Founder Stories: Robert Plenge, Genetic Dose Reponse Curves

[Speaker:JOSHUA ELKINGTON] Yeah, thank you very much. So Cecilia, this is just a perspective on how to use genetics for drug discovery, but it's a topic that I think a lot about. And, you know, what I thought I would do is just sort of start by defining what I consider to be the problem that we actually face in drug discovery, which is, you know, it's essentially— it's a pretty straightforward process. If you think about making a drug, you want to figure out how to perturb a target, the magnitude of that perturbation, and whether it's going to be therapeutically beneficial in humans. But you also want to go one step beyond that. You want to actually show that that target, if you perturb it, is going to differentiate from standard of care. And so there are a lot of different ways in which people actually think about how to use genetics to actually define drug targets. And so I'll describe one very briefly, which is this allelic series model. This has actually come up several times as we've had the discussion today. I'll then show you one example in immunology for a gene called TIK2. Which has been used in the treatment of psoriasis. And then finally, I'll just conclude by describing a few of the resources I think will be required in order to build out this model over time. And so the way I like to describe this— this is a very hugely oversimplified view of genetics and drug discovery— but along the x-axis is gene function, from gain of function to loss of function, and along the y-axis is a human phenotype. It could be risk of disease, it could be some of the hematopoietic phenotypes that Nicole just talked about, et cetera. But the first step that you actually have to think about is, is your phenotype a good surrogate for drug efficacy? And one way to actually do this is to simply ask, given the genetic associations to date, do those overlap with approved therapies? And this is a study we did a few years ago. Yukinori Okada, who's now in Japan, actually led this. And there was a very non-random overlap for genetics of rheumatoid arthritis and approved drugs for rheumatoid arthritis. So I think it's a good place to give you confidence that the phenotype that you're studying is a good surrogate for drug efficacy. But then you need to roll up your sleeves and do the human genetic studies to actually find those variants. Each X on this axis is meant to be an individual allele. Maybe it's one allele. Ideally, you would find multiple alleles that are associated with the trait of interest. But in this stage of the exercise, you really don't know anything about function. So you know an allele is associated with a trait. You might have a sense of the magnitude of that association in terms of the impact on the clinical phenotype, but then this is when you actually have to understand the, you know, the variant-to-function part of this particular discussion, and to understand where it actually distributes along this x-axis. So is it gain-of-function, is it loss-of-function, what cell types are involved, what pathways are involved? But if you start to do this, you start to really begin to understand again, a genetic perturbation on a target and how that actually potentially could be translated into a therapeutic hypothesis. And, again, this is the variant-to-function component of the discussion that we're actually having over the next couple of days. But I also think that one of the very, very interesting parts of human genetics is to be able to take those exact same alleles and to be— explore the pleiotropic consequences in large populations. And you can actually use this not just as a surrogate for efficacy, but you can use this as a surrogate for toxicity. So a surrogate for on-target adverse drug events. And in this way, you can begin to sort of map out these dose-response curves. And here is where I think the biobanks are incredibly, incredibly valuable, because you can actually go beyond any of the phenotypes you've actually studied in any individual case-control cohort. To understand the pleiotropic consequences. And if you can put all of this together, find the alleles, understand the function, understand the impact on clinical phenotypes, understand the pleiotropic consequences, begin to match modality and mechanism, again, I think this is a really good starting point for a drug discovery journey. And so I think one point that I do want to make has come up a few times, is not necessarily common variants or rare variants, and similarly, it's not necessarily protein coding or regulatory. It's really about having that range of alleles that perturb a target to estimate the impact in humans before you start this drug discovery journey. And so that, to me, is what we're trying to accomplish with human genetics, to guide not just the target, but the magnitude of the effect. So I'll go over one example in human immunology again, and then come back and just conclude with what are some of the resources that are required to build out this model over time. So the example of the series that I'll show is for a gene called TIK2. This is an intracellular signaling molecule, and I'll show you more about this on the next slide. There are rare mutations, complete human knockouts, and these mutations are associated with immunodeficiency and an increased risk for particular types of infections. There are also common protein coding alleles that affect the enzymatic function of TIK2, and these alleles are loss of function, and they protect from a variety of autoimmune diseases, including RA, lupus, type 1 diabetes, and IBD. And what I'll show you using UK Biobank as an example is that these same common alleles that protect from autoimmunity do not increase risk of infection. So you begin to see this concept of an allelic series in a therapeutic window. So just to tell you very quickly about TIK2 and where it lies within the— The signaling pathway is part of— again, it's an intracellular signaling molecule downstream of IL-23 and IL-12, but also other signaling molecules including interferon gamma. And there have been multiple genetic studies in psoriasis which have implicated different components of the signaling pathway. So the IL— P19 and P40 components of the ligand, one of the two subunits of the signaling receptor, and then TIK2 itself. Antibodies against the extra— the ligands themselves have been shown to be beneficial in treating psoriasis. So this is that first part using psoriasis and genetics of psoriasis as an acceptable surrogate for drug efficacy. I think the fact that you actually have drugs that are targeting components of this pathway provides that evidence. The question, of course, is will inhibitors of TIK2 be therapeutically beneficial? So a number of groups have actually done large-scale genetic studies to find multiple variants at TIK2. This is a paper that was published recently in Nature Genetics. It actually focused on type 1 diabetes and rheumatoid arthritis, a study by Shomari Chaudhary and John Todd and others. And what they found is that there are multiple alleles, protein coding, within TIK2 that protect from these different diseases. These same alleles have been shown to protect from psoriasis as well. There are also some other low-frequency variants. So again, this is this idea that you can find variants that are associated with a range of clinical phenotypes, in this case protection from these different autoimmune diseases. One of these has actually been studied in more detail, and this is a study from Kelly Dendroux and colleagues that actually isolated blood from patients who are carriers of these different alleles. They did different conditions where they actually activated different signaling pathways. Here I'm showing for interferon alpha signaling pathway, and you can actually see there's a dose-response between carriers of the risk allele and the protective allele. And more specifically, there is about 80% loss of function in this homozygous state is associated with protection from autoimmune diseases, again, including psoriasis. Again, now on this dose-response curve, we're beginning to see not just the associated alleles with clinical phenotypes, but the magnitude of effect and the effect on function specifically. So just in the last few minutes, turning towards biobanks and how you can use these same variants for pleiotropy scans. In that same paper, they looked and used the UK Biobank, and what they showed is that in otherwise healthy individuals, you have a particular allelic distribution. But if you look at specific individuals who have had different types of infections, you can ask the question, is there an association with risk of infection? And the answer for this one particular variant, the P1104A variant, there was not an association with an increased risk of infection. So, that is about 80% loss of function of TIK2 is associated with protection from autoimmunity, but not associated with increased risk of infection. Again, this begins to sort of look at this consequence of adverse drug events, in this case, risk of infection. So, if you put it together, you actually have A curve which is another version of these dose-response curves. If you completely knock out TIK2, you have immunodeficiency, you have increased risk of infection. Partial loss of function in the order of about 80% is associated with protection from autoimmunity, but no increased risk of infection. And so therefore, the therapeutic hypothesis is partial inhibition at a range of 50% to 70% should be therapeutically beneficial. Again, the starting point for a drug discovery journey. I've been telling this story for a few years, and then very recently, BMS published their clinical trial data in psoriasis. This was a few months ago in New England Journal of Medicine. And indeed, they were able to show 50% to 70% inhibition of TYK2 by a molecule that acts as an allosteric modulator had actually very, very strong and very, very beneficial effects on protecting from the psoriasis, not protecting from, treating the symptoms of psoriasis. And so these are our clinical scores called PASI. It basically measures the amount of psoriasis that you have on your skin, and that 50 to 70% inhibition of TIK2 led to PASI 75 scores in about 75% of individuals and PASI 90% scores, that is 90% of the lesions cleared on the skin, in about half of the population. So just in the last slide, just gaps remain, and the pre-reads, I like the— there's a term that I really like, the federated ecosystem of interoperable data custodians. So, how do we actually become this federated ecosystem of data custodians? And I'll just throw out a few examples of what I think are going to be important for building out this roadmap. First, I think we need to continue to understand the genetic architecture of complex traits. That is, of course, doing more GWAS and a variety of intermediate traits, but probably more importantly, it's really getting down to the fine mapping of those common variant signals, doing colocalization. We've heard about that today. But I also think it includes continued sequencing of rare Mendelian families, but also large-scale sequencing, and exome sequencing in particular, in case-control populations with these different common diseases to build out this allelic series. A component of this is also to look for rare loss-of-function mutations, whether they're in individuals who have been ascertained with disease or those who are part of these special populations, consanguineous populations, et cetera. So basically, again, building out the complete allelic series at any gene of interest. And then to do a variety of things to study the functional interrogation, high-throughput assessment of mutations, I think single-cell sequencing in cell types of interest. To really understand the cis-eQTL effects and the magnitude of those cis-eQTL effects. You can really understand, again, the effect of any genetic perturbation on function. And Mendelian randomization for a variety of quantitative traits, PQTLS is one example, but I think there are a variety of intermediate traits that can actually be useful in this process. Building out these biobanks and be able to go from not just hundreds of thousands, but really tens of millions, and be able to federate that, so anytime you mutation of interest, you can very quickly understand the phenotypic consequence. And we need to do that much more quickly than we can today. I think something that we haven't really touched upon today that I think is an interesting consideration are the— really the data analysis skills that are used to generate these function phenotype maps. It's not just colocalization, it's not just understanding cis-eQTLs, but really to try to tease apart all components of this dose-response model. And to be able to integrate this in a way that allows you to quickly visualize any gene of interest in the function phenotype map itself. This is a model, and I think there are lots of limitations to this particular model, and this is something that we can actually, I think, discuss over the next day and a half, but it's very reductionist, it's linear, you need multiple variants, it doesn't really take into consideration the phenotypic consequences at the cell or the pathway level, and I think there are a lot of interesting approaches that one can use there for drug discovery as well, including phenotypic screens based upon polygenic risk scores as a simple example. I think something that we haven't talked at all about, and it's probably not exactly the purview of this particular group, is how do you really get into matching modality and mechanism to guide a drug discovery process, but I think that's a real challenge that we actually face. Last comment, and then I'll finish up. I think one of the— in the pre-reads, it was, how do we provide incentives for making these types of datasets available? How are we going to incentivize individuals who have done these large-scale genetic studies to put them into a common federated ecosystem? And I think if you had a portal that could run a number of these analyses, including the ability to actually generate these function phenotype maps, I think that would be potentially an interesting incentive, because there are so many different components. That it's very hard for any one individual to do it, but hopefully we can actually all come together to enable the tools to allow this to happen. So with that, I'll be happy to take any questions.

Great. Any questions for Chris?

Manolis?

So, Manolis, I would like to— so one of the One of the aspects about drug discovery is that we're looking at something that we want to treat that may be too late when it's relative to the cause of the disease. There are some diseases where once you have your diagnosis, there's some irreversible effects that no matter what the genetic cause is, it's just too late to treat, to target that genetic factor that led to this. So, and this is not specific to specific to you, but just to raise this point that if we are interested in actually addressing that last phase of getting to targets that would be relevant for drug development, we don't only need to understand the mechanism of getting the disease, but we need to start thinking about phenotypes that are relevant for drug development that are probably have to do with progression, have to do with specific complications, have to do with treating biology that was not present in the pre-disease state, but is now present in the post-disease state and probably has a different genetic architecture from the causality of the disease. So I'm not sure exactly how to place the— how much one can place resources on the one versus the other, but it is important to balance as we develop this discussion of how we put several investment components into this landscape.

Yeah, and that's why I always start— it's important to start with what phenotypes are good surrogates for drug efficacy. And I think if you see the overlap of approved drugs and traits, whether it's risk of disease or some other quantitative trait, I think it's a good place to start. But you're right, I think you need to think very carefully and deeply about that, and it's not necessarily straightforward. There's a longer answer to that, of course, but that's, I think, a quick answer.

Robert? Yeah, Robert, that's a great point. That last piece about incentivization, I do feel that there is this very— this other sort of rather straightforward thing, which We as a community insist that we deposit— that everybody deposits their summary statistics, you know, post-GWAS. Each time we make it easy to do, and that's the— that's kind of the basic research molecular biology tradition, is to insist on publication and data deposition on publication. And previously there have been these concerns around privacy for summary statistics, But I think that has faded away. And so we're in a position to, as a community, say, these are the rules of the game of this community. And I definitely think that should be on the table, as well as making it easy, as well as producing incentivized— if you do this, certain other things are easier.

Yeah, I would certainly agree with that. I'm sure probably other people are in a better position to comment on how to make that happen. At least in the pre-reads, it's a relatively small proportion of individuals who have done these studies that have actually deposited their data. I don't know if it's 20% or what that number is, but we need to do a better job somehow of basically bringing all this together.

[INAUDIBLE] [Speaker:KATHY_KLEINER] Yeah, so Kathy Kleiner at Sanofi. So I agree with you, but it's also important to know that while we have a decent tradition in genetics— and the genomics community is sharing data, depositing it in various kinds of repositories. It is an uneven tradition. And the rules are there. We all know that some of the very prominent journals enforce those rules with various degrees of rigor, depending on how important they think the paper is, and who the labs are, and et cetera. So, if you want to have those standards, I agree with you, that is one of the ways— but instead under quality control. Then we're going to have to do a better job of figuring out how to not only promote those standards, but to get the communities around us to help enforce those standards.

[Speaker:DR. JOSHUA ELKINGTON] Right. [Speaker:DR. JOSHUA ELKINGTON] Yes, sir.

[Speaker:DR. JOSHUA ELKINGTON] There we go. [Speaker:DR. JOSHUA ELKINGTON] Thank you.

[Speaker:DR. JOSHUA ELKINGTON] Major funders being a more effective way of influencing this rather than relying on each journal to adhere to it, because, you know, If it were simply the case that NIH and Wellcome Trust said, you've done GWAS meta-analysis in an important disease with our support, you must put the summary statistics from that meta-analysis in a public repository, or we're not going to fund that effort any further.

I mean, there's—

But I mean, I guess all of this, it's not an either/or, right? We're allowed to use— all of these different techniques to incentivize. And I think, you know, the major funders at the moment should be celebrated that they insist on raw data deposition to these protected managed access databases. But there's clearly this value in summary statistics that go way, way more useful in many ways.