Round 1: Is DLL3 a Clinically Actionable Predictive Biomarker in NEC?

Early clinical data suggest a potential relationship between delta-like ligand 3 (DLL3) expression and response to DLL3-directed therapies, but predictive utility has not yet been established. In Round 1, faculty examine what current clinical evidence does—and does not—show, while considering whether DLL3 can be measured and interpreted with sufficient reproducibility to support predictive use. 

To hear arguments related to who to test and when, see Round 2. 

To hear arguments related to clinical actionability and implementation, see Round 3.

To view the full series, click here.

Transcript

Jonathan Strosberg, MD: Is DLL3 a clinically actionable predictive marker in NEC? So there's been 1 completed phase 3 study, the DeLLphi study of tarlatamab versus placebo in small cell lung cancer after first-line chemotherapy. The primary endpoint was overall survival and it pretty strongly met its results with median overall survival of 13.6 months with tarlatamab versus 8.3 months. Obrixtamig is another T-cell engager. It's being evaluated in small cell lung cancer as well as extrapulmonary neuroendocrine carcinoma. And in this phase 1 study of obrixtamig, the results were stratified by DLL3 expression with DLL3 high being defined as greater than 50%, DLL3 low less than 50%. There was no mention of intensity of staining, just percentage expression. 

And you can see there was a pretty dramatic cutoff between DLL3 high results with response rate of 40% versus only 3% or just 1 responder in a DLL3 low population. Another study of obrixtamig plus chemotherapy in the first-line setting, DAREON-7, showed a median progression-free survival of 7.7 months. It required DLL3 positive staining, but any positivity. There was no threshold and there was no stratification by intensity or percentage. Alveltamig is actually a trispecific T-cell engager. It binds 2 epitopes on DLL3 as well as CD3. And here they did stratify by DLL3 expression greater than 50% versus less than 50%. And responses were seen almost exclusively in the greater than 50%. So the response rate was 33% in 10 milligrams, 56% for 30 milligrams. For DLL3 less than 50%, the response rate was 0% in the low dose and 7% in the higher dose. 

So we have 2 studies that show fairly significant demarcation between DLL3 high, more than 50% in DLL3 low, less than 50%. Gocatamig is another trispecific T-cell engager. It binds DLL3, CD3, and also albumin, which extends the half-life. And it was studied both in neuroendocrine prostate cancer as well as in other types of neuroendocrine carcinoma. We only have the results here for neuroendocrine prostate cancer. So the response rate was 15% and 88% of patients were DLL3 positive, but the response didn't correlate with DLL3 expression, but this is really small numbers. So we're starting to get some indication that at least in the extra pulmonary neuroendocrine carcinoma space, percentage DLL3 expression seems to make a difference, but certainly not definitive evidence. So I'd like to invite Dr Raj to come up and talk about the predictive utility of DLL3. 

Nitya Raj, MD: Hi everyone. Good morning. Okay, so with all of that evidence as a background and a foundation, the question is what these findings do in terms of establishing predictive utility of DLL3. And so I am going to speak from the viewpoint of the predictive utility of DLL3 has not yet been established in neuroendocrine carcinoma. So Dr Strosberg walked through this data a little bit already, but here's the beautiful overall survival curve from the DeLLphi-304 study. And so DeLLphi-304, again, was done in small cell lung cancer. And because DLL3 expression is near universal in small cell lung cancer, there was no DLL3 testing requirement to enter the study. So this is any previously treated progressive small cell lung cancer patient. And what they saw in DeLLphi-304 is a very, very clear overall survival benefit with the use of tarlatamab over chemotherapy. 

So median overall survival was 13.6 months with use of tarlatamab and 8.3 months with standard chemotherapy. So this is very, very promising data. Obviously now tarlatamab is approved in progressive small cell lung cancer. But the question is because DLL3 expression is near universal in small cell lung cancer, these data establish therapeutic relevance, but they really don't help us in determining whether expression level predicts benefit. And that's the big question for our extrapulmonary neuroendocrine carcinoma patients.  

And so this is digging into some of the early extrapulmonary neuroendocrine carcinoma data in a little bit more detail. And so what we're starting to see in the studies that have been done is that there's an expression-response relationship that we are trying to understand and learn more about. So here's the data from the alveltamig study. And so what you can see in this study is that they, again, defined an expression of greater than or equal to 50%. And you're seeing potentially a trend towards increased benefit in that cohort that is higher expression with the 50% cutoff. 

And then here too, we have some of the early data with obrixtamig, and as Dr Strosberg had explained, tumors were cut off with this DLL3 high expression of greater than or equal to 50% or DLL3 low, which was less than 50%. And again, you're seeing this greater benefit with the T-cell engager in the population of tumors that are DLL3 high. And so these are 2 separate programs, 2 different drugs. And what we're seeing in some of the early data is that the preliminary expression associated response differences really are playing out amongst different therapeutic treatment options.  

And so, I think the big question and what we're trying to dig through and understand a little bit more is that we really need more clinical evidence and standardization to establish the predictive utility of DLL3 in our extrapulmonary patients. And so what are the limitations of some of the data that we've heard about over the past few slides? 

And I think the things that we're thinking through is that these are nonrandomized controlled trials. So these aren't randomized studies. They're earlier phase, phase 1, phase 2. And so they're not yet powered to validate the predictive utility irrespective of cutoffs or no cutoffs. And so that's the big challenge right now that we're grappling with. Response did not correlate with DLL3 expression in the small gocatamig neuroendocrine prostate cancer cohort. But again, that's a very, very small sample size. The trial is relatively early. So we're waiting for more patients to be treated, following patients longer, seeing how that data plays out.  

And again, as you've heard a few times now, really we don't have any approved therapy that is currently using a defined DLL3 expression threshold. So a lot of these studies are now subdividing the tumors by this greater than or less than 50% cutoff, but we don't have anything approved yet. 

So we're still learning, we're still gaining the data. In terms of standardization, while many studies are using this 50% cutoff, there is no universally accepted definition of DLL3 positive or DLL3 high. So even now when we ask our pathologists, "Could you test the tumor for DLL3 and do an IHC test?" We get a readout in the clinical report of some percentage. And in the clinic, we're still struggling to figure out, well, is that DLL3 high? Is that enough expression for me to even try to get tarlatamab compassionately? That's what we're playing with versus trying to get them on a clinical trial as more trials are becoming options for our patients.  

And all of these scoring methods, these expression thresholds, they are so much subjective still, I would say, and they are varying across the studies, varying across institutions. I'm sure Dr Roden can speak to the comfort of the pathologist too and doing this testing on the tumors and putting these results into the pathology reports. Okay. So I'm going to pass it back to Dr Roden. 

Anja Roden, MD: Yeah, thank you. So the next question is, is DLL3 sufficiently well defined and reproducibly measured to support the clinical interpretation? So most of the studies and trials use the Ventana DLL3 SP347 assay. And that is good because we learned from the almost disaster with the PD-L1 where we have multiple clones. They're all associated with different trials. It is a nightmare for the lab to actually establish 4 different clones and also for the pathologist to have different readouts for each clone.  

So far, a systematic review has also shown that most commonly used and extensively validated DLL3 assay is the SP347. We had performed a inter-observer variability study for the expression of DLL3 amongst 3 pathologists. And at that time, we used a cutoff of 50% of tumor cells expressing DLL3. And using this cutoff, the agreement was actually pretty good. It was almost 83%. And here are 2 examples. So both examples are small cell carcinomas. You can appreciate the cells. 

They are less than 3 times of lymphocytes in size. They have a high nuclear to cytoplasmic ratio, meaning they only have very little cytoplasm. And you can even see nuclear molding here. And on the upper right-hand side, this small cell carcinoma diffusely expressed DLL3 in a 3+ intensity, and all 3 pathologists agreed more than 50% of tumor cells expressed the DLL3 in a 3+ fashion. On the lower left-hand side, this is another small cell carcinoma, and all 3 pathologists agreed that less than 50% of the tumor cells expressed DLL3 in a 3+ and 2+ fashion.  

There was also an abstract a few years ago that showed lab reproducibility for the analytic agreement across various lab conditions and assays, and this was over 90%. So these data are promising. However, they have not been validated and this lab study has not been published. So I think there is much more to do. And then even though most of these studies and trials use the same assay, there are differences in expression patterns between the various tumors.  

And there are also, as you just heard, different thresholds that have been used for DLS trials. And these thresholds include, we include all the patients that have at least some staining of DLL3 to having at least 50% of the tumor cells showing staining of DLL3. And then this study here on the right side actually also shows you very nicely the different expression amongst different tumors. So above the line are the lung tumors and in blue are the small cell carcinomas of the lung. And they looked at an age score and the mean age score here was about 124 lung small cell carcinomas. In green are the lung large cell neuroendocrine carcinomas, they have a very similar mean age score.

If we look at the extrapulmonary neuroendocrine carcinomas, here's an example of the uterine cervix small cell carcinomas. They had a mean age score of about 150, but the uterine large cell neuroendocrine carcinomas had a mean age score of only about 70. Similarly, if you look at the Merkel cell carcinomas, the mean age score was about 150 while the medullary thyroid carcinomas had a very low mean age score. So there are differences in the DLL3 expression amongst various tumors, even from the same site. 

Dr Strosberg: All right. So maybe we can have a little discussion among the panel. So based on the current evidence, what can DLL3 expression reliably tell us today and what additional evidence would establish predictive utility? And I guess one question perhaps for you, Dr Roden, is do you think percentage expression will ultimately be more important or H score or any predictions on that? 

Dr Roden: So, I don't think we can really answer this question yet without the clinical trials. I mean, we can, I would say, pretty reliable tell you the percentage of tumor cell expression, at least if we would use a 50% cutoff. I don't know about other cutoffs. We would have to actually look into that.  

And you also know there will be some inter-observer variability if we go for intensity. So my 1+ might be my colleagues’ 2+ or my 3+ might be my colleagues’ 2+. And on top of it, there will be variation in intensity even within 1 tumor. So if you see a resection specimen in 1 area, you will see 3+. In another area, you will see 2+. So we can report all of that, whether we can report it very reliably, needs to be studied. But what that really means as a predictive marker, we don't know until we really have these all in clinical trials, I would say. 

Dr Strosberg: Great. And let me ask you 1 more question. To what extent does DLL3 expression vary, you think, inter-tumorally and intra-tumorally if you happen to have more than 1 site? 

Dr Roden: So this other good question. There is 1 small study that looked at small cell lung carcinomas at the resection specimen and then at the recurrence. And there was about an 88% concordance. However, this is only 1 study and we have not looked into the distinction between primary tumor and metastasis and primary tumor and then recurrence in extrapulmonary neuroendocrine carcinomas. And also the study on small cell carcinomas has not been validated. Based on my experience, we do see different intensities within 1 tumor in resection specimens. Now, mostly we get biopsies, so you can't really talk about that, but there will be variation, I'm pretty sure. 

Dr Strosberg: Got it. And Dr Raj, am I correct in saying that they're doing DLL3 imaging studies at Memorial? I think there's a PET scan that's being developed. 

Dr Raj: Yeah. I mean, there's a lot of interesting early work going into DLL3 PET scans as a means to look by imaging at the distribution of the receptor. And I think it speaks to the question you just asked about heterogeneity, like amongst what we may see by IHC testing in the primary, the metastases at recurrence and whatnot. And so there's some early work that we're doing at Memorial looking at expression on PET scans. And then the next step is to look at that in the context of response to therapies targeting DLL. 

Dr Strosberg: Great. And what clinical evidence would convince you that DLL3 expression is predictive? 

Dr Raj: Yeah. So this is an exciting time for extrapulmonary neuroendocrine carcinoma. I mean, we have not had many therapies. We've had platinum-based chemotherapy and then we extrapolate our second-line regimens from site of origin specific options for non-neuroendocrine carcinoma disease. And so for extrapulmonary neuroendocrine, the fact that we may have a biomarker that we might be able to look at and target is exciting. So some of the evidence we have already that we've talked about is very promising in terms of some of the response rates and PFS data that we're seeing. But what we need now is some of the bigger phase 3 data, which is hopefully coming soon, in the next few years. But that randomized data taken together with some of the phase 2 data, specifically looking at different thresholds is what's going to help us decide what is truly predictive in terms of DLL3 expression. 

Dr Strosberg: Great. Thank you.