Good morning, everyone. Thanks so much for joining us. I'm Salveen Richter, biotechnology analyst at Goldman Sachs. We're really pleased to have the team from Relay here with us. We have Sanjiv Patel, CEO, we have Don Bergstrom, President of R&D, and Peter Rahmer, Chief Corporate Officer. With that, maybe I'll just start with a broad question, given what we've been seeing in kind of the AI field, and not just in, you know, tech, but also in healthcare. It seems like the terms artificial intelligence and machine learning are being used in many different ways. That being said, it seems to be grouped under one umbrella. Help us understand how you're differentiated with this motion-based design, with your Dynamo platforms.
You know, just in general, it'd be helpful to understand how to think about the application of these technologies in drug discovery and, you know, potential advantages and areas of caution.
Well, thank you first of all for the invitation. It's a pleasure to be here. As you know, there's nothing at the moment that gets released as on the wire without the terms AI or machine learning, generative AI, ChatGPT, it's everywhere. There is a huge amount of excitement, rightly so. I think the excitement is there because we know that in the past, those last few decades, we've seen computational approaches change whole industries, retail, nobody goes to the mall anymore. Everyone does what they need to do online and travel. I think the time has come for using these approaches to really change the way that healthcare is delivered and created. We're equally cautious in that this is a very difficult undertaking. Biology is complicated, and making medicines is not straightforward.
I think, you know, to answer the questions like, how can we apply it and, you know, where are the areas of excitement and caution, I think there are a few learnings for us. We were one of the first companies back in 2016 to deploy computational approaches in the discovery of new medicines. I think the first learning for us is, to be successful, you really have to focus. The application of these technologies is on very small steps, and there are many hundreds of steps in the discovery of a new medicine, all the way from understanding what target to work on, figuring out how to understand and modulate your protein of interest. How do you find a chemical starting point? How do you optimize those chemical starting points into drug-like molecules?
All of which can be broken down and enhanced using computational approaches, ML, AI, generative large language models. The key is understanding where to focus, breaking down the process into small steps, and then having real tools. I urge, you know, the audience here to, if you come across any of these terms, is to say, "Look, where do you apply the AI or generative large language models, and then what specific tools are you using?" We've been able to deploy those tools across those steps over these last few years to put now four medicines in the clinic. The second thing that we've learned is you need really good datasets. For each of these steps and each of the tools that we use, we've been over seven years, developed really clear, clean, proprietary datasets.
The better the datasets you have, the better the models are gonna be, and then the better the predictions are gonna be. The third thing we've learned is people are important. You know, there's a lot of hype around people will be gone, and it's all going to be done at the press of a button. That's not our belief. The chemists, the wet lab scientists, are producing data that is needed to then feed the models, make these models more effective, and then again, you know, test the models. The age of it all being done by computers is not realistic.
Look, the final thing is just the cultural aspect of having wet lab scientists working with computational scientists and large language models is not intuitive and takes time and effort to build up that culture where both sides can work together. Our approach is now been successful over 7 years, and that we've got defined tools, we have defined datasets, we have a tool that is used to working in a culture that's digital native. That's why we think that there's a huge amount of excitement is justified. It's just gonna take time, and we're talking about a timescale of decades, not days. That's why we're excited but cautious around this.
Across your portfolio of assets, remind us what we're going to be seeing over the next year or so.
The key milestones ahead of us for the rest of the next 12 months. We're in the H2 of this year for our RLY-4008, which is our selective FGFR2 inhibitor. We will fully enroll the pivotal cohort. The second thing that you'll see is data outside of cholangiocarcinoma fusion patients. There's a lot of interest in understanding how big an opportunity RLY-4008 is, and we'll show data around non-cholangiocarcinoma fusion patients, as well as FGFR2 altered amplifications and mutations. For RLY-2608, which is our mutant selective PI3Kα inhibitor, we will start the expansion cohorts, and then in 2024, show further data.
For our CDK2 inhibitor, it will enter the clinic in 2024, and for our ERα degrader, we will select a development candidate. Plenty of milestones ahead, and we'll continue to prosecute our preclinical portfolio. We have $938 million at the end of Q1, so we have plenty of runway to provide meaningful value for our shareholders.
...Let's start with the PI3Kα program, given we saw some data recently. How do you think about the early cut of data that we saw? Explain to us that confidence that you have that as you kind of follow these patients out or you use additional levers, that you will get to that optimal profile that you want to show in this class.
First of all, it was very similar to our RLY-4008 disclosures that we made. The first disclosure we made for 4008 was back in 2021, where we showed proof of concept, which was that we could dose above a certain threshold level, we could avoid the off-target toxicities associated with the non-selective inhibitors, and then we showed early signs of target engagement. A year later, we came back and showed, you know, the first interpretable efficacy data. Obviously, the first disclosure gave us great confidence that we would get to a positive second disclosure.
We followed exactly the same pathway here with RLY-2608, in that the disclosure objective was achieved, which is we could dose above the IC80 consistently over a 24-hour period, which is we believe to be a kind of therapeutically active threshold. Secondly, is we could avoid the off-target toxicities that have been associated with reduced dose intensity in the non-selective inhibitors. The key off-target toxicities, diarrhea, hyperglycemia, that we pretty clearly showed that we avoided. Then early signs of target engagement, and obviously, that comes in a variety of forms from ctDNA to resist measurements. So we feel very confident that the profile of RLY-2608 is exactly what we'd hoped for.
Now, as we go forward over the next year or so, we will continue to backfill these dose escalation cohorts, select the dose, open the expansion cohorts, and then provide the definitive efficacy that we all hope to see for these patients in need. Don if you want to just comment on...
Right. You know, I think the other pertinent consideration here is as we continue to develop RLY-2608, ultimately the registration endpoint in breast cancer is going to be progression-free survival. That is a contrast with the RLY-4008, where the cholangiocarcinoma patient population response rate and durability of response are the registration endpoints. As we're looking towards that ultimate goal of where we'll be to show differentiated efficacy for RLY-2608 in PI3K-mutated breast cancer patients, the questions become not just: Are tumors shrinking? The questions are: Are patients tolerating the drug? Can they stay on the drug chronically, and can you maintain dose intensity chronically? The toxicity profile of the non-selective PI3Kα inhibitors has been problematic for patients being able to stay on drug.
In the SOLAR-1 trial of Alpelisib that ultimately led to its registration, median progression-free survival was 11 months. Median duration of therapy was only 5.5 months, with at least a quarter of patients discontinuing therapy due to adverse events. I think as we look at the leading data that are coming out on 2608, and the data that we just disclosed, we were able to show that at some of the earlier dose levels, we were able to actually keep patients on drug for a long period of time, for 8 months and longer, and patients were staying on drug at their scheduled dose with a delivered dose intensity in excess of 98% of scheduled dose.
I think as we're looking to see our data continue to mature, we're going to be continuing asking that question about dose intensity. Not only can we hit the target hard acutely, but can we maintain that target inhibition chronically over the course of, you know, going into double digit months, because ultimately, that's what the field is telling us they want in our patient population. We'll be looking at the ability to control disease over a long period of time, not just looking at response rate, but looking at clinical benefit rate, how many patients have at least 6 months of disease control? Ultimately, as we build the database over time, we'll be able to answer the progression-free survival question, which again, is the ultimate registration endpoint in this patient population.
Could you just expand on that in the context of, you know, how to think about or how you think about this as a, we need to follow over time or go to higher doses versus there being some kind of fundamental chemical barrier to potency and selectivity? Because I think that was a question that came up.
Yes.
I don't know if you want to take it.
you know, first of all, you know, I think the important thing to consider is that PI3Kα is a validated target in our patient population, and there is an approved drug targeting PI3Kα in these patients. This is not a target where I think we're wondering about the biology. I think it's a target where we have a high degree of conviction, that if you can hit the target with sufficient inhibition, it can translate into clinical benefit. What's the challenge then in this field is that for non-selective inhibitors to achieve that level of target inhibition, it's come with significant toxicity burden. That's ultimately limited not only the degree of target inhibition you can achieve, but how durable the target inhibition can be.
I think what we've just disclosed from this initial RLY-2608 data set is that we can in fact hit the target at the threshold which we had set, based on our preclinical experience and based on the clinical experience of others. We can do it in such a way that we don't see any of the hallmark toxicities that lead to the discontinuation of the non-selective PI3Kα inhibitors. These data give us the confidence that we are in the target exposure range that we need to be. It's a validated target. You know, I think there's been some question around whether you actually need to have some dirtiness, whether you need to hit wild-type PI3Kα.
In our experience and what's been published by others, we've yet to see any credible data that would suggest that you need to have potent activity against wild-type PI3Kα to be able to get clinical benefit in these patients. I think where we sit today, with the data that we have in hand, is indicative that 2608 is doing what we designed it to do, and with additional patients and additional follow-up, I think we'll see this profile translate ultimately into a differentiated efficacy profile.
The only thing I would add, too, is that questioning of potency, do we have the right target coverage? All stems from the fact that we have not seen a partial response in these early data sets in the combination patients. We obviously saw a very dramatic response in our monotherapy double mutant patient. I think if you were to take the lens that Don just provided us as to how to properly assess a drug in this setting, it would lead you to the conclusion that the data set as Sanjiv started with, saying that it really is meeting the very profile that we hope to achieve at these early stages.
Getting too fixated on, you know, a few more millimeters of tumor shrinkage, at this point, would be, you know, could potentially allow you to miss the main objective that we're ultimately trying to accomplish here.
Do you expect to see PRs over time? I guess, if you increase dosing, do you think that'll translate into additional efficacy?
Do you want me to take that? You know, I, again, I think our anticipation would be that we should see responses over time, but again, that's not the primary goal here, right? I think if we were to see responses without being able to maintain dose intensity or keep patients on drug, that would not be success. I think we will see, over time, responses based on everything we know about the target and our preclinical data. Ultimately, I think what keeps us excited is the potential to be able to actually translate that ultimately into clinical benefit, long-term clinical benefit, and ultimately, PFS. I don't necessarily know that we need to have more drug on board to be able to get there.
Our data would suggest, consistent with what's been published in the literature, that the level of target inhibition we're at is associated with full efficacy. We are seeing tumor shrink in the patient population that we disclosed on, the number was still small, and the data were still early. I think we need to get more experience. We will look at a higher dose, largely to complete the exercise of the dose escalation. I think as we continue to move forward, the doses that we've identified that are giving us exposures above our target threshold, 400 mg, 600 mg, 800 mg, are likely going to be the dose range that makes sense moving forward.
How long do you think we would need follow-up to get this interpretable or understanding of the profile?
Maybe Pete goes in.
Yeah, I think it's going to come in steps, as it always does. In this context, where the ultimate regulatory and clinical endpoint is progression-free survival, one of the early interpreted barometers of that for a safe drug is clinical benefit rate, which is the measurement of stable disease or better at the 24-week time period. I think the early interpretable outcome output that we'll get, which we'll have some of that data, we'll probably have the beginnings of that data next year, is around CBR at relevant doses.
I think it is certainly going to take more time and more of a either from the full expansion cohort maturation and/or all the way into some sort of randomized study to ultimately see the definition of what the potential median PFS could be for this agent. I think, we're going to see a good progression of the data next year that should be, we think, going to be interpretable against what we could ultimately achieve with 2608.
There is still confidence here of hitting mutant versus wild type in the approach?
Yes. Yes. I think when we look across the totality of the data that we've presented, we look at the PK data, we look at our pharmacodynamic assay, we look at circulating tumor DNA. Then we look at tumor measurements. All of them would point to us having significant inhibition of mutant PI3Kα in the setting where we're not seeing any grade 3 hyperglycemia. We're not seeing any significant rates of diarrhea and rash, which would be indicative that we are sparing the wild-type PI3Kα or other PI3K family member inhibition that's been associated with the toxicity with agents like Alpelisib or Inavolisib.
Is there anything, when you look at target engagement, that would prevent you from getting towards, you know, getting the additional 20%, getting towards 100%?
Yeah, I mean, I think that's a largely a technical question. The assay that we've used in our ex vivo pharmacodynamic assay measures phospho-Akt, which is downstream of PI3Kα. It's not the direct substrate of PI3Kα. As we were qualifying that assay for use in our clinical trial, we did use Inavolisib spiked into human plasma to qualify the assay. Very potent non-mutant selective PI3Kα inhibitor from Genentech Roche. What we saw for peak inhibition for that agent in this assay was also in the 80s for inhibition. I think we are probably pretty close to the maximal inhibition that we'll be able to achieve measured by that assay.
Perhaps you could just walk through the physician feedback on the back of the data, and then, where you think this drug will optimally be used in the patient population.
Maybe, Pete, you want to just take the kind of broader feedback?
Yeah. I think we've been. One of the main goals behind making this initial data disclosure, as it was for RLY-4008, was to help establish confidence in the investigator community to agent, investigational agent. In the face of the fact that this is being, in our study, we preclude the use of Alpelisib, an agent with full approval in this setting. Showing some early data that starts to demonstrate this clear clinical proof of mechanism, we hypothesized would be helpful in terms of awareness and driving enrollment and furthering our continued clinical execution. I would say that's played out exactly as we had hoped. The feedback from the physician community has been overwhelmingly positive across the board.
You know, we as, on the, I would say, on the Wall Street side, generally are used to trying to make this very black and white comparison between early data in oncology, should yield response rate, that is then the only barometer of success here. You have to take into context the patient population that we're talking about here, and the physician community is obviously very well aware of that and what the ultimate clinical goal here, which is to maintain long-term disease control over sequential lines of therapy in these breast cancer patients. You have to remember that only 5% of these patients are diagnosed in the metastatic setting.
A lot of these patients are found in the very early course of the disease, which is the reason why it's one of the better controlled solid tumors we have to date. I think for most physicians, they're very encouraged by the ability to maintain dose intensity and to really mitigate some of the key off-target liabilities that has plagued the space.
Yeah. The second other question, maybe I'll start, and then maybe, Don, you can finish it. I mean, our goal with our platform is to try and create highly selective medicines. The reason for these hormone receptor-positive, HER2-negative breast cancer patients is that the treatment will come in combination. The issue at the moment is the drugs are non-selective in many ways, and therefore, the off-target toxicities either limit the ability to combine or provide a lack of tolerability of the combinations. The drugs are used sequentially. What we hope to do is try and create a franchise of highly selective therapies, the mutant selective PI3K-alpha being the foundation of that, but also, you know, we have behind it a selective CDK2, and then we have an unnamed set of programs behind it.
The goal is to provide a clean set of highly selective medicines that can be combined together and can be used not sequentially, but actually together in early lines of therapy. Maybe, Don, you can comment exactly on how we see it to be used.
Right. Our current development and the data that we've reported out has been exclusively in patients who have been previously treated with a CDK4/6 inhibitor, which is a standard of care in the frontline metastatic setting, an emerging standard of care in the adjuvant setting. I think as we look at the continued development of RLY-2608, certainly the post-CDK4/6 population is a population with immense unmet need. It's a population where the best progression-free survival numbers we've seen have been in the range of about seven months, and where the challenge that the KOLs have thrown out is they want to see double-digit PFS. That's clearly a place where there can continue to be development and there's high unmet need.
As Sanjiv pointed out, trying to move earlier and move into combination with CDK4/6 inhibitors or higher order combinations with other emerging inhibitors has been a challenge. We saw Novartis try to combine Alpelisib and Ribociclib and not be able to identify a dose to take forward. We've seen Roche Genentech combine Inavolisib, their PI3K inhibitor, with palbociclib. Part of that was dictated by the fact that Palbo has the least diarrhea liability of any of the CDK4/6 inhibitors, which has, you know, given the diarrhea liability of Inavolisib, has forced that direction.
I think the profile that we've seen with RLY-2608 so far is an incredibly combinable profile. I think it's a profile that gives you the option of combining with any of the CDK4/6 inhibitors to go into CDK4/6 naive patients, or as we continue to build our portfolio, to be able to achieve these higher order combinations that ultimately, we believe will be able to deliver more meaningful clinical benefits to patients earlier in their disease journey.
What are you trying to achieve with the second asset, with the PAM mutant?
You want to take that one, Pete?
Yeah. The initial reason for creating RLY-5836, which is an additional PAM mutant selective inhibitor, leveraging the same mechanism as RLY-2608, but a different chemical scaffold, was largely around mitigating any possible idiosyncratic tox that may arise with RLY-2608 in the clinic and making sure that we weren't, you know, starting from scratch if we were to ever see that. Obviously, that's been largely dispelled in the existing data that we just showed for RLY-2608, as we stand here today, RLY-2608 is likely the go forward molecule. That said, we have RLY-5836 ready to enter the clinic, and it's an extremely large opportunity. It's a very important one for us as a company, we will explore them both clinically for a period of time.
It is, in our mind, RLY-2608's to lose and a very high hurdle for RLY-5836 to displace RLY-2608. This is likely a decision, you know, we won't let linger on long, and so probably one we would make in 2024 at the latest.
Just a last question here. We look to that next data read, you've guided to 2024, but could it come this year? When we see that, what should we be looking for with this upcoming data set?
Yeah, as we spoke to earlier, you know, we think one of the interpretive pieces of this next disclosure will be early signs of the, on the clinical benefit rate. Therefore, that's gonna require having a good number of patients above the target exposure threshold with at least six months of follow-up, so to at least have the potential to have reached that threshold. That's what leads you to 2024. You know, so I think that's what we're hoping to focus on in that next data disclosure, the reason why 2024 is the guidance.
Maybe stepping back from a BD perspective, you've been, you know, acquiring and partnering around different technologies. Maybe explain why and how you're thinking about this and how you integrate them into your platform?
Yes. I mean, over the course of the last seven years, we've partnered and used a variety of both experimental tools, but mainly in and around structural biology and, you know, really imaging the dynamics of proteins, as well as on the computational side, a variety of new and emerging technologies and techniques. We made one acquisition of a company called Zebi AI, which is a company focused on creating a computational version of DNA-encoded library screens, which we found experimentally very successful and amended itself to a big data approach and being done ultimately without the experimental side. I think we've constantly tried to stay at the cutting edge of all the technologies around us. We believe there's no one kind of technology or approach or tool that's gonna be successful.
It's really creating a constellation of different tools and integrating them together. We will continue to be out there. If a service provider is out there with a tool, they want it to be used in the most effective way, and we're one of the companies that has the most experienced team at using any of these approaches. Ultimately, they want to try and use those approaches to make medicines, and again, we're a great place for them to come. I think we'll continue to incorporate as many of these emerging technologies as is possible. In terms of acquisitions, I think, you know, we sit on a very healthy balance sheet, but that healthy balance sheet allows us to prosecute our milestones well into a period where we can deliver significant value.
We have a very high threshold on acquisitions over the coming years. In terms of BD, Jim, well, generally, we're excited about trying to get these medicines to patients as rapidly as possible, and, you know, we're continually open-minded. You saw that with our RLY-1971, where we partnered that with Roche Genentech, and we're excited about the prospect of combining it with their KRAS G12C. I think we'll continue to remain open-minded about how to get these medicines to patients.
Any update you can provide on the SHP2 that you partnered with Roche Genentech and whether they've started those combination trials?
As you know, the disclosure, is entirely controlled by Roche Genentech. I mean, Pete, if you can comment on that.
Thanks for that disclaimer. What we can speak to is in the early fall of 2021, they initiated the combination study with their KRAS G12C inhibitor, GDC-6036. I believe it has a generic name now, too, that I can't remember. In the fall of 2022, they initiated the combination with their PD-L1 inhibitor, Atezolizumab. I think, you know, when we consummated the deal, we had aspirations to move broadly into other different combinations, I think we remain very happy with how they're prosecuting the molecule in the context of what we were hoping it could be at the time of the deal.
On your FGFR2 asset, you have different populations that you're studying the drug in. Maybe help us understand as we see these data sets over the year, how, what the, what we should be looking for in each of them?
Maybe, again, Pete, do you want to just take us through that?
To remind you what's coming up, in the H2 of this year, we will be showing for the first time the data from our non-CCA expansion cohorts. Those started in early 2022. The goal with target enrollment at that point in time, about 30 patients in each of those cohorts. One is fusion tumor agnostic, the other is focal amplifications, and the third is mutations, all outside of CCA. CCA development will stay isolated to that arm. We would imagine by the time we make this disclosure, that we should have those cohorts largely enrolled, so roughly 30 patients each. Again, the goal here is to start to discern some signals of where we could focus future development outside of CCA for RLY-4008.
We do believe there's gonna be an opportunity outside of CCA, and these data will start to show us where and to what magnitude. I think generally speaking, if we were to kind of rank order where the evidence would point us to the best chances of success, it's probably in a fusion tumor-agnostic setting. We've seen this play out a couple of different times where you've seen a small molecule work well in a specific fusion population, it turns out that that is generally usually generalizable across solid tumors. The fusion oncogenes tend to be generalizable across all tumors. Then next on amplifications mutations, there's less precedence in the space there, but as it pertains to the pan-FGFR inhibitors, we've seen some evidence of activity in the gastric amplification setting.
So that would be one of interest that we'll continue to watch. On the mutations, there's even less evidence from clinical evidence from the existing inhibitors. But we've been able to demonstrate two things on the mutation side. One, that the data we showed at ASCO recently, we showed for the first time really, our ability to drive to partial, to discernible response rate in KRAS carcinoma mutations, activating mutations. We had 14 of those patients at a 29% response rate in them, so that's a good early signal. We also, in an earlier data set, had shown a breast cancer mutation response, the first know of, that we know of, for an FGFR inhibitor. The last thing we showed at ASCO was a vignette from our expand...
mutation expansion cohort in a head and neck cancer patient, where we drove to a very dramatic partial response, I think a greater than 60% tumor shrinkage. highly really confident that there's gonna be an opportunity outside of CCA.
For your breast cancer portfolio, how are you thinking about the synergies within and what would be the next update we get there?
Maybe, Don, you can just talk about what we have.
Right. The next molecule that went through the clinic is our highly selective CDK2 inhibitor that went through the clinic in early 2024. We've chosen to focus on this target because we've seen robust biologic synergy between inhibition of MET and PI3Kα and inhibition of CDK2, especially in a post CDK4/6 setting. I think, you know, we're looking to build on the observation of that synergy again, with having two agents with the right profile to be combinable with each other, and we view that the CDK2 opportunity, you know, will be an important combination partner as we continue to prosecute our PI3K and our RLY-2608 portfolio.
Any questions from the audience?
Thank you. What are your thoughts on the profile of competitor CDK2 programs that, such as those that were presented at ASCO this year?
Maybe you want to start, maybe Don, you can comment on the data.
Yeah, I think it's important to reiterate what our objective is with our CDK2 inhibitor before Don puts the data in context, which is, our goal is not necessarily to create a best-in-class CDK2 inhibitor, but one that is optimized for combination with our PI3Kα franchise. And thus far, we've been able to achieve that type of profile in our preclinical exploration. We think, you know, as Don can speak to, preclinically, the profile looks to be as good, if not better, against certain parameters, largely against Pfizer CDK2, 'cause that molecule is in the public domain, so we could benchmark in head-to-head preclinical work. Over to Don on what are our take on the initial clinical data.
I think we were encouraged by the early signals that were shown that you can, you know, achieve some early evidence of antitumor activity with a safety profile that would be largely consistent with what you would expect for inhibition of CDK2. One of the questions that we have gotten a lot of questions about was specifically around ocular toxicity. I think it was encouraging to see, you know, the fairly comprehensive experience that Pfizer had with their molecule without seeing significant ocular toxicity. I think it would be consistent with our interpretation of the target that's not likely to be on target.
Great. With that, thank you so much.
Thank you.
Really appreciate it.
Thanks so much.
Thanks.