HOW SPECIFIC ARE ANTIBODY DRUGS?

Kristen:

Hello, thank you for joining us. I’m Kristen McPike, Program and Alliance Manager at the Antibody Society. Today’s webinar is one in a series designed to inform and educate our members in the broader scientific community about topics related to antibody discovery and development. Our expert speakers, Diana Norden and Chris Raghavan, will be discussing How Specific Are Antibody Drugs: Revealing Insights from a New Generation of Specificity Assays. Please note the webinar is being recorded. Please add any questions to the Q&A box in the viewer and the questions will be answered at the end of the webinar. Without further ado, I’ll turn the show over to our speakers.

Diana Norden:

Thank you, Kristen. Thank you to everyone attending our webinar. Today, we’ll be answering the question of how specific are antibody drugs. For our webinar, I will be starting us off. I will go over some key definitions and mechanisms of antibody off-target binding. Then I will discuss some key features of protein arrays as they are a new and emerging tool to better assess antibody specificity. Then we’ll answer the main question for today, which is how specific are antibodies? We’ll look at two datasets of preclinical as well as clinical antibodies and see the rates of off-target binding there. Then I’ll hand it over to my colleague Kris, who will discuss how Integral Molecular’s Membrane Proteome Array can help you in your specificity profiling needs. go over the workflow for MPA projects, and give some exciting updates in how we are working with the FDA to have the MPA qualified as a drug development tool. And then of course we’ll end with questions from the audience.

Let’s get started. Antibody therapeutics represent a large and growing pipeline. This graph here shows the number of antibodies entering first-in-human studies over the past 14 years. You see a very large increase in the last five years or so where there’s more and more antibodies entering clinical trials. And antibody therapeutics are increasing even more so considering the advancements in CAR-T cell therapeutics where antibodies are used as the targeting arm in the CAR-T cells.

Antibodies are so popular and valuable as targeted therapeutics because of their excellent specificity profile when compared to small molecules. This word cloud here on the left shows some key vocabulary that’s often used to describe antibodies such as Exquisite Specificity, Absolute Specificity, Precise Targeting. And a lot of this is because antibodies are often generated from immunizations with a very specific antigen. So in theory, they should only recognize that antigen. But we now know, as well as others in the field have started to recognize, that antibodies might not be as specific as previously thought. We recently published a paper last year where we presented some data to show that one in three antibodies actually have off-target binding. And this is the data that I’ll be presenting today. If you have more questions about the data or are interested in reading further, then the reference is shown here on the right.

What are some of the mechanisms by which antibodies can have off-target binding? Off-target binding can either be polyspecific or polyreactive. And these are two distinct phenomena. Polyspecific off-target binding is a CDR-mediated binding directed to a specific epitope, it’s just on the wrong target. Versus polyreactive off-target binding, it’s more of a general stickiness. This can be either CDR-mediated or non-CDR-mediated. These binding interactions are often due to hydrophobic or charge residues, and this is truly non-specific binding. It can be to proteins, carbohydrates, surfaces, or extracellular matrix. Polyreactivity is a big risk for developability in antibodies, but it’s not so much of a risk in safety. And polyreactivity has been well characterized in numerous other papers. So for today we’re only talking about polyspecific off-target binding, which again is a specific binding interaction just to the wrong epitope.

So then if we go further, how is antibody polyspecificity possible? There have been three different mechanisms reported. There can be molecular mimicry were critical epitope residues are mirrored in an unrelated protein. This can also occur through CDR plasticity or differential engagement. This is where the paratope can either be flexible to accommodate more than one antigen or there can actually be multiple paratopes in a single antibody. These two last examples are less common, but they have been reported. In our own studies, we’ve actually come across an interesting example of molecular mimicry. In our own lab, we generated a panel of antibodies against a target called GLUT4 or SLC2A4. And when we profiled this antibody, we found a really unexpected off-target binding interaction to a protein called Notch1. GLUT4 and Notch1 are not family members, they’re very unrelated, and if you look at the structure, they’re also not very similar at all and they share less than 7% in sequence identity.

But we also epitope mapped this antibody and we found that the antibody binds a loop-constrained epitope on GLUT4. So then when we went back and looked again at the structure and sequence of Notch1, knowing now what off-target is, we’ve actually found a very similar disulfide constrained loop with similar amino acids on Notch1. Even though the total protein structure is very different, this particular epitope was actually mirrored with Notch1. So it shows that it’s really important to profile specificity comprehensively and not just by family members. This type of polyspecificity can have really important implications, or severe implications in safety and can actually lead to toxicity. I have two examples that I want to show where off-target binding directly caused toxicity. One, is the story of camrelizumab. This is a PD-1 antibody. It entered clinical trials, and in Phase 1, patients actually presented with severe capillary hemangioma. This was very unexpected because there are multiple other PD-1 antibodies being tested in the clinic, but none of them had this same adverse event.

Developers had to profile the binding of camrelizumab further and they actually found an off-target binding interaction to a VEGF receptor. And through their investigations, they were able to determine that the adverse events were strictly caused by not only binding to the VEGF receptor, but also activation of the receptor. Through this investigation, clinical development was allowed to continue, but camrelizumab now has to be administered, or co-administered, together with a VEGF receptor inhibitor in order to account for that off-target binding interaction.

In another example, we see where off-target binding to a secreted protein can also lead to toxicity. Here’s an antibody called ABT-736. It was being developed for Alzheimer’s disease and it targets amyloid beta oligomers. But in monkey toxicity studies, there were severe adverse events noted. And again, through investigations into this binding profile, the developers were able to determine that it has off-target binding to both the monkey version and human version of platelet factor 4. And this was then able to explain the severe adverse events because PF-4 is known to be involved in heparin-induced thrombocytopenia in humans, which was consistent with the adverse events that had been seen in the monkeys. Development of this antibody was completely discontinued. These are two examples of really late stage molecules where safety issues arose directly because of the off-target binding, and it highlights an increased need for better tools to really evaluate specificity to improve safety.

Let’s look at the guidance documents for specificity testing. It is a requirement for first-in-human studies, the guidance document relevant for antibodies is the points to consider in the manufacture and testing of monoclonal antibody products for human use. This guidance document states that specificity testing should be performed using tissue cross-reactivity studies, or a TCR assay.

What is a TCR assay? It is an immunohistochemistry based screening assay where an antibody is applied to over 30 different human tissues, and antibody binding is then evaluated. When binding is consistent or expected, so consistent with the known expression profile of the target, then really no off-targets are suspected. Versus if you have unexpected staining, that’s when you can expect some off-target binding. While this can give some interesting information on which tissues your antibody might bind, using this assay to truly determine antibody binding specificity comes with a lot of limitations. For one, like I just mentioned, it only really tells you what tissues an antibody binds to, but it can’t tell you what protein it’s actually binding to. So even expected staining in a particular tissue can give you false assurance that there’s actually binding the correct protein and it might not be the case.

Tissues that are used in the TCR assay often undergo a lot of processing in order to preserve tissue integrity, such as being frozen, maybe dried, and applications of fixatives. But unfortunately, all of these can alter epitopes. There’s also a lot of variability and expression levels of various proteins within the tissues. There’s only three donors used typically, and there’s no way to control protein expression levels. There’s variability naturally, endogenously, in protein expression levels in tissues. It’s also a qualitative assessment, has low sensitivity, and can be prone to both false positives and false negatives, and it’s often incompatible with some of the newer biotherapeutic modalities.

What we see is that a lot of these technical limitations in the assay are actually reflected in the real world use of the TCR assay. This publication here to the right reports on a survey that was given out to developers at 26 different biotech and pharmaceutical companies. The developers were asked about their experience in TCR, how they’re using the data and how they’re incorporating it into their development pipelines.

Some interesting results was that over 90% of respondents said that the TCR had actually not impacted the design of their tox studies. Over 80% said that they had never seen a correlation of TCR data to preclinical or clinical toxicity. The survey of toxicologists reflects a lack of trust in the TCR assay to truly identify or de-risk antibody off-target binding, and it’s not giving any actionable results. They’re not really using the TCR results to progress or design the continued development. We know now, and the survey also reflects, that more and more developers are starting to use alternative methods besides tissue cross-reactivity in profiling the specificity of their antibodies.

Interestingly, these limitations of TCR are actually also reflected by the FDA. Even in the original guidance documents, in the Points to Consider guidance documents from ’97, there’s an interesting statement there that says that appropriate and newer technologies should be used instead of TCR as they become available and validated. And then last year, in January of last year, FDA put out new guidance documents for CAR-T development. And here, in addition to TCR, now protein arrays are also named as an appropriate specificity testing method to evaluate the binding of the antigen binding domain of the CAR. Now, this is the first time where FDA is really delineating an appropriate alternative to TCR, namely protein arrays.

There are two types of protein arrays. There are traditional peptide protein arrays, they’re often called protein microarrays or spotted protein arrays. This is where proteins are expressed in a non-mammalian system, cells are lysed, and those proteins are spotted onto slides, and this gives rise to a lot of surface interactions and denaturation of proteins, and they have not been shown to be reliable when it comes to profiling specificity of antibody therapeutics. And then now, there’s a new type of protein array called a cell-based protein array. And here, plasmids are transfected into cells, and then proteins are tested for binding within human cells. And this is the type of data that’s now being used for regulatory submissions. So from now on, we’ll only be talking about cell-based protein arrays in our webinar.

At Integral Molecular, we have developed our own cell-based protein array called the Membrane Proteome Array, or the MPA. This is the technology platform that we use to profile specificity of antibodies. This is the platform that we use for all of the data that I’ll be sharing in just a few minutes. In the MPA, we have over 6,000 human membrane proteins. They’re expressed in their native state within human cells. They’re arrayed in a 384-well plate format to allow for high-throughput screening. We use flow cytometry to quantify binding, and we always use unfixed cells, and this way, we’re able to test specificity across the full membrane proteome and identify precisely which proteins an antibody binds to. And I’ll show some key features of this platform, as well as cell-based protein arrays in general.

One of those is the careful curation of the library. This is the list of proteins that are included in the array. When we designed our MPA library, we actually designed it based on FDA guidance documents. So we looked in the guidance documents and identified the tissues that should be tested in a TCR assay, and then from those tissues, we used RNA sequencing and various online databases to identify precisely what membrane proteins are present or expected to be present in those tissues. So this way, the MPA library is a true reflection of what is to be tested in a TCR assay. But keep in mind that in our array, proteins are being overexpressed in cells to ensure that they’re actually being expressed and that they’re actually being tested for binding, versus in a TCR assay, you can’t control protein expression levels, and a lot of proteins might be expressed very lowly or maybe not at all.

And so, our MPA, like I said, it has about 6,000 proteins, it covers 94% of the membrane proteome. It also includes additional heterocomplexes. And then recently, we now have developed an additional library, a Secreted Proteome Library. This includes an additional 1,200 secreted or soluble proteins. And together, these two libraries now represent truly the most comprehensive testing of specificity across the human proteome. So by screening against both of these libraries, you can test for binding against any protein that an antibody therapeutic given to a person could potentially encounter in that person.

Another key feature of cell-based protein arrays is their compatibility with screening of various biotherapeutic modalities. On our MPA, we screen numerous scFv formats, we screen multi-specifics, VHHs, and even peptides. And a lot of this is because of the truly high sensitivity that the flow cytometry assay provides. When screening, it’s very important to always screen on unfixed cells. We don’t use any fixatives in any of our processes, whether it’s the screen or in the validation assay, to really ensure native epitopes. Applications of fixatives can alter epitopes, it can lead to cross-linking of proteins and even denaturation of epitopes, both of which can increase the chance of a false negative, which could have really important safety implications, based on the examples that we talked about earlier.

We actually have a good example of this. We screened a molecule targeting CD19 on both unfixed and fixed cells, and in both cases, the antibody bound the target, CD19, but we also identified an off-target binding interaction to an off-target only in the unfixed cells. In using fixed cells, we were not able to identify this interaction, which, again, really is why we always routinely use only unfixed cells in all of our experiments.

So back to the main question for today, how specific are antibodies really? Now that we have these new tools to truly assess antibody specificity, what is the rate of off-target binding? We have been very interested in this for a while, but there hadn’t been any large-scale studies to really look at the rates of off-target binding through development, and also, nobody has really looked at the effect of off-target binding on clinical development. So we performed two large systematic analyses of two different datasets. We had a dataset of preclinical antibody candidates, and then an additional one of clinical stage antibodies.

For our preclinical molecules, what we did was perform a retrospective analysis of customer-provided samples that have been screened on our MPA. This is mostly lead candidate molecules from various biopharmaceutical companies. Companies often send us a panel of lead candidates to help them in their lead selection process. And we looked at a total of 254 antibodies, and out of them we found off-target binding in 83 of the molecules, which comes to an off-target rate of 33%. I should note that here, we are not counting common Fc binding, such as to Fc gamma receptors, that does not count as polyspecificity in this analysis.

Of the antibodies that had off-target binding, most of them had off-target binding to one, maybe two proteins, some of them had off-target binding to up to five proteins. So this rate of off-target binding of 83 really shows that off-target binding might be more common than previously thought, and it also shows the true benefit of testing lead candidates sooner in development. So by testing a panel of lead candidates, you can select a lead that’s truly clean and doesn’t have any risk for safety issues due to off-target binding later in development.

And then, we were also very curious about clinical stage antibodies, so antibodies going into people, is there still off-target binding present in these antibodies? So here, we performed a prospective analysis. We produced biosimilars of clinical trial stage FDA-approved and withdrawn antibody drugs, and then also screened them on our MPA platform. We screened a total of 83 biosimilars that we produced, and still here, we found off-target binding in 15 of these molecules, which comes to an off-target rate of 18%.

And then, what we did was we divided them into their appropriate categories, so looking at approved antibodies, those in phase 2 and phase 3, and then in withdrawn. So even in approved antibodies, we were still seeing off-target binding, so we had a total of 40 antibodies in the approved category and six of them showed off-target binding, which comes to an off-target rate of 15%. And in phase 2 and phase 3, it was 20%, and in withdrawn, 22%. So we were truly detecting poly-specificity at all stages of clinical development. And when we looked at the different rates in off-target binding, it suggests that poly-specificity truly is a risk in development and a major factor in drug attrition.

And our data then challenge this assumption that antibodies are inherently specific, and also underscores a critical need for improved safety testing or for improved specificity testing. This is very surprising, because like I had just said, all molecules are required to be tested for specificity before entering clinical development, so why are we still seeing off-target binding in this dataset? This comes back to the issue of the tissue cross-reactivity studies, which likely missed some of these interactions. So that was our question, were these molecules tested by TCR, and did the TCR miss these off-target binding interactions?

So what we did was for the approved antibodies that had off-target binding, we pulled their BLA applications and looked at what the TCR results had said, and then we were able to find some discrepancies between what we found on the MPA to what the TCR had reported, and I have some examples to show for that. So here is an antibody that targets a plasma membrane protein that is primarily expressed on lymphocytes. We screened it on the MPA, like I just said, and here we found that the MPA identified the target, but it also identified two off-targets. When we looked at the BLA application, it said that the TCR staining was consistent with the known expression profile of the target. So here’s an example where the TCR study simply missed two off-targets. We’re not sure why that was, if it’s a true false negative, because of maybe staining of the tissue, or low sensitivity.

In the other example, here’s an antibody that targets a plasma membrane protein expressed on myeloid cells. Again, the MPA identified the target, but also identified an off-target. And interestingly, this off-target is actually a membrane protein expressed on the same cell type as the target, so not surprisingly, the BLA application stated that the TCR staining was consistent with a known expression profile of the target. So this is an example where TCR has a limited ability to identify some off-target proteins. The target and off-target were expressed on the same cell type so that there’s no way that the TCR would be able to distinguish binding between the two proteins.

And in the last example, we have an antibody that targets a protein that is expressed both on the plasma membrane, as well as in intracellular vesicles. This is an oncology target that is upregulated in numerous cancers, but in normal healthy tissue, it has low and widespread expression in numerous tissues. Again, the MPA identified the target, but also identified an off-target for this molecule. The BLA application stated that the TCR staining was consistent with the known expression profile of the target. And this is another example where TCR has limited ability to identify some off-target proteins. Because the target is expressed in numerous tissues, it’s likely that the widespread expression made it difficult for TCR to pick up any additional staining from the off-target.

So by looking at these case studies, comparing MPA data to TCR data, it’s pretty clear to see that some of these newer technologies, like our MPA, truly enable better specificity screening. Just to summarize some of that here, our MPA is able to precisely identify on a protein level any off-targets, versus the TCR only tells which tissue an antibody binds to. The MPA uses native conformations, we always use unfixed cells, we have a more comprehensive membrane protein representation, we also have secreted proteins now as well, versus there’s a lot of variability in expression in tissues used in TCR. MPA is highly quantitative and is high sensitivity, versus the qualitative assessment that a TCR uses. And there’s also differences in compatibility; we screen all the different biotherapeutic modalities on the MPA. It’s also a high throughput assay, our projects only take about four weeks to complete, and it’s significantly less costly compared to a TCR assay.

So now, I’ll hand it over to my colleague, Kris, who will go over some additional details on our MPA platform.

Kris Raghavan:

Thanks, Diana. So hi, everyone. Kris Raghavan. I am the Senior Application Scientist for the MPA. Essentially, I work very closely with Diana and I also work closer, I would say, with the customers that we have, mainly as a point of technical support as we move from planning stages of your project all the way to the post-project discussion of data. I’m going to take you through a little bit about what projects actually look like and get into a little bit more of how it’s being used.

As Diana very clearly demonstrated, there is a strong need for specificity testing during preclinical development to properly de-risk your campaigns. But we understand that our customers are coming to us at drastically different points in their development pipelines, and so we have designed services to really suit those different needs. Now, for those earlier on in their projects that may be coming to us looking for a lead selection from a cohort of molecules, we have our standard MPA service. But if you already have one or two leads and are looking to screen for off-targets before filing an IND submission, we, specifically Diana, will author a customized report package which is ready-made for you to submit directly to a regulatory body. This MPA+IND package also has prioritized timelines, so you’re not just getting more value in the data and report, but it’ll also be completed in about half the time.

So what does an actual project look like? Our studies are segmented into three distinct stages, starting with the assay setup where we are going to titrate your sample against a panel of controls performed in two parallel cell lines aimed to identify a single optimal screening concentration or condition. Once that is selected, we will screen your samples against our full library of proteins and any hits identified above a certain detection threshold will be advanced to the validation stage. At this validation stage, we’re going to again perform a titrated binding assay, which will confirm whether a protein interaction is reproducible or whether it was a false positive of the screen. This is also a stage where we can include an isotype or monovalent control to better understand what component of your molecule might be responsible for any off-target binding that we observe. Now, at the conclusion of this study, we will generate a report with all these results as our final deliverable.

Now, let’s say we did identify an off-target binding interaction and it’s your favorite molecule and you really need to know more about this interaction and characterize it further. We have developed a Risk Analysis package to do exactly that. This package includes a more rigorous set of experiments designed to better quantify the EC50 to get a better idea of binding strength, and the Bmax for protein expression and binding, and even subcellular localization to test your molecules interactions with this off-target protein. We will then put together our highly detailed risk analysis report aimed to provide a comprehensive picture of your molecule’s relationship to this off-target protein.

After all of this, we still have more offerings if you really want to continue to learn more about your test article. We can generate an upgraded IND-ready report for any molecule that has completed a standard MPA study and also suggest some follow-up services to really better understand your asset. That includes Integral Molecular Epitope Mapping and Protein Engineering services. Externally, we have Lonza Biologics we can recommend, which they can assess the potential for immunotoxicity from any molecules that do have off-targets, and Evitria, who specialize in antibody production services.

So now let’s jump to look at some case studies that really highlight how our data has been used most effectively, I would say. Here we see a great example of how the MPA has been able to aid lead selection. These assets were designed to bind CLDN6, but it was very important to identify the potential for polyspecificity, especially since CLDN family members are so similar. CLDN9 differs from CLDN6 by only about three extracellular amino acids. As you can see here on the right side of the slide, some of these molecules did demonstrate a fair amount of polyspecificity. But out of this cohort, we were able to identify one that was specific only to CLDN6, and that was advanced further for development.

Now, another affiliate neighbor that we have here, Cabaletta Bio, has also been incorporating us into their development processes for a while now. With them we have screened some novel modalities for cell therapies and they have used our data in successful IND submissions, which we are especially proud of. What is most interesting about this project, I’d say, is that we did identify an off-target interaction for this molecule. However, with this knowledge, Cabaletta performed some follow-up studies to prove in their IND submission that this potential interaction would not be a safety risk in humans. Importantly, they were successful in their IND submission without tissue cross-reactivity studies.

To speak more to what customers are telling us as we interact with them and do projects, over 50% of polled customers plan to use our data in IND filings. As I said before, the use of our data in IND submissions has been successful without TCR studies. Our IND submissions have been accepted not just by the FDA, but the EMA and NMPA. I would say most importantly, several of our customers report that in conversations with the FDA, they were requested to include new array-based specificity testing like the MPA in their filing.

Speaking of the FDA, we are midway through the FDA’s ISTAND program, which was created to validate new technologies like the MPA as a qualified drug development tool, or DDT. MPA data has already been routinely used in IND submissions. However, with this new qualification, the FDA will have a full understanding of our technology, its context of use, limitations, our quality control and record tracing, thereby, I would say, taking the pressure off of our customers to really explain those aspects of our platform in their submissions. The FDA will already be aware of it. We are really excited to say that we will be submitting the final full qualification package this year to hopefully receive a decision in 2026. Of course, dependent on the FDA, so please stay tuned for more news as that develops.

We are proud to be trusted by over 150 companies since we began the MPA and have been incorporating the MPA’s data into their drug development programs and their publications. Altogether we’ve screened over 2,000 molecules, with more coming in every day. So hopefully we’ve made a strong case for the need for better technologies for specificity testing. There are many reasons why the MPA should be attractive to those looking to de-risk their drug discovery and preclinical development programs. Our platform is compatible with a wide range of therapeutic modalities, and we are the only company screening against native, unfixed proteins. Critically, our customers have been successful in their IND submissions without TCR and we are optimistic that the FDA will qualify us as a validated DDT.

Just to conclude, to give a bit of a personal representation of our company, we are here at Integral Molecular in Philadelphia. A lot of people hear small biotech and think it’s a startup, but actually no. We’ve been established in the industry for over 20 years. We started as a spin-off from the University of Pennsylvania, and since then our expertise with membrane proteins and antibodies has really brought us to where we are today. So with that, I would like to thank you so much for your attention, and we would be more than happy to take all the questions that I think have been flooding in. Thank you very much.

Kristen:

Thank you so much, Diana and Kris. So yeah, I’ll get the questions kicked off here in just a second. So first, I think this question is targeted to Diana. With regards to polyspecificity of preclinical and clinical stage antibodies, can you annotate them for their source, like the display library, in vivo generated, humanized from non-human species, etc.?

Diana Norden:

Yes, that’s a great question. Thank you for that question. This is an active project, we have a lot of interest in this field obviously, and that is something that we are working to characterize right now. I don’t have an answer today, but hopefully we will be putting out a new publication in the near future, so stay tuned for that as well.

Kristen:

Okay, thank you so much for that. Then next question, are you able to test efficacy against the off-target?

Diana Norden:

Also in relation probably to the data I showed?

Kristen:

Yeah.

Diana Norden:

Yeah, another great question. Like I said, this is an ongoing project, and so what we’re doing right now is we’re trying to answer specifically that question. So what we’re doing is we have pulled out some antibodies from our data set that had really unexpected adverse events, or in the case of the withdrawn antibodies, any that were withdrawn due to safety. Any of those antibodies that also had off-target binding. What we’re investigating now is if we can link that adverse event specifically to the off-target binding that we found. So that is where this project now is going, and I think we will have some instances where the off-target binding directly causes safety. I would expect so based on what we’re seeing, but again, this is an ongoing project.

Kristen:

Perfect, okay. Then next we have, if you observe undescribed off-target binding with an antibody in clinical use, do you have to report this to the FDA?

Diana Norden:

No. I mean, the clinical studies are being monitored very closely, so this is out of our hands. Also, we are producing a biosimilar, so we can’t guarantee… We’re not testing the drug compound here, so that is a bit of a caveat. I don’t think it will affect the specificity profile anything, but that is something to keep it mind. We don’t have an obligation to report this to the clinical trial that’s being monitored very closely already.

Kristen:

Perfect. Okay, thank you so much. A reminder to everyone, they can drop additional questions in the chat if you have any. There was just a general comment that says, “Great tool,” so just wanted to give you all that shout out. Question for Kris, how do you ensure proteins are expressed?

Kris Raghavan:

Yeah, great question. So our library, part of its quality control, we include a V5-tag on all proteins, and as we do quality control on our library, we’re taking regular reads of that V5-tag. This allows us to determine to what degree each protein is being expressed, flag anything that is either poorly expressed or having issues there, and that just gives us a better understanding of protein expression in our library.

Kristen:

Great. Okay, more questions are spilling in as we go. One more for Kris. Is the initial MPA screening also performed on unfixed cells?

Kris Raghavan:

Yeah, so everything we do is in unfixed. That is our brand, I would say. So the reason why we’re doing that is we want to be as physiologically representative as possible to in vivo conditions where how a drug might actually interact with a cell. So not only the optimization, but the screening and validation are all done in unfixed cell conditions.

Kristen:

Great. Okay. What is the average false positive rate of the array? Not sure who that question is going to.

Kris Raghavan:

Yeah, I can handle that. So, I would say it always comes down to a molecule-specific question. In general, we have a very good optimization step, which we use to identify what is the concentration and cell line that is going to give us the lowest chance of false positive in the actual screen. We have a predictive measurement that takes into consideration the background, non-specific binding against a negative control. And so we use that to… That’s what we’re really talking about when we’re optimizing the conditions for screening. When we do screen, I would say anywhere between five and 10 potential false positives can come up. That 10 is on the very high side. More likely it’s between maybe two to five. We have the ability to test those all on validation. We really rely on our validation to be telling us what is reproducible versus what was a false positive. In any high throughput screen, you’re going to be at risk for false positives, but that is the benefit of our platform, that this is all packaged together, that de-risking.

Kristen:

Thank you so much for that. Okay. Now, Diana, going back to you, can MPA data only replace TCR data for CAR T-cells?

Diana Norden:

Yes, I can clarify that a little bit. The guidance document that specifically lists protein arrays as a method to determine specificity is the CAR T guidance document. And then for antibodies, that question is better approached on a case-by-case basis, whether or not to use TCR. A lot of our customers will discuss this with FDA at their pre-IND meeting or their interact meetings, and then come up with a plan whether or not to use TCR. Based on customer feedback, FDA is very receptive to them relying on MPA data and skipping the TCR.

Kristen:

Okay. We’ve got lots more questions, so happy to keep going. How amenable is this platform regarding discovery? Is it feasible to screen whole libraries? (thinking about the hype around AI de novo antibody synthesis in the recent paper of the group, Sandeep Kumar and Peter Tessier 2024).

Kris Raghavan:

I guess I can just answer that a bit. So not only are we using our platforms for specificity testing, but we can also use the same technology to identify orphan receptors, things like that. If you know your molecule has a functional response but you don’t know what it’s interacting with, we can definitely use it there. Diana, maybe you want to speak a little bit more to AI use.

Diana Norden:

I am not sure. That is not my area of expertise. That is not something that we are using at the moment. But we will be sure to look up that reference. So, thank you for sharing.

Kristen:

Okay. What about off-target effects of antibodies is useful for other diseases or cancers, other than the primary target? Do you have any examples?

Diana Norden:

I don’t think we have an example of that because the antibodies are usually developed for a very specific target, an indication. The interesting thing when looking at off-targets is that often, they’re very just unexpected and they can’t be predicted. And so usually, customers will either deprioritize that particular molecule or engineer it to remove the off-target binding interaction. I don’t think, at least as far as I know, an antibody has been completely redesigned for the off-target instead of the target.

Kristen:

What is the difference between an MPA and MPA+IND in terms of the data that clients get, timeline, cost, things of that nature?

Kris Raghavan:

Yeah. So yeah, they’re very similar. I would say they differ the most at the validation stage. We are going to include more replicates in that. Typically, IND molecules that are being used in our IND studies are cleaner, have less off-targets to manage in the validation stage. So because of that, we can provide more rigor to those studies. There are also some controls that are included by default as opposed to add-ons. So the isotype or monovalent arm controls that I mentioned are just by default, a part of that package. And of course the report itself is, I would say, triple the length, very, very detailed and provides a more thorough breakdown of the methods and use cases, things like that.

Kristen:

Perfect. Can you screen multiple molecules together or only one by one?

Kris Raghavan:

Yeah. So, I would say the best results come from single molecules being screened. We have tried pooled molecules and samples. Essentially someone provides us a tube, we’re going to test what’s in that tube. The problem with pooled samples is that when we’re doing our optimization step, we are really at the mercy of whatever sample is the dirtiest or has the most non-specific poly reactivity essentially is what we’re talking about. So if you have two molecules that are pristine and very clean and would screen very good by themselves, but you have a third one in that pooled sample, that it has a high degree of poly reactivity, the screen is not going to look great at all. And in fact, it would probably not return any data that would be actionable for the client.

Kristen:

And then one more for you, Kris. What’s included in the IND package? And can it be sent to the FDA as is?

Kris Raghavan:

Yeah. So I touched on this a bit, but the report itself has been tailor-made to send directly off to the FDA. We do this because we don’t want our customers getting bogged down in the minutia of our studies. We want to provide all of that to them so that they don’t really have to touch at all. We ask for more information from our customers in those IND projects. Things like background, things that we can put together essentially to give an executive summary. Things that are traceability, more aspects of traceability, like how the molecule was produced and its expiration dates, things like that. Everything that the FDA would be asking them for.

Kristen:

Perfect. Okay. Another one for Diana. If you choose not to do TCR, then how do you know which tissues your antibody binds to?

Diana Norden:

Oh, that’s actually a fairly simple question. There is a lot of information available online now. There’s a lot of databases that have this information. We routinely use the Human Protein Atlas because it’s continuously being updated. So for example, if we find an off-target, we often look at the tissue expression profile of that off-target on the Human Protein Atlas. So particularly for well-characterized proteins, that’s a great option. If an off-target comes up that’s not as well-characterized, then there’s also something called a tissue microarray that can be done. This is a simple assay, it’s pretty straightforward that can give you that information.

Kristen:

Have you had the opportunity of analyzing the specificity of antibodies designed to target intracellular epitopes?

Kris Raghavan:

Yeah. So we have, actually. Our library includes a pretty decent amount of intracellular proteins, things like nuclear, mitochondrial, endoplasmic reticulum, those types of membranes. We have a proprietary and nonchemical method of permeabilizing the cells to give the test article access to those intracellular compartments. We also do this as a way to control against receptor cycling. So there are customers that are very concerned with missing binding events because receptors are being cycled in or out of the cell. So, this is all done to make sure that we have access to the intracellular compartments. During the validation stage, we will do parallel conditions with permeabilized and non-permeabilized conditions, just to get a better idea of where that binding is occurring. I think that gives a better, more comprehensive understanding of these binding dynamics that our customers are very appreciative of, I would say.

Kristen:

Then can you screen viral modalities?

Kris Raghavan:

Yeah. We have successfully in the past, things like AAVs. We have the receptor in our library, so we have positive controls for them in a lot of contexts. I should add that in cases where a customer provides us a test article that has a target that is not in our library, we can accept plasmid DNA from them to transfect into our system. We have protocols by which we can add cassettes of custom transfected cells into our screening platform. We don’t accept outside cell lines, but again, we can accept plasmid DNA to transfect in our own hands.

Kristen:

Thank you. Okay. There’s one more question here, unless anybody else has anything they want to add. Besides the FDA, is there also a coordination happening with the EMA or maybe the ICH?

Kris Raghavan:

So yeah, I don’t believe we have an active filing with them, but they are typically, I would say, fast followers to the FDA. And we are working on getting those similar designations with those agencies.

Kristen:

Perfect. Thank you so much. Last call, everyone. If there are any additional questions, please add them to the chat. I’ll give you a minute or so to do that. Okay. Well, thank you both again, so much for this insightful presentation, Diana and Kris. The on-demand version of this webinar will be available very soon and a link will be sent by email to everyone who registered. Feel free to watch this or any other of our on-demand webinars when it’s convenient for you. And this will also be available on the Antibody Society YouTube channel. Thank you again and have a great rest of your day.

Kris Raghavan:

Thank you, everyone.

Diana Norden:

Thank you.