JCO PO author Dr. Samuel J. Klempner shares insights into his JCO PO article, “PD-L1 Immunohistochemistry in Gastric Cancer: Comparison of Combined Positive Score and Tumor Area Positivity across 28-8, 22C3, and SP263 assays”. Host Dr. Rafeh Naqash and Dr. Klempner discuss assessing the analytical comparability of three commercially available PD-L1 assays and two scoring algorithms used to assess PD-L1 status in gastric cancer samples.

TRANSCRIPT

Dr. Abdul Rafeh Naqash: Hello and welcome to JCO Precision Oncology Conversations, where we bring you engaging conversations with authors of clinically relevant and highly significant JCO PO articles. I am your host, Dr. Abdul Rafeh Naqash, Social Media Editor for JCO Precision Oncology and Assistant Professor at the OU Health Stephenson Cancer Center. Today we are excited to be joined by Dr. Samuel J. Klempner, Director of Gastro Esophageal Medical Oncology and Assistant Professor at Harvard Medical School Mass Gen Cancer Center and author of the JCO Precision Oncology article, “PD-L1 Immunohistochemistry in Gastric Cancer: Comparison of Combined Positive Score and Tumor Area Positivity Across 28-8, 22C3, and SP263 Assays.”

At the time of this recording, our guest disclosures will be linked in the transcript.

Dr. Klempner, welcome to our podcast and thanks for joining us today.

Dr. Samuel J. Klempner: Happy to be here. Thanks for having me.

Dr. Abdul Rafeh Naqash: For the sake of this podcast, we'll be using our first names. So, Sam, it was great to see you at ASCO recently, where I believe you presented these data as an abstract as well.

Dr. Samuel J. Klempner: Yes, we had a poster presentation for this paper, which was published in parallel with the meeting.

Dr. Abdul Rafeh Naqash: Congratulations, and I'm very happy that you chose JCO PO as the destination for these data. So we're going to be talking about a lot of different things today in the context of gastric cancer, which I know you treat very often in your clinic. So could you tell us what the treatment landscape for advanced gastric cancer currently is? Because that goes into the context of why I believe you and your colleagues went ahead with this project.

Dr. Samuel J. Klempner: Yeah, happy to. As you know, unfortunately, half or more of our patients, by the time they come to medical attention for a gastric or GE junction or esophageal adenocarcinomas, unfortunately have advanced disease, often metastatic at presentation. So we have this large population of patients with advanced disease, and over the last couple years, we've actually made some substantial advances in the management and survival of this population. This has been mainly driven by biomarker selection, whether it be adding immunotherapy on top of HER2 therapy, whether it be testing for claudin and seeing the results with claudin directed therapies. And perhaps the vast majority of patients are potentially eligible for immune checkpoint inhibitors. We've seen several phase three trials, perhaps highlighted by CheckMate 649, KEYNOTE 859, rationale studies confirming that there are populations of patients who derive significant survival advantages from the addition of anti PD-1 on top of chemotherapy. So the landscape has really evolved into a biomarker directed world, which is exactly what we hope, because ultimately, the goal is, of course, to match patients with the best drugs at the right time. And that's really the background of where this analytical effort came from.

Dr. Abdul Rafeh Naqash: Thank you for giving us that overview. Going to the second part, which, as you mentioned in your initial overview about the role of immunotherapy, and as we all know, immunotherapy has changed the treatment landscape for a lot of different tumor types. And as clinicians, we often see or ask, what is the PD-L1 positivity for, let's say, lung cancer, which is what I treat, and gastric cancer, which is what you treat. Some of the nuances that we don't necessarily go into when we're looking at those reports is the combined positivity score, the tumor proportion score, or the tumor area positivity. Could you give us an understanding, for the sake of our audience or for the sake of our trainees who might be listening to this podcast, what the CPS, or what the TAP mean and where they are used in the treatment landscape for biomarker selection in the context of gastric cancer? And how do you approach the different cutoffs for CPS when you're treating an individual in the standard of care setting for gastric cancer?

Dr. Samuel J. Klempner: For sure, happy to. So I think eventually it all comes back to patients. When we're sitting in a clinic room with the patient, we want to be able to have features about the tumor that's going to tell us if a therapy is more or less likely to work, maybe if there's a prognostic implication so we have predictive and prognostic biomarkers. And PD-L1 expression does not appear to be particularly prognostic, but it does appear to be predictive of benefit from immune checkpoint inhibitors. Therefore, all of the phase 3 trials that we've seen in some way have linked the biomarker expression to outcomes, whether it's the primary endpoint, whether it's post hoc retrospective analyses, etc. What we've seen is that all of these phase 3 trials have largely used different antibodies to define PD-L1 strata within the trial. So whether that's 22C3, whether it's 28-8, whether it's 263, those are the predominant antibody clones used to examine PD-L1 expression in tumor samples. And it's been pretty clear across these large phase 3 trials that there is a trend with increasing PD-L1 expression and increasing magnitude of benefit. We see this in the improved hazard ratios in the CPS greater than five or greater than ten versus less than one, etcetera.

However, the scoring systems have varied. There is TPS tumor positivity, which only accounts for tumor cells. There is combined positive score, which accounts for tumor cells and mononuclear infiltrates and involves counting cells. And then perhaps the most recent one is the tumor area positivity, which is essentially a non counting method to look broadly at the area of the sample that is expressing PD-L1. It was on this background that we said, is there analytical concordance among the main antibodies? Our work does not address whether there is difference in clinical outcomes between testing 28-8 and 22C3 and SP263. It is simply a pure analytical comparison of the three antibodies. Is a CPS 5, when you call it by 28-8, somewhat agreeable to a TPS or a TAP greater than five with the same antibody and with a different antibody. So we felt that this was kind of a question that hadn't really been fully addressed in the field and may help contextualize results for clinicians and ultimately cross trial comparisons.

Dr. Abdul Rafeh Naqash: Thank you for that explanation. And you bring forth a very important question. And I remember this example of a patient with lung cancer who had tissue NGS done, and they had a limited gene panel with PD-L1 testing sent that showed a PD-L1 of close to 15 or 20%, and then another NGS panel with a different antibody, suggesting that they had a PD-L1 of close to 60-70%, which significantly changes the overall approach for treatment in the context of blood cancer. Is that something that you experience in gastric cancer also, in terms of variability for CPS, determining what treatment combinations you might be able to put an individual patient on?

Dr. Samuel J. Klempner: It's rare that we have samples at any institution tested in multiple methods, but these types of papers and others had looked at some stuff similar and prior to our publication, but we know that there is both spatial heterogeneity. So if you test a tumor versus metastasis, you may have different PD-L1 scoring even in regions of large samples, like surgical resections, there will be some intra tumor heterogeneity in regions of expression. And then we also know that sometimes after therapy, for example, post radiation, there's some data that at the time of surgery, the PD-L1 expression may be higher than what the presurgical sample was. So there's a lot of variables that are factored in. But one thing that wasn't really well known is, across the standard antibodies, how well is the inter assay comparison? There had been some work from a group in Singapore, a very nice paper suggesting that at the higher cut points, the agreement was pretty good across the assays, CPS greater than 5 and greater than 10, and maybe slightly less so at the lower. They had used a different method, which was not really what is standard, and they had used multiplex immunofluorescence or IHC. This is not a validated method for PD-L1 scoring. So that was an open question, sort of. Although they laid a very important piece of data down, we wanted to use the most standard assays and essentially do a very similar analysis, but using the standard scoring criteria.

Dr. Abdul Rafeh Naqash: Very interesting. So, could you walk us through the approach of how you looked at this question, what kind of samples you used and what kind of testing algorithms you implemented to look at the cross validation of these three different antibodies?

Dr. Samuel J. Klempner: The antibodies were chosen primarily because those are the standard ones that either have companion diagnostics or have been used most commonly in phase 3 trials. So 22C3 has most commonly been linked to pembrolizumab, 28-8 to nivolumab, and 263 used with Roche and Genentech trials primarily. And so we selected the antibodies based on the common use. We selected the scoring systems of CPS and TAP, again based on the most commonly used and validated scoring algorithms in gastric cancer. And then, although most patients in clinic and metastatic disease present with biopsy samples from the primary tumor, there may be some limitations in biopsy samples in terms of small amount of material and ability to reliably count 100 cells, etc., for CPS. So we actually use surgically resected samples from a commercial biobank, 100 samples, and essentially 28-8 was really the reference. And we picked samples that, using 28-8 CPS PD-L1 expression represented the entire spectrum, meaning we had CPS less than 1, we had greater than 1 and less than 5, greater than 5 and less than 10, and greater than 10, so that we could compare across these different strata, because those are the most common strata that have been used in clinical trials and linked to magnitude of benefit.

Dr. Abdul Rafeh Naqash: And something that, interestingly, I see here when we go to some of the results, and I'm pretty sure you'll talk about the concordance, is the correlation coefficient seems to increase as the percentage positivity increases for a certain antibody. Could you try to help us understand why that might be the case? Is it because it's easier for the pathologist to look at the slide when there is a certain level of positivity that crosses a certain threshold? Or could there be some other factors that are not well understood.

Dr. Samuel J. Klempner: Yeah, it's a totally good question, and I think it's something that's seen in other IHC biomarkers as well. If you look at HER2, you'll see some similar trends. The agreement at IHC 3+ is pretty good and greater than it is at lower cut points. And having talked to multiple pathologists, and I'm not a pathologist, we had three pathologists scoring all of these samples, and essentially, it's what you might expect. It is just easier when there's a lot of the marker. It is easier to judge the high extremes of the strata. So the agreement at greater than 10 is quite good, and this has already been shown by others. It's just an easier thing to score for anyone. The agreement is better across all of the assays at higher cut points, whether it's TAP greater than 10% or CPS greater than 10%. And you can see that pretty clearly in our data, and it's also been shown in other data sets looking at roughly similar questions in other tumor types.

Dr. Abdul Rafeh Naqash: Going to the interesting results that you have in this paper, could you highlight for us some of the important findings that you had and put them into context of what their clinical implications may be?

Dr. Samuel J. Klempner: Yeah, I think I'll start with the clinical implications so that what clinicians, and we're both clinicians, what we want to know is, if I have a report that says the CPS is greater than 1 and it's done with a 22C3 test, is that also likely to be greater than one if it had been done with a 28-8 test or scored with a different algorithm - CPS versus TAP? So, essentially, some degree of confidence on the interchangeability between the assays themselves, that is really the clinical implication. And so, to accomplish this, we set out to basically do the comparisons you'd have to do to convince yourself that that is true. So you take samples against a reference range, in this case, across the PD-L1 strata, you pick a reference test, in this case, 28-8, you have one pathologist be the start, and then you compare other pathologists against each other and that person, and you look. And in the pathology literature, they have strata of agreement which tend to go from poor, moderate, good to excellent. And these are sort of accepted standards in the pathology world about inter reader agreement. So between one pathologist and another, and things that are moderate or good are considered essentially acceptable at interchangeable levels.

And so, as you suggested, at the higher cut points, the agreement is very good. The clinical interpretation of that is that if you get a TAP greater than 10% scored on a 22C3 antibody on a Dako staining system, you can feel relatively confident that that would also be called a TAP or a CPS greater than 10 by a 28-8 antibody, suggesting there is good agreement between the two antibodies at that cut point. As you move down, there is a little bit less agreement, and that is consistent with what's been shown before. But in our data set, the agreement was still pretty good across all three of the antibody clones, even at the lower cut point, so greater than 1% for TAP or CPS greater than 1. And that provides, I think, some reassurance to clinicians that whatever test their own pathology lab is using, if it's one of these three assays, they can provide some degree of confidence that what they're seeing would be similar to what they were seeing if it had been done with another test.

Dr. Abdul Rafeh Naqash: I think that that is very important, because even though we do want broad testing in general for metastatic tumors, as you probably will agree with, but there's a lot of practices still that institutions tend to do their own testing with limited gene panels or even IHCs. So I think to put that in the context of your study, as you said, if you have a certain antibody that is positive, as you've shown, then that also likely means that with another antibody that your institution may not test for, it's likely the tumor sample is likely going to be positive at a similar level.

So I think you also used digital pathology as part of this project, even though that may not be the most important aspect. As we move slowly and steadily towards artificial intelligence and machine learning, could you tell us how you incorporated the digital assessments and how you utilize them to correlate with the pathologist assessment and the futuristic perspective of how we could eventually try to incorporate digital pathology assessments for this kind of staining approach, which might limit interobserver operability differences as well as time constraints?

Dr. Samuel J. Klempner: I hope I can do this part justice, because, again, I'm not a pathologist. But the digital imaging analysis was really essentially used as a quality check and verification tool in our own paper. Our intent was not to establish DIA directly as a superior methodology to TAP or CPS, but simply to provide ourselves some degree of confidence in the staining pattern and distribution across the three assays, and whether or not this would generate significant differences in what the PD-L1 score would have been called. And so, the bottom line is, the digital imaging analysis suggested there were very minor differences across the three assays in terms of, like, percent cell positivity, which is one of the main readouts, and the mean difference was actually quite small. So we felt that the digital imaging analysis, which was really considered somewhat exploratory in our own work, supported what we saw with the pathology comparators read in traditional methods. I think it sets somewhat of an initial pilot data benchmark to say that maybe we can think about moving tools like digital imaging analyses forward in terms of PD-L1 scoring approaches in the future. But it does not provide adequate data to say that we can do this now or we have enough samples and enough comparisons to say that, “Hey, for sure, digital imaging is equivalent to pathology reading.” I think that we're getting there and our data supports that that may ultimately be the conclusion, but for us it was really essentially an orthogonal support and sanity check for our traditional approach, which is, of course, a pathologist based scoring. So supportive and suggestive, but not definitively conclusive.

Dr. Abdul Rafeh Naqash: Definitely early days for visual pathology assessments, but I think that it's a very rapidly evolving field, and hopefully we'll see more of this in the next few years, as well as incorporating some assessments into clinical trials.

Now, shifting away from your honorary pathologist role as part of this project to your actual role as a clinician investigator/clinician scientist, could you tell us your career trajectory, how you started, how you've self paced yourself, and how you've tried to mentor certain different individuals in your current role?

Dr. Samuel J. Klempner: Yeah, I remember my grandfather and other people telling me, just try to leave it a little bit better than you found it. And so that's, I think, a guiding principle. I hope that at the end of my own career, I can leave oncology a little bit better than when I started. I think the best way to do that is to mentor and train the next generation who are going to drive these practices. I started, like many others, personally touched by cancer in my family, which started me on a journey towards oncology, was somewhat frustrated by the lack of options available to my mom, and then became deeply interested in the science and how come we knew so little about cancer, so spent a fair amount of time in labs, and had a really formative experience with Lew Cantley looking at PI3 kinase resistance and signal transduction, and wanted to learn to speak the language and interact with people driving the lab based work. And that's been something I've tried to keep as central to my career as someone who has a very strong translational interest.

And so I try to think of ways that I think we can learn from every single patient and every subgroup. I mean, for example, in our own work here, it's very unclear if there's a biology linked to the different PD-L1 strata. So for example, does a PD-L1 CPS greater than 10 tumor have a very high interferon gene signature? Or are there features of the T cells that are different between a CPS 10 or higher versus a less than 1? So PD-L1 is a biomarker, but is it really telling us about biology? And so these are the types of questions that I try to stimulate in all the residents and fellows and hopefully it will drive translational projects. But I think just having the conversations and asking the questions and talking to people. I mean, I love the ASCO Career Lounge and always try to do that when possible. I know you do the same. I think staying curious is really the thing that I try to remain in life and also in my career and have fun and enjoy with your colleagues. And I think that will make us all better researchers and ultimately translate to better outcomes for our patients, which is, of course, why we all do this.

Dr. Abdul Rafeh Naqash: Wonderfully said Sam, thank you so much. Thanks again for choosing JCO PO as the final destination for your work. Hopefully we see more of the similar work that you do in your field in JCO PO. And thank you for talking to us about your journey as well.

Dr. Samuel J. Klempner: Yes, thanks for having me. I'll talk to you sometime soon.

Dr. Abdul Rafeh Naqash: Thank you for listening to JCO Precision Oncology Conversations. Don't forget to give us a rating or review, and be sure to subscribe so you never miss an episode. You can find all ASCO shows at asco.org/podcast.

The purpose of this podcast is to educate and to inform. This is not a substitute for professional medical care and is not intended for use in the diagnosis or treatment of individual conditions.

Guests on this podcast express their own opinions, experience and conclusions. Guest statements on the podcast do not express the opinions of ASCO. The mention of any product, service, organization, activity or therapy should not be construed as an ASCO endorsement.

Disclosures

Dr. Klempner

Stock and Ownership Interests
TP Therapeutics
Nuvalent, Inc

Honoraria
Merck Serono

Consulting or Advisory Role
Atellas Pharma
Bristol-Myers Squibb
Merck
Daiichi Sankyo/UCB Japan
Sanofi/Aventis
Mersana
Exact Sciences
Novartis
SERVIER
AstraZeneca
Amgen
I-Mab
iho Oncology

PD-L1 Assay Concordance in Gastric Cancer

JCO Precision Oncology Conversations