VIDEO: Jonathan O’Connell on High-Throughput Screening in Drug Discovery
| Author |  Director, In Vitro Pharmacology & HTS, Dewpoint Therapeutics |
|---|---|
| Type | Kitchen Table TalkFor Beginners |
| Topics | |
| Tags | |
| Share |
The Dewpoint scientists hosted a stimulating discussion about drug discovery for the 2023 class of interns and the condensates community on July 12, as part of the Kitchen Table Talk series. The speaker was my long-time mentor and manager Jonathan O’Connell, who is currently the VP & Head of Drug Discovery at Dewpoint. Jonathan has over 25 years of experience building teams and leveraging automation and technology to advance the science of drug discovery. Prior to Dewpoint, Jonathan built and lead teams at Valo Health, Forma Therapeutics, and Bristol-Myers Squibb. He started his drug discovery career at GSK after receiving a BS in biochemistry from the University of Manchester in England, and a PhD in biochemistry from the University of Glasgow in Scotland.
Jonathan has been an incredible mentor to me and many others. We hit it off from the very beginning, have similar backgrounds and interests in high-throughput screening (HTS), and work well together. In the video below, he shares many insights about the industry standards of HTS from his 25+ years in the drug discovery field.
Click here to catch up on the engaging talk and discussion below.
TRANSCRIPT
Kingsley Appiah
Hello. It’s my pleasure to introduce Jonathan O’Connell for today’s Kitchen Table Talk. Jonathan has spent greater than twenty-five years building teams and leveraging automation and technology to advance the science of drug discovery. He’s currently the Head of Discovery Platform at Dewpoint Therapeutics. In this capacity, he oversees the high-throughput screening, in vitro pharmacology, compound management, lab informatics, and lab operations in our Boston and in Dresden sites.
Prior to Dewpoint, he was the VP, and Head of Early Drug Discovery and Technology at Valo Health and Forma Therapeutics, where his role in addition to that aforementioned included admin, and early discovery chemistry, analytical sciences, and automated synthesis. Prior to joining in Forma Therapeutics, Jonathan spent several years at Bristol-Myers Squibb where he held positions of increasing responsibility, cumulating in the building and directing BMS’s centralized HTS laboratory to support lead discovery, lead optimization, in vitro admin across multiple therapeutic areas. Jonathan started career at GSK, where his role involved assay design to support high-throughput screening and lab automation. Jonathan is also a fellow of a Society of Lab Automation and Screening and currently a member of their board of directors.
During his more than twenty-five-year journey in the industry, he has developed careers and mentored several scientists along the way. I have personally benefited from his mentorship for a greater part of my career in the industry. He was my direct manager for thirteen and a half years at Bristol-Myers Squibb, and later at Valo Health and now at Dewpoint Therapeutics. So you can just imagine how the affinity that I have I have for him is… We hit it off my very first time that I talked to him on the phone when I was interviewing for Bristol-Myers Squibb. I was moving from Boston to Connecticut, and just the voice got me. And then, so I talked to him, I realized, we were thinking about things the same way. The background was very similar. So at that time, I had a job with Pfizer and AstraZeneca here in Boston. And then when he gave me the job, I just jumped at it, and then moved to Connecticut before coming back here after sixteen and a half years. So just imagine the relationship there.
Okay. Jonathan received his BS in biochemistry the University of Manchester in England, and a PhD in biochemistry at the University of Glasgow in Scotland, in the laboratory of Professor Miles Houslay, studying enzymology and cellular localization of type four phosodiesterase. He’s here today to share his thoughts about high-throughput screening in drug discovery. So I’m very excited to be the one introducing him today. So Jonathan, you’re welcome to the Kitchen Table Talk.
Jonathan O’Connell
Thank you, Kingsley, for that. Thank you, Kingsley. A very kind introduction. Really appreciate it. And good morning to everyone here in Boston. Good morning, good afternoon to everybody online depending on where you are. And you know, so as Kingsley alluded to, I’m gonna really be talking today about high-throughput screening in drug discovery. And it’s very difficult. This is a big topic. It’s very difficult to get into this in in any great detail in fairly short talk today. So bear in mind that many of the topics I cover are indeed whole talks on their own. So we will just be scratching the surface, but hopefully giving you some good context as we go through this, and I hope you’ll all enjoy it…
So I wanted to start here. This is kind of an old slide but I think it gets the point across quite nicely. So here is the structure of HIV protease. And within it, the active site. You can actually see Saquinavir from Roche binding exquisitely in the active site of that protein. It binds with high affinity, it binds very specifically, and that’s what makes a good drug molecule. And so It looks quite simple when you see it like this. So why can’t we all do that? So where do these drug molecules come from? How do we actually find drug molecules which bind so beautifully into active sites of proteins, and then can be turned around and developed into drugs.
So there are a few approaches to it. And, you know, I’ve kinda tried to bucket these into some very broad categories. I mean, this is oversimplifying I admit, but I hope it’ll get the point across. So it’s a fast follow on approach, and it’s I I kinda see this as kind of maybe cheating a little. It’s where somebody else has already identified a molecule and they’ve published on it or or they’ve patented it. And so you can actually leverage that information that’s in the literature and then design your own molecule, leveraging the information that’s already out there. And then hopefully, you can find stuff that gets you into a different patent space, and actually you can go ahead and then develop drug molecules by that approach.
It works very well, actually. It’s very successful, but it is high risk. And it’s high risk because that information’s out there. Everybody has the same information. And so therefore, you know, you may be going down a very sort of strong path and putting a lot of resources and effort into it, and another company may come along and just publish on the exact same information. They found same information as you, and if they patent it, you’re dead in the water. So while it works, it’s high risk.
The structure-based design, which usually requires some sort of knowledge of the crystal structure of the protein so you can do it. So you can actually design molecules that fit into that active site. And when you know what the three-dimensional pocket looks like and you know the the amino acid residues that present themselves in that pocket, you can design molecules to fit. This is a little harder. It does work. But actually, this is really used more for later stage drug discovery when you have a start point. And you can then actually visualize that molecule in the active site of molecules through a crystal structure, and then use that information to enhance your affinity and to be able to move your molecule forward.
But the third approach on this slide is what we’re gonna be talking about today, which is high-throughput screening. And and this is basically a random search. And the idea is to throw as many molecules as you can at it and see what sticks essentially. And that’s what really defines HTS. It relies on serendipity. The beauty of it is completely unbiased. So no prior knowledge going in, you don’t you’re not saying, I want these sorts of molecules or these sorts of molecules, you’re basically going with as diverse a set of molecules as you possibly can to really see what which one works. And you need to test as many as possible for this diversity approach. So this random search that we do through HTS is really gonna be the focus of of today’s talk.
So if we go back to the slide that we had originally, you know, really looking at that. So here trying to illustrate that we are just gonna throw as many different molecules as we possibly can and test as many different molecules as we can. We don’t really throw it. We test them to really see what what fits. And and hopefully some of those will come out as hits. But I wanna just, I’m gonna mention this a few times during the talk: HTS is a process to find a starting point to develop a drug. It is not a process to discover a drug. That would be kinda remarkable if you got that right out of an HTS campaign. So the idea is, you know, you find something that does bind. Maybe the affinity is not great. The potency isn’t great, but then we can leverage a number of approaches as we refine those from a chemistry perspective, modify the molecules, and increase that potency over time.
So as I was just alluding to, the concept of HTS is to really test as many molecules as you possibly can. So in big pharma, that’s millions of compounds. You know, one or two is typical. In biotech such as Dewpoint, you’re typically in the hundreds thousands of molecules for your HTS campaign. You can also take a DNA-encoded library approach. We’re not gonna be talking about that today because we can’t quite squeeze that into the time we have. But in DNA-encoded libraries, you’re testing not not hundreds of thousands, not millions, but billions of molecules for activity against your target.
And and the whole idea is that we basically you know, we really need to, because of the scale that we’re talking about, is we leverage automation and cutting-edge-type technology. To really facilitate that entire process. That enables the millions of tests, enables us to shrink down the assay volumes to be really small and much more efficient. Increases the throughput and really reduces that overall cost as I was saying. You do need a highly advanced skill set as part of your team to be able to develop these assays.
This is not a simple process, and there’s a lot of factors to be taken into account. So a typical HTS teams such as we have here a Dewpoint, we we have a number of really skilled assay developers to actually build these assays and to actually be able to run these high-throughput screens. But essentially because of the scale, you’re testing these hundreds of thousands of molecules to look for hits. The whole point is you’re really trying to look for that needle in a haystack. It’s like needle in a haystack, which is hopefully gonna be your the lead molecule which becomes a drug in the future.
So I wanna spend a little bit of time–I know it’s a boring slide with just a few bullets on it–but I wanna spend a little bit of time on this point on this slide because there’s a few important points on it. So in HTS, each compound is tested once. That’s just once. One replicate, one concentration, once. So we don’t typically do replicates. And the reason is if you’re doing a million compounds, that’s essentially a million wells or a million tubes, however you’re doing the assay. So if you wanna do duplicates, that’s two million. Right? And that’s that’s a big difference.
And, you know, basically the cost of an HTS campaign is, you know, it varies according to the assays, but somewhere in the region of fifty thousand dollars to a hundred thousand dollars in reagents. That’s to do it with one replicate. So if you do duplicates, you double that cost. So it becomes very expensive very quickly. The process, this is the most important point I wanna I wanna get across on this slide. The process is designed. So there is more benefit from testing more compounds, than there is for getting a very accurate read on an individual compound.
So by that I mean, if one compound with that one replicate is a little off and you miss it, well you’ve missed it. But because you’ve not done duplicates, you’ve tested so many more compounds, you actually find so many more hits, which have greater diversity, greater span of chemical space. And so therefore, you actually find a better starting point. So it’s better to test more compounds than it is to really home in on any individual one in that primary HTS campaign. And we’re gonna come back to that point as we go through the talk and we’ll talk about the rest of the process.
Okay. So this is a typical HTS process, which is kind of laid out in this sort of funnel diagram here. So I’ve already referred to the primary HTS campaign. That’s where we do the one replicate. That’s typically at ten micromolar. That’s a good start concentration for most libraries. And, you know, again, in that hundreds of thousands to millions of compounds being tested at that time. So when we have the hits from that primary campaign, and I’ll show you how we select these on the next few slides, you basically need to then go and confirm they’re really active because you remember you only did one replicate. Everybody in science knows you do replicates.
So we then go ahead and we do the replicates to confirm that activity. That’s what we call a retest process. So typically then we’re gonna do triplicates. And we may do it in multiple assays. So we may have a null assay to look for interfering compounds to make sure there’s no interference and they’re not false positives through interference. We may look for selectivity. Say you’re doing a kinase and wanna make sure you don’t hit this other kinase at the same time. We can do selectivity at this same time on those retests. And that allows you to select the best possible molecules to move into next stage of your process, which is CRCs.
Otherwise, known as concentration response curves or IC50s, EC50s, whichever works for you, but it’s like concentration response curve is the next step in the process. And there we were doing triplicates typically with ten point curves, and again in multiple assays to really give us a greater breadth of information around those compounds so that we know which ones we want to select to to be able to move forward in the process.
So once we’ve had that and we now have a real potency for those individual compounds in that CRC process, we move on to the confirmation stage. So that’s where we, you know, typically we’re gonna re-synthesize the compound. So obviously we tested it. It was solubilized. It’s been in solution. It may have gone off. It may have degraded. So we wanna make absolutely sure that that hit that we have is really what we think it is. So we will remake those compounds and retest them.
And once you get through this process, that’s really when you can say you’ve got a lead. You’ve proved it’s active. It’s got the profile, the selectivity profile, all the other parameters you’re interested in for a very early lead, for a drug discovery program. And then you really start focusing on that lead, and our chemistry colleagues really start to get to work on it, making modifications, and really trying to enhance the properties of that molecule. To turn it into a drug.
Guanhua He
Can I ask a question? I’m really surprised I’m very naive about high-throughput screening, but I’m surprised that you don’t worry about false negatives. Then you prioritize, you know, avoiding false positive because it seems to me that if you have false negatives, then you miss the gold in the million compound, then it’s more serious. And then if you have false positive, you can always filter them out in later replicates.
Jonathan O’Connell
Yes. You’re you’re absolutely right. And we don’t worry about false positives. False positives are actually, you know, something that we weed out later in the process. You’re absolutely absolutely right. So what we try to do when we design these assays is to minimize the potential of false negatives as much as we possibly can. You know, but you are always gonna have some degree of false negatives because every assay has variation. And the point is that we do worry about it, we try to minimize them. We try to cast the net as wide as we possibly can, but you can never eliminate them completely.
But you know, if you miss one compound, but can test more wells and find another ten, then it’s okay to miss one in order to find ten different ones. That’s really the point I was trying to get across. Right? So You could really focus in on making sure you always catch that one, but then you’re gonna miss the other ten because you can’t afford to do that number of wells in order to test it. So I hope that makes sense.
Guanhua He
Yeah. That makes sense. Thank you.
Jonathan O’Connell
Okay. So I’m gonna talk about the data you get out of an HTS campaign a little and basically how we select hits and move forward. So here’s what we call a scatter plot. This was an HTS. It was an enzyme inhibition assay. And what you can see here is that on the on the y axis, in this case is percent inhibition, and on the x axis, it’s just compound number. So that’s just basically the compounds spread across, so all the compounds tested spread across the x axis. So most of the compounds are inactive. You can see those in blue. And the ones we’re interested in that actually inhibit this particular enzyme are those that are colored in red. So hopefully, it’s one of those red colored compounds on this, that is the hit, that is that needle in the haystack that you were looking for. So that’s essentially, you know, the hits that come out of the HTS campaign.
When you actually look at literature and you look at how this has been done many times in the past, people talk about using three standard deviations as a guide. And you see histograms parted such as this which really show your distribution of your activity. Personally, the way that I try to do things, I find this very unhelpful because basically, when you when you look at this, all of your compounds are here, because that’s most of your hundreds of thousand or two million or so compounds. The compounds you’re interested in are out here, and you can’t see them because of the scale of what you’re working with.
So that’s why we go with this scatter plot approach, which is much more useful because it basically smears them across the x axis and allows you to focus in on the area such as this that you’re actually interested in. And so let’s talk a little bit more about the three standard deviation thing. So this is real data out of real HTS. It’s a little older data, but I find it gets this point across really well. So, you know, you’re really looking for something that stands out. Right? You you can’t retest everything because again, you’re screening hundreds of thousands here.
So you’ve got to pick what you want to move forward. Anything that you move forward, you have to cherry pick, you have to go find that compound, you have to pick it, you have to put it in a separate tube and do a separate assay. There’s a lot of work in the downstream processes. So we gotta be careful about not saturating those downstream processes and really thinking about how we select those hits. So a good literature guide is anything outside three standard deviations is a good guide, and it’s a great guide to get started.
However, I did pick this dataset for a reason. There’s a strong negative bias in this particular assay. You see a lot of negative percent inhibition, which could be considered activation. But in this case, it’s actually fluorescent interference in this particular assay. So some compounds are intrinsically fluorescent, and they create interference in this assay. So you end up with a number of compounds that are sort of on the negative side of this distribution. And what that does is push out your three standard deviation lines because of the increased variation there.
So in actual fact, when we look at this data a different way, so here’s the three standard deviation line from the the previous chart, and I’ve just cut off the negative data to zoom in and make it easier to see. Now you can actually, by a really scientific method, which is known as dropping the threshold by eye, we lower that threshold, and we actually bring it down a little bit. To see that, you know, where where we can bring it to be outside the noise where most of the hits are, but still bring some, you know, very interesting compounds forward.
So you see if we use three standard deviation, in this particular case, there’s 360 hits on this assay, just dropping it like six or seven percent, doubles the number of hits, and we’re able to take a lot more chemical matter through to the next. And we can play with this line. You can go deeper in the noise. You can make it more selective by pushing it up. And it really depends on what we’re looking for, what those compounds look like, and there’s a lot more information goes into this, and I’m kind of alluding to here. But but this is kind of a, you know, a good illustration of that.
Okay. So I’m gonna switch gears now and talk a little bit about automation, the role of automation in high-throughput screening. So why do we automate? Why do we use robots and and high-tech equipment? I mean, it’s pretty simple really. We’re doing hundreds of thousands millions of tests, right? And that’s pretty monotonous and a lot of work to be doing individually. So if you’re doing a million compounds, if this was done in, you know, eppendorf tubes, that’d be a million eppendorf tubes. So obviously, we’re using plates and we’re compressing the assays down.
And then we’re, you know, basically using a lot a lot of automated instruments to to run the assay. So, you know, a million compounds is lot of steps. Right? If it’s an enzyme based assay, it’s a million compound additions, a million enzyme additions, a million substrate additions, it goes on. Right? You get the point. So the automation really eliminates that repetitive nature of a scientist’s work. It improves consistency because all plates get treated exactly the same.
Great thing about robots: they don’t have families. So they stay here all night, and they stay here all weekend. And they keep working. Right? So automation really helps helps facilitate, you know, the extended work week. And so a robot is actually less efficient than any of the scientists here in an eight hour period or in an hour period. But in a twenty four hour period, it is much more efficient. Right? And so that’s the key to remember. A lot of people say, I can go faster than the robot.”I say, “Yeah, but you need to eat and go home and see your family and all that stuff.” The robot doesn’t. Right?
So that’s where the automation really comes in. But most importantly, and all joking aside, the really important factor is that it frees up scientists that do science, not repetitive nature work. Right? So we’ve got a phenomenal team here at Dewpont with some really talented individuals. And if they’re sitting there all day pipetting like this, I’m not really maximizing their education or their contribution to our drug discovery programs. So the robot can do that repetitive work. Scientists can do high-level science. They can be analyzing data. They can be designing new assays. They can do the mechanistic work to follow-up on their hits, and and doing that higher work while the robot does all the grunt work for us.
So very quickly on this slide, talking about plates because I think most people are aware of this, but I still think this is a a useful slide to talk about. So back in the ’80s, I think everybody’s familiar with the 96-well plate. Most assays were done in this. So 96 wells in a plate, assay volumes were sort of 50-200 microliter range. By the time we got to the sort of the the mid-1990s to to the very early 2000s, we’d evolved to the 384-well plate, which those of you that are good at math will be on this, but four times 96 is 384. And really, what we’ve done is basically shrank down the the volume. So you’re cramming more wells into the same footprint. So now we’re down to 20-50 microliters an assay.
But with the scale of libraries and the number of compounds that we were testing, it was really becoming still very expensive in a 384-well plate. So the industry evolved then to the 1536-well plate, which is four times 384. And so now we have really tiny wells with typical assay volumes in the two to ten microliter. So usually you’re doing a three to five microliter assay in these plates. And that really just significantly reduces the cost and everything else of the operation because you’re putting so much less volume in each of these wells, but you do require highly specialized instruments in order to be able to measure effect, pipette into these wells, and to be able to run your assay in this format.
And one of the key challenges to miniaturization, initially in the industry, was actually adding compound. So we didn’t really talk about this yet, but we talked about the fact that we have hundreds of thousands of compounds to millions of compounds in the library. They have to be dispensable to your assay. An HTS library is by its nature diverse. That’s what we want. We want it to be diverse, lots of chemical structures. So it’s very diverse. That also means your compounds have very diverse physiochemical properties. One of those is solubility. And so we have to use something that’s gonna try and get all those compounds into solution.
And the universal solvent that has been used since day one is DMSO. Trust me, we’ve tried many different alternatives. We still use DMSO because it’s the best. So DMSO is great as a solvent, It’s horrendous for an assay. And so your typical tolerance maxes is out at about one percent on volume. Right? So in terms of your assay. So that means, you know, if you’ve got a three microliter assay, you can only have thirty nanoliters of compound. And it’s important to add it in at hundred percent DMSO so because if you dilute that in aqueous first, often your compounds will crash out of solution. So we like to go in with a hundred percent DMSO.
So we needed a technology that would enable us to be able to dispense in a hundred percent DMSO in those nanoliter range volumes. And the solution to that was developed around about 2003. And that was the Echo from Labcyte, which is now owned by Beckman. And the Echo enables us to dispense down to two and a half nanoliters using acoustics. So all will become clear. So I actually have a a quick video here, which I’ll play most to the way through.
So this is a commercial video from Labcyte. So here’s your compound plate going into it. Typically, your compounds are not colored. There may be a couple of colors in there. They don’t always look like that. So your compound plate goes into the Echo, and then your assay plate that you’re gonna dispense into goes into that machine–into the Echo, and it goes in upside down above it. So what the echo does is basically send out a pulse of ultrasound into every well. What it’s looking for is the meniscus. So it gets a reflection from the bottom of the plate, from the bottom of the well, and then it gets a reflection of that ultrasound from the meniscus. So it uses that to basically figure out where that meniscus is.
So it doesn’t matter if you have different volumes in every well in your source plate because it basically scans the whole plate. And then, as you can see here in this illustration, looks for for where that meniscus is. So we basically will run you can see the the difference in time there in milliseconds for that reflection to happen based on the the different depths of that that fluid volume. So it it so it’s it’s really, really essential that it can find that meniscus very accurately. So what the the instrument then does is is basically scan the whole plate. And you can see on this illustration here that the transducer is actually coupled to the bottom of the plate with water, which allows very you know, a very efficient transfer of those sound waves at the bottom of the plate.
So once it’s done, it goes back to the first well it’s gonna dispense. And now it focuses a beam of ultrasound at the meniscus, and that pinches off a two and a half nanoliter droplet, which flies straight up through the air, and it goes into your destination plate and sticks. So if you try pipetting by hand two and a half nanoliters, it wouldn’t work because surface tension makes that tiny, tiny, tiny droplets stick to your pipette tip. Now surface tension is your friend because it makes it stick in the plate. So it only fires two and a half nanoliter droplets, and if you want 25 nanoliters, you fire ten droplets. So here you can see the Echo actually dispensing different volumes to different wells and allows you to do that. So this is how we add compounds to our assays in HTS typically. There are other ways of doing it, but across the industry, this is by far the heaviest used instrument in order to be able to to dispense these these tiny volumes.
I was lucky enough to be in involved in this right from day one. So at Bristol Myers, at the time when I was there, we actually beat a partner with Labcyte on the development of the instrument. So we had the instruments before they were out in the market, and we were working on the final stage testing and refinement. So remember that you know, this was around about 2003. So technology wasn’t quite what it is today.
So if you forgive this video, this is an old video. And this is actually using flash photography. And here you can actually see at this time, it was actually a five nanometer drop that being pinched off. You see the injection pool which forms down here and then the droplet sort of pinches off and goes straight up through the air. This is a real droplet being formed in the Echo and starting to shoot upwards. So it almost looks like a backwards video of a droplet hitting water, but this is how the ultrasound works, pinches off that droplet and and moves it forward.
And and on this side, you can see droplets being formed where my mouse is. So this is 50 nanoliters being made up of 10 five nanoliter droplets, and you can see these sort of droplets appear to sort of grow in the center of the well. This is 384-well plates you can’t really see it in a 1536. These are old videos, but it just really helps to illustrate it. So I thought I would include those here.
So here’s the Echo, and it’s actually on one of our robots here at Dewpoint. So so this is Roberto in action. And this is our newest robot that’s put together for us by HighRes at Dewpoint. And this is the robot that runs at Dewpoint all of our phenotypic, cellular, high-content assays to look for condensates and condensate-modifying drugs within cells. So we fully automate the entire process and I’m gonna have some slides on that later on.
Here you can see the robot is basically moving a 1536 plate around the robot. It’s putting that plate in all of the instruments that you would use if you were doing this at the bench, It’s just doing it for you. It treats every plate identically, incredible consistency, and and, you know, we pretty much run these things you know, twenty four hours a day most of the time. So very, very efficient way of working. And in the background, actually, you can see there’s a second robot here. We have a few of these at Dewpoint to help drive our high-throughput screens.
So this is how we get the hundreds of thousands of tests on in high-throughput screening. It’s by automating these assays, and you see a plate just disappeared into the Echo there. So this is, we just have a camera above the system. This is just a bird’s eye view of the system and showing this running.
Okay. So I’m gonna move on and talk a little bit now about assays. So there are multiple assay formats that you can use for high-throughput screening. So you know, on the biochemical side, you know, I mean, pretty much any assay that you consider has now been adaptable to high-throughput screening. But they’re typically fluorescent and luminescent technologies You know, ideally using time-resolved fluorescence because that helps to eliminate interfering compounds. But also, you know, luciferase, be it firefly or renilla, and then FP assays.
Cellular assays use all the same technologies. It’s just within a cellular context. So reporter gene assays using the luciferase or beta-lact, something like that, fluorescent dyes to measure ion flux used for ion tunnels and GPCRs, and then luminescent dyes for things like cyclic AMP measurements, you know, certain GPCRs, etcetera.
Also, you can do high-content in high-throughput, and that is a predominant format here at Dewpoint, where basically cellular events can be measured with a fluorescent microscope. And so we use that, you know, we can look at protein translocation, expression levels, lipid accumulation, things like that. And here at Dewpoint, we can visualize condensates and actually see condensates being modified and condensate behavior within the cell. And as I said before, I’ll come back to that later.
And then there are some other formats that were historically low-throughput, but now are becoming high-throughput, such as qPCR, which is now possible in 1536 as well. And mass spec. So, you know, mass spec has really been rate limited historically due to the need to clean up the fluid there by doing liquid chromatography first. But now with new technologies such as acoustic droplet ejection, electro spray mass spec, you can actually I’m get sample rates of up to three samples a second with with mass spec, which is really quite remarkable. Still not easy, but it’s remarkable we can actually be able to do that now.
In terms of an assay, it’s you know, people say to me, “what makes a good assay for HTS,” and it sounds kinda silly to say, “well, it’s an assay which achieves your goal.” Right? But that is actually really, really important to think about. You need to design an assay to discover the types of molecules you are interested in finding. And you need to put a lot of effort into that, and that’s why we have specialized assay designers that really help build our assays for high-throughput screening.
So I’ll just use one example here if you look on the the bottom right. So this is, if you think about doing an enzyme-based assay, the green line is uninhibited, so it’s basically a standard sort of almost what would be like a Km curve with rates on the y axis and substrate concentration on the x axis. So here you can see the uninhibited in green here. If you have a competitive inhibitor, so if you want molecules that bind to the active site of that molecule, it competes with substrate, then you’re gonna want a competitive inhibitor. And then you’re gonna want to screen at Km or less for your substrate concentration because you need to be able to distinguish between uninhibited and inhibited and be able to see a difference. That’s how you’re gonna find it in your assay. If you have too high a substrate concentration, you start to compress those lines and not be able to see the difference. Right? So that’s why you have to be very careful.
But if you want to say go for an allosteric inhibitor, and you’re not interested in active site inhibitors, then you could bump your substrate concentration way up because the red line on this graph shows noncompetetive inhibition like an allosteric compound would be. And you can actually see that even at high substrate concentrations, you will still find your hits because you’ll still see a difference between the red and green lines here on this, but you’re now starting to filter out your competitive compounds because, you know, they’re at sort of lower compound concentrations, because they can’t compete with very high substrate concentrations for that active site. So so that’s just one thing to think about.
You know, you need to think about on rates, so a lot of molecules may bind slowly. Think about pre-incubation. So half hour pre-incubation makes a huge difference in an assay to allow your compound residence time to bind to your enzyme to bind to your receptor, whatever assay you’re doing before you present ligand or substrate. And then you also wanna make sure you’re focusing on minimizing compound interference, which we we don’t have a lot of time to get into.
So other than that, we’d really wanna make sure that you can see a hit. So so how will you make sure that your assay is statistically robust enough And I’ll getting to that on the next slide, and we can you know, the cost effective piece is kind of obvious, you know, ten cents a well doesn’t sound like much, but you multiply that by a million, that’s a hundred thousand dollars to run your assay. So we’re usually focusing on aiming for a cost around about three or five cents a well when design an HTS assay.
So let’s talk about the stats over the next couple of slides. These slides are really quite quick. The next few So everybody’s familiar with signal to background. And I hear it all the time. Let’s try and make sure we got a solid signal to background ratio. That defines a good assay. Here’s real data from a 1536-well plate. It’s just a single plate, and the plate was made up of half the plate with high controls, half the plate with low controls. You can see your high controls here. You can see all your low controls here. The average of your high over the average of your low gives it you a value of eight. So everyone knows signal the background of a, great assay, good to go.
And so but when I look at it, I look say, well, if I have a hit, it’s gonna be somewhere in the middle and I wanna make sure I can see it. Well, clearly with this separation, if I have a hit in the middle, I’m gonna be able to identify that, and I’m gonna be able to find it. But often what happens is, you know, you use a different plate type, different plates have different backgrounds, instruments may have different backgrounds. So if you just bring up the background in this–it’s exactly the same data with a higher background–now, all of a sudden, the signal to background has been reduced significantly to less than four. Which many people would say is not a good assay. And yet, when I look at this, I would say, well I could still find a hit. So I’ve still got great separation on my signal and my background. That’s the important piece.
So the parameter that we use in HTS is is called Z’, and the formula’s there at the top. I won’t go through it. But basically, that looks at the separation of the variation around your signal and the variation around your background, and makes sure that those two signals come separated and you have overlapping distributions. And so if you look at the the first set of data I showed you on the left, signal to background of 8, gives you a Z’ of 0.85. An acceptable Z’ is between 0.5 and 1. So this shows it’s a great assay just like we determined. When we had a higher background, you can see the signal to background fell apart as we already talked about, but the Z’ stays exactly the same because that’s measuring the separation of these two distributions. So that’s why we use Z’ as a parameter.
Looking at it another way. If your assay was noisy, you can still have a great signal to background. So on the right graph here, you can see a much noisier assay. So now if I have a hit here, I’m not sure. Is that really a hit? Or is it just high signal that’s noisy? So I’m gonna end up with a huge number of false positives if I were to run this one due to the noise in that signal. But the signal to background’s still 8 because the average of those values is still the same. It’s just that it’s a lot noisier. So if you then look at this data the same way into a Z’ calculation, now you can see your Z’ has fallen apart and it’s 0.4.
So that would be to me an unacceptable Z’ because it’s tough. Now, to a number of people out there who I see and I see in literature and I see in presentations, yeah, you’re okay with the Z’ of 0.3. This is 0.4; it’s not good enough in my mind. Right? You really need to focus in on building a solid assay with good stats so you can reliably identify those hits. Remember we’re testing millions of compounds, and so it only takes a little bit of noise to overwhelm you with false positives. Right? So you’ve got to be very careful, you know, to make sure that you build a robust assay.
We also build a lot of QC into the process when we’re running HTS campaigns. And so typically, we bookend our campaign with control plates. So one of those would be an empty plate. And so there what we’re looking for is dispense patterns, any sort of systematic error in in the process that we’re doing. We would then have a control plate which has compounds which have known activity against your target or your assay, so we can make sure they’re performing as expected. So we put those on each end and then with all the plates that you run as your assay plates that have your compounds in them, we actually make sure that all of the the know, your high controls, your low controls, everything is performing, and the stats like Z’ are all performing as expected in those.
So we’re coming to the end of the talk here. I just wanted to spend a couple of slides talking about what we’re doing here at Dewpoint. So I referred to this a few times for our external audience in terms of, you know, we’re doing phenotypic HTS here are Dewpoint. So I am not gonna talk about condensates today. In this lecture series, there are lots of talks about condensates, which you can go find. And that will give you the background. So I don’t have time to get in that, but I just thought I’d show as a quick reminder.
In the video on the left, you can see condensates forming within a cell, That’s what we wanna measure. That’s what we’re gonna measure in our HTS campaign, either condensates forming, condensates dissolving, or different changes to condensate behavior. Also, to remind everybody the molecular animation on the right, it’s there to indicate that condensates are are not simple structures. Okay. They’re made up of many, many different proteins and typically genetic material as well. And so, it’s a very complex environment within a condensate.
So our goal is to really find small molecules that can modify the behavior of condensates. So we call those c-mods. And so we’re gonna set up HTS campaigns to discover c-mods, and that’s essentially what we do. And because they are such complex structures, you cannot effectively replicate those outside of a cell. Sure you can form a condensate but it won’t have all of the right elements, and it won’t have all of those right elements at the right concentration, the right proportions. So we believe that the only way to really find truly effective condensate-modifying drugs is to do all of our drug discovery, even our early HTS drug discovery within the context of a cell.
And so to do that, we actually had to build really basically an industry-leading infrastructure to be able to fully automate. With everything I talked about previously in the talk, that’s one thing to do that for an enzyme-based assay. It’s another thing to do it for a high-content assay where you’re using fluorescent microscopy to measure the output of your screen and do that in a fully automated way. But that’s what we’ve done here, and we’ve been able to pull that together.
So we built, you know, multiple systems that basically enables us to do the entire assay from cell incubation, compound treatment, washing, fixing, fluorescent staining all in an automated way. And I just included the image on the left there to show that this is a 406 washer, and we use this to fix our cells, but when you’re adding parafomaldehyde, you can’t do that in an open lab. So this is basically its own little fume hood on the system where this this box is is vented and allows us to completely automate that process on the robot at the same time.
Here is some real data from Dewpoint, we’re looking at a scatter plot from an HTS that was done here. This particular campaign was 375,000 compounds. And here’s a scatter plot just like I showed you before. You can see our cutoff threshold there, which delivered us, you know, basically 1700 hits. And looking at the Z’s, as we ran that assay, so every plate that was run, showed a Z’ of above 0.5, you know, matching all the criteria. And this was a very effective campaign that has, you know, led to multiple leads we’re moving forward through our programs.
Just showing some of the info from that. So this was automated on the robot that you saw earlier in the video. We were able to hit an amazing throughput of about a hundred thousand compounds a week. Now that might not sound much for those of you who used to high-throughput of enzyme and other types of assays, but for these types of assays, we’re using a fluorescent microscope. This is this is really quite remarkable. And so, it’s important to note that really our limiting factor was just imaging time.
We use the Phenix plus from PerkinElmer here at Dewpoint, and this was about two hours a plate. So 80 plates is 160 hours. There’s only a 168 hours in a week. So now you can see how far we were pushing this automation in order to be able to run this assay. We can’t, you know, we we don’t have a TARDIS or anything else to play with time. So we gotta work within the constraints of a week and that was about the best we could do with a single Phenix for this assay.
Okay. Last couple of slides, the last two minutes. I wanted to talk a little bit about chemical libraries and what should you screen. So if you think about all of chemical space, it’s supposed to be somewhere around about ten to the sixty molecules. So if you could test every molecule that could be made, there’d be ten sixty of them and that would be a wonderful library, but that’d be massive. So it’s not feasible to test all of chemical space. So what we try and do is sample chemical space in a traditional HTS library as broadly as possible.
So these yellow dots here are really supposed to show how we’re sampling that chemical space within our library. So each of those dots is a molecule in our library. It would be a lot smaller than that in reality, but you wouldn’t be able to see on the slide. And then if you look in red here, this is our target chemical space. So the idea is that if your target chemical space overlaps with the chemical space in the library, you’re gonna find some hits. And that’s the whole goal of running the HTS campaign. So you want as diverse a library as possible, then hopefully, for every target you throw at it, you’re gonna find some hits.
There are other ways of looking at this. You can use focused libraries, especially if you work in say kinases or GPCRs. There’s a lot of knowledge out there about what sort of molecules hit those targets. And you can build libraries that are focused to hit those particular targets. So now instead of pinpricks, you’ve got these sort of soccer balls like in chemical space, right, where you’ve got a lot of related molecules, which you know are gonna hit kinases or GPCRs. And then hopefully, you know, you still have a really good chance of finding a hit.
So, you’ve got a really good chance of finding a hit with this approach with this approach. The downside is you already know that you’ve got a lot of molecules in there that hit GPCRs and kinases, so you’re gonna have a selectivity issue out of the gates because you’re gonna you already know you hit the number of other targets. That’s why the molecules in the library in the first place.
And then you’ve got your DNA-encoded library approach, which we didn’t talk about today, but I thought I would leave this slide in just for context because DNA-encoded libraries are used heavily in HTS. So here in DNA encoded libraries, you’re making libraries of related molecules. So diversity in a DNA-encoded library approach comes from the number of libraries that you actually have. And then within each library, you have millions of molecules. So you sample very deep within that chemical space, but your diversity comes from the number of libraries you have. And so so now instead of pinpricks, you’ve got sort of tennis balls spread across the chemical space. Right?
And so again, looking at this, so when DEL (DNA-encoded library) works well, such as in this example, you get amazingly potent hits out of your initial screen. You can get single-digit nanomolar potency out of your DEL screen. If you just brush it, you’ll still find some good hits, they’ll just be less potent, probably. But you’ve also got a reasonable chance of missing completely because you don’t have the diversity in a DEL library that you do from an HTS library. So when DEL works, DEL works fantastically.
When HTS works–it often works actually, and it works very well–but you tend to find lower potency hits. So, you know, combined approaches, thinking about the best approach for the sorts of targets you’re gonna be screening is always gonna be important to consider.
So again, I said that I would scratch the surface. I covered a lot of topics today and, you know, so from a high level. Many of these will be, you know, can be additional talks, and I think we’ll be focusing on actually having a follow-up talk in this series looking at automating phenotypic HTS assays in a lot greater detail than I was able to do today and so everybody should look out for that as a future talk in this series. I would just say at this point, thanks for listening. Happy to take any questions.
Estaban Terzo
Hi. My name is Estaban. I thank you for your talk. It was really enlightening. So I have worked on some sort of condensate during my PhD, and I always thought of how tricky it would be to try to target one molecule out of a condensate since given the fact that you’re trying to disassemble or assemble something that is large, right, or complex. So a simple question, right? I mean, I don’t want to make it take too long, but how much do you need to care and when about the binding to all the other components of the complex to target molecule x.
Jonathan O’Connell
So very good question and, this is something that we thought about a lot as we were getting started. Right? So I kinda jump to the end as a simplest way of answering, right? And say it actually works really well. Right? So so we can find small molecules that basically do disassemble condensates very, very effectively. And, you know, in some cases, because of the way we’re going after this, we know how they work. We may know the target protein. In some cases, we actually don’t know how it works yet. But we can show that that molecule can actually just disassemble or actually promote assembly of a condensate.
So you don’t really need to worry too much about where it’s binding, how many things it’s binding. You really need to focus on the effect that it’s having. And the challenge then comes from driving your SAR and driving your drug discovery program if you don’t know the exact target because, you know, t’s very difficult to refine that. But it is very, very possible, and we’re we’re having a lot of success doing that here at Dewpoint.
Michael Fenn
Great talk. Really interesting. So I just had a question on the chemical diversity there at the that you were talking about. So when you’re saying about false negatives versus false positives, when you design one of these screens and the library that you’re gonna use, in terms of kind of thinking about false negatives in particular. I mean, do you consider like going deeper in certain particular areas of chemical space that might have a higher probability of being hits, or what’s the trade off versus just being much higher diversity. Right? I imagine it’s easier to pick out those false negatives if you have kind of more similar compounds versus those that are not. Is that right?
Jonathan O’Connell
Yeah, I mean, so not so much on the the false negative side, but, you can really design the library in any which way you want. Right? Just like I talked about assay development, you really need to think about having a library for the types of targets that you wanna screen at your company or wherever you are. Right? So we think a lot about, you know, just structural diversity because that’s important. We think about things like Lipinski’s rules. We think about things like solubility, right, to make sure that, because if a compound crashes out, you can’t get it in solution, there’s no point in it being in your library. If it’s crashed out, it’s not really in your library.
So, you know, so we focus a lot on that side of things. You can say, if your company is gonna be heavily focused on a particular target class. Of course, you can bias your your molecules to work on that, to work in that space. From from all the literature, you can use that information to help you pick those sorts of molecules. Here, we are also focused on permeability. We’re in phenotypic assays. We also want to make sure that the reasonable probability of molecules getting into the cell that are in our library since all of our assays are phenotypic.
Michael Fenn
Are there in silico tools to help you optimize that?
Jonathan O’Connell
There are, yeah. Personally I’m not very familiar with those, but a number of our chemists here at Dewpoint are very familiar. Of course they can really help sort of do a lot of those calculated properties of molecules in order to you know, to select the field library.
Sidharth Sirdeshmukh
That was an awesome talk. And this is a question that kind of follows up from what Mike was asking about. We’re just talking about permeability, but I know that in our assays, we run different kinds of cells. Like you said, some are tagged, some are not. But also these different cells require different kinds of media. And so I just wanted to know if you had any comments on whether different cell types or different media types bias hit calling across your phenotypic screens. Is that a problem? Is that something that we think about?
Jonathan O’Connell
So anything that you put into your assay will bias your hit calling, pretty much. And it’s one of those eternal challenges of HTS. Right? So if you have a lot of protein, if you have, say, BSA, media, binds compounds. Right? So if it binds up compounds, it’s not available to your assay. So it has to be considered. This all goes into the assay design. Right? So we want to basically kinda work to the edge to lower those levels as much possible to make as much compound available to your assay as possible. Different cell types of course.
Jonathan O’Connell
But, you know, the most important thing to think about is your target. Right? What is the most disease relevant cell type that you can run with. That’s gonna have the most impact on your end results. Let’s not worry about the fact that if you run a different cell, you might get different hits, that’s very true. So let’s focus on most disease relevant cell type, because that’s the one that, you know, that will ultimately deliver you the the best bang for the buck at the end of the day.
Sidharth Sirdeshmukh
Okay. Cool. So then if we have to use different cell types, is there any interest in using the same media for all cell types?
Jonathan O’Connell
I don’t know. I mean, yeah, sure. It’s something that you could do, but I’d struggle to understand why you would want to do that. Right? You know, versus really having your disease relevant cell line as optimized as is possible. And that, again, it’s because you could find different hits. But once you get into your animal models or whatever, they may not work because you’re using you know, a non ideal environment upfront in order to discover that. So you’ve got to get it as disease relevant as possible. And so personally, I wouldn’t drive after going with the same media for everything because I don’t want to optimize it.
Sidharth Sirdeshmukh
Thank you.
Kingsley Appiah
There’s a question here from Sujay. He said he’s asking why ten micromolar is a good starting point? Can it lead to less specificity?
Jonathan O’Connell
Why ten micromolar? That’s a great question. And so there’s a there’s a simple logistics answer to that. Right?And, you know, it’s just one of those things that’s evolved over many, many decades of running HTS. And it seems to work. Right? It seems to find just about the right number of hits. If you go with a higher concentration, you tend to basically overwhelm the system, so you find way too many hits. You can’t see the woods through the trees. If you go with a lower concentration, you may find some very specific molecules, but you also eliminate a lot of weaker ones. And the weaker ones often add the diversity.
And so it’s just a number that’s been revised, refined over many years of testing. You know, sometimes if you’re going with say, a focus library as I talked about at the end, say kinase library, you might actually wanna drop the concentration for that because you know you’re gonna find a lot of hits. So sometimes then you’ll go lower. If you’re going with with a fragment screen, using very tiny molecules–much smaller, not not full molecules, but fragments of larger molecules–then you may wanna go with a much higher concentration, and we can screen those as high as millimolar concentrations. So it really depends on what you’re using. But for a general library, for a general diverse HTS library, ten micromolar is just a number that’s just evolved, you know, we’ve come through evolution of doing HTS. I wish to give you a more scientific answer than that, but that that’s a reality.
Kingsley Appiah
Great. Second one from online from Peng. Can you please unmute, Peng and ask your question? I’m gonna ask your question, I see it here. So the question is: wondering how good the microfluidic approach making reactions to be in the nanomole scale is to reduce first step HTS?
Jonathan O’Connell
So the question is, if I understand correctly, it’s really to comment on using microfluidics to run HTS. Right?
Kingsley Appiah
Potentially.
Jonathan O’Connell
Yeah. So it’s great question. And actually, it would be amazing if we could do less. And honestly, we’ve looked at this in the past, and I I’ve done some work personally using microfluidics to run HTS campaigns. There’s a big challenge with using microfluidics for HTS, and, you know, certainly the microfluidic approach I’ve looked at use of like a fluorinated oil in order to form the the droplets in the microfluidics. And what happens when you have these droplets…
So you can run an assay in the droplet. You can even have cells in the droplet. You can create a signal. You can measure it. You can sort your droplets. It all works beautifully. The problem is once you add your library, the vast majority of your compound library partitions out of the droplet and into the oil, and is then no longer available to your assay. So it creates a real hurdle in the microfluidic approach to to HTS. When you know the properties of your molecule and if you know they stay in aqueous, it works great. But microfluidics for general HTS, we’ve really not been able to break that barrier yet.
Kingsley Appiah
I’m gonna go with one more online here with Lennart?
Lennart Enders
Yes, thank you. And I was wondering you mentioned that you use fixation and then do your microscopy readout. And I mean, there’s a lot of papers out there that say that you quite introduce artifacts, especially for condensates when you use fixation. So have you ever thought about using fluorescently labeled proteins instead and then do live imaging because, I mean, you could also save on the reagents, antibodies, and so on, and you might, yeah, miss out on things that that are hidden by the fixation process?
Jonathan O’Connell
Yeah. It’s a good question. And so we do some work with live cell. We’ve not really used that for true HTS as yet, but we certainly do live cell for a lot of our follow-up. And so absolutely you you can do it. The challenge is, you know, we gotta engineer the lines with fluorescent proteins. Right? There’s quite a lead time when you go down that approach. And so far, we’ve not really noticed any significant difference, at least in the approach we’ve taken. Right?
So we’ve been able to find many, many molecules that really modify the behavior of the condensates. But it is something that we’re keeping an eye on and we are looking at. So it’s a really good comment, really good question. And I don’t know if maybe someone else at Dewpoint has an additional comment to that, but it’s definitely something that is possible and is possible in our environment.
Lennart Enders
Great. Thank you very much for your answer, and to talk. Thank you.
Mubtasim Akhyar
Hi. I had just a quick question. So especially during, like, a very large drug screen, do you have any way you can combat kind of issues like with different viscosities of different compounds or such because during automation, I’m assuming that certain errors can come from, like, a compound either, like, I don’t know, not being mixed well or being more viscous than something else. How do you combat that in the middle of a large screen like that?
Jonathan O’Connell
So couple of answers to your question. So compound viscosity is not really an issue because of the the concentrations we’re talking about. Right? So we’re only really a ten micromolar, so we don’t really struggle with compound viscosity. But reagent viscosity is a real issue, and you’ve got to be very careful, make sure you can pipette your reagents and prepare those reliably. So it depends on what you need to dissolve your reagents in. But we do a lot of testing and a lot of upfront work to make sure we can do all of those precisely and accurately and repeatedly, you know, before we get started with the campaign. So we kinda do all that work as part of the assay development.
But you also reference compound mixing. That is critical. Right? So you know, really it’s one of the biggest areas of variation. Right? Especially if you’re working in a 1536-well plate, tiny well, right? You can’t shake that. Shaking doesn’t help. Right? Because surface tension just keeps everything together. You don’t really mix anything. It’s not like a 96-well where you can create some movement in fluid.
So we try to do things, you know, we try to engineer things in certain ways to help with that. And there’s many tactics. In a traditional sort of enzyme-based HTS, compound goes in first. You’ve got a massive overlay. That creates good mixing. When you’ve gotta cellular assay, you have to have liquid in the cell before you add compound to keep your cells happy, but then we still try to have a larger overlay on top of that to again help with the mixing. But also cellular assays tend to be longer duration. So you have plenty of time for diffusion within the well in addition rather than your short-lived enzyme-based assays. So we do have a number of tactics to try and minimize those sort of, you know, variabilities. That’s a good question.
Kingsley Appiah
A lot of questions going around. Let’s go back to Shulin. She’s on online? Then I’ll read it. It says, when a compound is tested to be partial inactive at ten micromolar, would you later increase the concentration to confirm the IC50?
Jonathan O’Connell
So the compound’s partially inactive at ten micromolar. Right? So by that, you mean it’s giving something like forty percent inhibition or something like that, I assume. Right. So, sure. So, you know, you’ve really got to determine what hits you have what the distribution of those hits you have, and what’s interesting, you know, what the chemical properties are though. So if a compound only gives forty percent inhibition at ten micromolar, that still could be a great starting point.
So absolutely, we would do an IC50 on that, get a potency, see if there’s related molecules we can test, see if those are generating anything that’s more potent. And then, you know, if it’s still really interesting, we’re gonna start modifying that molecule and see if we can drive that potency down. So, yes, we will absolutely go ahead and do an IC50 if the chemical structure was interesting.
Sidharth Sirdeshmukh
So this is kinda like I don’t wanna ask the last question, but this kinda sounds like the last question. But, like, through your eyes, what does the future of phenotypic screening look like?
Jonathan O’Connell
What does the future look like? So, that’s a really interesting question we can probably talk for a couple of hours on. Right? But the reality is that, you know, phenotypic assays have been around in HTS for quite a while. And there’s been numerous approaches. Right? Not necessarily with condensates, but, you know, just other phenotypic assays. And the challenge has been deconvolution. Right? So really figure out if you’re looking at the results of a pathway, how do you figure out what the real hits are in terms of, you know, and what those hits are targeting. And so, you know, with condensates, there’s a little difference. We know we’re targeting the condensate. We might not know what in the condensate, but we know we’re targeting the condensate. So, you know, so we have a little bit of an advantage with this approach.
But, you know, really the challenge with phenotypic assays right now is just throughput. Right? So it’s great for follow-up. Today, as you alluded to in the slides. Right? You know, we’re maxing out at about hundred thousand a week. It would be great to push those boundaries. So I think as imaging technology, instrumentation improves, and then data analysis on the back end, so to be able to use much more efficient data analysis, to be able to use AI data analysis, which we are doing here to really find different hits, different mechanisms. That’s really for me is where the future is going is in terms of the, you know, the efficiency with the instrumentation and then the image analysis piece post screen, post running the assay.
Jill Bouchard
Great answer. Just wanna clear up. We still have questions, and hate to burst your bubble on wrapping us up, but we still have more. Go ahead.
Kingsley Appiah
So a question, as a follow-up from Sujay. Can you unmute and ask?
Sujay Guha
Yeah. So I wanted to ask with a system like echo for dispensing which is more like acoustic. If you have adherent cell lines with media, how does that work? Because it looks like it gets inverted. Are there other systems as well where they dispense from top? Like, how’s how do you mitigate that challenge?
Jonathan O’Connell
So could you just repeat that? So I got the fact that you were asking about sort of using the echo for cellular assays?
Sujay Guha
Yes. Adherent cell lines with media?
Jonathan O’Connell
Yeah. So it works fine. So the only difference is that Instead of in the video that you saw earlier where we were dispensing into a dry plate, basically, we have the cells in there. And we have media on top of those cells, and then we dispatch right into the liquid. Right? So that droplet just merges with the liquid. So you can still use the echo with, you know, with adherent cell lines. It works absolutely fine. It works really well, actually.
Sujay Guha
Okay. Sounds good. Thank you. No problem. Thanks. Great.
Kingsley Appiah
Next question from Biplab K C.
Biplab K C
Okay. Thank you very much for your talk. I was just wondering about the phenotypic screening because your compounds need to pass through the cells first and then goes into the condensate. So I think that the selectivity comes from whether the compound can first internalize into cells or not. So I was thinking whether there is some features of the hits that contradict with the cellular internalization process and ability to partition into condensate. So do you have any…
Jonathan O’Connell
I think I caught most of that. Sorry. Wasn’t super clear, but so you’re absolutely right. For the phenotypic assays, cell permeability is a must. Right? If the compounds don’t get into the cells, then they’re never gonna be active. So, you know, it’s just one of those challenges that we face. Right? So we try to have a library that contains molecules that we believe to be cell permeant, but that’s calculated. It’s not necessarily a given. And we may miss stuff that if it wasn’t within a you know, if it could have got in the cell it might have been active, but there’s nothing we can do about that because if we try and create the environment outside a cell, then we’re missing so many components of the condensate that we take ourself off in tangents that aren’t necessarily productive. So, yeah, it is a downside of phenotypic assays that you do need cell permeability, but it also means that our series have cell permeability from the start, and we don’t need to engineer it in later.
Biplab K C
Yeah. Because as per my understanding in the condensate usually because of the modifying interactions, there might be some kind of charge, molecules, or some amphiphilic. But in the case of cellular internalization the requirement might be a little bit different. So Yeah. Thank you very much for that.
Jonathan O’Connell
Oh, thank you.
Kingsley Appiah
Okay. The next question, I’ll just jump over there. Last one?
Jill Bouchard
Nope. Sorry. One more.
Kingsley Appiah
Oh, there’s one more? Okay. Alright. This is from Kadi Saar. Can you unmute, Kadi?
Kadi Saar
Yep. Yes. I’ve just unmuted myself. Hope you can hear me. Yeah. Thank you, thank you for the talk. And I had a quick question about, you know, when when people are doing phenotypic screening by, for instance, HTS like you are. How much effort do you put on, you know, really understanding what the molecular mechanism or the target was because, you know, I guess on the one hand, it would help you a lot in terms of being able to drive SAR, but then it’s also quite a bit of work to to set these assays up that could help you to understand it. of course, you can do it, but it’s work. So I kind of, you know, philosophically, how do you approach it, at what stage it becomes really important for, you know, either you at Dewpoint or others in the area.
Jonathan O’Connell
Yeah. Those I mean, it’s a great concept. We wrestle with this all the time, obviously. And and so clearly, you’re absolutely right. If we can identify the target, then fantastic. That really helps us drive SAR because we can develop those assays and you know you could look for binding to that target, really work on focusing on enhancing that while in parallel testing in the the cellular context, and it really helps. So we do put a lot of effort into it, and I kinda mentioned a little earlier, I think, you know, sometimes we get it. Right? So in terms of the condensate, we know some of the molecules within a condensate, but we don’t know them all. So if we can build an assay, we will, and then we’ll test against it. And sometimes that can really help us.
And in some cases, we don’t know the target, but we’re still working on trying to figure out what that is, in order to help move things forward. But you can effectively drive SAR with these phenotypic assays, you just have to think of it in a multi-parametric way. But you have to think about, you know, from multiple aspects because you know, you may change a molecule and then lose a little bit of cell permeability, which might affect your SAR and then you just gotta really think broadly when you’re looking at SAR, if you’re gonna do it from this perspective.
Kadi Saar
Makes sense. Yeah. Thanks.
Kingsley Appiah
The next question is from Suguna.
Suguna Rachakonda
So the question is: irrespective of the capabilities of HTS and automation and assays we have, the quality of the hit you’re going to get is dependent on the quality of the library. And how do we get to improve that quality and how can we use AI to combine with HTS to get a better starting point for drug discovery?
Jonathan O’Connell
Ah Suguna, that was a big question. So, yeah. So you’re absolutely right. You know? Put rubbish into your library, you’ll get rubbish out your your library. Right? I mean, it’s quite simple. So basically, we run multiple screens. So each screen is a learning experience. Sort of accumulate that data, and we will take molecules out of the library, but maybe initially just informatically, if they’re known to be bad actors, for example, so that you can eliminate those false positives from bad acting compounds. So we use that to sort of refine the active library.
We also look at areas where we struggle and we try to think of, like, you know, use a chemistry perspective to really look for new molecules that we could add to the library, which will hopefully plug some gaps and address those targets. Right? That another approach that you can take to enhancing it. Simple diversity is one of the key factors–we alluded to that in the screen. The more compounds you can test, the better the output. It’s a very linear equation in HTS. So if you can double the size of your library, you double your chance of success. And that’s been shown in many publications and many efforts over the years. So the simplest way is to add more diversity.
And there’s a lot of chemistry tools that allow you to look at that and to measure diversity and look for separation of one molecule from another in your library. So we tend to do that. The AI piece that you alluded to we do, in terms of selecting molecules, we do that really mostly from an image analysis perspective, and that’s really where the most power comes from the AI piece at least with what we’re doing currently. So it’s basically taking the the output, the images that come out of your HTS, I said earlier we use the Phenix Plus, so Harmony does the analysis. That’s on a predetermined analysis.
So we can take an unbiased analysis of those images and look for clusters of molecules using AI to find molecules that may have an effect on the condensate that we weren’t looking for. And that allows us to identify novel hits that we never thought we would find, and never knew we would find until we ran that screen. And that’s proving to be incredibly powerful. The ability to use these AI tools to actually analyze the images, take that data, and find unique sets of hits that then we move forward into our drug discovery programs. We can also use AI to take the data we did get and that we used by conventional methods to actually look for related molecules and things that we think may be a hit. If this is a hit, maybe this should also be a hit. And actually funnel those in as well. So we have a few approaches to tackle that. I hope I answered all your question there.
Suguna Rachakonda
Thank you.
Kingsley Appiah
Okay. Is there any more question from in house here?
Aya Abdelbaky
Okay. So there’s something I felt a little bit honestly confusing. So you were saying that’s really hard to, like, form again a condensate outside the cell. But does this mean that you expect find the very similar or mostly identical condensate in the same cell line you’re targeting or whatever the disease is? Because I thought all those condensates won’t be a very similar identical every time you find it. Right?
Jonathan O’Connell
I’m sorry. I didn’t quite catch all of your question. So you asked about.. did you ask about forming condensate inside a cell and outside a cell or between different cell lines? I was missing that. Sorry.
Aya Abdelbaky
You you were saying it’s part of it being difficult as target that you can’t really like form it yourself. Yeah. And it has to be there, in yeah. But does this also mean that you expect to see something similar every time you target it and I find this a little bit, it’s not how it is. Right?
Jonathan O’Connell
So one of the things that, the approaches that we take is that, obviously, we understand, we have a lot of condensate expertise, obviously within Dewpoint. Right? But we’re able to form these condensates in a very reliable manner, in a very predictable manner. And one thing that we can say for sure is that with the condensate-modifying drugs that we found is that the effect on those condensates is really, really reproducible. So we can see that time in, time out. So even though the condensates themselves may be perhaps subtly different with different sizes or with different constitutions, at least the molecules that we’re discovering seem to have a really reproducible effect on those condensates. And that seems to be holding in terms of our drug discovery efforts.
Aya Abdelbaky
Okay. Thank you.
Kingsley Appiah
A follow-up question from Esteban.
Esteban Terzo
Yeah. I’m sorry. And also kind of like on the heels of this good question that you were asked now. I mean, I understand the importance of standardizing, you know, assays, especially in high-throughput screen. But we also have to bear in mind that not all condensates are going to react the same way. So my question was, like, do you you know, for the purposes of one disease, do you aim for the c-mod effect to be irreversible, or is it better to be reversible? So then you can actually look into the effect on the disease.
Jonathan O’Connell
So that’s really speaking more to sort of condensates for drug discovery versus condensates for HTS, which is what we’re, you know, focusing on today. So I’m sure there’s people who are much better able to to answer that that question than I, but, in terms of the types of molecules, we don’t really go for an irreversible approach. We go for more reversible. And, we’ve been able to show that, you know, on removal of drug that you can get condensates reforming. So most of our approach has been on sort of the reversible side of things. In terms of the way that we go after our program versus irreversible.
Esteban Terzo
Alright. Thank you so much.
Kingsley Appiah
I think that’s our last question.
Jill Bouchard
Yeah. I think that might have been it. What a crazy good–oh, you have another one? One more. Okay. Great. Turn it on first and then give it a couple of seconds.
Casey McGuire
Hello? I was wondering when you’re doing HTS screen across different programs, how do you decide the treatment time? Just because in some cases, when you’re dissolving a condensate, for example, you’re gonna end up causing apoptosis. And so you might not be able to capture some of that condensate dissolution if you’re just kind of killing all the cells. And so being able to figure out what time for treatment is ideal when you’re doing a screen and you’re not able to kind of do those time courses later on.
Jonathan O’Connell
Yeah. So, there’s multiple things we do here. So one is it’s a great partnership between the HTS scientists and the disease biologists who are experts in that particular disease and the condensates. So really, we do a lot of work on figuring out what time points we wanna actually use for the assay, and we can do it multiple ways. Sometimes even if we’re unsure, we can even run a focus screen where we’ll actually look at not to refocus more of a pilot screen, where we’ll actually look at just testing maybe twenty or thirty thousand compounds, and then do that at multiple time points. So then really dig into the hits that are found to really, you know, try and elucidate that. So that’s one thing we will do.
The other thing is that the beauty of the fact that we’re doing phenotypic assays and we’re using fluorescent microscopy is we don’t necessarily just look for the feature that we’re interested in. So we can also look for cell toxicity, so we can look for multiple other things, actually in that assay. So for example, nucleolus is a condensate. So we can actually lock and track the nucleolus as well and make sure that we’re dissolving the nucleolus. And we can look for various toxic events that the drugs may have and filter those out to focus on the the cleaner condensate-modifying drugs right out of that initial screen.
Jill Bouchard
What an incredible conversation I’d have to say. This was awesome. Thank you for a fantastic talk.
Kingsley Appiah
Thank you.
Jonathan O’Connell
Thank you, everybody.
Jill Bouchard
Yeah. For everybody on Zoom, thanks for coming, and also, obviously, people still at the kitchen table. And come again the next time. We’ll let you know when we have another one.