

The same goes with the neuronal subtype, which I identified the four cases with neuroendocrine differentiation. But beyond that, this classification allowed us to identify more tumors with similar molecular profile, but do not have the specific morphologic features that would place them into a morphology category. And that goes, as you can see, nicely examples of basal subtype that did not have squamous differentiation, and neuronal cases that did not have neuroendocrine differentiation. Again, probably showing that the light microscopy by itself may not be able to identify more of these cases that would fall into a unique molecular subtype, and this way such molecular classification might benefit particularly these tumors.

And the way you'll be able to overcome some of these problems is by actually testing more, which can increase always the power of the analysis. This is the paper that looked at, that combined all the databases of previous molecular classification tumors of urothelial carcinoma, and re-analyzed the samples by the bladder cancer molecular taxonomy group. And what this paper showed that the subtypes were further expanded and refined. And, that can also provide more insights into the biology, such as what pathways or genes that are more likely to play a significant role in these subtypes. And it can also provide different types of certifications, such as outcome, as you can see, in the Kaplan–Meier curves here. Also, it can give you insights into what type of treatment one can devise for patients, based on the molecular profile of their tumors.

Now, we are familiar with the criticism related to bladder cancer, which is the molecular classification, urothelial carcinoma, which is the heterogeneity of the disease. Depending on what component of the tumor you profile, there's a chance that you'd be able to assign a different subtype, which is evident in this paper that was discussed also earlier by Eva. That is dealing with primary tumors, but also similar problem could have been if you're dealing with tumor primary metastatic pairs, where you can see some discrepant cases.

But looking at these papers in more depth, you can see in the first, in the primary tumor setting, there is still actually some concordance, not all cases were discordant, so that's one thing. The other thing is when you have more than one area of those same tumor showing different molecular profiling, it's no different from saying this is a urothelial carcinoma with squamous differentiation. And what you need to do is basically test more of the tumor, different areas of the tumor to be able to identify the full spectrum of how the tumor can look like.

And when it comes to the primary met pairs, the authors here with this nice paper, when they found discordant cases, and went back to the primary tumor and map that primary tumor, they were able to find areas that could be matching or concordant with the discordant metastasis. And that in a way could be very helpful actually, when you look at it in a different way that by doing more of these cases, you'd be able to identify areas of the tumor that might be actually the source of metastasis, and can help you prospectively identify these cases in the future, and devise a better outcome and risk stratification for patients.

Another criticism is, you have different classification system from the time this started and use different names into these classification systems. But while that is true, the way you can look at it as well, that this is still an evolving field, and it's still in its infancy, and naturally, the more you test, the more information you've got, and the more details you'll get, and you'll be able to more refine the classification and make it more consistent. And this is exactly what happened when we started with the earlier systems, fewer cases existed. When you have more cases, you'll be able to actually produce more subtypes, and produce more consistent results. And even with this, there is still a good level of harmonization and concordance among these subtypes, even if you look at them just as they were from the first classifications.

Other things that can be used against the molecular classification, and as was mentioned earlier, is how costly it is, and how it might not be available everywhere. While that might be true now, we all know that it's all a matter of time before more technology advancements are happening, more competition is happening in this field that can help drive the cost down and also help make this testing widely available in less developed parts of the world. So that's one of the issues that can help overcome.

The other thing is, if you're looking at histology only, that's one thing, but the minute you start adding other ancillary tests to any histology evaluation, the price, or the cost of that analysis, will significantly increase very quickly. Be it multiplex immunohistochemistry, or any other like FISH or any other molecular technique to help you characterize these tumors.
The other thing is, we acknowledge the lack of agreement, or the interrupts over disagreement, an issue, important issue of urothelial carcinoma, which most of the times, expert pathologic review is required for proper classification of these tumors. But we know that expert pathology review is not always available and it's not practical for the most part. So these are issues that will argue against the pure histology classification.

Of course, there are important issues that were mentioned, and I'll read, which is lymphovascular invasion, sub-staging of these tumors, assessing of margins. While some of these are relevant histologic findings that we mentioned, their actual clinical value is still debatable, or is not completely solved, particularly vascular invasion, for example. But for all of this, it could be that all that is needed is more testing of cases with the focus on this particular parameter. And maybe we were able to find a molecular signature that can help identify cases with, that are more likely to have lymphovascular invasion. There may be another technology that will help us identify tumor at the margin, and the staging as well. Maybe some signatures may help you correlate with staging, better than the systems that we have, and maybe even more consistently applied than morphology parameters.
So other things that would be needed for, to help improve the molecular classification, is the technology that we're using. We have so many techniques that are becoming available and if they're well refined and improved, it would be a matter of time before they become implemented at a wide scale. We know that there are some studies showing single-cell analysis can address subpopulations within the same tumor and address heterogeneity. There are more sophisticated techniques such as spatial transcriptomic, and proteomic, where you can provide insight to relationships with different components of the same tumor on the same slide. We know the applications of the digital pathology and artificial intelligence, where there are some studies showing if you develop the right algorithms and link them with molecular features, by using the basic morphology features from a slide, to identify specific molecular abnormalities.

So these are all things that are happening at a small scale now, but with more work, they can be easily become widely applicable, to be used at a very large scale.
So in conclusion, molecular classification will continue to improve and will be refined. And while histology has served us very well for a long time, I think it will soon undergo a major overhaul that will reshape and redefine its role. And the more we incorporate molecular testing into the clinical practice, the more it will help us understand the diseases better. We'll provide more consistent classification and provide even more helpful and important information that can help guide specific and personalized treatments for these patients. Thank you.

Ashish Kamat: That was great Hikmat. So yeah, this is one of the problems in some ways, when we have two such experts, such as you and Eva, the role of the moderator becomes minimized because you guys covered pretty much everything that I was going to ask you, but let me ask you some things anyway.

So, if you focus on the molecular subtyping versus immunohistochemistry at the level that the two of you are, clearly the points that you raised in the debate are very valid. But let's translate to, for example, the average person in the United States, the patient, the doctors, that are being seen in the community. These labs, and even some universities in the U.S., don't have access to the latest technology, or if they do, it's only for the research lab. So in these situations, how do you propose the GU pathologist, or the clinician, should triage which tumors or which patients are selected for molecular subtyping versus pure immunohistochemistry. Eva, if I could have you start with that question, how would you triage these patients? Who should go for routine immunohistochemistry, and who should go for molecular subtyping to get the additional information that's important?

Eva Comperat: Well, this is a very tricky question, because really most of my tumors are really sorted out with normal pathology. If I have doubts for example, about a neuro-endocrine component, which is quite important to recognize, for example, of course, I do some immunos. But it's always immuno, this is written, the recommendation is always based on morphology. So if you do not have the morphology, don't put any immunos just on the tumor, because you can't afford it. This is the first thing. The second thing is probably that if you have really very tricky tumor, which is really very heterogeneous and mixed types, probably these guys should really get some molecular testing, to be sure that you really test the right component of the tumor. And maybe you see something more with molecular, which you do not see with the standard histology, or immunohistochemistry probably. But afterward, I would say that 90% of my tumors are really sorted them out with just pathology only.

Ashish Kamat: 
Hikmat, what about in the U.S., from the U.S. perspective? What would you advise our practitioners here in the US that may not have access to subtyping techniques or technology? Should they send these tumors out? Should they rely on morphology alone?

Hikmat Al-Ahmadie: As I mentioned, and as Eva mentioned, morphology still provides a great deal of information, that should be sufficient for a large number of patients and proportion of patients. But that is mostly because our standard treatments now are based only on morphology. We have not incorporated what molecular classification can do. And maybe if the standard is molecular classification we'd be talking something completely different.

But, for the sake of this discussion, yes, most of the time, if a patient is healthy and no other comorbidities, maybe then you just apply them with the standard procedures, regular histology, irregular morphology, any subtypings, and you can do it. But then when you start dealing with a patient, for example, in bad renal function, and then chemotherapy may not necessarily be the best option for them. Can they benefit from something else maybe? Does the tumor have a specific alteration that can help them respond better to one treatment or the other? Would they be eligible for immune therapy? So these are the situations that once they become issues with any particular patients, maybe additional steps could be warranted, such as molecular classification, or additional immunohistic chemical studies that can help triage this patient in a better way.

Ashish Kamat: Yeah. So you raise an important point, and I do want to clarify for the audience that when we're talking about molecular classification and subtyping, we have to sometimes keep in mind that they may mean the same thing, or they may mean different things. Right? So what is your experience, Eva, and your recommendations to the community of researchers? Maybe the fellows in GU pathology. Should there be efforts focused on identifying IOT based markers that would help with subtyping, or should the field focus more on actually next-gen sequencing techniques to identify the different tub types? Which one, in your opinion, is worth expending research, money, dollars, and effort on?

Eva Comperat: Well, I'm more a fan of NGS to be honest, but probably because we have more experience, we have more data on NGS with the treatment of these direct cancers. But, I think also all the other options are good options, but we need more experience quite clearly. I mean, we're working now for nearly 10 years on the molecular and it's still not a gold standard, which I really regret that it's not enough introduced and integrated into treatment. So I think with all the new technologies, which are really upcoming, we have to be of course, careful, but we always are careful with our research and giving data and so on. And I really think this is also all the other types and techniques that are upcoming. And I think within five years, I hope so at least, they will be really integrated into the treatment. I don't know which is Hikmat's opinion on this.

Ashish Kamat: Yeah. Hikmat, I was just going to ask you, do you think that the push towards newer technologies is more relevant, or the quest to find basic IOT based markers that can help put subtype?

Hikmat Al-Ahmadie: I think we need more technology because IHC is helpful and powerful, but unless you get to a point where you can do one marker that can help you, like be specific in one entity versus the other one or two markers perhaps. I think doing 12, 13, 25, is not practical. And we're talking about the cost, that can increase the cost significantly, that can make the interpretation more difficult. Basically you're going to still rely on an algorithm or a computer, to help you make the decision at the end, while this is not the whole idea of practical classification. I think the technology is what is going to move us into this proper consistent and reproducible classification.

Ashish Kamat: Yeah. I think it's an important point that both of you raise is that just because we can do it with a cheaper modality such as IHC, doesn't mean that it will actually be more cost-effective because you'll have to use more markers, more panels, use technology algorithms anyways. And the cost of course, of next-gen and others, is dropping to a point where it actually is not as cost-prohibitive as it used to be.

I could chat with you guys forever on this topic, but in the interest of time, we will have to wrap it up. So let me ask each one of you to give the audience your closing thoughts here, your sort of summary of the high-level situation as to where we are today. Hikmat, let me start with you and give the lady the last word.

Hikmat Al-Ahmadie: Absolutely. So, I think morphology is a great, an important function of classification. It has served us for the longest time. It's still a great attribute. We recognize some of its shortcomings, and molecular is the way of the future. So we'll have to be smart about how to do it and to do more of this in the right way, and how to use it to the best of our advantage and help improve the classification systems that have been good for a long time, but again, they were not perfect. Molecular classification will help us improve our classification, our risk certification, and at the end, the benefit for everyone.

Ashish Kamat: Eva?

Eva Comperat: Well, I actually had joined pretty much Hikmat's opinion. I think we should really, in the end, have a double classification. One, which is really based on morphology, and the other one, which is really based on the molecular. Put them together, and this will be the most beneficial for the patient, in my opinion.

Ashish Kamat: 
Great. Thank you so much, both of you, for taking time off from your busy schedules and joining us, and the Uro Today community, in sharing your thoughts. This has been great. Stay well and stay safe.

Hikmat Al-Ahmadie: 
Thank you.

Eva Comperat: 
Thank you.
email news signup