Unlocking 100+ Undruggable Targets

Kristen McPike 0:06
Hello, everyone, and thank you for joining us. I’m Kristen McPike, Program and Alliance Manager at the Antibody Society. Today’s scientific briefing is one in a series designed to inform and educate our members and the broader scientific community about topics relating to antibody discovery and development. Our expert speaker, Dr. Ross Chambers from Integral Molecular, will be discussing 100+ undruggable targets unlocked through parallel MAB engineering with us today. Please note that this briefing is being recorded, and you can add any questions to the Q&A box at the top of the screen, and the questions will be answered at the end of the presentation. Finally, there’ll be a few poll questions throughout the presentation that we’d love your participation in.

Without further ado, I’ll turn the show over to our speaker, Dr. Chambers. Thank you.

Ross Chambers 0:58
Great. Thanks, Kristen, and welcome everyone to today’s webinar.

I’m Ross Chambers. I lead the antibody discovery group here at Integral Molecular. And the story I want to tell you today is really about a new approach to drug discovery that we’re pursuing. And the goal is to be able to advance undruggable targets, previously undruggable targets into drug discovery, and also to increase the efficiency of the overall process. So without further ado, I’ll get going.

For those that don’t know who we are, Integral Molecular is a company focused on membrane proteins. And we have a number of technologies that we offer as services, and they’re all focused really around membrane proteins. We’ve been in business more than 25 years and it’s really focusing on antibody discovery against membrane proteins, engineering, and epitope and paratope mapping. We offer a number of solutions out there, well known for the Membrane Proteome Array for specificity testing, for Lipoparticles or VLPs, which is one of our founding technologies for a variety of membrane protein applications. We did epitope and paratope mapping. But the story I’m going to tell you today is really focused on therapeutic antibodies. We have our own pipeline. And we also have a sister company called Cell Surface Bio that is also providing antibodies for the very important step of enabling validation of drug targets, because for many of the targets I’m going to tell you about today, there are no tools, no antibodies for them, so it makes it very difficult to advance them into discovery. We also have a virology arm that I won’t be talking about today that we make reporter viruses for a variety of purposes. So that’s who we are. We’ve been around a long time, as I said, 25 years. And in those 25 years, we’ve worked with hundreds of companies and we’re very proud of doing, high quality science. It’s resulted in many patents and publications and we’re very proud to be able to publish in high-tier journals with our science.

All right, so the story I’m going to tell you today, as I said, is how do we advance access to undruggable targets and do it more efficiently? And then once you hit antibodies, these targets, there’s other steps you’ve got to worry about too, which is how do we more efficiently optimize these antibodies through engineering? And how do we make sure that they advance through preclinical development more successfully? Oh, Kristen, there is a polling. This is a good time for the polling question.

Kristen McPike 4:11
Okay, so everyone, the poll has been shared.

Ross Chambers 4:21
The next section we’re going to talk about is around antibody discovery. So we’re interested in enabling access to previously undruggable targets. And these really are multi-pass membrane proteins, which we’ll tell you more about. They’re very difficult for a number of reasons. And we want to be able to make them as accessible as for all the normal kinds of targets. We have the poll result in.

Kristen McPike 4:54
We do. We’re still getting some responses to this so far. 37% of people have said that their biggest hurdle that they’ve experienced during antibody discovery campaigns is getting functional antibodies. Yes.

Ross Chambers 4:56
Getting functional antibodies. Yeah, I would agree that that can be a tough one. And really that’s, you need diversity of epitopes to enable that and hopefully part of our story today will help, will touch on that problem.

Kristen McPike 5:29
There’s a tie also between presenting antigens and their native confirmation and poor antigen expression, and you got a total of 23 responses.

Ross Chambers 5:35
Well, hopefully the problems I’m going to talk about now resonate with people based on that response. So first off, you look at drug discovery as an industry, one big question is why do we go after the same targets? You can see this illustrated here in this graph which is looking at the number of publications ranked against all of the 20,000 human genes. And you can see there’s an extreme bias on a very small number of genes. We have a very strong bias in the drug discovery industry going after the same targets over and over again. You know all those targets. lots of drug candidates to the same target. Part of that is for, well, there’s a number of reasons behind that, but the reason I want to touch on today is this untapped part of the proteome that really hasn’t been well explored. There are lots of genes in there, proteins, that have strong connections to disease. So it’s not like we’re focusing on all the important targets, is that there are important targets distributed across the proteome, but they’re just not being explored. And one of those bottlenecks is the fact that there’s this little biology known, and there aren’t reagents available for us to explore them and understand their role in disease and advance them as candidates. This is illustrated by one of the NIH programs, the IDG program, Illuminating the Druggable Genome, where they identified a panel of hundreds of targets that were under study, but they believe are going to be important therapeutic drug targets. They include more than 100 ion channels and more than 100 GPCRs, both of which are multi-pass membrane proteins. Importantly, less than 10% of those proteins on that list have a high quality antibody available. That’s part of the bottleneck. So part of that dark proteome that I’m talking about here make up multi-pass membrane proteins in three very important families that everyone understands is important because of small molecule success in drugging them is the GPCRs. the ion channels and SLC transporters. These make up, you know, more than 1000 of the 20,000 human proteins and very important drug targets, but again, very few of them have antibodies available. Now, the reason for that is not just because of no reason. There’s important technical reasons why we don’t have antibodies to these proteins. Multi-pass membrane proteins in general are very difficult targets. And it’s not that they have one problem, it’s that they have multiple independent problems that you have to solve. Now, here at Integral, we have been focused on multi-pass membrane proteins for more than 25 years, and we have been tasked with systematically solving these problems. Some of these problems, as in that poll, suggests people are already familiar with these. One of the big ones is that these proteins, because they weave through the membrane, multiple times. Their structure is dependent on the membrane. So in order to present native epitopes to the immune system or to any kind of antibody discovery system, you have to maintain them in that structure. You’ve got to keep them in the membrane. That’s technically very difficult. They’re also very difficult to express, very difficult to get them at high levels, or they can be toxic when you do overexpress them. And the extracellular regions can be very, very small. So there’s not much for an antibody to grab onto. And many of them are very highly conserved, which makes them very poorly immunogenic. So anyway, there’s multiple independent problems there that prevent someone from easily making antibodies to these proteins. So we have been working for many years here at assembling technologies that systematically solve all of these problems. And three of the sort of core aspects of these technologies are presented here that enable us to access these very, very difficult targets. The first one is in engineering. Some of these targets just need to be engineered to get to high levels of expression, but you’ve got to do that at the same time as maintaining those native extracellular epitomes. So it’s quite a technical challenge. The second thing is you must present the antigen in a form that’s in the membrane, and we use mRNA and virus-like particles. And these are not only highly immunogenic, but importantly, they present the antigen in their native conformational state. And the third arm is around the host animal that we generate the antibodies in. So, you know, I’ve used mice and rabbits and all kinds of normal animals for many years, but we have pivoted to chickens because many of the proteins that we’re talking about here are highly conserved. And chickens and they lost to solve that problem. and access you can’t get from systems. So those are sort of three of the big pillars, and I’ll go into more depth on these in the next few slides.

So, first on antigen engineering. This is to solve the expression problem. So you need to get the protein at the surface of the cell at high levels in order to create antigen, as well as creating a sensitive screening assay. Now, it’s quite difficult because multi-pass membrane proteins transit through the secretary path. And this is a multi-stage process with lots of potential bottlenecks. It may have trouble in translating and folding, in assembling together in the ER, in trafficking through the various compartments through to the surface, and also staying on the surface at high levels. And what we have done here is, over the years, developed a diverse toolkit to enable us to systematically solve this problem. And it has to be a diverse toolkit because we have diverse proteins. And in general, I would describe it as a number of techniques for engineering the antigen itself or various chaperones to help them get through their ER or trafficking, and special expression vectors that help us do this, solve these problems. This is illustrated here in some examples for you. So here’s 3 examples of three proteins that proved to be very difficult to express in their wild type state. So we have a transporter, a 6TM, and a GPCR. On the top row is the wild type expression. So these are epitope tag proteins, transiently transfected into cells and human cells and overexpressed. You can see here in this flow histogram that you see very, very poor expression. A little bit of expression for the transporter, but the other two are essentially 0 expression. We engineered all of these proteins with a variety of methods described below and able to successfully get these high levels. So the first one, the transport, we had to create what we call a chimera where the inside part is from a different protein and the outside part is the target. And we had to add point mutations. That was successful at getting higher expression. In the 6TM, we had to try many more things with special signal sequence, a trafficking motif, deleting parts of it and point mutations, again getting high expression, and the last one, a signal sequence and a trafficking motif. So we are now at the point now where we can you know, successfully solve expression problems from the majority of these kinds of proteins. The second part I mentioned was the antigen itself. This is super important in order to get antibodies that recognize the native structure. So it’s very easy to get antibodies to the denatured structure. It’s the native structure that’s important. So as I said, we use two antigens. We use mRNA and we use Lipoparticles or virus-like particles, as otherwise known. So RNA, previously we used to use DNA immunization, but RNA is even better. It enables very high levels of expression in vivo. So you inject the RNA into the animal. The animal expresses the protein in vivo. It’s presented on the membranes, the plasma membranes of those cells and to the immune system in a native state. Very highly immunogenic. Lipoparticles are also required for the most efficient system. We use them in combination with mRNA. So we prime them with mRNA and boost them with VLPs, which are very successful together at getting very high title responses. We also use them for phage panning, so when we pull out the monoclonals, you need something to pan on, and VLPs or Lipoparticles are very efficient for doing that. Here’s an example of the power of this prime-boost strategy. So you can get responses with RNA alone for some proteins, but some of the more difficult proteins really require this dual immunization system of priming with RNA and Lipoparticle boosting. So here’s one example shown here where we prime with RNA and we didn’t see a titer in this flow histogram. But when you boost it with Lipoparticles, it triggers the production of plasma cells. You now see a strong titer pop up. So together, RNA and Lipoparticles are very powerful for giving access to these difficult proteins. The third arm I mentioned was chickens. Now, as you now know, most people use rodents, mice, and I’ve used many mice in my career as well. But they do have serious limitations. If you look at the conservation of drug targets between humans and mice, more than half of drug targets are highly conserved to humans, more than 90% identical to humans, which means that when we immunize those into a mouse, the mouse will see it as self and won’t generate an immune response. And if it does generate an immune response, it might be very weak. Which makes it difficult to discover antibodies, or it’s focusing on a very small subset of epitopes. And it may not be the epitopes that you want, especially important for finding functional antibodies, because for functional antibodies, you need diverse epitopes to make sure you hit that right spot if it’s an agonist or an antagonist. So mice are a problem. And this problem isn’t really solved if you go to another mammal. So the problem with mammals in general is they’re all closely related to humans. Mammals as a group are all closely related. So you really have to go to some very distantly evolved species, such as chickens, which is more than 300 million years distant from the evolution of mammals. And what this means is, is that most of these drug targets now are not highly conserved to chickens. So you see in that graph on the right that now only 15% of drug targets are more than 90% identical to humans. So most of the targets are now accessible and giving diverse immune responses when you immunize into chickens. This is what drove us to chickens and doing these difficult targets. Let me show you some examples for this. So here’s 4 examples of some highly conserved proteins that we’ve worked on over the years. Clauden 6, KV1.3, CB1 and SLC2A4. All four of these proteins are very highly conserved, especially in the extracellular domains, the part that you need a B cell epitope to. But when you look at the conservation to chickens, very often it’s 10 to 20% lower. And this is a big deal. Just getting 10% lower difference enables the immune system to really start to recognize it. And in all these cases, you can see here that we got strong immune responses to all these proteins using chickens. Whereas in mice, they really struggle. There’s another aspect of chickens I want to highlight, and it may sound like a bit of a tangent, but it’s really around the story of longer HCDR3s. So as you know, the heavy chain third CDR is the most length variable and most important CDR of an antibody. And it varies in length considerably. And important to recognize is that the length of the CDR3 controls the topology of the antibody, the kinds of epitopes it can access. So for example, very short CDRs create a concave paratope. which enables it to bind things that protrude, like a nivolumab binding PD-1. CDRs at around 11 amino acids or so, they tend to be pretty flat. And then things longer than, say, 14 amino acids, they tend to protrude. And what this means is that these longer CDRs enable them to bind into pockets, which is very important, especially for functional antibodies, because often those are in recessed regions. So it’s opening up new epitopes that we wouldn’t otherwise be able to recognize. And why this is relevant is because chickens, it turns out, are similar to llamas. Camelids have longer HCDR3s. And in fact, mice are the odd man out here. They have quite short CDRs. You can see this on the graph on the right, where we’ve analyzed the length of the therapeutic antibodies out there. And you can see here, on the average length, is around 10 amino acids. And that reflects that most of these antibodies from therapeutics come from mice. That’s changing now. More and more animal antibodies are coming from fully human libraries, and they are longer because humans have longer CDRs as well. But you can see on the line graph on the right, in the orange line, is the length distribution of chicken MAMs, and they are considerably longer. And what this means is that we’re able to access epitopes that you can’t access easily with other systems. Now, to illustrate that point, here’s a project that we worked on some years ago and we published in PNAS on this difficult target, SLC2A4. It’s a glucose transporter. Now, glucose transporters, these are 12 TM proteins, very small extracellular regions, as part of their functional mechanism, they teeter-totter between an inward open state and an outward open state as part of their transport function. And you’re able to lock these in these different states, either using small molecules like cytokylasin B or fluoretin, or genetically with mutations. And we discovered a panel of antibodies and we wanted to understand how they saw these different conformational states. And we’re very excited to find that we’re able to find an antibody like LM43 that was specific for the inward open state but couldn’t bind the outward open state. Conversely, LM48, the one in the middle, could bind the outward open state and not the inward open state. And we found antibodies that didn’t care at all. But importantly, the interesting thing about this antibody was it had a very long CDR, 26 amino acids. And when we epitope mapped it, it appeared to bind in this pocket, this of the outward open state, suggesting that this long CDR may have been important for being abel to find these kinds of antibodies, because certainly other people don’t find these kinds of antibodies. All right, so we have assembled this platform, as I’ve described you to, using antigen engineering, mRNA and VLPs, and using this chicken system and other technologies together into a system to routinely make antibodies to multi-pass membrane proteins. So we’re now at the point that we can efficiently make antibodies to these targets. And this is the system that we have together now. So we’ll get a target, we’ll make antigen-immunized chickens, make phase display libraries from those immunized animals, screen them using high throughput flow cytometry, and then convert them into IgGs into all the normal characterization. And this is what we’re using now to access all these impossible targets. And we’re making them available as reagents through a sister company that we spun off recently called Cell Surface Bio. So if you need antibodies to these difficult targets, they can be available through Cell Surface Bio, either as catalog reagents or as a custom project. So we have the system. So what we decided recently is let’s go and start making antibodies to these undruggable targets. So we assembled a list, the top 100 of possible targets. So we combed through the literature and various databases and assembled a list of important multi-pass membrane proteins that people haven’t been able to make antibodies to, but have very strong associations with disease and high quality journals and very strong evidence that they’re important for disease, but people haven’t been able to advance them because they’re difficult membrane proteins. We also selected some of these proteins because they were very selective in their tissue expression. So it’s very important for some drugs to be able to target them to very specific tissues. So we also, many of these proteins, these multi-pass proteins have very selective expression in various tissues. So we assembled this list and we set off in the last year to make antibodies to all of them, to make them available to advance drug discovery. Here’s one example that we’re very proud of, aquaporin 5. So last December, there was a very important paper came out in Science showing that aquaporin 5, this is a water channel, had a very important role in gastric cancer. There’s been a number of these high profile papers around this target, very strong scientific validation, and it’s a cancer driver. So very important target. However, when you look at it as a target, very, very challenging. It’s A multi-pass membrane protein. It’s got very small loops, highly conserved, and most difficult to express, very difficult to express. We put it through our system and we’re very happy to see that we are able to isolate panels of antibodies from this. Here’s some more examples of that 100. We’re able to now, on a wide range of different targets, be able to generate antibodies that work in flow cytometry, so druggable candidates for this diverse range of proteins. And the higher level view is this is our sort of our project board for the year. Of the 100 targets, they’re at various stages. We’ve got about 40 done in terms of we’ve got binders done. There’s another 30 in, we’ve got titer, and we’re in the process of pulling out the monoclonals, and there’s another 20 or so that are still in the immunization stage. So we’re working through this list of 100, and of course, this is just this year. We’re going to be continuing next year and doing more and more targets like this, but it’s proving that the system can function at this level. The next step is to advance these as drug candidates, and this is going to depend, the modality is going to depend on the disease. So for example, for some of the cancer targets, we’re making T-cell engagers, like for aquaporin 5, we’ve made a T-cell engager. For other ones, ADC might be appropriate. And for some targets like antagonists, we want to bring down the function, we’ll create a CD3 engager like we’re showing here for Kv1.3.

Okay, so that’s the discovery part. If you want to post the next poll, Kristen, and I’ll talk about the next section of the talk, which is you have a lead candidate, but often in discovery, you know, you have multiple things for a drug specification that you have to fulfill. It’s got to have the right affinity, it’s got to have the right specificity, it’s got to be able to be manufactured, and there are many other aspects like that that are required to make a great drug candidate, and they don’t always come in the first hits you pull out of discovery. So, oftentimes, you have to engineer them, and this can be… a big task. It can be very complicated, slow and expensive to get through engineering. So what we wanted to do is to see if we can radically improve the throughput of the ability to engineer antibodies to make them into lead candidates. Do we have the poll answers up now, Kristen?

Kristen McPike 27:11
We do, and you can also see the answers live, everyone, in the chat, if you’re curious. The answer, when optimizing 1 antibody property, how often do you negatively impact another? Often enough that it slows us down and occasionally are tied.

Ross Chambers 27:31
That has been my experience. We’ve been engineering antibodies here for many, many years, which is why we wanted to change it. So, as I said, engineering lead candidates can be a frustrating business. You fix the affinity, but you bust the developability or something else. And it can feel a little bit like that whack-a-mole game. And what we wanted to do is be able to engineer everything at the same time in parallel in a one-shot deal. And there’s a list of things of things on the left there you might want to consider. You know, a lot of things you want to change in an antibody and make sure it’s a great drug candidate. But how do you do this much more efficiently? The platform that we recently developed is a high-throughput parallel engineering platform. And it’s a little bit different than most other people’s approaches. Many people, as you may be aware, go straight into AI machine learning and design a lead candidate based on generalized data. The approach that we’re taking here is basing on experimental data for that particular antibody. So what we’re proposing and what we are doing is creating comprehensive CDR scanning data. So every position in the CDR mutated to every possible amino acid, using that data to feed into an MLAI system, to create the optimal combinations to finish with an optimized lead candidate all at once in parallel without all this back and forth whack-a-mole business to try and fix your molecule. This is the platform as it looks. This is called Paratope-PLUS. This is available as a service. And what we do is you submit the sequence and we’ll do an assay setup, make sure we have an assay, obviously. And then we’ll go through, we’ve developed a very high throughput, efficient cloning system to enable us to make 1000 mutations very quickly. We’ll make every possible position, every possible amino acid, and express those 1000 variants, and then measure them for both expression and for binding, and that might be multiple assays depending on your program, and that ends up giving you a bunch of data, so it gives you all the paratope residues. But all the permissive variants that you can change, things that improve the binding or other characteristics that you measure for, and then that feeds into AIML to make these combinations. One other thing it gives you is around IP, for stronger IP, and I’ll touch on that at the end. That may not be on people’s radar. I’m going to walk through this with an internal example from our GPRC5D program. This GPRC5D is a GPCR. It’s a target for cancer for multiple myeloma. And there are some clinical antibodies out there that are approved and it’s a brilliant target, really incredible clinical results. We created a panel internally. We used our platform. This was a little earlier. So this is using Lipoparticles in DNA, but we now would do that today in RNA and using chickens. And we created a panel of antibodies and created various bispecifics and trispecifics and tested them in mice and showed that it could kill off the cancer. But there were some things that we weren’t super happy about. We wanted to tune the affinity more, to tune some of the species cross-reactivity to help through preclinical and create a stronger IP that I’ll explain why that’s the case later. So with that, here’s some of the data, what it looks like. This is just the heavy chain data. There’s also the light chain, but for brevity, I’ll just show you the heavy chain data. So this is creating 1000 mutations. And on the y-axis is the positions in the antibody, and on the x-axis is the 20 possible amino acids. In blue are the residues, the mutations that generally don’t have an effect. In white are things that significantly hurt the antibody, and in orange are things that improve the antibody. So this is a comprehensive picture of your antibody that is the foundation or blueprint for building that better antibody. And you have all the data. One little piece of it is shown on the right here is a little graph showing one position. And this is, as an antibody engineer myself, it’s super valuable to see this granular detail of all the data. For example, this position here seems to be an important position for improved binding. But you have a number of candidates that can be used for improving the affinity. You have a lysine, you have a methionine, and you have an arginine. My personal preference would be for the lysine because there’s less drama with that, but methionine, you know, has a developability risk. So this gives you this option to do multi-parallel engineering of your antibody and to dodge potential problems. Here’s a summary of this program, so this GPRC5D that was optimized in this multi-parametric way. We improved the affinity, some developability issues or concerns, a species cross-reactivity problem we had, it was humanized and improved the IP, which I’ll describe. And we changed 25% of all the CDR residues. So with this multiple combination thing, you can really make a lot of improvements all at once. Here’s a look at some of the affinity data. So here’s a set of 400 clones that we created combinations of various numbers from 2 to 6 mutations and screened them for improvements in binding. You see that more than 95% of them showed improvements over wild type. And some of the best ones were 20 to 30 fold improvement in affinity. And it was already a pretty good antibody. So this is, you know, this is now way down on the picomolar affinity. But all these mutations were fed into AI machine learning to create the optimal combinations for all the parameters. And one other interesting story from this was, in general, when we use chickens, we get very good species cross-reactivity. We often get antibodies that bind to human, cyno, and mouse. But it can be a little quirky. It can only take one amino as a difference to disrupt a species binding. And in this case, it was. Like this antibody bound human and mouse, but didn’t happen to bind cyno very well. We are able to find a combination mutation that while it didn’t also improve the affinity to the humans, on the graph on the left, you see the parental antibody in the gray line and the improved affinity on the blue line. But for the cynomolgus, the parental was almost dead on cyno. But we were able to make a variant that had almost equal affinity to the human, which enabled us for the preclinical studies. The last aspect of this engineering that may not be on everyone’s radar is around intellectual property. A few years ago, there was an important ruling, this case between Amgen and Sanofi, and the result of which meant that you could no longer get these broad functional claims for antibodies. So now IP for antibodies is very narrow, easier to work around. So we proposed a solution last year, and we published this in Nature Biotech, where we described this approach of CDR scanning as a way to give you broad claims for patents. So it goes like this, is that you create all the permissive mutations, and you end up with all the, a broad genius of all the possible antibodies that combined. And you put those in the claims. And we tested this with the USPTO because we filed a patent around that GPRC5D antibody using all this data. This data fulfills the enablement and written description requirements for patents. And we’re very pleased that this year that we had our claims accepted. And importantly, they not only included the claims around the single substitutions, because we provide the data for single substitutions, not the combos. But they accepted all the combinations as well. So we have a very broad claim around this asset, where basically we own all the space around all the variants of this antibody. Okay, last section I want to talk about. You want to put the next poll up, Kristen?

Ross Chambers 36:29
All right, the last section I want to talk about, so you’ve got your antibody, you’ve engineered it and it’s optimal, but now you have to get through pre-clinical and clinical. And there’s many bumps along the road there too that I want to talk about. One of these key things is around specificity of antibodies that I want to talk about in this last section of this webinar. Do we have the poll results in?

Kristen McPike 37:03
Approximately what percentage of antibodies do you think are nonspecific? Five to 30% is the highest response. And again, you can see these responses in the chat, everyone.

Ross Chambers 37:14
Yeah, I think, there’s a sort of a general thought that antibodies are amazingly specific. And they are compared to small molecules. It’s one of the strengths of antibodies over small molecules is they’re incredibly specific. But they are still molecules. And at the end of the day, they do have the ability to be to be non-specific. One of the things that we have done over the years is we have a platform that I’ll describe in a moment that measures specificity. And we have tested hundreds and hundreds of molecules, pre-clinical and clinical molecules on this platform for specificity. The surprising thing is that one in three antibodies show off-target binding. The reason that’s surprising, well, surprising to most people, is that because antibodies are well known for specificity. And we published this a number of years ago, a couple of years ago in MABs describing this. So what do you do about this? This is a major problem for safety and could cause major bottlenecks in pre-clinical and obviously in clinical. So a number of years ago, we developed this platform called the Membrane Proteome Array, or the MPA. And this platform is a collection of essentially all of the membrane proteins and secreted proteins, things that antibodies will interact with when injected into the human body. And we screen these, express and screen the antibodies, your antibody, this is a service, against all these proteins. And we quantify the binding. We use flow cytometry and ELISA. And then we read out for that to show, is there any off-target binding? So the case on the right example, you see in blue dot, it’s binding to the target, but unfortunately it also has an off-target binding. There are some other platforms out there, but the important thing I want to stress here is this is the only platform that uses native proteins, unfixed cells, to screen at each step. That’s important because that’s what’s relevant to the human body, is native proteins, not fixed proteins. And it’s also the only platform that uses quantitative assays for measuring that interaction. So it’s more actionable data. So this very popular platform.

We’ve been using it to look at current clinical antibodies, things in clinical testing or FDA approved antibodies. We looked at a set here of 35 FDA approved antibodies that also had publicly available TCR data. You may know the current standard in, or the grandfathered method in for testing for specificity, is the tissue cross-reactivity assay. This is an IHC approach, so denatured antigen, a denatured protein, and it’s tissues. You don’t know which protein is off target to. So this is a bunch of tissues with an IHC, and you look for reactivity. So we took those where we had the results. We took those 35 antibodies and we ran them on the MPA. We found seven of those antibodies that had off-target binding and the rest were clean. If you look at the TCR results, though, the seven that we found off-target binding for, the TCR said were clean. And they also found a bunch that had unexpected cross-reactivity that we said were clean and some that were inconclusive. So what’s interesting is when you look at now the performance of these antibodies in the clinic, those antibodies at the top, the seven antibodies that TCR said were clean, but the MPA, our MPA had off target binding, well, there’s some red flags there. Four of them had black box warnings. Two of them had severe side effects, and one had a brain off-target binding and it had no toxicity and you wouldn’t expect that for an antibody. So, and the ones that TCR showed had some reactivity. They were clearly false positives because we saw they were clean and also there’s no clinical safety issues. So we thought this was interesting that seemingly the MPA better correlates with safety than the classic TCR result. Now, another way of looking at this is how does this reactivity change as you move from preclinical through to discovery? Because obviously, as you go through testing, safety issues that will pop up will take a drug out. So if you look at approved antibodies, this actually had the lowest rate of off-target binding, which is what you would expect for a molecule that successfully got through all the testing and was found to be safe and effective. So 15% of those had off-target binding. If you go down to now lower to earlier stages, phase two and three, that goes now up to 20%, and withdrawn drugs, 22%. If you go even earlier to preclinical leads, it’s 33%. So you can see there a trend that MPA is predicting safety issues that seem to correlate with as it progresses through clinical studies. Okay, so that is the story. I just want to summarize here through the lens of one of our assets from our pipeline, which is a Claudin 6 MAb. The molecule that’s in the clinic now is CTIM-76. So this was a complex membrane protein, membrane embedded, highly conserved. And had an additional challenge that not only it had a very close member, Claudin 9, that was expressed in normal tissue. This is a cancer target. We made it an antibody asset, made it bispecific. This is now in clinical trials. We developed this antibody using the discovery platform I described. It was using VLPs and DNA. And this is some time ago now, it would be RNA today. Got a very diverse collection of antibodies out that really bombed that surface of that protein with many diverse epitopes covered. We were looking for that magic antibody that was incredibly specific, are really focusing around a single amino acid it was discriminating from to give it that specificity. We used the MPA to screen candidates for specificity, resulting in CTIM-76 that is specific to Claudin 6, does not bind to Claudin 9 or any other protein in the proteome. And then, of course, we had to fine tune it to make the final candidate. We did antibody engineering, as I described with Paratope-PLUS, and ended up with a very potent candidate that’s now in phase one.

So that’s our story. If any of this interest you, a lot of the stuff is available to you. The Membrane Proteome Array, protein engineering, epitope and paratope mapping or Paratope-PLUS, and antibody reagents is available as reagents and services. And the antibody reagents and VLPs are available through Cell Surface Bio, our sister company. If you want to develop antibodies with us through a partnership or target identification through the MPA. And we have a number of therapeutic antibodies available for licensing. And if you’re specifically interested in the 100 plus targets I described today that we’re developing, as reagents, these are available through Cell Surface Bio, and as therapeutic assets that are available for licensing through Integral Molecular. Okay, so that’s the discussion today, and I’m happy to answer any questions that anyone may have.

Kristen McPike 45:02
Thank you so much, Dr. Chambers. So we do have a couple of questions. I will start with, when performing the CDR scanning, do you include all the CDRs or just focus on HCDR3?

Ross Chambers 45:18
Yeah, good question. We’re flexible, we can do whatever you want, but generally, we do all the CDRs. The thesis is that if you want the comprehensive dataset to really give you no blind spots for your engineering. But, you know, it’s a flexible program. The positions that you want mutated, we can accommodate whatever you’re interested in. But generally, we do all of them. The method is very efficient for mutagenesis, so it’s very easy for us to generate 1000 mutations. So it’s not that big a lift.

Kristen McPike 45:54
Thank you so much. And then following that, following the multi-parameter lead optimization strategy, have you ever seen a slight change in epitope, i.e. epitope drift?

Ross Chambers 46:06
That’s a good question. No, I haven’t got a particular example like that, but I’ve seen examples in the literature of, you can definitely shift the, make new contacts on one side and potentially make some alterations. But yeah, no, I haven’t got any particular internal stories, but I know there are some examples in the literature.

Kristen McPike 46:27
Thank you. Then 2 pretty similar questions about chickens being transgenic or humanized. Could you clarify whether the antibodies are generated from transgenic chickens or whether they were first humanized and then undergo affinity maturation?

Ross Chambers 46:45
There are transgenic chickens out there, but we prefer using wild type chickens. So we’re using wild type chickens. And the reason is, wild type chickens generally give very strong responses and very diverse responses. We see that all the time. And humanization is, obviously, a step you have to do for this approach, you have to humanize the antibodies. But it turns out that humanizing chicken antibodies is very straightforward. Chickens, like rabbits, use a single germline. They don’t have all that wide panel of germlines, and it’s a well-behaved germline for developability. So a simple CDR graft onto a human framework, and the frameworks are very highly conserved to humans. So a simple CDR graft is very straightforward and often results in minimal or no affinity loss. And there’s one possible back mutation that we use in general, but yeah, very, very easy to humanize. So we’ve taken the approach of using wild type animals for the strength of stronger immune responses and diverse epitope responses, because we know that the humanization is very straightforward downstream.

Kristen McPike 48:02
Thank you so much. Next, what kind of expression system do you use to make your VLPs? And what kind of target protein slash VLP concentration do you need for successful phage panning against the VLPs?

Ross Chambers 48:18
We have a couple of different systems for making VLPs. We use HEK 293 cells a lot of the time. They’re great cells for producing high levels of proteins. But we also have an avian cell line. And in many cases that we find that protein that will struggle in human cells will express great in an avian cell line, presumably it’s bypassing some negative regulatory system. So, we have those two systems for producing VLPs. And the second part of the question was it the level of expression required for success?

Kristen McPike 48:57
Wwhat kind of target protein slash VLP concentration do you need for successful phage panning against them?

Ross Chambers 49:02
Yeah, that’s an important one. We have learned that number by hard work over many years. So we quantitate the expression of the protein in the VLPs and we aim for at least 50 picomoles per mg. And we have shown that if you can get 50 picomoles per mg, you have a very high success rate in phage panning. The higher the better, obviously, but you need at least 50 to have a very high chance of success, and below that, the success really drops off steeply.

Kristen McPike 49:40
Thank you so much. And just a reminder, everyone, to add any additional questions you have. I have a few more to ask. For CDR scanning, once you’ve scanned all the single mutations, how do you know which ones to combine? Do you have to test every pairing?

Ross Chambers 49:57
Oh, well, that’s where AI comes in. Ten years ago, that would be a huge task, right? How to combine all these mutations. We’re trying to optimize multiple things at once. So we’ve built a platform that allows us to optimize all of those parameters at the same time. So we’ve created a system, but we’ve feed all the 1000 mutations, and you create those 1000 mutations. You might have a bunch of screening data that depends on how many assays you want to run the 1000 panel across, feed it in and run it through and it will come up with all the combos, combo mutations for us. Otherwise, that could otherwise be a gargantuan task to do all those combinations. But this is one of the great powers of AI, is to speed up the combination design.

Kristen McPike 50:52
And then a follow-up to that, how are you deciding which mutations to test?

Ross Chambers 50:58
Which ones, we’re testing them all. So yeah, that’s an important distinction. We’re testing all the mutations. So you know the complete data sets. You can then combine things with faith that you know what the outcome’s going to be.

Kristen McPike 51:19
That makes a lot of sense. So then for MPA, how do you ensure proteins are expressed?

Ross Chambers 51:27
Yes, our library of plasmids is built on many years of experience in expressing proteins. And yes, so we’ve got optimized constructs that express well, and importantly, we quantitate them all. So when we build the arrays, we quantitate the expression of all the proteins to ensure that they’re expressed.

Kristen McPike 51:49
And then the last question, at what stage should I screen my therapeutics for off targets?

Ross Chambers 51:57
People are testing them at multiple stages. I think as soon as you have a panel of antibodies that you’re interested in advancing, maybe you’ve just got out of discovery and you’ve got maybe half a dozen candidates that you’re interested in that fit all the other requirements, I would definitely want to test it then. That’s what we do. Because, you know, off-target binding is a pretty important criterion to know about before you advance it. And it also informs what you do for engineering. So you can fix these. Having an off-target isn’t necessarily the end of the world. You can fix it. You can either mutate it out, or maybe it’s an off-target that’s not super important, but you want to know about it. Obviously, after you’ve engineered, every time you change the antibody, you want to test it. So the final candidate that’s going to advance into preclinical and GLP tox, when it gets expensive, you definitely want to test it then too. And we have a different version of MPA, a more sophisticated version, or more detailed for the IND to give you more information there. But yeah, I would test it at least after the discovery panel and then your final candidate and maybe in between if you’re doing a lot of engineering to help guide you.