Founder Stories: David Reese, CTO at Amgen

This is the BioCentury Show. Brought to you by BioEquity Europe. BioEquity is Europe's premier international showcase for financial dealmakers and biopharma executives to meet rising biotechs.

Hello and welcome to the BioCentury Show. I'm Simone Fishburne, editor-in-chief, and today I'm joined by David Rhys, Executive Vice President of R&D at Amgen. This is going to be an interesting conversation largely about omics. Dr. Rhys is a fair long-timer, I would say, at Amgen. He joined in 2005 and has held various leadership roles covering discovery, development, regulatory affairs, and of course translational sciences and oncology. All of that is relevant to some of the questions I've got for you in our conversation today. We're gonna talk about some of Amgen's R&D priorities. And as I mentioned, we're gonna delve into genomics, more precisely omics. And I know that that's a topic close to your heart. So let's start with this because you and I have chatted recently and you have talked about the idea that I'm quoting you, we are entering an era of human data. Maybe you'd like to expand on what that means and how it's playing into your thinking at Amgen.

Sure, Simona, and, and it's wonderful to be here with you. This is an expansive topic, you know, in human data, you know, I think for our purposes, you know, how do we think of human data? We think of human data, everything from now in the, post-DeepMind era to having the structures of now hundreds of thousands or millions of proteins and starting to untangle protein-protein interactions and being able to predict them in silico to multi-omic data. So starting with genomic data, but now transcriptomic data, proteomic data, metabolomic data, Layered on top of that, the data that we generate from clinical trials. So not only basic safety and efficacy data, but in some cases dense biomarker data and incorporating omics into clinical development. And then finally, at the other end of the funnel, an absolute tsunami of real-world evidence and real-world data from electronic health records, from insurance claims databases, and other often publicly accessible possible sources. When you put all of this together, to me, that, that's all data on human disease, human pathobiology, and how humans respond to therapies. That's really what's going to power drug discovery and drug development in the 21st century, as opposed to animal models. Will we still use those? Of course. But the primary engine here will be human data in all of its varied manifestations.

So I want to get into a couple of the bottlenecks that are constraining our ability or need to be solved. Now, one of them is data quality. That is always a big issue. Correlations and causality are often mixed, misinterpreted. And I'm sure that in Amgen you're all over that. I'm sure you don't let people just say, you know, I saw this correlate with that, ergo, there's a connection there. Do you have thoughts about how that needs to be addressed in a sort of broader field-wide way? And how do you sort of create discipline around that at Amgen?

Yeah, it's a great question. It's actually, I think, a huge question facing the field right now in terms of data integrity, but also more broadly than the reliability of the scientific literature. And there are different facets of this. Let's start with one, you know, so the databases themselves, of course, need to be as representative as possible of populations with diseases. And we know we need to do a better job, for example, in incorporating minority groups into genetic studies or groups that have just historically been underrepresented. Genomic studies still have a predominantly Northern European background. That clearly is insufficient. So one has to be cognizant of what's going in in terms of the actual data. Now, the good news is humans are much more alike than they are different. But those differences can be very, very important in terms of disease causation, progression, and responses to therapy. What I think you're also pointing to then is the reliability of the outputs. And, you know, let's take the simplest use case, which is, say, you know, generating a finding that gene X is associated with disease Y. And often, you know, the first thing we do with major findings in the literature is reassess that. Against our internal database, which now has over 200 petabytes of data. And there are many, many methodologic pitfalls, but, you know, one of the most common is underpowering of these sorts of studies. And we can very quickly look, often in a matter of days, and understand that in fact Gene X has nothing to do with Disease Y. It's a false positive, often from an underpowered study. And these papers litter, you know, they litter the scientific literature right now. And so, so the question is, how as a field do we create some sort of quality control around that so that there aren't so many rabbit holes or false associations that people are chasing? One thing that comes to mind for me, you know, for those who are around and early era of microarrays. Once the machines became widely available, folks were doing all sorts of gene expression studies, but the quality control was poor. Ultimately, the journals instituted requirements around quality control, and that greatly cleaned up the literature. I think we're probably due for one of those periodic rethinkings, and this has got to be across the entire field— academia, industry, regulatory authorities— to say, Hey, let's not waste precious resources on pursuing false associations.

So going from there and talking about the vast amounts of data that are going to start being generated largely because we have the technologies to do it, real-world data, real-world evidence is obviously another dynamic that's coming into play. And many hurdles exist there— interoperability, very fragmented systems, in particular in the US. Is this something that we're going to need different skill sets? You and I, many people are trained as biologists. We're trained— I don't know. I was trained on dose-response curves, but not on large statistical or even more technological aspects. And so what is your thought about our need for different training or skill sets And I suppose at this point I have to bring in the idea that tech has just laid off a ton of people. Is that an opportunity for our ecosystem?

I think it's a huge opportunity. And what we now call data scientists broadly, I think is what you're referring to. They are going— are and are going to be an even more important integral part of the drug discovery and development teams going forward. In fact, the mathematician in some cases may be the most important person on the team. And the mathematician who has some biologic insight will be worth his or her weight in gold. There's no question about that in bringing those disciplines together. I think one of the things we have to do is very simple, which is widen the aperture for the types of people. We have people on our teams now, they've come from everything from finance to theoretical physics, to computer science in these sorts of data scientist positions. And to me, the old notions of credentialing that we look for aren't particularly useful. What I'm much more interested in, certainly if I'm recruiting someone into leadership positions, is what's the substrate, what's the toolkit. And the letters after the name, they may mean something, but not a lot. Right now, not a lot going forward. And so, you know, I think it's just as much on us to free up our thinking as to the types of people that can really contribute to drug discovery, you know, than we have in the past. And we're really at a turning point, I think, where that's going to be crucially important.

So before we go to a break, I want to dig back into something that you said when we chatted previously. And it also sort of goes back a little bit historically with pharma pharmacogenomics where you sort of felt that maybe that hadn't paid off as much as we'd hoped for generally. I think you mean across the field. Maybe you can elaborate on that and tell me what the issues are and that these new skill sets that are going to come in and new data available, is there still hope for pharmacogenomics?

Yeah, I, I think its time will come, but like many things, I, I think we probably underappreciated the complexity. We have a few examples where we know a certain genotype is associated either with response to drugs or shouldn't be given, for example, for safety reasons. But it's literally like a handful or less of very common examples like that. And 20 years ago, we really thought that that was an area that was about to break open.

And so you're talking, are you, about like metabolic enzymes?

Like, exactly, metabolic enzymes, I think, is the— here, the the, you know, the simplest example. You know, so now I think, and this is where I think, you know, the, the multiple forms of human data that I described will come into play to give us much more complicated models about why do certain individuals respond to drugs where others are resistant. And it's going to be, it'll be a combination of the actual pharmacology and perhaps differences in pharmacology, something as as metabolizing enzymes, but also many other host factors. In autoimmune diseases, we know diseases like lupus must be multiple diseases on a molecular level. And the response to therapies is very different. But we have this very crude phenomenologic clinical classification of the disease. You literally have to have x out of 11 things on a checklist in order to qualify for a diagnosis of lupus. If you look at that checklist, tells us almost nothing about the disorder. So, so I think, you know, we're just ready to go to that, that next layer. And that will then lead us to a form of pharmacogenomics that's a little different than metabolizing enzymes, but a much more holistic view.

And so you use the word holistic, but is pharmacogenomics a prerequisite or an integral part then of personalized medicines?

Oh yeah, I think it's probably necessary but not sufficient.

Okay.

That's how I would put it.

All right. Well, that's a great place for us to take a break. So we'll take a brief break. And when we come back, we're going to talk about some of Amgen's programs and priorities.

The 23rd Bioequity Europe Conference heads to Dublin, Ireland in May 2023. With a program focused on how to create a new playbook for biotech success in Europe. In 2022, over 800 industry leaders from 30 countries joined BioEquity in Milan, Italy for 2+ days of strategic panels, one-to-one partnering meetings, and C-level networking. More than 150 biotechs selected by BioCentury presented their story to investors and potential partners. Don't wait. Two of the last three in-person Bioequity conferences sold out. Register today and join BioCentury EBD and our Dublin regional host committee at the industry's premier C-suite and investor conference. Visit bioequityeurope.com for more information.

Welcome back. I'm here with David Rhys from Amgen. David, so if I have this right, December was the 10-year anniversary of Amgen's deal with Decode. Some of our viewers might be young enough not to need a reminder of what that brought to Amgen. Can you talk about how that acquisition has played out, what did and didn't live up to expectations, and how the unit's playing in your plans going forward?

Yeah, no, that's great. I mean, deCODE Genetics, I think, is one of the world's great resources in this field, founded by Kari Stefansson in the mid to late 1990s in Iceland, initially started with genotyping of a fraction of the Icelandic population has now grown to a database that encompasses about 2.5 million individuals who've been genotyped, several hundred thousand whole genome sequences, each of these being added to now on a monthly basis, but also now in the hundreds of thousands with near-complete proteomic profiling, sometimes longitudinal, as well as tens of thousands with transcriptomic profiling. Pro profiling. As I mentioned earlier, all told, this is now well north of 200 petabytes worth of data. And, you know, we use this every day as part of the drug discovery and development process. You know, the simplest use would be target validation, in the sense that there's a potential disease target of interest, maybe it came from genetic finding, maybe it came from other observations, but we want to put it through this multi-omic analysis to say, is this really a driver of this disease? That should lead us then to a therapeutic hypothesis. Where should we interdict that pathway to potentially alter profoundly the natural history of that disease? We do this now routinely with all of our programs in this This is something that Kari and his team working with the teams in California and elsewhere, you know, have just integrated across that drug discovery, target discovery in particular, and validation effort. You know, to me, the next big frontier is doing this routinely in clinical development. I think ultimately that is the big payoff. We've got programs in cardiovascular disease, for example, a program that targets Lp(a). Elevated Lp(a) levels are associated with a fraction of atherosclerotic cardiovascular disease. It is an independent risk factor of LDL cholesterol. It is not modifiable by diet, exercise, or really currently available pharmacologic interventions. There's a very strong genetic component. There's very strong epidemiology linking elevated Lp(a) levels to atherosclerosis. We use this multi-omic profiling in the development program and we've now just launched a Phase 3 cardiovascular outcomes trial with a small interfering RNA molecule called Opassiran that profoundly lowers Lp(a) levels. The design of that trial was directly influenced by the multi- biomic profiling that I'm talking about. And we will be collecting samples in that trial to further do profiling to now start to understand why did some patients potentially benefit, whereas others didn't. This I think is the real future of this sort of profiling. The sky's the limit there. We're really just in the very, very early innings.

So the resource, let's call it Decode, has it translated into picking better targets? And is there a way that you follow metrics on, you know, discovery efficiency or something like that?

Yeah, you know, absolutely. I, I mean, you know, to me, the, you know, there are few existential challenges facing research and development organizations in this industry. One is reducing the cycle time, you know, from the 10 to 14 years it takes from, you know, idea in the lab until marketing authorization. You know, patients can't wait that long and it's not economically viable. Going forward for it to continue to take that long. Second is improving the success rate. Meaning roughly at best, 1 out of 10 molecules that goes into the clinic actually makes it across the finish line. We have to do better than that. I think this multi-omic profiling, which is a way you can think of it as trying to load the dice. Because you're taking experiments that Mother Nature has done for you. If there's a genetic association, for example, to tell you that something is important. So if we can load the dice, if we could simply double the success rate from 10% to 20%, you know, that would be absolutely game-changing. Think about that. We would still fail 80% of the time, and it would be a massive success. You know, my son's a law student. He thinks that's sort of crazy thinking.

He's not a math student.

But, you know, I said, look, if I succeeded 20% of the time, I'd be the best there ever was in this field. But it's not pie in the sky anymore.

So, you know, on that note, let's just talk a little bit about the expectations that sort of could have come that you started out with. And have you learned things along the way with Decode that are guidances to the sort of limitations of this? You know, obviously with the Human Genome Project, people thought that it would just be a matter of years and we'd just be, you know, minting drugs, as it were.

Yeah.

So what are the limitations that you're learning?

You know, the old truism probably applies here that we overestimate the importance of things in the short term and underestimate the importance in the long term. And I think for, you know, genetics, genomics, and the sort of multiomics that we're talking about, that will almost certainly undoubtedly prove to be the case. You know, one of the critical things that Decode has brought to us that you never see anywhere, and it never shows up on a balance sheet, or any other sort of accounting is that we don't do what I would say are stupid things, meaning there are reports in the literature about potential disease targets. And when we take a look at it with this sort of intense multi-omic analysis, we say, you know, we just, that's just not true. And so we elect not to do it. And so the avoidance of that opportunity cost is an immense, that alone, alone is worth the price of admission and more. Just avoiding doing dumb things that in the end will have no hope of succeeding is tremendously valuable. Now, where I think we all got it wrong was that we hoped there would be a number of targets like PCSK9, where there's clearly a genetic finding in a single gene that leads you to a very simple clinical hypothesis. But the number of those sort of monogenic cases that are going to lead to drugs like, you know, Repatha, PCSK9 inhibitor, uh, or is probably small. Uh, it's probably small, and things are more complicated than that. So that was the short-term overestimation. We thought there would be more of those, but in the long term, this sort of infusion of what we're doing into clinical development and much more complex risk predictors, I think will probably have a much greater payoff than we currently appreciate.

So I'm going to switch to, you know, one of— talk a little bit more about some of your programs. Now, a top issue for many of the watchers, our audience, will be your thoughts and your plans for Lumachrome.

Ras.

So PD-1 combos have not really been encouraging there. And I, you know, again, you know, speaking to you previously, you used this phrase that has stuck with me, which I think is great, is like, ras is a different beast.

Yeah.

So what's your strategy now? Are you thinking about different modalities or small molecules for different KRAS mutations? How are you thinking about this program?

Yeah, so Lumicras targets the KRAS G12C mutation. It's It's present in about 15% of adenocarcinomas of the lung, 4% or so of colorectal cancer, and then at a low frequency in a variety of other tumors. I think one thing that's really important to keep in mind here is it took 40 years to actually get the first RAS inhibitor approved by the FDA. Now there are two. Underestimate the magnitude of that accomplishment. When I was a fellow in training as a medical oncologist, I'm an oncologist by way of background. The first thing I started studying was the RAS MAP kinase pathway. And of course we thought, hey, all we need to do is figure out how to get a RAS inhibitor and a lot of these problems will be solved. And of course it took 40 years to actually see that come to fruition. When I say it's a different beast, what I mean is it's a different beast than something like EGFR, epidermal growth factor receptor or other receptor tyrosine kinases where there are specific suites of mutations and we can now predict what they will be that confer resistance. RAS is a downstream signaling molecule in this node. It's structurally very different. It's very clear that there aren't clean resistance patterns. The work that we've done suggests that there are many ways that tumor cells can take to circumvent RAS inhibition. And so there's not a simple answer. And so there's a lot of biology to sort out, and there's going to be a lot of, I think, now hard empirical work in the clinic. You know, we're looking, for example, to move into earlier lines of therapy in lung cancer with a chemotherapy-lumacrast combination, in particular in those tumors that are are PD-L1 negative. So these are non-expressors. They derive a very modest benefit, if at all, from checkpoint inhibitors. So it's now what I would consider the sort of hard spadework of building on these initial findings, but at least having the tools to do that.

So let's stay with cancer for a minute. We wrote last week an article about dose optimization. Moves by FDA via Project Optimus, you know, shifting from what's really been standard in the field, maybe in an overdue way, from maximum tolerated dose to sort of studies earlier on in development that will optimize the dose and reduce toxicity. So how does that square with the way you've done sort of rapid development, you know, that strategy and What do you also think it means for smaller companies? Because presumably this is going to be a very resource-intensive or costly approach to coming up with optimum dose and drug development.

Yeah, the first observation I'd make is that it may be hard to generalize here. It will probably depend on the target and the drug in development in question. I would agree with the premise that it's time that we move past the simple maximum maximum tolerated dose approach, which remember was introduced decades ago now when the drugs that we had available were cytotoxic agents where the notion of MTD actually made sense. You wanted to give as much as you possibly could to maximize an antitumor response while having an acceptable therapeutic window. But for a targeted therapy, for biologics like a bispecific antibody, there may be an optimally effective biologic dose that may or may not be the maximum tolerated dose. So I think mandating exploration of that is a good thing for the field. It's a good thing for patients. You know, the key is how to build in this exploration in parallel in the development program so that you don't slow the availability of agents to patients. And there are potential pitfalls, uh, as. Because for example, with targeted therapies, if you're looking at a short-term readout like response rates, you may see that different doses have a relatively similar response rate, but the important endpoint may be the duration of that response or progression-free survival or things that are, you know, what we call time-to-event endpoints. And it takes a while to sort that out.

Out.

One of the— and the other pitfall is with cancer agents, how do we actually breed resistance to tumor cells in the laboratory? We give them a marginally effective but slightly subtherapeutic dose of the agent for a long period of time, say in cell culture, and then eventually resistant clones grow out. We have to make absolutely sure that we're not doing that in the clinic. And so So it's going to be a constant balancing act between speed of development and making these agents available to the largest number of patients possible with getting that right answer. And so I just don't think there's gonna be a generic approach. It's going to be very target and drug dependent.

So let's talk for a minute about Olpasiran, which you referred to earlier. And that is also a new modality for Amgen. SiRNA. I think you have one other RNAi therapy for NASH in your pipeline. So, you know, does going into this kind of a new modality bring different challenges for you? And should I expect, should we expect to see more from Engen in nucleic acid modalities?

Yeah, we're absolutely interested in nucleic acid therapeutics. You know, we chose to, you know, get out of the gate with small interfering RNAs, so RNA therapeutics. But this is something that I would expect us to develop over time. Opassiran itself looks great so far. I think it will definitively test the Lp(a) hypothesis. We presented Phase 2b data at the American Heart Association meeting last November showing profound reductions on the order of 95% at most doses tested. Of Lp(a) levels in individuals with elevated Lp(a). And so, you know, as I mentioned, we're in a phase 3 cardiovascular outcomes trial now. But the, you know, the drug clearly profoundly lowers Lp(a) levels, and if that affects event rates, we're going to know. I don't think there's any question about that. But, you know, your broader question, are we interested in RNA therapeutics?

Absolutely. Absolutely. So let me just end now by, you know, we've talked about some of your flagship programs, but I'm interested in you talking about either other programs or other innovations that are maybe less in the limelight that you are particularly excited about.

Yeah. You know, maybe I'll end with, you know, with a platform that, that we're developing that we call, you know, proximity biology or the induced proximity platform. And, and this is predicated on the belief Part of the future will be multi-specific molecules. For most of this industry's 130-so-year history, there's been a single drug and a single target, and they interact in some way, and the drug either increases or decreases the activity of the target generally. I think we're now entering the era of multi-specifics. A BiTE molecule, bispecific T-cell engager, is an example. One hand grabs, pumps a tumor antigen, the other hand grabs the T cell, brings them into proximity, the T cell is activated and kills the tumor cell. We are creating a platform built around both small molecules and large molecules that will create these sorts of multi-specific molecules. So a PROTAC, you know, protein, you know, targeted protein that induces targeted protein degradation. One piece of it grabs a protein target of interest, the other grabs a ubiquitin ligase. The target protein is ubiquitininated, which is a tag for that to be destroyed in the cellular disposal. There are huge vistas this opens up because you don't have to necessarily modulate the activity of the target, you just have to be able to bind to it. And so we have incorporated technologies around DNA-encoded libraries about generating antibody fragments, all of which have as a purpose the ability to put together like Legos, multi-specific molecules that do exactly what you want. And they can target proteins, they can target RNAs. So we're developing RNA targeting, you know, what are called, you know, ribotax that can, target an RNA and selectively degrade it. Why am I going on and on about this? You may want to—

I think he's also written about it, so I'm delighted for you to do that.

Let me end with this thought. 80-85% of the current targets of interest are not druggable. They're the so-called undruggables with current modalities, and we believe that multispecifics will be useful, and in fact, many cases required to open up that undruggable universe. And so that, I think, is a big part of the future, and that's a platform we're really investing a lot in. I have no doubt it will be just as important 10, 15 years from now as monoclonal antibodies are now, and which took a while to develop.

David, this was a great conversation. Thank you very much for your time, and we'll certainly be following those.

Well, thanks so much for having me.

The BioCentury Show is brought to you by Bioequity Europe. Bioequity is Europe's premier international showcase for financial dealmakers and biopharma executives to meet rising biotechs.