Good morning, everybody. My name is Daina Graybosch. I'm a Senior Managing Director here at Leerink Partners, and I cover immuno-oncology, and I have the pleasure of covering Compugen for several years. Thank you this morning to Eran, who's the CSO of Compugen. We're going to do a fireside chat, and I'm going to turn it over to Eran to give a quick intro of the company, and then we'll get into the questions.
Thank you, Dana. Thank you for having us. Compugen is a clinical stage company focusing on immuno-oncology and also a pioneer in computational target discovery. We use a computational platform called Unigen to identify novel drug targets in the field of immuno-oncology. Actually, we discovered TIGIT around the same time Genentech did, and we discovered fewer layers to PVRL2 and other first-in-class assets. We have two wholly owned assets: COM701, blocking PVRLR, and COM902, which blocks TIGIT, a potential best-in-class antibody. We also licensed to AstraZeneca the right to use our COM902 as part of their TIGIT PD-1 bispecific, and we are going to discuss this trial soon. Also, more recently, we identified another first-in-class IL-18 binding protein, and we developed a blocker antibody, COM503, which was licensed to Gilead. We also have some more non-disclosed assets, all of them stemming from our computational discovery platform.
Awesome. I want to start with the program, the bispecific that you've licensed to AstraZeneca, because they have gotten incredibly aggressive with that program in the last year. Both the milestones and the potential royalties are pretty meaningful to Compugen. Rilvegostomig, do I say that correctly?
Rilvego, the nickname.
We'll call it, you say, rilvegostomig.
Rilvego, yes.
Rilvego.
Easier.
Rilvegostomig. It's a TIGIT PD-1 bispecific. It's derived from your TIGIT antibody. They have many phase three combos, and they've started a lot of them in the last year. I wonder, as you look across their portfolio of programs, is there any approach or trial that you have relatively more confidence in, given the data they've published to date, or by other TIGIT molecules you think are relevant to theirs?
Yes, I think the most established data from them and from others is in the non-small cell lung cancer, PD-1 above 50. They actually showed really nice results, encouraging more than 60% ORR in these settings. TROPION-L ung 10 probably is an interesting one, in which they combine rilvegostomig or rilvegostomig plus datopotamab deruxtecan, their ADC, TROP2 ADC, versus pembrolizumab in these settings. I think this is definitely one to follow. They also showed some nice data with gastric, also some other TIGIT assets. I think the gastric trial, for example, is ARTEMIDE- Gastric 01, in which they combine rilvegostomig with Enhertu, which is a very powerful drug by itself, also an interesting one. I think also the others, I mean, rilvegostomig, I mean, definitely, as you said, the aggressive strategy, kind of replacing the IO backbone with this PD-1 TIGIT bispecific is really fascinating for us.
Actually, maybe let's talk about a couple of those. You mentioned the troponin lung 10, and we talked about it actually after your last earnings, which is an interesting design from a milestone perspective, because I think it's clinically de-risked, even if TIGIT's not necessarily adding anything, because they have the bispecific plus their TROP2 ADC, Datopotamab deruxtecan, versus pembrolizumab. You are really stacking the deck. You have three mechanisms versus one. Clinically, it seems very likely to be successful. I wonder if there is some regulatory or commercial risk to that. Will regulators ask about contribution to components of both sides of the bispecific? Though they do not typically, it sort of feels like this is going to push that question. If any trial pushes it, this might be one.
Yeah, actually, looking across the seven phase three trials, most of them not necessarily compare rilvegostomig to pembrolizumab. For example, in this Artemide Gastric 01, they are combining rilvegostomig with Enhertu and chemo, and they compare it to pembrolizumab plus Herceptin plus chemo. Also in this trial, they're not really comparing directly rilvegostomig to pembrolizumab, and also in most of the other trials. First of all, I don't know. I'm not part of the discussion with AstraZeneca and FDA. I just assume that they are building these trials toward approval. They do expect to see that if the risk-benefit profile and what is great about rilvegostomig, not only it's effective, but it's also having a non-active Fc, has a very good safety profile from what we've seen until now. It's a very good combination partner.
Overall, probably what they will have to show is that if the risk-benefit of adding rilvegostomig to TROP2 ADC versus pembrolizumab is going to win, both in terms of benefit, but also in terms of the risk, they'll probably get approval. Again, this is not, we're not part of the discussion with the FDA and AstraZeneca. Just our view on that.
There's another interesting component of this, which I'd love your opinion on, because you do so much work biomarker and AI-wise. They're using this biomarker, the TROP2 QCS, which they developed with AI also to enrich patients in that trial. I think some of the concern we've heard from investors and KOLs is that they developed that biomarker in a post-hoc manner, and that then they're using it prospectively here. How validated is it really? Could they be chasing a signal? I wonder if you have any thoughts on that.
Yes, so normally, every biomarker starts by post-hoc analysis and defining criteria. Obviously, at certain points, you need to prospectively select patients based on the biomarker. There is always a risk. I think, from my experience, since you shouldn't have much of bias doing this retrospective analysis if you're doing it right, and I'm sure they are doing it right. I think the risk is not huge. Yes, you definitely need to see how it works in prospective selection of patients based on the biomarker and the sophisticated AI-based approach. We'll have to see how it works, but I'm positive on that in general.
Both the trials you mentioned, the gastric one and the lung, have an ADC component as they're taking it forward. Have we seen any data with rilvegostomig and ADCs? More theoretically, or from your preclinical work, do you expect mechanistic synergies with TIGIT and ADCs? Anything different than PD-1 alone with ADCs?
Yeah, first of all, we will see this year, according to AstraZeneca, initial data from combining rilvegostomig with ADCs. Definitely eager to see that. I'm not sure if there's something specific. We know that the ADCs, which are eventually smart chemotherapies, are inducing immunogenic cell death and reduce tumor burden. There's synergistic with checkpoints in general. I'm not sure if some chemo might increase the expression of PVR, but I'm not sure specifically these ADCs. Overall, I would assume it should be synergistic, at least like with PD-1 and other checkpoints. I'm not sure of something specific for TIGIT there.
That's interesting. Maybe we should tell everybody what PVR is.
PVR is the ligand of TIGIT. Yeah.
Yeah.
Thanks.
Interesting. Let's, I guess, move into your own COM902 program. Sort of what read-through do you see from rilvegostomig to your own program, given it's half of the bispecific? As you think forward, if the bispecific works, it has all these regulatory advantages, and it has AstraZeneca investing behind it, where could you find success with COM902 that wouldn't have already been taken by rilvegostomig?
I think there is definitely a direct impact on 902, because 902, like rilvegostomig, is an Fc-reduced binding, so the same class of TIGIT antibodies, and eventually it's the same TIGIT antibody. Success with rilvegostomig is definitely going to reflect on 902 and actually make it a validated clinical asset. There are quite a few opportunities. Obviously, other companies who might want a good best-in-class potential TIGIT antibody could have an interest there. For our own combination, when we add COM701, we'll discuss this soon, that's two, PD-1 and TIGIT. This actually opens the door to take what we believe and have started to show, taking TIGIT and PD-1 into less inflamed indications. This is something that we could do with COM902.
I think it will be great for COM902, both from our own internal validation and moving with the target, and also as a potential asset for collaboration with others.
Let's talk about your PVRIG antibody COM701, and maybe we come back to TIGIT, but I want to make sure we talk about it. Can you summarize the data that supports that PVRIG will prove to be an important IO target in ovarian cancer, which is where you are developing it, particularly as other IO therapies have had a little bit of activity in ovarian, like a glimmer, multiple of them, PD-1, IL-2, CTLA-4, but that really hasn't justified ever later phase development?
Yeah, ovarian is definitely a tough one for immuno-oncology, and there are many attempts and not many success. There are a few good reasons why we think we should be different with COM701. First is biology. The PVRIG pathway is highly expressed in ovarian. It was a target indication for us from the beginning, and we think that this is a good indication to test PVRIG blockade in general. Looking at the biology of specifically PVRIG compared to other checkpoints, it has a very different biology. Its biology, which relates to its expression of stem-like memory T cells and dendritic cells, has shown to increase inflammation in less inflamed tumor types across few indications. This unique biology should be able to drive T cells in a less inflamed indication like ovarian cancer.
What we saw across all our trials was signal activity in places where checkpoints are typically not working. There is a biological reason to think why PVRIG blockade could work where other checkpoints work. The clinical data, which is probably the most important, the proof is in the pudding. We did not test many patients in monotherapy, only six, but one of them responded, long-term duration of response to PVRIG to COM701, and three more were stabilized by monotherapy treatment, and all the translation signals showing that we were able to modulate very effectively tumor environment in ovarian cancer patients, last line platinum-resistant patients with monotherapy. Looking at the triplets, I mean, historically, as you said, PD-1, even PD-1 plus TIGIT showed 8% response rate in platinum-resistant ovarian cancer. We tested 40 patients, and we had roughly 20% response rates.
The reason we did 40 patients was to make sure that, yes, we are above that signal of the PD-1 plus TIGIT. We have the biology, and we have the clinical signals that COM701 blockade and mono or combination is active in platinum-resistant ovarian cancer patients. These are heavily pretreated patients. Some of them attempt a few clinical trials before even entering our trial. Now the rationale is, let's go earlier into the platinum-sensitive settings. We identified this group of patients, which are second or third line. They already received maintenance with BEV, already received maintenance with SPAR, already cannot tolerate it. These patients received the platinum chemotherapy like the six cycles, and then they just sit and wait to become platinum-resistant. They do not have any maintenance option there.
We have COM701, which is clinical signal, excellent safety profile, and the patient who did respond in our trial showed really extremely long duration of response. This is exactly the type of drug you want in these maintenance settings, and this is why we're doing our trial in these settings. Also, by the way, the competitive landscape in the platinum-sensitive setting is completely different from the platinum-resistant, which the ADCs are now very moving aggressively into that indication.
Maybe if you follow up, do you know why ovarian present with high PVRIG implies something about the immune cell composition within the tumor?
Yeah, it's a good point. We are not sure. What we've seen is that in hormone-driven indications like breast, endometrial, and ovarian, for example, we do see upregulation of the ligand, PVRL2. I remind you that, for example, we had complete response in HR-positive breast cancer patients. We had responses in endometrial. We do see also clinical signals in these indications. PVRIG itself, with its expression of stem-like memory T cells, the new publication showed that platinum actually induces TLS and stem-like memory T cells. This also suggests why we think it will synthesize the tumor to respond to COM701. This might explain also why PVRIG itself is relatively high in ovarian cancer. We're not sure, but definitely we see dominance of the pathway in ovarian cancer in general.
Actually, I don't actually understand. Is it both the ligand, PVRL2, and PVRIG?
Both of them are relatively high, yes.
That are relatively high.
Yeah.
Interesting. You early on had pretty interesting biomarker data that showed that patients that expressed the ligand, you had an enrichment of clinical benefit. That was in the first 20 patients, I think, or an early cohort. Are you still developing that biomarker as you move forward? What can we learn from this next study in the platinum-sensitive population about that biomarker and enrichment strategy?
Yes, so obviously, again, going back to our results, we had these patients who responded really nice to the drug and some who have actually less benefit. We looked retrospectively. We used all our tools, computational analysis, because obviously the Holy Grail is to identify this population of patients. We have initial observation of PVRL2, the ligand of PVRIG, being able to be highly expressed on patients who responded. We had similar observation in a small breast cancer cohort. It was interesting, extremely important, but at this point in time, we do not have sufficient data to really start prospectively selecting the patients. We are going to continue to follow. It is a bit more challenging in the platinum-sensitive because the availability of really pre-treated biopsies is not as abundant as you can get in the late light settings.
We are definitely going to continue to follow up on that because this is critically important and the signals were there, and we just need to substantiate it much more before we start prospectively selecting.
You're not going to require biopsies to enroll because you don't want to slow down?
We cannot. Yeah, we cannot. In these settings, we're probably going to have some archival and whatever we can get, but it's not going to be a mandatory pre-treated biopsy, probably.
Is there anything you can look at sort of in ctDNA or in liquid biopsy, the CTCs?
We didn't disclose all of that, but obviously, as we always do, we put a lot of efforts into translational. ctDNAs are not clinically validated in ovarian, but it definitely could be an exploratory readout we might consider to explore to support the clinical readouts.
Got it. What gives you confidence that COM701 will have single-agent activity in this platinum-sensitive? You previously were building and building to doublets to triplets, and now going back to a monotherapy.
First, the biology and the clinical signals we have seen in monotherapy in platinum-resistant ovarian cancer. Then moving into the earlier settings, we know that looking at the suite of plots in our study, you see some of the patients responding really long-term and deep responses. Some of them in the platinum-resistant settings, again, after so many trials, so many failures, they progress so fast. I mean, they probably never had the chance to respond to a treatment which is chemotherapy-free, like the one we used. Now, going to the platinum-sensitive, having debulking first, we're going to take only the patients who responded to the treatment. They have low tumor burden. They've been synthesized by the platinum probably to respond better to COM701 due to this TLS and TSCM. In this earlier line, the immune system in general is probably more functional.
I think that while we're not able to enrich by a biomarker PVRL2, we think that we are going to enrich for responders by looking at this line of treatment after responding already and have low tumor burden to the platinum. This is the rationale, and I think we have a good, I mean, we have a good reason to believe it should work, but definitely we didn't test it yet in the platinum-sensitive, and we have to wait and see.
Maintenance feels like a very natural place for any checkpoint inhibitor or immune therapy because we know that tumor burden in patients with high tumor burden is, I think, maybe consistently the most correlative clinical signal to lack of response. We have not seen that many companies use this path. Now, there is adjuvant post-surgery, so early stage, that is certainly a path. Merck, Coga, and Pfizer used it in bladder cancer. It has not been used very many other places, and I wonder if you could speculate on why, and does that introduce any risk that it is a less trodden path?
I think in general, I mean, for checkpoints and also for other clinical development, many times you look for the signal in the last line settings, and then you move earlier. I mean, also in the last three, it was not maintenance, but they saw some signals in the relapse settings, and then they moved to the first line melanoma trial. I think it makes sense. I think that specifically that population that we identified in ovarian, we had tons of discussion with KOL to really identify this exact population, exact unmet need, and we're really glad we're able to identify it. I think it's, and there are some trials, for example, the GLORIOSA trial, they tried ADC plus BEV versus BEV, not exact same patient population, but people are looking at different attempts for maintenance.
Specifically for checkpoints, I mean, there are some approvals, and again, in a way, it resembled the adjuvant settings, but I think this is definitely a place where it makes sense. Specifically, we believe that PVRIG is such a checkpoint that should work in ovarian where other checkpoints failed in maintenance settings in ovarian till now.
One concern we have about the study is you're using PFS as a primary endpoint. It's randomized, which is nice, but you only have 60 patients randomized two to one with PFS, and PFS overall, but in this setting, if you look historically, is quite variable. I guess the worry is that you don't end up with a balanced population in the two arms, and it will be difficult to interpret. Why am I wrong?
Yeah, you're not wrong. Obviously, if we do a 300-patient study, then we have more robust results, and we could maybe do clearer conclusions. To think, even though we're quite confident in the signals that you've seen in monotherapy and in the biology, we're now entering for the first time into the platinum-sensitive settings. This patient population is really defined. We're really focusing on unique, so we think that the chances for bias are relatively small. The historical control for this population is very stable across few phase three trials, so we know what to expect. We also have the internal randomized control. Based on our discussions with KOLs, with Biostats running the simulation, we think that this size of study should be sufficient to enable us to reach into a conclusion.
The conclusion with a relatively, you also have the risk of false positive, false negative, but statistically, it's going to be sufficient for us to make a decision with relatively high certainty. You know, we can do a few things because this is an adaptive trial design. Either the trial is futile, okay, so it's good to know that we don't see a signal there. If we do see a signal, we expect for three months' prorogation of PFS. If we do see a signal, then we can do a few things. The way the trial is designed, we can continue to enroll more patients and consider now going with regulatory authorities to go into the fast track approval. We can stop the trial and say, okay, now we have an estimation of the magnitude of effect.
We can start now, design now properly phase three study, or we can now add other combinations. This is exactly the adaptive trial. We can add BEV, we can add triplet, and see what happens. I think this trial design is exactly sufficient for us to make an early conclusion, to see where it goes, and we are quite flexible on the next steps.
There is one other IO drug in phase three right now in ovarian. It's Mural Oncology. It's nemvaleukin alfa, which is a not alpha IL-2. They're combining with pembro in platinum-resistant ovarian cancer. If that happens to be successful, and I think it reads out this year, would that change anything about how you're thinking about COM701?
First, I think they're going to the platinum-resistant settings. We're going early to platinum-sensitive. There's not going to be direct competition. Looking at the decision-making they made, it's interesting to see because I think overall they treated, they had data for 14 patients, and the confirmed response rate, as far as I've seen, is roughly as ours, 20%. They made the decision, which is obviously, I mean, they took a risk, but obviously a decision that we could have taken as well to start a big phase three trial in these settings of platinum-resistant. We took a different route of saying, okay, we think that it's a bit more challenging for a chemotherapy-free, especially given the ADC's competition to remain in that settings. We thought we have higher chances to move into the earlier settings in the maintenance settings.
First, it's not a direct competition for us. Definitely interesting to see, but just a different strategy.
Got it. You wouldn't then combine with IL-2, for instance, or you could consider that later on anyway.
Maybe with COM701 with PVRL2.
In your adaptive trial. Yes. Let's move on to TIGIT. A couple of questions there because I do want to get to your IL-18 binding protein as well. What happened with Merck? You've been in TIGIT and thinking about TIGIT for a very long time, and we haven't seen the data yet. I think a lot of investors say, look, Dana, isn't the simplest explanation likely true? TIGIT is just not that good of a target.
First of all, not all TIGIT antibodies are the same. We started discussing the Fc issue. Genentech and Merck, that were the first movers in the field, also generated the first important results, have an active Fc. An active Fc is associated, from what we see, with—nobody did a direct comparison, but across trials, different safety profile, higher discontinuation rate compared to the non-active Fc like we and AstraZeneca and Arcus have. Not all TIGIT are the same. You also know that TIGIT probably is active. I mean, few randomized phase two studies have shown that TIGIT is an active molecule. The question is how active it is. Given there are also some safety issues, especially with Fc non-active, the risk-benefit of using TIGIT becoming more challenging, especially when you have an active Fc. This is my take on that.
Especially for Merck with the decoff formulation, they cannot even discontinue TIGIT alone because it's co-formulated. I think definitely the activity of TIGIT is not what we have seen in the mice, synergistic with PD-1 and melting tumors, but it is an active molecule. With the right Fc, it's a very good combination partner for PD-1 and AZ strategy. What we are trying to do, I think, is different from what Merck and Genentech try to do.
Is there anything in particular you're looking to validate that hypothesis when Merck does share the data?
I mean, we saw quite extensively the discontinuation rate also from other companies with active Fc. I assume we're going to see the same. We're going to see the safety. We're going to see discontinuation. Also in terms of efficacy, one might imagine that if you have an active Fc that has the potential to deplete the same cells you want to reinvigorate by TIGIT blockade, maybe even the efficacy is harmed by the active Fc. I assume we're going to see more of the same in terms of the active Fc for Merck and Corpher-Breaker.
Yeah, you'd really like to see all their T cells and how the compartments change. They're probably not going to show us that.
Probably.
You probably didn't even measure that.
Not in the phase three trials.
Not in the phase three. Let's talk about COM503. That's your anti-IL-18 binding protein that you've licensed to Gilead, but you're running the phase one trial. I wonder just, could you tell us why you're excited about this approach and any expectations we should have for the phase one in terms of the kind of signal you expect?
Yeah, so IL-18BP is a great example of demonstration of computational capabilities because we looked for resistant mechanisms of TAMs in tumor environments computationally. We're not looking for a cytokine therapy at all. And we identified in an unbiased manner that we have a resistant mechanism in tumor environment called the IL-18BP. It was identified before, but we identified that by using a blocker antibody, we could have a very different cytokine approach compared to all the other cytokines because cytokines are amazing in inducing in vitro stimulation of immunity. But when you give them to the blood, normally in high levels, they have a real challenge, and practically so many developments failed, like the Nektar and others, trying to really reach sufficient amount of cytokine in the tumor environment before you induce first toxicity in the periphery.
These are all the failures of cytokines and probably challenges that also the next generation IL-18 molecules, which does not bind IL-10BP, might face as well because they are also given systemically to the blood. What we identified is that in the tumor environment, you have IL-10 naturally there because it is induced by inflammasome signaling, very low in the periphery, but it is not active. You have very potent cytokine floating in the tumor environment bound by this IL-10 binding protein molecule, which inhibits it. We use a blocker antibody, a drug moiety. We inhibit the inhibitor, so we displace IL-10BP away from IL-10. By that, we induce localized activity of the cytokine, mostly almost only preclinically in tumor environment. We started now the phase three trial.
This asset was, as mentioned, licensed to Gilead, but we are leading the phase I trial, and we just started dose escalation. From our clinical data, we expect to see a very differentiated activity, probably TME localized effect, hopefully to see some monotherapy signal as we've seen preclinically. Yeah, we're excited for this study.
Are you getting the paired biopsies to observe that local TME effect and understand your dose window?
We did not disclose all of that. You can see in clinicaltrials.gov that we have a backfill cohort that typically people are doing in backfill, some dose optimization. Looking for what they did for other studies, one might speculate that we are going to put a lot of efforts, as we did for COM701, in untreated biopsies, sequencing, using our computational capabilities to really understand the drug MOA, biomarkers, and potential PD markers. This is not disclosed specifically for this program yet. This is in cooperation with Gilead, so we will see about disclosures.
Yeah. Are there any particular indications that have high IL-18 binding protein that you're trying to enrich those, or would you enrich in the future?
Actually, we have seen and also published that IL-18BP and IL-18, I mean, this complex of IL-18 mostly bound to IL-18BP is quite abundant in many indications. We are not sure if the levels of IL-18 or IL-18BP are going to be limiting. We did see that preclinically, you need some amount of T or NK cells to respond to the initial burst of IL-18 released by the drug. Probably a good start will be not completely desert tumors, but the opportunities are quite broad across many indications.
Got it. Maybe last question. Your discovery platform, you talked already about how the computation helped you discover IL-18. Can you talk more in a forward-looking way? How is your platform and your AI work different from what we're seeing others do right now with an AI explosion? You've been doing it forever.
Exactly. We are doing it for quite a while. I think it is sometimes difficult to explain the exact tools, but what is easier to see is that our platform is validated. I mean, many people claim that they can discover new drug targets by using AI. Not many have shown. We discovered TIGIT, we discovered PVRIG, we discovered ILDR2, we discovered IL-10BP as a drug target for a blocker antibody. Multiple times we have shown that using computational capabilities, we can identify and bring novel drug targets. There are very few, if any, companies who have such track record computationally. I think eventually the proof is in the pudding. This is what differentiates us, that this is a validated platform to identify novel drug targets.
If I could get one last question, I guess one last question back on TIGIT. We talked about how AstraZeneca as well as Arcus-Gilead have Fc-incompetent TIGITs. I do know that COM902 is differentiated from Arcus' TIGIT. I wonder if you could just tell us a moment of how it is differentiated in case both those are successful and how that differentiation could prove out for you.
I'm not sure we compared directly to Arcus, but we did test in the past the affinity of the COM902 versus most of the other leading benchmarks. It's just a very good antibody. Probably that's why AstraZeneca chose it. It has very good binding affinities, blocker capabilities. The Fc is not exactly the same, but I'm not thinking it's very different. It's IgG4 versus mute IgG1, so maybe have some very low Fc binding. Clinically, I'm not sure if all of that is going to translate to meaningful differences, but we'll have to see.
Got it. Okay. Thank you very much.
Thank you, Dana.
For taking the time. Thank you, everyone.