Good morning, everyone. I'm Ken Shields, Vice President in Equity Research, and part of the targeted oncology team for Leerink Partners. Thanks for joining us on the first day of our annual Global Biopharma Conference. Very excited this morning to have with us from Relay Therapeutics Don Bergstrom, President of R&D, and Megan Goulart, Vice President of Investor Relations and Communications. So welcome.
Thank you. Yeah, thanks for inviting us. Always nice to come down to Miami from Boston this time of year.
Yeah. So yeah, just to kick things off, maybe we can just start with some general questions. Can you provide a brief overview of the company and the Dynamo platform?
Yeah. So Relay is a company that was founded in 2016, and the founding vision was really to sit at the intersection of computation and experimentation, and to really be able to use emerging computational tools, which included both physics-based simulations of proteins coupled with artificial intelligence and other computational technologies, to be able to drive our experiments forward in a manner that would make them more efficient and more effective at discovering new medicines. We are really focusing our computational and experimental platform on the discovery of new medicines as opposed to some others in this space that are using AI to elucidate novel biology.
So we made the decision from the outset that we would initially focus on very well-validated target space with either pharmacologic or genetic validation so that we'd have high confidence if we could make a medicine that met our target product profile that once we got into clinical development, we'd be able to have an increased probability of success of generating positive clinical data. We sought out targets where, while there was some validation of the targets, there were clear unmet needs in the profile of the drugs that had been developed to date. Among those early targets that we started working on, you saw us go after FGFR2, clear oncogenic driver, but we saw that there were limitations to inhibiting all of the FGFR pathway.
And then I think very significantly looking at PI3 kinase, the most frequently mutated kinase in cancer, very significant potential commercial opportunity, validation in the breast cancer space, but where attempts at targeting PI3 kinase have largely been limited by the fact that non-selective inhibitors also inhibit wild-type PI3 kinase, which bring a significant amount of toxicity with it that ultimately we feel is efficacy limiting. So that was the initial founding vision of the company. We've been going over the past eight years. We've continued to build out our platform and continue to make investments in the platform and are continuing, obviously, to apply the platform across a broader range of targets as time has gone by.
Okay. Thanks for that. And I know you kind of touched on it a little bit, but how does this platform differ in terms of AI drug development efforts? We're seeing some more of that. And then also, how much and how successfully do you think big pharma or other industry players are starting to adopt or invest in AI technology? And where do you guys fit in?
Yeah. So I'll start with the first part of the question. So I think, again, we've really focused our platform on the discovery of new chemical matter and new small molecule drugs. And we really do this in a three-step process. And we start with the application of the platform to understanding the targets we're going after and how we would want to drug them to be able to get the profile that we want. And again, this is a combination of both experimental and computational methods. We found that there's not a single computational tool that you can use that eliminates the need to run an experiment. The computational tools help us generate hypotheses that then we can go and experimentally validate. We can go back and refine our computational models and drive a program forward.
So in that first step, target modulation hypothesis, we call it, you're really diving deep to understand the target. You're producing full-length protein. You're experimentally solving structures, in many cases, using experimental methods that give you an insight into how the protein behaves, things like cryoelectron microscopy or multitemperature X-ray crystallography. You're then using that to build computational models of the protein and identifying novel ways that you could go after a target. So again, in our PI3K program, we used this to actually identify what we thought could be a novel allosteric pocket that we'd be able to target for discovery of a mutant selective PI3K alpha inhibitor. And we're able to take that insight that we had and use it to drive, ultimately, our drug discovery program forward. So the first step is really understanding the target.
The second step is, once you have a hypothesis for how you want to go after a target, it is finding chemical starting points to start a drug discovery program. And this can be done both through conventional experimental means, but those are limited in terms of the amount of chemical space that you can look at. So we also use significant computational hit finding, including both virtual screens, which are completely computational, as well as a hybrid approach that we call machine learning DNA- encoded library screens, where we'll run a DNA- encoded library screen, which is an experimental method for generating a very large quantity of data, which then can be used to actually train a machine learning model to be able to identify the parameters of a small molecule that may lead it to be a potent binder of your protein of interest.
So we can then use those models to go beyond just the initial library that was used in the screen to look basically at the entirety of the synthesis- on- demand chemical space, which is now exceeding 30 billion compounds, which can be synthesized- on- demand, to find hits computationally that could be followed up on as starting points. And then finally, once you have a hit, you need to turn it into a drug. So we've really been focusing on building tools to make our lead optimization process more effective and more efficient than conventional experimental means.
It includes having tools both for predicting potency or selectivity, as well as tools for predicting ADME profiles for how a drug will behave in an organism so that we can use multiparametric optimization as we go through our lead optimization steps to ultimately, when we get to the lab, synthesize the compounds that are most likely to meet our target profile. I think it's an approach that the way the computational world has been moving is very logical and consistent with that. We see now a number of people talking about applying machine learning in the drug discovery process. The challenge here is that there's actually quite a cultural barrier to really being a fully integrated computational experimental shop. Relay, since its inception, was really founded to have computation and experimentation be equal partners in drug discovery.
And I think for legacy organizations that have been largely experimentally driven, there is a cultural skepticism that needs to be overcome for the experimental scientists to look at the computational scientists as equal partners in drug discovery. And I think similarly, on the pure tech side, there's a discounting of the importance of experimentation for being able to validate hypotheses and refine models. So it's going to be challenging for anybody to build this technology either internally, organically, or even to bolt it on. If you're a large organization to try to bolt on something like what we do at Relay, we'll have significant cultural barriers in a traditional experimentation-driven organization. So I think there's been a lot of focus on this. Obviously, data is the other critical point, having the data that you can use to build models. Just having a lot of data is not sufficient.
It's got to be data that was generated intentionally, that's appropriately documented and appropriately consistent, that you can use it to build models, and the models just aren't being trained on the noise that exists in the data. So it's challenging, even for organizations with large legacy databases, to really be able to leverage those data to build highly informative models. So if you're now talking about building the data on your own, obviously, that's a significant barrier. And it's a barrier that we've been working at overcoming over the past eight years, where we've been very consciously building our data and building our algorithms to be able to inform our drug discovery process.
Okay. Well, thank you for that. So why don't we jump into the PI3 kinase asset, 2608? Can you provide a brief overview of this and how it may fit into the HR-positive, HER2-negative landscape?
Yeah. So PI3 kinase is, as I mentioned earlier, the most frequently mutated kinase across all solid tumors. And in hormone receptor-positive HER2-negative breast cancer, we estimate that between 35%-40% of all HR-positive HER2-negative patients carry a mutation in the PI3 kinase gene. Now, when we started working on the program, the initial validation for the clinical relevance of PI3K mutations had been generated by alpelisib, marketed as Piqray by Novartis, that initially ran a trial combining alpelisib with fulvestrant in breast cancer patients who had received prior endocrine therapy, and showing that the addition of alpelisib to that regimen resulted in a significantly longer progression-free survival rate. So we knew that this was an important target in that patient population.
We've seen further validation that came at the end of 2023, where Roche, for their non-selective PI3K inhibitor, inavolisib, showed a positive phase III trial in a different disease setting. So this was earlier in the treatment with patients who were endocrine-resistant from their adjuvant therapy, had not yet been treated with a CDK4/6 inhibitor, and Roche combined inavolisib with palbociclib and fulvestrant in this endocrine-resistant population and showed clinically meaningful improvement in progression-free survival, again, further validating the target in hormone receptor-positive breast cancer and validating the additional combination benefit of putting PI3K inhibition on top of CDK4/6 inhibition. Now, what those positive efficacy results have come with is significant tolerability challenges. So in the case of alpelisib, high rates of hyperglycemia, diarrhea, and rash leading to treatment discontinuation.
The median time that patients can stay on treatment with these agents in clinical trials is about five and a half months. You see over a 25% rate of discontinuation due to AEs. And one of the frequent AEs that leads to discontinuation is hyperglycemia. So patients who receive these drugs, you get a blockade of wild-type PI3K signaling, which is needed for normal glucose metabolism. These patients essentially have drug-induced diabetes that, once you get to grade 3, is insulin-dependent. And this is a place where many oncologists are no longer comfortable managing that AE. They're either calling in endocrinology consults or they're just taking, in the real world, just taking patients off alpelisib rather than manage that AE. And that, again, is due to inhibition of the wild-type PI3K.
Our hypothesis was if you could identify a drug that could give you full inhibition of the mutant PI3K that's driving the tumor while sparing the wild-type PI3K that's required for normal glucose metabolism and other important normal functions, you should be able to thread this needle where you could push to efficacy comparable to or greater than what you see for the non-selective PI3K inhibitors while maintaining tolerability and the ability to keep patients on drug.
Okay. Yeah. You mentioned the Roche data, the INAVO120. Was there something about the exclusion criteria there that couldn't allow for metabolically unhealthy patients? And how might that impact your future development in the front line?
Yeah. So just for everybody's guidance here, the INAVO120 trial that Roche read out last December was focused on patients that had to be very metabolically fit. So they required a hemoglobin A1c of less than six and a fasting plasma glucose less than 126, which, per CDC data, would actually exclude about 50% of the Western population from qualifying for treatment in that clinical trial. Consequently, what you saw in that trial was that the enrollments of the trial skewed towards Asia. So a higher rate of Asian patients were represented compared to what you typically see in a global phase III trial. It also skewed younger. The patients in general were younger than you'd expect to see in this patient population, again, because it was pushing towards more metabolically fit patients. These patients were also patients who were required to be endocrine-resistant.
They had to progress within a certain amount of time of their adjuvant endocrine therapy. Consequently, the totality of their inclusion criteria probably represents only about 20% of hormone receptor-positive HER2-negative patients being treated in the front line. We think that that leaves ample opportunity for development of alternative regimens in more broadly defined patient populations in the front line setting.
Yeah. So then maybe we can take that into the data that you guys have shown and then expectations ahead of some of the combination datasets this year. Some people in the investment community are potentially worried about the efficacy. What gives you confidence that it can be really competitive in terms of efficacy?
So I'll start with the data and then hand over to Megan to talk about our guidance for next data. But the first data that we disclosed from the program was at AACR last year, where I think our goal was to show fairly definitive data that we're able to identify doses that were giving us inhibition of PI3K in the therapeutic range while sparing the wild-type toxicities. And I think we're able to show that with no grade 3 hyperglycemia, reduced rates of any grade hyperglycemia, reduced rates of diarrhea, and reduced rates of rash. We then updated the data in August of last year in conjunction with announcing that we were opening our first expansion cohort. So this is now picking a dose to go into a broader range of patients. At that time, we started to have more mature efficacy data.
We chose to move forward at a 600 mg b.i.d. dose based on the totality of the data, so both on our PK projections, on pharmacodynamic markers like ctDNA clearance, and then on early efficacy data, which included looking not only at response rate, which actually is not a terribly relevant endpoint in this patient population, but looking at clinical benefit rate because ultimately, this is a drug that will be approved based on progression-free survival. What we were seeing at that cut that we did that we disclosed last August is that of the five efficacy evaluable patients, we had seen tumor regression in four of the five and an objective response, confirmed objective response in one of the five.
We saw that of the 17 patients that we had enrolled at 600 mg at the time of the data cut, 15 of the 17 patients were still on treatment. For the patients who had at least six months follow-up and were eligible for a clinical benefit rate assessment, six of the seven met the criteria for clinical benefit rate. That would be an estimate at that point of a clinical benefit rate in the 80s%. The comparators for the non-selective PI3K inhibitors in similar populations are between 46% and 48%. I think the totality of the data that we're seeing and the trends that we're seeing are consistent with what we thought we'd see for a mutant selective PI3K alpha inhibitor. I think we continue to be enthusiastic about the profile of the agent.
Yeah. And then as far as our guidance for what we're going to see this year, as Don just talked through, we believe we've really showed differentiated safety and tolerability so far. And so the outstanding question is really showing longer-term efficacy. And so as Don mentioned, we opened our initial 600 mg expansion cohort in July. That, in addition to the patients from the dose escalation at 600 mg, we think we'll have nearly 40 patients at least six months of duration. In addition to, we opened another 600 mg dose expansion cohort late last year of another 20 patients that'll have a little shorter duration. So we'll have about 40 patients that will have had the opportunity to be on for at least six months.
We'll be able to have a clinical benefit rate, which is a measure of stable disease at 24 weeks in that cohort of patients. Then we'll probably also be able to start doing an early read on some landmark analyses at six and maybe nine months.
Okay. Thanks. And then, yeah, you mentioned the initiation of that other 600-mg cohort. I mean, is it that there also a 400-mg cohort? Something to do with Project Optimus, or could you provide some color on that?
Exactly. So the totality of the data we're seeing is pushing us towards 600 mg as representing an optimal dose. We anticipate that as we initiate regulatory interactions to discuss the design of a pivotal trial, that dose is going to be a key question. We see this in the U.S. with Project Optimus. FDA is typically looking for the lowest dose that gives you an optimal risk-benefit profile. So we made the decision to also open a 400 mg cohort, which is the lowest dose that meets our target coverage threshold that we're looking for. But on average, it meets it, which means some patients are above, some patients are below the target, as opposed to 600 mg, which the vast majority of patients are above the target.
But we feel it reflects a potential lowest biologically relevant dose that would give us a reasonable comparator to justify the 600 mg dose as we move forward. So we've opened a 20-patient expansion cohort. We've got about 8-10 patients from dose escalation at 400 mg. So we anticipate we'll have a body of data that allows for meaningful interactions with regulators as we move forward.
Okay. But you're pretty confident in the 600-mg?
Yeah. I think we feel 600 reflects an optimal dose.
Okay. And then I guess just looking ahead, oh, actually, also, was there some triplet data as well expected this year?
Yeah. We also, late last year, started initial dose escalation of the triplet arms. That's really just dose finding. At this point, we have guided to second-half data, but that'll be really initial safety data working to identify a dose. We've started with Ribo at this point, but the protocol also allows for us to use Palbo as well. Looking to probably do that in the future too.
Okay. And then just looking ahead, obviously, we're eagerly anticipating the data. What could a potential regulatory strategy be, a registrational strategy? What could be the comparator, capivasertib or alpelisib in combo? What are the factors that might influence what comparator is used?
So today, there are two regulatory standards of care in a post-CDK4/6 setting, both with full approval. alpelisib and then capivasertib, which is a multitargeted kinase inhibitor that includes AKT among its targets, that was approved at the end of last year for the treatment of post-CDK4/6 patients in combination with fulvestrant. That label was restricted to patients with PI3K pathway alterations. Those represent the two regulatory standards of care. I think we anticipate that to get a label in the immediate post-CDK4/6 population, we likely would need to run a comparator trial against one of those. We are hearing, as we talk to investigators and KOLs, that the sentiment is moving away from alpelisib and towards capivasertib, although we don't yet have any sales data for capivasertib. We don't have any quantitative confirmation of the qualitative feedback we've been getting.
But I think assuming that that does play out over the course of 2023, that Capi would be an emerging clinical standard of care, that that could be a likely comparator that we'd go against. But this is all, obviously, with the caveat that we still need to have regulatory interactions with the FDA and with other regulators to confirm what the ultimate registration paradigm would look like.
Okay. Thank you. Maybe we can move to lirafugratinib. Can you provide an overview of this asset and its differentiation from the previous FGFR inhibitors?
Yeah. So this was a program that, when we started, there was data that was supporting that FGFR2 was, in fact, a driver oncogene. The initial clinical data was in patients with cholangiocarcinoma with FGFR2 fusions. But FGFR2 is altered across a broader range of solid tumors as well. So we thought it looked like a very interesting, well-validated precision oncology target. Now, the drawback is all of the attempts to target FGFR2 historically have been done with multitargeted kinase inhibitors. The initial generation of inhibitors were very broad, included VEGF activity that brought a lot of the VEGF toxicity with it. But then the generation that preceded us was more selective for the FGFR family, but still was essentially equipotent against FGFR1, 2, and 3. And some of the agents also hit FGFR4. So that actually brought a lot of toxicity that was unrelated to FGFR2 inhibition.
In the case of FGFR1, the hallmark toxicity is hyperphosphatemia. In fact, some sponsors have chosen to address this with personalized dosing. Janssen with erdafitinib, their pan-FGFR inhibitor, actually have a labeled dose that starts at 8 mg. If patients don't have hyperphosphatemia after two weeks of dosing, they can dose escalate to 9 mg . That's the way they're trying to eke out a little bit more target coverage. We felt that a much more rational way of optimizing for FGFR2 inhibition would be to design a selective FGFR2 inhibitor. Nobody had been able to do this up to the point that we started working on the target. We used our platform, again, to really drive a deep understanding of the dynamics of the FGFR2 protein and to elucidate a mechanism that we could use to selectively target FGFR2 while sparing the other FGFR family members.
That's exactly what we were able to do with lirafugratinib, RLY-4008, where we're several hundredfold more selective for FGFR2 over the other FGFR family members. We've shown very low rates of hyperphosphatemia in the clinic, consistent with sparing FGFR1, and very low rates of diarrhea in the clinic, consistent with sparing FGFR4.
Okay. So yeah, maybe could you provide a recap of the clinical data to date and which indications are most promising? What is the status of the cholangiocarcinoma development and?
Yeah. So I'll start with a recap of the data and then hand over to Megan. The initial data that we generated and disclosed for lirafugratinib was in cholangiocarcinoma with patients with FGFR2 fusions or rearrangements. Again, this was the patient population where the existing proof of concept had been established. Response rates that you're looking at for non-selective FGFR inhibitors in that population is between 36%-42% for the two approved agents. As we came in, we felt that looked low for a fusion driver oncogene compared to what we've seen in other disease settings. As we continued to get more mature data on 4008, what we saw was that our response rate that we reported at ESMO in 2022 was between 58% and 82%, depending on what dose you looked at.
So we were seeing what looked to be numerically higher response rates compared to what would be reported for the non-selective inhibitors, again, consistent with what our hypothesis would be for achieving selective inhibition of FGFR2. We also then expanded that clinical development program to look at other solid tumors with FGFR2 alterations. And rather than define it by tumor histology, we defined those cohorts by molecular alteration. So patients with FGFR2 fusions and rearrangements, patients with FGFR2 gene amplification, and patients with FGFR2 mutations. We reported the initial data out from those cohorts at the Triple Meeting last year. And consistent with our hypotheses, the place where we saw the most activity was in FGFR2 fusions, where we reported out a 35% response rate. Of note of interest in that disclosure that we made last year is we did also do some tumor type by tumor type analysis.
So if we looked specifically at patients with breast cancer, we had 10 patients we enrolled who were hormone receptor-positive, HER2-negative breast cancer, who had FGFR2 alterations across the three cohorts. And of those 10 patients, all very heavily pretreated, four of the 10 patients had achieved very good durability, good quality partial responses. So I think we're seeing both some evidence of potential tumor-agnostic activity, especially in the FGFR2 fusion or rearrangement population, and also seeing some early evidence of potential interesting signal in specific tumor types such as breast cancer.
Yeah. And then in terms of where we are from a development perspective, last October, we announced that we had fully enrolled the pivotal cholangiocarcinoma cohort. And so we're letting those data continue to mature. And then, as Don mentioned, we also, in October, disclosed the initial tumor-agnostic data, which were promising but early. So we've continued to enroll those cohorts and actually just closed enrollment because we enrolled the patients we needed. And so we're letting those data mature as well. And we've guided to reporting data from the tumor-agnostic cohorts as well as a regulatory update in the second half of the year.
Okay. Yeah. And then speaking to that regulatory update, I mean, are you guys going to try to decide to go forward cholangiocarcinoma versus tumor-agnostic? What are sort of the puts and takes in terms of that decision? And what do you think could be the risks of the tumor-agnostic label commercially?
Yeah. So part of the guidance, actually, that we issued in October was that we were pausing near-term commercialization efforts for cholangio to allow the tumor-agnostic part of the program to catch up, just given the differences in population size. Cholangio is about a little less than 1,000 patients a year, whereas the fusions that Don mentioned is about 7,000 patients a year. And so just given the size of the opportunity, wanted to align the timing on those. And then from a tumor-agnostic standpoint, with the caveat that we haven't had any regulatory interactions at this point, what we've heard the FDA say, the standard of care in that area is about a 0%-15% response rate.
What we've seen the FDA guide to in a number of different scenarios is that they want to see a study designed with 15% excluded from the lower bound of the 95% confidence interval. So you'd be looking at, say, in a 75-100 patient study, about a 30% response rate. And that's in addition to having durability, at least out to the 6-month point, but then also in addition to having good diversity of tumor types. So you can't have one or two tumor types driving the responses. You need to see them across a good number of tumor types. So that's, again, general guidance from the FDA, part of what we're letting the data mature, and we'll take them to the FDA to have conversations on potential paths forward. And that's what we'll report back on in the second half.
Okay. Thanks. And then just looking at the competitive landscape, are there other emerging FGFR inhibitor assets that you guys are aware of? How should you be viewed relative to those? I mean, we cover Abbisko, it's a Chinese company. They have some FGFR assets, but they may not be specifically FGFR2 specific, but maybe you could speak to the competition.
Yeah. So I'll start with just a general comment that we were the first using our platform, our Dynamo platform, the first to be able to figure out how to get a selective FGFR2 inhibitor. Others had been working on it, and other very, very good drug discovery shops had announced that they were working on discovering FGFR2 selective inhibitors, and nobody had been able to crack the problem. I think you really need the type of toolbox that we have on the Dynamo platform to be able to crack that problem. Of note, even after we've disclosed 4008, after the structure was clear from patent filings, we've not really seen anybody else following us be able to get the same type of profile that we're able to generate with 4008.
Again, which I think speaks to the power of being able to drive drug discovery the way we do at Relay. With regard to Abbisko specifically, I know they just reported data a couple of days ago. So we have not necessarily profiled that molecule extensively. But I think in the initial profile that we're seeing from that trial, they're probably still not achieving the same type of profile that we've gotten for 4008.
Okay. Well, it looks like we're running up on time, but maybe one last quick hitter. You guys are also going to disclose some new preclinical programs later this year. Maybe can you provide any just little color on that?
Yeah. So we've previously disclosed that we have programs in oncology as well as genetic disease. So it'll come from one of the two programs, from one of those two sides, looking to disclose it this year. And also the other guidance we've given is just that it's going to be in the ilk of a PI3K and FGFR2 somewhere where we're looking to have a best-in-class or first-in-class type profile.
Okay. Well, with that, we'll conclude. Thank you.
Thanks.