VIDEO: Mark Murcko on A Condensate Mindset in Drug Discovery

Dewpoint scientists hosted experienced drug-hunter Dr. Mark Murcko to deliver the inaugural 2024 Kitchen Table Talk. Mark discussed the evolution of drug discovery strategies that have led to successful therapies and how a condensate mindset can reshape how we find the drugs of the future.

Mark is a seasoned drug hunter who contributed to the discovery and development of nine marketed drugs that today help treat glaucoma, cystic, fibrosis, HIV and hepatitis C virus infections. He has also authored over 90 papers and is a co-inventor on over 50 patents.

Mark is a chemist with a PhD from Yale University and has worked in the pharmaceutical and biotech industry for several decades. For example, at Merck, Mark contributed to the discovery of the glaucoma drug dorzolamide, which is the first marketed drug that resulted from a structure-based drug design.
Mark was also a founding scientist at Vertex Pharmaceuticals and while his roles and responsibilities grew and he became a CTO and chair of the SAB, he contributed to over a dozen clinical candidates, as well as all the seven marketed drugs from the company.

Mark is also a recognizable name in the startup biotech circles, where he contributed to the inception and disruptions of over 15 organizations. Among these he was a founding CSO at Relay Therapeutics as well as Dewpoint Therapeutics, and currently sits on the board for both.

Click here to view the engaging discussion on a condensate mindset to drug discovery. 

TRANSCRIPT

Diana Mitrea:

It is a great honor to introduce today’s KTT speaker and my mentor, Dr. Mark Murcko. Mark is a seasoned drug hunter who contributed to the discovery and development of 9 marketed drugs that today help treat glaucoma, cystic fibrosis, HIV and hepatitis C virus infections. He has also authored over 90 papers and is a co-inventor on over 50 patents.

By training, Mark is a chemist with a PhD from Yale University and has worked in the pharmaceutical and biotech industry for several decades. For example, at Merck, Mark contributed to the discovery of the glaucoma drug dorzolamide, which is the first marketed drug that resulted from a structure-based drug design.

Mark was also a founding scientist at Vertex Pharmaceuticals and while his roles and responsibilities grew and he became a CTO and chair of the SAB, he contributed to over a dozen clinical candidates, as well as all the seven marketed drugs from the company.

Mark is also a recognizable name in the startup biotech circles, where he contributed to the inception and disruptions of over 15 organizations. Among these he was a founding CSO at Relay Therapeutics as well as Dewpoint Therapeutics, and currently sits on the board for both.

I could go on and on, but Mark asked me to keep it short, so I’ll just add that he’s a wonderful teacher and mentor, loves hiking, and he loves building trebuchets. So, Mark, thank you for being here with us today, and the floor is yours.

Mark Murcko:
Thanks Diana. And hi, everybody. So, I am pleased to be here.

And basically, I think that there’s all kinds of opportunities here in the condensate space to help drug discovery. And that’s the perspective that I bring today, is I am a drug discovery guy. And I’m always looking for ways to improve our ability to discover new medicines, because we have to begin from a standpoint of humility. All of us collectively in the world, we only produce a few dozen breakthrough medicines a year, and that’s from what everybody in the world collectively is doing.

And so I’m always looking for ways to make drug discovery just a little bit more effective, you know, to bring different approaches to things. And so, I’d like to share some of my thoughts on how a condensate mindset can be useful in this regard.

And, as Diana mentioned, I helped to build out the plan and to launch Dewpoint, and I was involved with the company for the first three years, and I’m still on the board, still spending a lot of time with the scientific team at Dewpoint, and I’m really excited about all the progress that is being made. And so in a few places in the talk, I will highlight some of the points I’m trying to make, using some data from Dewpoint. So just be on the lookout for the Dewpoint logo for those specific examples. And of course, it’s been fun over the last six plus years to be interacting with the whole team, and that includes the internal folks as well as the founders. You know, Tony and Rick and Phil, and the whole SAB, and in particular, getting ready for this talk, I would like to thank Diana who is tremendously helpful in getting my own thinking organized, and also had the chance to get some feedback from Bede and Violeta, and from Isaac. So, thanks to all of you for that.

So now, I just want to maybe just take a second to say that despite all of that, I’m really only representing myself today, I’m not here on behalf of Dewpoint or any other organization. And I say that because what I’ve attempted to do today is to come up with a very personal sort of talk. It’s about my evolution. I hope it’s evolution in thinking about the trade-offs between target driven and phenotypic drug discovery. And then how condensates play into that. And that’s what I’d like to cover today.

And I think, since the Kitchen Table Talks by design, they’re supposed to celebrate advances in our understanding of biomolecular condensates. I think it’s appropriate that in that spirit today I would like to talk about how condensates and a condensate mindset may help us to discover new medicines.

Now, as I mentioned before, drug discovery is really hard, and we mostly fail. That’s what we do, and there are many reasons for failure. We may not make the right molecule, or we may not deliver enough of the molecule to the right tissues and sufficient concentration where we may simply not understand the disease biology. And so we pursue the wrong targets in the first place, or we may run the wrong clinical trial.

And all of these challenges, of course, are exacerbated by ignorance. And ignorance is not stupidity. Ignorance is our global collective lack of knowledge about how the body works and how molecules interact with each other. So these are all very big, messy challenges. And so I think it’s useful to start by asking, how have drugs been discovered in the past. And how are they now discovered?

Historically, people from around the world through careful observation, you know, they made all kinds of interesting discoveries, such as the fact on the left that you can take willow bark and make a tea out of it, and it helps you with inflammation.

Nobody knew why, but those kinds of folk remedies, I think we still have a lot to learn from. And then, in the last century, of course, has been the rise of the pharmaceutical industry, and from the early 1900s to about 1970 drugs were discovered either through serendipity, like penicillin, was, or through a process known as classical pharmacology, in which libraries of compounds, including natural products, are screened in cells or animals to find substances that did something desirable. Although the reasons why those compounds were working were never understood. And more recently the use of cellular phenotype has become widespread.

But the term phenotype is actually used in a variety of ways. It may mean a mechanistic readout on a specific target or pathway, such as a reporter assay in the GPCR. Or it may have a more functional pharmacologic kind of meaning. So I think a good example of that would be if you take an epithelial cell from the lung of a cystic fibrosis patient and measure the amount of chloride that’s effluxed out of that cell. That would be an example of a functional readout; and then on the right, it may simply be a phenomenon that you’re measuring some kind of change in cellular behavior, where you don’t yet know exactly how it may tie to a disease, or even to a target or pathway. So an example of that might be, I don’t know a response to oxidative stress, or maybe a change in a cell painting image. Okay? So in any drug discovery project, what you’d like to do is combine these. You’d like to understand the interplay between them and to understand how they relate to clinical response, because in the end the goal is a medicine, and to ensure medical relevance you’ve got to find some way of linking all the way through from the cellular phenotypes to a relevant, approvable clinical endpoint. That’s the goal.

So I think a good example of this linkage of using this end to end thinking in a phenotypic drug discovery project comes from the Vertex work in cystic fibrosis, and so far that has produced 4 pretty revolutionary medicines. And we were lucky because we understood the genotype-phenotype relationships and the natural history of individuals with the disease. We had varying mutations, and all of that information taught us that if we could get back to even just 10% of the normal amount of chloride being efflux from the cells through CFTR that could be medically beneficial, and we had access to relevant epithelial cells from patients with different mutations, and we could run an early proof of concept clinical study, where all we had to do was measure the amount of chloride in the sweat of the patients.

And then longer term, were functionally relevant readouts, such as the ability to breathe. You know actual amount of, you know, oxygen expelled from the lungs or exacerbations leading to hospitalization, all of those things also could be measured. So it was this end to end interpretability that really enabled the project to succeed.

Now, of course, therefore, as you’d expect, I’m a huge fan of phenotypic drug discovery. But of course, there are some challenges which I boil down to interpretation and optimization. So screening hits. Usually in the beginning they have very weak activity, not always, but usually, and so small changes in chemical structures can wipe out the activity which is very confusing.

And then another challenge is, you really need to be confident in the disease or elements of the screen, you run.

And then another challenge is target deconvolution. So you find something that works in a phenotypic assay. But you don’t know why. And so how much effort should you put into that. And it can be devastatingly hard to figure out the targets, target, or targets. And it may be targets. It may be plural, in which case the chemical optimization is even harder, because you’re trying to simultaneously make a drug effectively engaging with more than one target.

So clearly, all of these things are really quite challenging. And then you get to optimizing chemical matter. That’s just as bad. So consider a case where you have two chemically related compounds. They’re nearly the same structure, but one works better than the other in cells. Well, why is that? It might be that one molecule gets into the cell better, it has better permeability of the two compounds. Two compounds may be equally able to get into the cells. But maybe one of them is simply not engaging the target as effectively.

And then, as I mentioned before, a target may be hitting a compound may be hitting multiple targets, that is to say, its mechanism of action may be polypharmacological, and that, of course, leads to more complexity and in chemistry.

But at the same time you don’t want to be hitting too many targets. Sometimes the early hits you get in a phenotypic program are promiscuous, and that, of course, can lead to toxicities. So as you can imagine sorting all this out can be challenging.

So there is another approach that’s been popular now for about 50 years, ever since the advent of molecular biology. And that’s target-based discovery. And the challenges here are overlapping, of course, because it’s still drug discovery. But they’re a little different.

An essential point for me is the precise target product profile. And that’s because you’re running a program. Typically if it’s target based by looking at molecules in a test tube. Right? And so you’re running a significant risk that you’re measuring something that may not be relevant to what’s actually going on in cells. Because think about it. Proteins and cells. They have multiple functions.

They’re engaging with lots of other partner proteins. Maybe you only want to block one protein interaction, or you only want to block one of the many functions of a particular protein. Or maybe you’re not accounting for any of a large number of other things that may be true in a cell. What’s the posttranslational state of that protein? Is it engaged with? With some kind of a lipid membrane?

All of these things can have an enormous effect on the actual functioning of that target in the cell, and you may be missing all of that in your biochemical screen. Right? So therefore, you have to find relevant chemical starting points, things that will work, not just in a test tube, but also in a cell. And so then that depends on the ability, first of all to run a good test tube assay and then to have the right follow-up assays, cellular assays. You have to have all of this worked out. So you have to be able to somehow find molecules that will disrupt the appropriate function, and that can sometimes be quite challenging.

This is often referred to as the ligandability of a target, some targets. It’s just very hard to ever find chemical matter to begin with, right? And then the other challenge, of course, is, can you run the biochemical screen at all? Sometimes you can’t.

And we were very fortunate at Dewpoint we had Ben Cravatt give a talk not that long ago, and he highlighted some of these critical points. So you can see how challenging this is and finding those relevant chemical starting points also just depends on the library you use. So how large of a library, and what chemical structures, and how drug-like should they be? Now, these are also challenges that we really don’t always understand.

And then, of course, if you’re trying to then optimize your compounds and turn them into drug candidates. You have the usual med chem challenges that we’re all used to: pharmacokinetics, selectivity. Now, I’m an old structure-based drug design guy, I’m a big fan of using X-ray crystallography in cases where you know the target. That’s been very good to me. But you have to be careful, because it’s critical to make sure that the structure you’re looking at, that 3D structure is actually relevant because maybe you crystallize just one domain of the protein right, and if you were to look at that same protein in the cell, it might look completely different.

For all the reasons that I was describing before something as simple as the posttranslational state of a protein can dramatically change its shape and its properties.

The pockets you’re designing into looking at that isolated crystal structure. Those pockets may not be there in the cell. Right? And so all of these things have to be addressed.

Now, here’s an example from my Vertex days that I think highlights some of these challenges. So we wished to identify drugs against influenza, and we had a hypothesis that host factors might be desirable targets. So we ran a phenotypic assay, using a 1,200 compound kinase inhibitor library, and we found single-digit micromolar hits the the molecule on the upper right hand of the slide, which looked to my eyes to be pretty drug-like, you know. Pretty good looking compounds.

And they sure look like kinase inhibitors. And so we assumed that probably we were right, and that we had identified some kind of host kinase. But then, of course, we ran mutagenesis on the virus to find out what the actual target was, and lo and behold! It was actually the cap snatching domain of the viral polymerase, which is a GTPase very closely related to an ATPase, right? So not a shock, but not a kinase.

Right? And so we then, since we knew the target, we converted the project into a structure based target-based project, got crystal structures and optimized the molecule into Pimodivir. The bottom middle of the slide, which you can see, looks pretty close to where we started. So very, very nice example of a screen producing a high-quality starting point. And I think that’s one of the many good lessons here for us.

I think the phenotypic assay it was disease relevant. It found us relevant chemical matter. But our assumption about the target was only half right. It wasn’t a human host kinase. It was a related kind of target, a GTPase from the viral polymerase.

We were able to figure that out because we could cheat right? We could use viral mutation as a way of helping us to discover the target. But normally it’s really quite challenging, as I mentioned before, to do target deconvolution.

It’s worth adding that it’s not required to figure out the target. There’s no law that says you must know the target of your drug.

However, you know it can be useful because it helps to guide your chemical exploration. It gives you a little bit more information to focus on. It may help you to also understand toxicities that you’re seeing. So I think all of these lessons are relevant as we think about the pros and cons of phenotypic versus target-based drug discovery.

And I, you know, I just concluded a long time ago, based on all of these kinds of examples that it’s just… it’s foolish to think about phenotype and structure based or target-based discovery as somehow enemies, a lot of people, you still hear that out in the industry. It’s kind of silly, I think, because they complement each other. And you can actually run a single project like we did with the Vertex flu program using elements of both kinds of approaches. And I think in both cases you need a deep understanding the disease process. And therefore you need a deep target product profile. You need to know what your molecule needs to do in order to be a potentially useful medicine. So therefore, you have to really understand the cellular context of disease, you have to have relevant assays. So if you’re running a phenotypic program, it’s useful to try to figure out the target if you can, because that may help your chemical exploration and may help you understand toxicity.

And, on the other hand, if you’re running a target driven program, you would better get comfortable thinking about the subtleties of the cellular context and think about a diverse range of cellular phenotypes in diverse cell types under diverse conditions.

And I think all of this makes total sense. When you know, when you think about the fact that whether we’re approaching a program through a target based or a phenotypic mindset. We just have to acknowledge that drugs are operating in a horrifically complex cellular environment. That’s just the reality that we must face. And so all of this finally brings us to condensates, which I don’t need to explain to anybody on this zoom call. But for years I had been interested in disordered domains of proteins. Because remember, I said, I’ve always done a lot of X-ray crystallography work, and we always chop off disordered domains of proteins because they make it hard to grow crystals.

And so disorder was a problem, right? But disordered domains are evolutionarily conserved. They’re doing something, although their functions were often mysterious. And so I had already been in industry for quite a while when Tony and Cliff were publishing their first papers.

They’re making their initial observations on condensates so clearly. I mean, you can see the gray hair. I am not a condensate native, like many of you are but I intuitively found this idea really attractive for a bunch of reasons. It helped to rationalize one of the reasons why disordered domains are important.

You know, this idea of localization? Because the way cells can use subcellular localization to achieve the desired rates of biochemical reactions. That was always just a huge mystery to me.

And then also, I was fortunate because my original training in physical organic chemistry I had done thermodynamics. I had done protein simulations, and so I was, I think, kind of naturally wired to be attracted to this idea of cellular condensates. And it I love how it really is a high. The condensate mindset is highly integrative, right? Because it reminds you that biology is just chemistry. It’s just molecules interacting. And that in turn is a function of physics. Right? It’s a function of these sort of soft matter physics

behavior is occurring on the mesoscale in the context of cells. That’s what all of this is about. And so a condensate as a community of proteins and nucleic acids, all interacting in complex ways that really gives us an entirely new way to think about the way to approach drug discovery.

And of course, also because of my experience with CF and influenza, I was already a huge fan of phenotypic screening. So all of this just seemed like a very powerful coming together of diverse viewpoints.

So, as we were forming Dewpoint back in 2018, 2019, naturally, as a budding drug company, we were pondering condensates from the drug hunting perspective.

So we evolved 2 concepts that I think are worth highlighting. A condensateopathy is an aberrant condensate behavior that drives a disease and a c-mod is a molecule hopefully someday, a drug that modulates a condensate behavior somehow to reverse a disease process. And I say, somehow, because a c-mod at an atomic level, at a molecular level can operate in many different ways. Right? They can change the physical properties or the composition or the localization, the dynamics of the condensate. And then, if you go even deeper to look at it, it could be because of a posttranslational modification, or the change in the shape of a protein, or blocking a single protein, protein, interaction.

And all of this could happen either inside the condensate or outside of it. Right? And so a hugely complex set of possible ways to produce c-mods, which again requires us to think very differently. And there are some really quite interesting subtleties about both of these concepts. So in some cases we might hypothesize that the condensatopathy is a downstream event.

Okay, that it follows from it is driven by a diverse range of cellular events that could be caused by mutations or other insults, and in that sense an aberrant condensate might be a kind of central node in disease.

Right? A common feature that occurs late in the disease process. And if that’s true, restoring normal condensate function, then could correct for a wide range of cellular insults.

That’s pretty cool. And then there’s an also an interesting wrinkle to the c-mod idea, because, as I just mentioned, we have to think very broadly about the community of biomolecules that are present in a condensate. So the whole idea of what is the molecular target? We need to think much more broadly than certainly I was trained, and so because it’s a community, a normal c-mod, what it does is it’s restoring normal condensate behavior. But how it does that at the atomic level can take many different forms as I just described. So, in short, a c-mod is affecting the emergent properties of the condensate community.

And so our view of the molecular events that we’re attempting to influence needs to get much broader.

Now, I think a good example of that comes from ALS, helps us to reframe how we think about disease, call out causality, and what constitutes a central node of a disease. Now, as I’m sure many of you know, there are not yet any disease modifying drugs for ALS.

And a major challenge has been the genetic diversity within the patient population. But if we shift the focus from genetics to condensates, it’s interesting that on the bottom you can see that the vast majority, about 97% of ALS patients, they share a common TDP-43 condensatopathy. And this appears to be a central node in this disease. So, regardless of the other factors, such as genetics, environment, injury, age, what have you, that’s an interesting way to reframe our thinking about this disease. So this observation led Dewpoint to a specific condensate hypothesis, which is dissolve the aberrant cytoplasmic condensates, restore the normal nuclear function of TDP-43. Wipe out the toxic gain of function. That comes with the cytosolic TDP-43, and ultimately we hope to restore systemic neuronal health.

And Violeta and the ALS team will have more to say about all of this in the coming months. But you know I’d like to, just at this for this talk, just say that you know this comes back to a point that I’ve already made several times. You want as much confidence as possible in your disease hypothesis. You’re looking for an unbroken chain of causality.

Of translatability, right? Because that supports your thinking from cells to animals, to humans to support the hypothesis that in this case an aberrant condensate is playing that central causal role in the disease and ALS is a great example of this, because the evidence ranges from patient derived motor neurons that demonstrated TDP-43 condensopathy to mouse brain sections, spinal cords from ALS patients increasingly complex models and increasingly relevant to the human disease state right? And this gives us confidence in our translatability. You know that the c-mods that we discover in the cell assays are more likely to have a similar impact on the central node of the disease in the humans.

And, as I said, Violeta and the team will have more to say about this soon. So I think it’s a good example now. But there’s many examples in the literature, and there’s probably 60 or more by now. Many of them come from viruses, and so on the left, we have the RSV example, which demonstrates viral replication, occurs in condensates, small molecules, act as c-mods by hardening the condensates, and that in turn inhibits, the viral replication, and the same process occurs in mice and on the right the COVID example. Similarly it demonstrates that condensates containing the viral N-protein, they play a critical role. Of course, in the life cycle, small molecules act as c-mods by hardening these condensates and a wide range of molecules with a variety of mechanisms of action can affect these condensate behaviors and have antiviral effects.

And there’s, as I said, more than 60 examples now that are all collected on the website. And there’s an initial scraping process which is semi-automated. And then Jill and Diana and others have really put a lot of effort into that manual curation. So I think it’s a great place to go look for examples of papers that are highly relevant to thinking about the role of condensates in drug discovery, and those of you who know Diana, know how rigorous she is, so trust me, these are good papers you should be reading.

Now, another important area of research is resistance to oncology drugs. It’s one of the saddest things is you create a medicine. You actually see that it provides some benefit for a cancer patient. But then gradually the tumors become resistant to that drug.

It turns out there are many recent papers, and just a few of them are shown on this slide that highlight the role of condensates in the development of resistance across multiple tumor types across multiple classes of drugs. So, for example, stress granules seem to play an important role in some of these cases. Paper on the left shows this effect with paclitaxel.

And on the right there’s a very recent paper that demonstrates that you can combine a number of different unapproved medicines, test medicines if you will, that dissolve or prevent the formation of stress granules which were leading to the resistance. And so what that does is, it restores, it rescues the sensitivity of the resistant cancer cells to the original anticancer drug.

That’s pretty encouraging. And I think this is a whole area for research that we should all be paying close attention to. So then a slightly different example. If we think about mechanism of action, one aspect of that is safety right when we discover a c-mod, and we want to know how it works. We’re also making sure, we hope that it’s not doing things we don’t want. So a condensate model can also help us to think about off target toxicity. And of course all drugs have some amount of safety risk associated with them. For example, doxorubicin has been known for a long time to cause cardiac toxicity, but the mechanism has not been well understood, and so in this study it was discovered that what it does is, it induces heterochromatin condensates, and it partitions in both transformed and untransformed cells, including cardiac cells.

And these condensates, they cause a systemic deregulation of the chromatin structure, and they mess with the transcriptional programming. And this was not known before. So this is an encouraging sign that we can use a condensate lens to think about, not just how to optimize drugs, but also to minimize the off-target effects. And so in the future, least in my fevered brain, you know what I imagine is building a set of appropriate, condensate assays that are used at the discovery phase to catch problems like these by looking for changes in condensate behavior among early drug candidates, and I’ll come back to that at the end of the talk.

So now I’d like to bring up another concept that I like. I call it sentinels, and so let’s use this SHP2 example to make the point. So mutant SHP2. What it does, of course, is, it drives hyperactivation of the phosphatase activity of SHP2. And that leads to upregulated ERK signaling. And it turns out this happens through the formation of aberrant condensates.

And this gain of function phenotype, it’s associated with cancer. It’s associated with number of other diseases. There’s actually a bunch of SHP2 inhibitors in the clinic, now for cancer.

And as a structure guy, I was especially excited by this because it turns out that this formation of the SHP2 condensate is also tied to a structurally open, 3 dimensional structure of SHP2, and small molecules that close ship to they act as c-mod dissolvers, and that in turn reduces the downstream ERK hyperactivation.

So that’s a pretty way, pretty cool way of combining the phenotype and the conformational change. They’re completely linked.

And it’s intriguing to think about how this same mechanism could apply potentially to other phosphatases. And that’s actually a point that the authors of this paper make in the discussion section. And I love that suggestion, and you can take it even further. You could imagine this being true for kinases or other enzyme classes.

And so I call this SHP2 example, a sentinel, because since we’re in this new space of condensate research, I think we should always be on the lookout for these harbingers of mechanistic modalities related to condensate, because this then suggests that if we’re working on, if we wanted to go work on another phosphatase, we already have kind of a plan of attack where we could use this result as a way of focusing our investigation of a condensate mechanism for that other phosphatase.

So I think it’s always fruitful to look for these opportunities of information reuse.

Now, another theme that I’ve gotten really excited about is human genetics, and there have been three landmark papers in the last couple of years that really demonstrates strong association between condensates and genetics, and there are additional papers beyond these three. But just to focus on these, you know, Banani et al. They searched publicly available disease-associated mutation databases. They were looking for mutations, that map onto proteins that are likely to be involved in condensation and features of those proteins that are often associated with condensation.

This was based on what is already known about the likelihood of a protein being involved with condensates, and this work involved, took quite a while for them to map it all out because they were thousands of diseases that are potentially going to end up involving some kind of a condensate mechanism. And then Hnisz et al, they focused on rare diseases. They were looking for insertions and deletions leading to frame shifts that lead to poly-Arg. And then poly-Arg, of course, can mess with aberrant localization of proteins because you’re changing the nuclear localization signal, the nucleolar rather localization signal.

And then from Richard Kriwacki’s work, and I think Diana is a co-author, on this paper they did a search of public databases to identify fusion oncogenes associated with cancers, many of those cancers being pediatric, but not all. And the paper really identified a range of other sequence and structural features that may help us to better understand the conditions under which condensates are formed or changes in protein structure and sequence that alter the likelihood of condensate formation.

So if you take this work together, it demonstrates the broad implications of condensate dysregulation across many diseases and gives us a framework for identifying these opportunities and also thinking about how to follow up on them.

So I think that’s very exciting as an area. So it’s clear that a condensate mindset it enables very creative, and I think powerful ways to approach drug discovery, and, as you can imagine, condensates are so different. They’re so complex.

Each drug discovery project is unique. And so the way we apply condensate thinking, it’ll be diverse and bespoke. It’ll depend on the specific project and the challenges that we’re facing that we need to address.

So on the left, perhaps your concern is finding a chemical starting point. So here’s an example of a Dewpoint phenotypic screen coming back to ALS again. It’s looking for changes in the TDP-43 positive cytoplasmic condensates. And, of course, that involves high content high throughput cellular screening, automated segmentation and image analysis. A lot of machine vision, a lot of machine learning. And that enables us to find compounds that dissolve these aberrant TDP-43 condensates, and they may work through diverse mechanisms. Because, again, remember, that’s the whole point. We’re trying to broaden this definition of what we mean by target or in the middle, with those molecules in hand, you might want to explore their mechanism, one aspect of which, as I’ve already mentioned, is off-target effects and selectivity.

So you might wish to know whether a drugs activity is specific to the condensate that’s most relevant to the disease you’re trying to treat. And so in the central panel, we’re looking at comparing a specific c-mod to a more promiscuous c-mod in an assay that profiles.

It’s a panel of condensate markers and high content images like these, coupled with the appropriate machine learning can give you deep insights into the specificity of each and every c-mod that you discover. Or on the right, perhaps your challenge is, you’ve already got a good molecule, and now you’re trying to turn it into a drug candidate. You’re trying to optimize it. So this right hand panel shows a fluorescent c-mod that partitions into condensates, and the partitioning could be studied either in a cellular context or using reconstituted droplets.

However, having said all of that, I’m reminded of how challenging it is to bring fresh thinking into a complex process. So if you want to sup-up your car and you want to get more performance out of it, you might think: Well, let me just add a whole another exhaust system to the car. You might just like weld on a second muffler. But if you do stuff like that, you’ll get maybe a 1% increase in the performance of your car.

You can’t just bolt on new parts to a complex process or a complex machine. That’s just not how it works. And I think that’s a good analogy, because studying condensates is, you know, I don’t have to tell anybody in the room or on the call that it’s really hard. Okay, it’s a, these are subtle multifaceted processes. And what I’ve been trying to do in this talk is demonstrate examples where, if you can go deep, if you can think critically about condensates, about the assays, the functions of the aberrant condensates the mechanisms of action the way the molecules are working, if you do all of that, and if you’re supporting it all with appropriate machine learning, meaning, you have lots of good data, and you’re pointing the machine learning at the right problems, because that’s the only way machine learning works, if you do all of that, it can be incredibly powerful. But you can’t take a traditional drug discovery mindset, and just say, Okay, let’s hire one or 2 condensate experts and then expect magic will happen. That’s just not going to work, because this is an entirely different way, I think, for thinking about drug discovery.

So it’s not a casual approach. It’s not a bolt on. But I think the examples that I’ve been showing, I think they suggest that a deep awareness of condensates offers a pretty powerful new approach to drug discovery. The first step is to always be looking for causal relationships between a condensate and a disease, what we call condensetopathies.

I’ve shown a few examples. I’ve highlighted the genetic work that suggests thousands of other potential opportunities, and the sentinel cases like SHP2. I think they implicate a condensate mechanism in a way that suggests other members of the same gene family might likewise employ similar strategies. And so you can go looking for those. Now, of course, there are still lots of questions, lots of things we don’t understand yet about condensate drug discovery.

So how best to repair an aberrant condensate. There’s obviously different ways of interfering with them. I, you know, often return to this really nice figure from Simon Alberti’s 2017 review. You know, are some of these approaches intrinsically better than others, or is in each situation is it going to be a different approach? That’s really optimal. I don’t think we know, is one approach going to produce a more useful medicine than another approach, and then, secondly, condensate modifying drugs. What we call c-mods. It’s a new concept. And so every new drug, we have to be asking as many questions as we can about the mechanisms of action targets. They hit subcellular localization, kinetic selectivity, etc., etc. because learning more about how c-mods work informs us for the next project and gives us additional tools and an understanding of how to do this effectively. And then, of course, the optimization of how we deliver drugs to specific condensates that in that in itself is its own challenge and and not a small one. So these are all things that we need to keep working on. And then, looking ahead, I think that it’s also true that condensate behaviors are going to play a huge role in how we approach clinical trials.

I think first of all, we should be able to use condensate response ex vivo in relevant cells from patients to select those patients for the clinical trials, and also to provide additional biomarkers of response which it should make clinical trial enrollment and response rate higher and make the trials faster. So that’s very powerful. And then, more broadly, the point that I made before. I can imagine that there will be some kind of broad, condensate fingerprint that we want to generate. It’ll be a very informative test by which we evaluate all new drug candidates. Right now, we carry out routine toxicology testing in animals, and we take a molecule and we profile it in a panel of receptors, typically 100 or a couple of 100 receptors. Those are things we do now routinely on every drug candidate. All of us in the world do this. So I don’t see any reason why we couldn’t also evaluate any new potential drug candidate by asking, what effect does it have on 20 different condensates across a hundred different cell types and make that a routine part of the drug evaluation process.

I think that’s something that we may see in the future.

So I hope this has been interesting for at least some of you. I’ve tried to provide a high-level, personal perspective on how the things we’re learning about condensates can and should alter the ways that we think about how to tackle these many and diverse challenges in drug discovery. So obviously, I’m looking forward to seeing what the coming years will teach us about how we can put these concepts into practice and create better medicines.

Thank you.

Diana Mitrea:
Thank you so much, Mark, for this insightful talk, and for leaving us with more questions and ideas for innovation.

So please, if you have questions, type them in the chat, and I’ll pass this on to Bede to field the questions.

Any questions here?

Q&A

Christina Curran:

When you were talking about condensate role in RSV, you mentioned that the condensates harden. What did you mean like that is, is that like a form of aggregation or something else?

Mark Murcko:
Yeah, they basically, they’re less liquid. They’re they hang around longer. That’s the that’s the non-technical answer.

Diana Mitrea:
Okay, we have a question from the chat. Gonzalo, do you want to unmute and answer your question.

Gonzalo Prat Gay:
Oh, thanks for your excellent talk! Actually a lecture, including all your vast experience along these years. It’s just by consent I pick up the point of the lady that did the first question about our RSV, which is actually probably is one of the few examples with a biological output for a drug candidate against condensates. And this was the candidate is Cyclopamine, which is a general individual of a hedgehod pathway which involves a signal various single pathways. Now this effect was shown like in 2016, and since then, even though the 4 to 5 proteins that built up the condensate from the viral side, they found no targeting or no evidence on how this works. So this is a general drug with a general wide effect, affecting and hurting a condensate in the virus, which is true, but it seems that it could be some sort of indirect affect, since you don’t know exactly what is targeting in the virus. So there’s so many things in the middle that can happen. So it gets to a very important issue, because this is a very strong example on on how a drug candidate might work on a with a biological outcome which is actually viral antiviral effect in and even in a whole being, in a mouse.

Mark Murcko:
It’s a good question. And it highlights the fact as I was describing in the talk, how challenging it can be to actually figure out the mechanism of action. In the first place, I think that’s right. I do think in the case of RSV, I think some of those, there was a complex, polycyclic, natural product hedgehog inhibitors that I think it was. It’s Ralf’s work right? He’s on. Oh, yeah. So I think I think there actually is some understanding of the mechanism of how those compounds work. I don’t think it’s through the hedgehog signaling. I think it’s through a different target.

Bede Portz:

Yeah, the nice thing about these talks being public and well attended is that there’s a lot of external expertise. So, Ralf, do you want to unmute and comment?

Ralf Altmeyer:
Indeed we have. There is a very specific mechanism. It’s not hedgehog related. So that is very clear, but it’s not that diffuse, vague kind of interaction. It’s a very precise action on in increasing the interaction strength between components of the transcription machinery and that leads to an inability of the RNA to be transported from the periphery to the ibag a sub condensate. And as a consequence, we have a liquid to solid phase transition. So a pocket. A small molecule very, very precise action, and the consequence is the physical. The painting of the physical properties meaning the viscoelasticity change leading to coincident hardening. So it is very precisely, probably is a natural product of 36 nanomolar hit, which is being improved, has been improved. So there is a not in common with traditional drug discovery based on molecular targets, and it translates extremely well and very precisely to the in vivo situation in the mouse.

Mark Murcko:
That’s a great example of the target deconvolution working in practice. So thanks for that, Ralph. So then it becomes you’ve got a clear target. You have a clear mechanism, and then it becomes it switches back to a chemical problem of then optimizing the molecule and then putting that into clinical trials and seeing how well it works. But now there’s a clear hypothesis. So, thanks for that.

Bede Portz:

Allysa, you’d like to unmute and ask a question.

Allysa Kemraj:
Hi, thank you so much. Absolutely incredible talk. I previously worked in clinical research. And I now work on the molecular level, like in test tubes. And for people like us who are like entrenched in this field, we understand the pipeline from like molecular level to clinical research. But do you have any recommendations on communicating to patients the relevance and translatability of studying condensates, of studying things on a very physical level?

Mark Murcko:
First of all, I’m really intrigued by the idea of you moving from clinical back to research. I think that’s incredibly valuable because I think that research teams sometimes don’t have as much awareness of the clinical side of things as they need to. And so I think it’s I’m sure that your clinical experience is incredibly helpful to the research folks that you’re working with now. So that’s a pretty cool thing right there and then to answer your question. I think that there are now a lot of people trying to figure this out. I don’t think there’s an answer yet, but you know, as I said it in my very last slide, I do think that in the future we will be able to use condensate response in patient cells as one way at least in some clinical settings has one way of selecting patients for the trials. And so obviously, we’ll need to prove that that works. We’ll need to demonstrate that that actually does increase the likelihood of a response in that subset of patients that have the appropriate condensate response. So that’s work to be done. But I think that’s the direction it has to go. So I don’t think it’s worked out yet, but that’s how I see it evolving.

Allysa Kemraj:
Absolutely thank you. And I actually worked on the clinical trial for Trikafta for CF patients. So I was super excited when you mentioned that. Thanks.

Mark Murcko:
Thanks for that. I mean, that’s been such a such a rewarding thing to see just how well those drugs have been helping those folks.

Jian Guo Ren:
Okay, thanks, Mark, for the wonderful talk. Actually, we have a very similar idea. I like the phenotype screening. My question is that this phenotype screening for drug development is really a very beautiful platform. However, I saw you also think about how to go back to the traditional targeted for the deconvolution. So my question is that, should that zoom into this is a beautiful with drug screening platform. More important, say, is, should we more focus on, identify the disease-related phenotype. I’m not sure. I say the clearly, for my question is that because somehow, I think, is that if we go back to the traditional target identify which factors is more important for our screening. But somehow I I single that’s not for me. It’s not important.

Mark Murcko:
Well, I think that’s a really good point. That’s kind of what I was trying to straddle this in my talk, because I see advantages to knowing about the targets. But I also see that potentially as a trap, because if you’re if you think about a condensate as a fundamentally different kind of beast, as a different kind of, let’s call it the target, is the condensate rather than any individual protein or nucleic acid in the condensate. Right? If you yeah, reduce it. If you go back to that reductionist mindset of only thinking about one target in isolation, I think you may completely miss. That’s the risk. Right? Is that you miss the magic of this downstream central node idea of what a condensate does in a disease process. And so I’m a fan of trying to figure out what we can learn about the target or targets that our c-mods are hitting. But I don’t think it’s a good idea to go back to a target centric approach. Do you see what I mean?

Jian Guo Ren:
Yeah, I actually, I understand what you mean.

Mark Murcko:

I’m trying to sound like I’m not waffling, because I’m not waffling. I’m saying that there’s a value in having a little bit of both in your thinking. But you can’t ignore the fact that a condensate is just a fundamentally different thing. And a c-mod will do things that we don’t normally think of drugs as doing. Quite, quite likely that will be true in many cases. And so that’s the magic that we have to be careful to not lose.

Jian Guo Ren:
Yeah, I agree with you. I think we should find a way to identify the disease-related phenotype, for example, maybe Myc, maybe beta catenin, maybe it’s like some just combination. We don’t know. But that’s the most important thing for you.

Mark Murcko:

Exactly. And you know, just to elaborate a tiny bit more. So when at Vertex we were working on cystic fibrosis, we did not understand at a molecular level how our molecules worked. We knew they were somehow engaging CFTR, but we didn’t know precisely where or whether there were other proteins involved, and we never had animal pharmacology models. We just had cellular models, and we went directly from those cellular models directly into patients, because the cellular models we understood at a physiological level, at a pharmacological level, we knew what those molecules were doing, and it mapped to the disease. 

Sidharth Sirdeshmukh:
Alright. That was an awesome talk and thanks, as always. So you’ve demonstrated structural approaches like the original ones, right? Moved into molecular dynamics and then into condensates. So kind of like at a higher level. How do you maintain this flexibility and openness throughout your career? Like to your thinking? And how do you evaluate emerging biological, informational spaces.

Mark Murcko:
How do I evaluate emerging biological spaces? Badly? I think this stuff is really hard. So you know, I think there’s not a lot of difference, I don’t think, between you know, sort of what I’ve seen over the course of my career is what a lot of people have. I mean, there’s new information coming out all the time. And you just you react to it. You constantly hopefully update. I said, evolve. I hope it’s evolution thinking that’s what you have to do, you know? So I certainly was not thinking about condensates 10 year to say 15 years ago, right? But it fits perfectly into that idea that cells are highly complex entities. They only work by massively complex interactions between all kinds of biomolecules. The only way to really understand that is not through an isolated protein in a test tube.

Diana Mitrea:
Thank you very much, Mark, and thank you, all of you who joined us for the talk today. Be on the lookout for the recording on Condensates.com, and we’ll see you next time. Thank you.