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.
Sapna P. Patel, MD: Hello. I'm Dr Sapna Patel. Welcome to Medscape's InDiscussion series on melanoma. This is episode 4. Today we'll be discussing liquid biopsy in melanoma. Let me introduce my guest, Dr David Polsky, the Alfred W. Kopf, MD, Professor of Dermatologic Oncology at the Ronald O. Perelman Department of Dermatology at New York University Grossman School of Medicine, where he's also the research vice chair and director of Pigmented Lesion Service. He's also a member of the Perlmutter Cancer Center, New York University Langone Health. Welcome to InDiscussion, David.
David Polsky, MD, PhD: Thank you.
Patel: You obviously do a lot of work in liquid biopsy. There's no denying it's becoming more utilized in clinical trials and to some degree in the real world, though I worry that we're not carefully using it, or we're throwing it out there and having difficulty explaining why we did it to patients. Maybe before we talk about the clinical trial data, we can think about biomarker development. I've seen you give talks as recently as last week. You gave a great talk going over the basics of sensitivity, specificity, positive predictive value, and negative predictive value. Can you give us a refresher on these definitions and how you think we should be thinking of our liquid biopsy tools in terms of sensitivity, specificity, and negative predictive value?
Polsky: First of all, let's not confuse the assay sensitivity with its clinical sensitivity. The assay sensitivity is what we do in the lab with limiting dilution of reagents, and we say, what's the least amount of molecules we can detect? The clinical sensitivity is the proportion of patients who are positive by some other test that turned out to be positive with your test. So, the correct answer, right? And specificity is also the correct answer, which is how many of the patients are truly negative. A healthy donor shouldn't have a BRAF mutation in their plasma, and we know from our own work that that's extraordinarily rare. So, that's a way to think about the liquid biopsy space, sensitivity and specificity.
The positive predictive value is clinically useful because it's the proportion of the test positives that actually are positive for the disease. As a clinician, you say, "I ordered this test, and it's positive," and then say, "Okay, this is the real deal, this is really happening."
Likewise, the negative predictive value is also useful in a decision. We've talked a lot about who should get adjuvant therapy. It's controversial. The survival benefits are there for distant metastasis–free survival and relapse-free survival but not necessarily for overall survival. Are there people who can wait safely? And, of course, there are people who are truly surgically cured who are going to get adjuvant therapy. But if you knew with 100% certainty that those patients are surgically cured, you could spare them. So, a negative predictive value would be if the test is negative, how many of them are truly negative? How many are truly surgically cured? So, the negative predictive value is very important.
With liquid biopsy in melanoma, one of the things that we think about is the false-negative rate. We know that, especially in the adjuvant setting, there's a proportion of patients who are cured, who are never going to be positive in their circulating tumor DNA (ctDNA).
Then there's a portion of patients who have such a little amount of minimal residual disease (MRD). It's below the assay detection limit, but they recur later. So, they might be initially negative — say, a baseline or even a few weeks or months in — but then they may turn positive later on. Or they may recur and not be positive at all because of the location of their recurrence. It might be in the lung, which doesn't shed very well, or especially in the brain, where it's really hard to detect ctDNA from brain tumors. So, the negative predictive value is another important feature when we talk about liquid biopsy and in terms of disease monitoring.
Patel: This idea of false-positive, false-negative, positive predictive value, and negative predictive is important as a clinician orders this test, especially if they order something like the FoundationOne CDx or the Oncomine. You're getting me all these genes across multiple cancer types, but what you really should be saying to the patient is, "I'm interested in the genes that I know are related to your melanoma." So, when it comes back with a CHEK2 mutation or a very random EGFR mutation, is that really a positive predictive value? The test is positive. Do they really have an EGFR-mutant neoplasm brewing? Not likely if you're not picking that up on a scan. I get concerned when that test comes back, and then people think there's a lot of actionable things to follow for melanoma — the results may not really be relevant. And as you said, these are panels that you can't tailor. The nice thing about digital droplet is that you really can tailor that for your disease and genes of interest, which I think is great.
You bring up this idea that there may be sites that have low shed. There may be contexts like the adjuvant setting where somebody is not necessarily positive at the time of adjuvant therapy or down the road turns positive. Do you think there's any utility in a single-time-point liquid biopsy, or is it really a tool that's best used for longitudinal monitoring?
Polsky: If you have to put me in a corner, I suppose if you could only have one measurement, certainly the baseline measurement is useful for indicating the tumor burden in the patient and their likelihood to respond in the metastatic setting. If you have a high ctDNA at baseline, that's a patient who's going to be challenged in terms of responding to treatment. If you have a high ctDNA in the adjuvant setting, that means you've gone through surgery, you're radiographically disease free, but you have MRD. So if you had to pick one, I suppose that would be one that has the most utility. But I really believe that it needs to be done in a longitudinal setting, because we've seen that when patients clear their ctDNA with treatment, they have much longer survival in the metastatic setting. We showed that in COMBI-d and COMBI-MB in our The Lancet Oncology paper from 2021, and then also recently in our ASCO poster with the CheckMate 238 data showing that patients who started off positive but became undetectable did better than patients who remained detectable even on treatment 3 weeks, 7 weeks later. We found that those patients, especially if they're a positive baseline and positive at 7 weeks, 80% of the patients who recurred at 6 months had that kind of kinetics — positive at baseline, positive at 7 weeks.
Patel: Your data certainly support this idea of longitudinal following. We'll dive deep into the data here in a bit, because what I want to circle around is exactly what you just landed on, which is your COMBI-d and COMBI-MB data in metastatic disease, where we know that there's a higher tumor burden actively happening. Logically, that means more shed, more positivity at baseline, as opposed to the CheckMate 238 data, which were in the adjuvant setting with two active therapy arms and where many people are cured, as you noted, so the positivity rate is lower. They're different settings, yet this idea of serially following, and if you're positive-positive, you'll have a worse outcome at that short interval of 9 weeks or 7 weeks or whatever it is. That's important. Do you feel that there's a role both in this adjuvant setting for MRD and for monitoring response to therapy? Are you more excited about one of these settings as giving us this kind of proof of principle and where to look for liquid biopsy?
Polsky: Obviously, I've got my biases. I think right now there's such excitement in the adjuvant setting because now with stage IIB and C patients receiving adjuvant therapy, we really see the question rising up even more dramatically — now we have even more people who are cured surgically who should be getting treatment here. We haven't done any work in ctDNA in stage II; we're doing tumor-based work there. But I think that with the controversies around adjuvant therapy, there's a potential utility in not only who should get adjuvant therapy but perhaps escalation — so, those patients who are positive. And let's say they're on single-agent PD-1 therapy, and they're persistently positive at 3 and 7 weeks or so, even despite PD-1 therapy, maybe we need to do a trial where they get randomized to some kind of doublet vs staying on the single agent. There's a lot more excitement in the adjuvant space for what we do, or I should say I'm more excited because of what I see.
In the metastatic setting, we have work in patients who are getting immunotherapy. We haven't published it yet. There are some studies out there. Our data are on frontline BRAF-MEK therapy, which is becoming less common. So, before you can generalize what to do with ctDNA in the metastatic setting, we need to do more studies on patients receiving immunotherapy.
Patel: My big concern or my big caveat is that in the adjuvant setting, such a low portion of patients are positive. And if we're using something like droplet digital technology, it's going to be even less of those patients, because now we're just sorting for the mutation-positive group. A tumor-informed signature in stage II and stage III melanoma, I suspect is going to be highly challenging because these are millimeters worth of tumor, and it will be fairly hard. It's probably not a surprise that we haven't seen a lot of Signatera data published in melanoma, and that is probably not limited to the fact that there's a tissue limitation here.
There were two recent publications showing it is feasible. You can do it in stage III melanomas, maybe even stage II and certainly in stage IV, but it requires getting that tissue. Then in that tissue, you do a 16-clonal fingerprint analysis for the top 16 hits of clonally amplified genes, of which BRAF and NRAS are not typically going to be included in there. So, then you might exhaust all your tissue doing that for a liquid biopsy assay, and then clinically not have your BRAF and NRAS data. Alternatively, you might spend all your tissue obtaining a BRAF status and then not have enough for one of these tumor-informed assays. I think in the metastatic setting, there's always more tumor to get. There's a higher shed rate.
I'm intrigued by this idea of saying at time point zero of either starting metastatic therapy or adjuvant therapy, can we at least start with the ctDNA-positive patients and think whether these are the ones where we need to follow and see if there's a signal? That's maybe how it's playing out in other disease types. Certainly in MRD, part of the benefit is monitoring for somebody who's negative turning positive.
There's one other area, David, that makes sense to talk about in liquid biopsy. We will often think of ctDNA as having one faction of investigators and researchers, but there's also people who still believe in circulating tumor cells (CTCs), with the idea that we don't know the source of DNA in a cell-free DNA assay. We don't really know if it's from replicating cells, dying cells post-radiation, and though the half-life of DNA is short, we just really don't know the source of that. On the other hand, if you find a CTC in plasma, the concern is that's a metastasis-competent structure that then can extravasate into an organ, multiply, and boom, you've got a metastatic tumor. And there's newer, trendier factions of exosome work and other things. What are your thoughts of these different tools in liquid biopsy monitoring?
Polsky: I think that for melanoma, people have been talking about CTCs for probably decades. One of the problems is the sensitivity, because those cells are super rare in circulation. So, how can you collect enough blood to purify those cells out? And other cancers have an antibody-capture methodology because there's a cell surface protein and you can use, say, a magnetic bead or something like that to try and capture the cells prior to analyzing them to determine if they are CTCs or not. There's a cell search assay, which is an FDA-cleared assay that's actually used for patient monitoring, but in melanoma, we don't have that. We don't have a really good way of capturing the cells to enrich the population before then going and analyzing what it is that you've captured. That's why CTCs in melanoma have lagged compared with other diseases.
With respect to exosomes, it's very exciting. All the biological information coming about it from exosomes and some of the studies showing that the tumor cells secrete these packages of DNA, RNA, and protein, and the early work saying that it might actually precondition the metastatic niche and all that kind of stuff from experimental animals. It's really quite fascinating. The purification strategies are a little bit variable. If you're going to bring it to the clinic, you've got to do all the analytical validation work in order to get it into a clinical lab. So, it's very exciting. A lot more work needs to be done in terms of standardizing these assays, optimizing them, validating them, and then testing them in a clinical trial. Maybe they will be the answer, but a lot more work needs to be done.
Patel: I agree. I have to say, I'm a believer in all of it. Our group at MD Anderson has published with Anthony Lucci's lab some of the cell search work in melanoma. This came from his breast cancer work, because he was one of the proponents and investigators who got cell search for breast cancer going. It turns out you can capture circulating melanoma cells. Like you said, it's really a magnetic bead assay. And we saw that in the adjuvant setting, CTC positivity obviously correlated with early relapse, and it segregated stage by stage. So, in stage IIIB or stage IIIC, the positives did worse than the negatives. In stage IV, it was interesting. Similarly, we showed even one positive CTC at baseline in a stage IV patient. Regardless of what therapy they go on to, they have a worse outcome. That's really these one-time-point captures.
I'm interested in this idea of longitudinal following. Even in uveal melanoma, we did something similar where we said, "Listen, CTCs are shed more in metastatic patients than in early patients." It's what I call SOTO — science of the obvious. The nonmetastatic patients are not shedding quite as much, maybe one CTC per 7.5 mL of blood. And the metastatic patients are shedding more of an average of nine or so CTCs. It turns out that even the presence of one CTC in a uveal melanoma patient after primary treatment is indicative or correlates with decreased survival. There's probably an algorithm that could incorporate both — a CTC and a ctDNA — and then really start to engage whether you have dying cells or whether you're in an early warning for disease progression, etc. The extracellular vesicles, these exosomes, are interesting. And because these really are just tubes of blood, I do think we're going to start to see more of this in our clinical trials.
Maybe you can take us through some of the clinical trial work you've done, the COMBI-d and COMBI-MB, we were talking about a little bit earlier. For those who may not be terribly familiar, can you describe what you did by getting those blood samples in patients at the start of treatment and then early on in treatment?
Polsky: Basically, COMBI-d was single-agent dabrafenib vs a combination of dabrafenib plus trametinib. We had samples from 345 of the 423 patients in the trial at baseline. And then we had samples only at one time point, only at 4 weeks on treatment, available from about 50% of the patients. So, we had a good sample set that was representative of the overall trial population, the biomarker subset. We found that we could detect the BRAF mutation. Of course, everybody had to have a BRAF V600E or V600K mutation. They had done that as part of their tumor analysis. They provided us with the data of which assay to use, V600E or V600K. Ninety-three percent of the patients had a detectable mutation in their plasmid. At the time, this was a very high detection rate. It's still probably the highest or one of the highest detection rates in any similar trial in melanoma. And then very strikingly, I think it was something like 40% or 60% became zero converted. They became undetectable at the 4-week time point. So, a very early on-treatment signal for the clinician to say, "Okay, is the treatment working or not?" We looked at that, and we saw it correlated with increases in progression-free survival of about a doubling of progression-free survival, a tripling of overall survival, and was an independent predictor of survival in a multivariable analysis, both as a continuous variable. So, for every increment increase in ctDNA quantity, it was a predictor of shorter survival. Then we made a cut-point of 64 copies/mL, and they fortunately had samples available from the COMBI-MB trial, which was a brain metastases trial, and they chose cohort A, which was asymptomatic patients and was the closest to the COMBI-d, which excluded brain metastasis trial patients. But in COMBI-MB, there was extracranial disease as well as intracranial disease. We were able to show, similarly in COMBI-MB, that if they zero converted, they had a better survival, and that 64 copies/mL was a valid cut-point for that.
Patel: That 93% detectable is an important benchmark for us. Now that you've been able to do it, we would like to see that positivity rate, regardless of what we're looking for in the metastatic setting. It makes me think of the tebentafusp data with ctDNA. Recognizing that you're not a uveal melanoma person, I'm not sure if you've seen those data.
Polsky: I did. I looked at them.
Patel: Interestingly, there was a very high positivity rate there as well with this kind of bespoke ctDNA assay, as we were talking about. Uveal melanomas have these conserved hotspot mutations in 99.9% of patients. While there wasn't a high rate of zero conversion because it's just a tougher disease to treat, they at first looked at log level reduction — so half a log, one log, two log, three log level. Then in a follow-up study, they looked at 50% reduction, 90% reduction. As you were saying, each incremental increase in ctDNA might portend a shorter treatment progression-free and overall survival. They looked at it similarly that each incremental decrease in ctDNA seemed to portend a longer overall survival — maybe not progression-free in uveal melanoma with tebentafusp, but certainly overall. So, it's really nice to see that these two datasets and approaches, even though they're different platforms, the proof is there. In the metastatic setting, there's a high positivity rate. If you're potentially using the right assay, you should be able to find it. And decreases/increases are likely to correlate then with some survival. Maybe not with response, although your COMBI-d and COMBI-MB data…
Polsky: Yeah, there are other measures of response.
Patel: And in tebentafusp, we don't see a lot of responses. So, objectively, you wouldn't want to say that decreases mean you're going to have a response, because objectively we just don't see a lot with that drug, but you do see this survival benefit. That should give listeners hope that in the metastatic setting, we're onto something in BRAF-positive melanoma. And of course, you've made panels now for NRAS, TERT1, and TERT2 , which we know is going to capture now beyond more than 70% of melanomas.
Polsky: Our panel captures almost 90%. Those were our data. It was up to around 87%. Almost 90% of patients will have one of those seven mutations in their tumor. That's from CheckMate 238.
Patel: Can you walk us through CheckMate 238, because now we're in the adjuvant setting?
Polsky: We got DNA from all the patients in CheckMate 238 and identified which one of those seven mutations they had. We did it with droplet digital PCR (ddPCR), so we needed very little DNA. There were even some samples where the assay was not run for the clinical trial because they said there was insufficient tumor in the section for the clinical assay that they needed in order to do the approved assay, and we were able to get an answer on almost 98 out of 100 of those, roughly.
Polsky: Positivity rate at baseline was about 12%, and we looked at it by substage, so it was up around 15% or so for stage IV resected and IIIC, and a little bit less for IIIB resected. What that's telling us, since baseline positivity is patients with MRD, is that for these patients who you would predict are going to recur more frequently, you can already see it at baseline in a subset of them. So it makes sense that the higher the substage, the higher the detection rate at baseline.
Patel: And then did you do longitudinal monitoring?
Polsky: Yes, we did a lot of longitudinal monitoring there. We're very fortunate that Bristol Myers Squibb collected lots of samples. We had baseline, and then we had week 3, week 7, week 12, and then every 3 months after that, out to 1 year or 2 years even. We're still assembling all the data, but we had enough to put the ASCO poster together, which focused on the earlier time points. As I mentioned before, these are two active treatment arms, but we did see the superiority of nivolumab over ipilimumab in our biomarker subset. And if people were persistently positive after baseline, they recurred by 6 months. They're really patients who are the high-risk patients — maybe you think about more treatment intensification, maybe putting them on a combo instead of a monotherapy — it's something that we could explore.
Patel: That's interesting. I think that was explored in CheckMate 915, which was adjuvant nivo vs adjuvant nivo plus ipi. It'd be similarly collected and some analysis of ctDNA was done; the positivity rate in that study at baseline was about 16%, and these low positivity rates are not surprising. We're in the adjuvant setting. We don't expect a lot of shedding of active tumor DNA.
What I found interesting was that the study overall was a negative adjuvant study. We do not have an indication for adjuvant nivo plus ipi because it was not demonstrated to be superior over adjuvant nivo. So, you start to wonder, Well, if we had parsed it apart by ctDNA, would we have seen something? It obviously makes us think of those IMvigor010 data in urothelial cancer. This was adjuvant atezolizumab vs observation, or maybe they had a matched placebo, but regardless, it was an inactive control. What they found was that was a negative study. So, adjuvant atezolizumab really is not indicated in urothelial carcinoma. And then Tom Powles' group went back and actually got the plasma from these patients and did the Signatera tumor-informed test by getting the tumor tissue sequencing and getting the top 16 clonal-amplified genes and then searching for that in the blood of these patients. If you looked at study entry at baseline again, they had a low positivity rate — something in the 30% range — it's twice as high as it is in melanoma, but they're in the adjuvant setting, so not every patient will be positive. The data starts to segregate by that ctDNA-positive group of patients. In the ctDNA-negative group of patients, adjuvant atezolizumab (atezo) really doesn't seem to have a benefit over observation or placebo.
But in the ctDNA-positive group, it absolutely has a benefit. Strangely, in CheckMate 915, that hasn't parsed out as well. In the ctDNA-positive group at baseline, maybe it's because it's small numbers here — it's 16%, and then half of those go on to nivo and half go on to who knows, and maybe it's not even half and half. It doesn't seem to tell us whom to treat. It remains lacking in predictive value, predicting who might benefit from therapy and still stays as a prognostic tool as you're alluding to. Somebody who's positive likely is going to recur sooner. In somebody who's positive, who stays positive on whatever therapy you do, it is likely that therapy is not working, and they need some sort of escalation. What strikes me as some recent data from ASCO on KEYNOTE-942, which is the pembrolizumab vs pembrolizumab plus personalized cancer vaccine, or I think as somebody is trying to say lately, individualized neoantigen therapy, avoiding the word vaccine. But those data seem to do the inverse; they said that if you're ctDNA-negative, pembro plus vaccine might be doing better than pembro. It was fully exploratory and underpowered, and truly the ctDNA-positive group in that study had only two pembro patients on it. So, I think we're probably needing to see more data in that setting because it really stretches the mind to believe that ctDNA-negative is where intervention would be useful. I think we would all believe that it's in the adjuvant setting where you're ctDNA positive, you turn positive, or you remain positive, that therapeutic intervention is necessary. I'm not sure we can make much of that, and so maybe more to come on that.
In the last few minutes, maybe we can talk about this idea of low positivity in the adjuvant setting, which means the metastatic setting is low-hanging fruit. There's now evidence in cutaneous melanoma with BRAF mutation and then in uveal melanoma that the metastatic-positivity rate is far higher with ctDNA. We've got some evidence that even CTCs are being shed more in the metastatic setting, and that positive at baseline may portend worse outcome regardless of therapy. I wonder what you make of that. If we're low positive in the adjuvant setting, what can we say about that? Biologically it makes sense, but we know that more than 15% of melanoma patients in the adjuvant setting end up recurring. Does that speak to this being a tool for MRD for monitoring longitudinally and maybe not selection of adjuvant therapy?
Polsky: That's right. It should be explored further in terms of monitoring. If you get a signal, maybe it prompts you to do a scan. Maybe you can change scanning intervals and prolong them based on it. We have to see.
The other thing is we need assays, perhaps that are clinically more sensitive to detect MRD better. We're working on that. When we have seen people who are negative and then turn positive, there are different patterns. Some of them are kind of like a hockey stick — all of a sudden, they go up quickly. It may be that it's just the nature of the metastatic deposit that it has to get to a critical mass before there's enough shedding for us to detect it. More work needs to be done there to improve the clinical sensitivity.
Patel: Because there are these different sort of phenotypes of turning positive, when does it correlate with a scan? Are people turning positive almost in advance of the scan?
Polsky: Right. There are some data out there. Evan Lipson has some early data on that from years ago. We're hoping to put that together with the CheckMate 238 data. We've also looked at COMBI-AD, and that's in process. We presented some early data at the Society for Melanoma Research (SMR) last year. We're hoping to be able to have that with clinical trial data and our assays later this year to help inform that question.
In the metastatic setting, some of the questions that always get thrown back to me are, "What are you going to do with the information? How much lead time do you need? Because if it's only a month before they were going to have a scan anyway, then that doesn't help me. If it's 2 months, maybe it'll help me." And then one of the thoughts in the metastatic setting was about BRAF-mutant patients on immunotherapy — could you switch them earlier to targeted therapy if they're progressing? It's a study we've been wanting to do but haven't gotten the traction to do yet. But that would be an interesting study. Because right now, the current combo, the current studies that have looked at that have just switched to targeted therapy at the time of radiographic progression. And we would argue that if you can do it before, but you can tell from ctDNA that these patients are going to be progressing because they see a rise in their ctDNA, switch them earlier and maybe they'll have a longer survival on the targeted therapy arm.
Patel: You just described a potential SWOG study right there. I like that.
Polsky: Okay, great. We'll talk later.
Patel: That's probably a great place for us to wrap things up here. There's so much more to talk about in liquid biopsy, because we're really just at the top of the glacier of what we know and how it can inform. I think these tumor-informed assays make sense in melanoma where we may want to sequence beyond just the known oncogenes, but it becomes a problem of tissue. In melanoma, we are the only cancer measured in millimeters, not centimeters. So, I think we're going to have to get better at some of that.
Let's wrap up. I'd like to thank you, David, for talking with us today about liquid biopsy. We touched on biomarker development in this space, key platforms that are in use, the data behind liquid biopsy monitoring in melanoma both in the early adjuvant and metastatic settings, and lessons learned from urothelial carcinoma.
I'd like to thank everyone for tuning in. If you haven't done so already, please take a moment to download the Medscape mobile app to listen and subscribe to this podcast series on melanoma. This is Sapna Patel 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: Liquid Biopsy in Melanoma - Medscape - Aug 17, 2023.