This transcript has been edited for clarity. For more episodes, download the Medscape app or subscribe to the podcast on Apple Podcasts, Spotify, or your preferred podcast provider.
Sumanta Pal, MD: Hi. My name is Monty Pal, and I am a medical oncologist at City of Hope in Los Angeles, California. Welcome, everyone, to season 2 of Medscape's InDiscussion on renal cell carcinoma. In season 1 of this podcast, we covered a wide range of topics ranging from frontline therapy, to second line treatment and beyond. We're back again to cover a new series of concepts.
Today we're going to discuss tissue biomarkers in kidney cancer and whether or not we're ready for prime time. First, let me introduce my guest, Dr Scott Haake. Dr Haake is an assistant professor of medicine and a physician scientist at Vanderbilt University. He also works at the National VA Hospital. His clinical and laboratory interests focus on kidney cancer. Welcome, Scott. How are you?
Scott Haake, MD, PhD: Hey, Monty. Thanks for having me. I really appreciate it.
Pal: You have such an interesting career setup — the classic physician-scientist model, as you have described it. Tell us about how much time you spend in the lab, how much time you spend seeing patients, and so forth.
Haake: I've really modeled my time split from my mentors in the field, which include prominently Kim Rathmell. Traditional physician scientists are 75%-80% lab-based and 20%-25% clinic-based. I spend most of my time running my lab, but then I also see patients with genitourinary (GU) cancer, with a focus on kidney cancer, here in Nashville.
Pal: I'm always in awe of clinicians that can straddle the line and do both. I imagine it can be challenging at times, but it also probably gives you a great perspective on how to take issues that plague us in the laboratory and turn them into options for patients in the clinic and vice versa. You can take clinical scenarios that we struggle with and investigate them in the laboratory. That's really what we're going to focus on a lot today.
I wanted to ask you, why kidney cancer? What got you into the field of kidney cancer specifically?
Haake: To be completely honest, it was in large part serendipity. My first attending on my first day as a medicine intern at the University of North Carolina at Chapel Hill was Kim Rathmell. I told her that I wanted to work in cancer and that I wanted to work in a lab. She said, Oh, you should join my lab and study kidney cancer. So I did. I've been a fan of her ever since. She's been a role model of mine ever since.
Pal: She's just tremendous, isn't she? Did you have a lot of lab-based experience prior to joining the Rathmell lab, or was that your first introduction to lab science?
Haake: I didn't have a PhD at that point, but I had spent a lot of time in both college and med school in wet labs, and so I came in with a decent amount of wet lab experience, but I had never done any translational oncology work. In Kim Rathmell's lab we studied RNA-seq gene expression signatures. I've continued to study those throughout my career. I imagine we'll be talking a bit about that today.
Pal: Excellent. I'm looking forward to addressing some of those topics today here. We have a whole multitude of therapies for frontline kidney cancer, ranging from ipilimumab plus nivolumab (nivo-ipi), a pure immunotherapy regimen, to targeted therapy and immunotherapy combinations. We've dealt with this actually on prior Medscape podcasts and had some really terrific guests outlining those issues. Are we at a point where there's one de facto standard for kidney cancer, or are you of the mindset that there's multiple potential options for patients frontline and beyond?
Haake: I do think you can say that essentially all parties would agree that we need to include immune checkpoint inhibitors in the first line as a multidrug treatment. But that's about where the consensus ends. There are a number of regimens which contain one or two of those drugs, and there is a lot of debate within the field in terms of which patient should get which one of those therapies.
Pal: I totally agree with you. I think that there is a challenge in terms of the clinical data. The response rates, for instance. If you look at targeted therapy in immunotherapy, they are fairly balanced across cohorts. Side effect profiles are fairly balanced. I have argued in the past about differences in quality of life among some of the regimens. But I think that everybody feels that the holy grail, if there is one, is probably going to be biomarker-based work. That leads us into your area of expertise. Tell me a little bit about your perspective on the biomarkers that we have available now from existing datasets in renal cell carcinoma.
Haake: We really have nothing within renal cell carcinoma in terms of tissue-based biomarkers that we can rely on. There are US Food and Drug Administration (FDA) approved biomarkers and companion diagnostics within the medical oncology field that are paired to immunotherapies and other solid tumors. But for a variety of reasons, none of those have really proven useful within the space of kidney cancer. We can go through those examples, and I can provide some conjecture or ideas in terms of why they are not useful. But the simple statement is that we have no reliable tissue-based predictive biomarker to choose among these various therapies within kidney cancer.
Pal: I couldn't agree with you more, but the one biomarker that everybody sort of seems to be coming back to in tissue at least is programmed death ligand 1 (PD-L1). It's something that's discussed often in the context of lung cancer, where they have these really sophisticated trials that look at PD-L1 thresholds: 1%, 50%, and so on. In kidney cancer, I'm of the mindset that we haven't met the mark for clinical utility with PD-L1. Is that accurate, Scott?
Haake: I would totally agree. People have tried really hard to make these tissue-based biomarkers work. Since these drugs target immune checkpoint inhibitors like PD1 and PD-L1, there's been a lot of focus on those specifically. Unfortunately, we don't see reliable discrimination and important clinical outcomes between the tumors that are positive for these tissue-based biomarkers and tumors that are negative, which is really disappointing because as you mentioned, there are other fields within solid tumor oncology where these have proven really useful. That's been a big limitation for the field.
Some of it may quite frankly have to do with the nature of therapy within kidney cancer. For instance, in lung cancer, there was a nice trial showing that in PD-L1 high versus vs low disease, you can choose immune therapy vs chemotherapy. It's not really a relevant question within our field whether you should be including immune checkpoint inhibitors or not. Everyone's going to get those. The question is, do they get one or two and do you pair it with something else? That may be part of the problem within our field. It's not the whole story.
Pal: That makes perfect sense. Now we've come to the more sophisticated biomarkers. You have been heavily involved in this body of research, but out of some of the trials that haven't really made their way into clinical implementation — trials looking at regimens like bevacizumab and atezolizumab, those looking at axitinib and avelumab. There's been some very deep and thoughtful biomarker work done. I wonder if you can start by highlighting some of that for our audience.
Haake: Just to zoom out for a minute, using gene expression signatures, these are assays where we measure gene expression across a variety of cell types that make up a tumor. These are immune cells, tumor cells, endothelial cells, fibroblasts, and a variety of cells. This process of measuring gene expression and using that to identify relevant biology has a long track record within solid tumor oncology. I think perhaps the best case for that is breast cancer. In a seminal work within the field, Chuck Perou and others used these sorts of platforms to identify unique subsets of breast cancer which preferentially respond to or are defined by HER2 expression, which is obviously a very relevant clinical target within breast cancer. There's a long track record of these tools being used to identify subsets of cancers within a larger umbrella.
The kidney cancer field has worked very hard to do this as well. There have been a number of examples. I think part of this gets down into how those gene expression signatures are discovered or designed. In some examples, we have relied on the investigators to identify important genes from prior knowledge in prior literature and design with their own thoughts driving the process of what a good gene expression signature would consist of. That's valuable because it narrows in on a very specific biology.
However, in 2023, with artificial intelligence and machine learning, and the vast size of these signatures measuring tens of thousands of genes, the question is, are we missing out on some important biology if we're relying on our own understanding rather than larger analyses of the data? That has led into other studies. For instance, there's been really nice work by the JAVELIN Renal 101 cohort, where they asked a very specific question of the data. They asked which genes correlate with response to the drugs tested in JAVELIN Renal 101. That identified a subset of genes. Those gene expression signatures worked quite well for answering the question they were designed to answer: Which genes predict for response to an anti–PD-L1 drug like avelumab, paired with an anti-angiogenesis drug, vs the control arm sunitinib? But when you ask the question which genes correlate with response to this drug, now you have really tailored it to that one drug in that one clinical context, which is the combination of avelumab plus axitinib vs sunitinib within JAVELIN Renal 101.
We've talked about two ways of establishing these signatures. One is where the investigator is really picking out relevant genes. That's sort of what they did in IMmotion150 vs what we call a supervised clustering, asking which genes when expressed correlate with response to specific drugs. That's what they did within JAVELIN Renal 101. These are both great studies and highly valuable.
A third way of analyzing these huge datasets is what we call unsupervised clustering. That's when you don't worry about your a priori knowledge. You don't worry about predicting response to the particular drug. You simply ask, what are the inherent, unique, distinct biological subgroups within this large cohort? That's what they did in IMmotion151; I was not part of that study, but this is what you call an unsupervised clustering analysis where they had a huge dataset, they had 800 tumors, which is bigger than what the Cancer Genome Atlas (TCGA) had, right? They didn't ask which genes correlate with response to the drugs they were testing in IMmotion151. They simply asked, "Name the inherent subgroups within this very large dataset." The advantage of that is that it's relatively unbiased. It's empirical; you get pure biology. The downside is that you may identify very distinct subsets of kidney cancer, but they may not be biologically relevant or they may not predict for drug response. Within the context of that study, there was some correlation between these distinct subsets and drug response.
That's why we were interested in these data. We had some hypotheses come out of that analysis. That's what we're trying to test in our phase 2 study.
Pal: I think that it's probably going to be important for us to dive deep into the IMmotion151 data that you just highlighted. Just for the audience, that's a trial of bevacizumab with atezolizumab vs sunitinib, The trial is published in The Lancet. I think it's very important data for the field. But frankly speaking, the trial was not really implemented in clinical practice because it didn't show an overall survival advantage ultimately. The question becomes, what can we really garner from this data? Start with the biologic piece of it. They subdivided the population into seven different clusters. Tell us a little bit about that clustering approach and what those clusters tell us.
Haake: This was called unsupervised clustering. It was not guided by the expression of which genes predict for drug therapy. It was simply identifying distinct subsets. The authors identified seven clusters. One was very small. We won't really discuss it. That was cluster seven, which is characterized by a very unique subset of genes which are expressed.
But the other six clusters represent the vast majority of cases, and you can lump the clusters into larger subgroups based on response to therapy and based on the biology of those subsets. There were a couple of clusters with very high expression of genes correlated with angiogenesis, and those patients tended to do relatively well, possibly because both arms within IMmotion151 received anti-angiogenesis agents. Both arms were treating some of the core biology of a couple of those clusters. A couple of the clusters responded really well to immunotherapy. That's not surprising when you look at the biology of those clusters, because you have relatively high expression of immune related genes such as CD8, which correlates with cytotoxic T cells, which are the T cells that attack cancer and interferon-gamma signaling. That was a separate cohort which responded really well to immunotherapy. When you look back at the biology of those clusters, it sort of makes sense.
Then there were a couple of clusters that didn't fit into either of those buckets and frankly did really poorly on this clinical trial and progressed pretty quickly when treated with either arm. Those tumors represent a black box and are an area where we need to commit time and resources to really understanding that biology better, so that we can better tailor therapies to target the unique biology of those clusters.
That's a high-level overview of those seven clusters and how we thought about the data and in terms of generating a prospective testable hypothesis.
Pal: I love it. This really brings us to the opportunity that we have based on that data to run a prospective clinical trial. And again, hats off to you and the Vanderbilt group for being the first to do this. Can you walk us through the design of OPTIC, the prospective study that you and others are leading there? You really helped define this opportunity. Full disclosure, I'm going to be one of the OPTIC investigators alongside you here, so I've fully bought into the concept.
Haake: I think this is a really exciting concept. I've obviously been working in the field of bulk RNA-seq or gene expression signatures for a long time. I think the example of breast cancer is compelling and I think this is a good resource to identify unique biology that can be actionable. This is a challenging study. It has not really been done before, to my knowledge, in solid tumor oncology. We're taking biopsies of tumors, preferably the metastatic tumor from patients with metastatic clear cell renal cell carcinoma. We're measuring gene expression. We're running it through an algorithm to assign it to one of these clusters, which we've discussed before, which we think will predict for response to therapy. Then we're tailoring our therapies to the biology of these clusters.
It's a challenging study, because it is biomarker driven and there are some logistics that have to be worked out with those biomarkers in terms of getting patients consented and on therapy as quickly as possible. But we think it's really exciting and is a new way forward in terms of biomarker development within this disease.
Pal: I am going to address the queries of the skeptics here, because there are individuals who say, wait a minute — we're taking a profile that was built in the context of bevacizumab and atezolizumab, and we don't use that regimen in the clinics anymore, so we're limited by that, but we're trying to force that particular algorithm into the OPTIC trial, which allocates patients to cabozantinib-nivolumab or nivolumab-ipilimumab. What would I say, for instance, to a very astute patient who brings that to my attention?
Haake: That's a great question. It's something that's come up before. I think it goes back to how these signatures were defined. Frankly, it wouldn't matter if IMmotion151 was randomizing patients to peanut butter and jelly vs ham sandwiches. The way these signatures were developed in no way reflects the drugs which were tested. This was an unsupervised cluster analysis. This is the same sort of analysis that they did in TCGA, where many of those patients didn't even have metastatic disease. You're not asking which genes when expressed predict for response to bevacizumab plus atezolizumab vs sunitinib. You're asking in an unsupervised way, "Just tell me the unique core biological subgroups within this disease." We're not asking for the machine learning or artificial intelligence or computer learning to draw any comparisons to the therapies. If you do that well, then much like the breast cancer story, subsets will follow out with actionable molecular targets. In breast cancer, it was HER2 and estrogen receptor signaling. We think in kidney cancer, maybe it's immunotherapy in some subgroups, and in others, immunotherapy plus the anti-angiogenesis therapies with the tumors with really high angiogenic signaling, requiring hitting that target in order to control disease.
Pal: Interesting. I think that's a brilliant response. In Los Angeles, and I'm sure everywhere, we have this cohort of incredibly brilliant patients who just know the ins and outs of the science and so forth. I can imagine these patients will very appropriately be asking these questions. Of course, we ask these questions in the investigative community as well.
With OPTIC, we've got modest numbers, I would say. Is there an opportunity there to build this out into a larger, more definitive study if we start seeing some encouraging results?
Haake: OPTIC in its current iteration is not the be-all, end-all study. Those cluster one and two tumors with the really high angiogenic gene expression signaling pathway — you really want to randomize those patients to, say, immunotherapy plus a tyrosine kinase inhibitor (IO-TKI) vs immunotherapy plus immunotherapy (IO-IO), right? Cabozantinib-nivolumab vs nivolumab-ipilimumab. That's really the definitive study. We're not doing that right now because we need to learn how to implement this biomarker. Frankly, that's a much larger study that's out of the scope for what we're able to tackle within OPTIC. But I think asking those prospective randomized questions will be the next iteration.
In addition, I think this is really a platform study, and so in 10 years, we're going to have new drug targets that are targeting new biology. I think you'll go back to the same question: "What's the core biology driving the growth of these clear cell renal cell carcinomas out of these various groups and which of those clusters makes the most sense for new drug X?" Moving forward, I think we'll be continually thinking about matching drugs to the unique biology of these various subsets.
Pal: Everybody in the field views you as being a rising star. You're doing a fantastic job out there, Scott. What do you view as being critical out there if you're a fellow or junior faculty member hoping to get in the field of kidney cancer? Give us a couple of words of advice.
Haake: I have two pieces of advice. One is, mentorship is really important. I think you need to identify what you want to be when you grow up. Do you want to be a clinical trialist, a translational researcher, a phase 1 investigator, or basic science researcher? I think you need to have that vision in your mind, and then you need to attach yourself to a mentor who you can model your career after and look to for guidance. That mentor may not even be at your institution. That's okay. I feel like that is number one.
Number two is to make sure you're at an institution that supports you in that vision; if you want to be a physician-scientist and your institution wants you to be a trialist, that's probably not going to work out for you. Have a vision, find a mentor, and find an institution that will support you in that vision.
Pal: That's a terrific, succinct summary. Scott, thanks so much for joining us here on Medscape InDiscussion. What a great way to start out season 2 with a bang. Today we've talked to Dr Scott Haake about biology of kidney cancer and how to incorporate that into prospective studies.
Thanks so much for tuning in. Please take a moment to download the Medscape app to listen and subscribe to this podcast series on renal cell carcinoma. This is Monty Pal for InDiscussion.
Medscape © 2023 WebMD, LLC
Any views expressed above are the author's own and do not necessarily reflect the views of WebMD or Medscape.
Cite this: Tissue Biomarkers in Renal Cell Carcinoma: Are We Ready for Prime Time? - Medscape - May 03, 2023.