Accelerating Drug Discovery at a Pandemic Pace

0:00:00.0 Jayshan Carpen,: Hello, and welcome to this Nature Research Custom Media webcast titled, "Accelerating Drug Discovery at a Pandemic Pace Using High-Throughput Surface Plasmon Resonance in the Discovery of Therapeutic Monoclonal Antibodies." My name is Jayshan Carpen, and I will be your moderator. Today's webcast is produced by Carterra. We'll begin the webcast with a presentation from Denisa Foster, biochemist at Lilly Protein Biosciences. We'll then end with a Q&A session with Denisa and Daniel Bedinger, application science team lead from Carterra. To ask the speakers a question, just type it in where it says, "Ask a Question", and then press "Submit" at any point during the webcast and we will answer them today. And now, over to Denisa.

Denisa Foster and Daniel Bedinger, application science team lead from Carterra

0:00:49.9 Denisa Foster: Thank you for the introduction. Hello, and thank you for joining me today as I discuss the use of Carterra's LSA high-throughput SPR for the discovery of antibodies against COVID-19. Before I get started, I would like to introduce the Lilly Biotechnology Center. We are located in San Diego, California. And among other things, we specialize in antibody discovery and engineering. I'm part of the Protein Biosciences group where we focus on molecular biology, expression and purification, biophysical characterization including SPR and high-throughput characterization of the epitope. For today's talk, I would like to focus on presenting the timeline of the discovery of the first clinical antibody against COVID-19 and the tools that helped to select it. The experimental design using Carterra LSA in the discovery process, selection of the lead antibodies and applications of high-throughput SPR after the discovery of Bamlanivimab. Lilly began collaboration with AbCellera at the beginning of March 2020, and we were able to achieve first human dose only three months later.

0:02:16.9 DF: The urgency to bring forth a monoclonal antibody therapy can be seen in the timeline shown here as cases in the US and across the world were increasing rapidly, escalating to a pandemic in only a matter of months, very little was available in terms of therapeutics. Remdesivir was approved for use in hospitalized patients in May, then in several months, convalescent plasma was granted approval. Lilly's first monoclonal antibody was granted emergency use authorization in November, then Regeneron's antibody cocktail, and these two were followed by exciting approval of multiple vaccines. The antibody discovery process began with a blood sample from an individual recovering from COVID-19 after severe illness. PBMCs were screened for bind into the SARS-CoV-2 spike protein that was expressed on membranes of mammalian cells, as well as spike protein that was covalently bound to beads. From here, machine learning was used to analyze and rank the hits and over 2000 cells were recovered for sequencing. From these, 440 unique pair chains were identified, and 187 antibodies were prioritized for cloning and expression and high-throughput purification. And all of these steps were completed by day 17.

Important goals were to determine affinities of these antibodies to the spike protein

0:04:19.7 DF: Next, as part of the biophysical characterization, important goals were to determine affinities of these antibodies to the spike protein. We wanted to gain knowledge of which domain they bind, assess epitope diversity, and whether the antibody engages the same epitope as the ACE2 receptor, which is known to be the receptor by which SARS-CoV-2 enters a cell. Biophysical and functional data would be used to triage down to 24 antibodies and all this data needed to be generated in five days time. So here, in this time span, is the focus of the majority of the work that I will be speaking of today. In order to obtain kinetic and epitope data on large numbers of antibodies in a short period of time, both Lilly and AbCellera utilized high-throughput SPR Carterra LSA. The LSA allows the user to immobilize 96 antibodies or samples simultaneously using a microfluidic continuous flow technology, allowing for up to 384 microspots on a single SPR chip.

0:05:46.7 DF: This continuous flow can achieve high density of antibodies on the surface using very little sample. Then, once the antibodies are captured or coupled on the chip surface, a single flow cell can apply analyte over the entire surface, and this provides interaction data for up to 384 samples simultaneously. This one analyte over many antibodies approach not only helps save time, but it also has the advantage of using small amount of reagents. And early in the discovery, the reagents can be very limiting.

0:06:35.6 DF: For my experiments, I tested 192 antibodies. So, in addition to the 187 test antibodies, we included benchmarks with known binding domain specificity. The antibodies were coupled on the surface of the chip at two different concentrations. We used the lower density microspots to determine kinetic rates and the higher density for binning experiments. We then determined affinity to the spike... Trimeric spike protein, which was injected at seven different dilutions. The S1 and S2 specific proteins were limited to three dilutions and were used only as a yes-or-no binding to determine domain specificity, and this allowed for shorter experimental time.

Binning experiments were run using a pre-mix format

0:07:33.3 DF: And the binning experiments were run using a pre-mix format. The trimeric spike protein was pre-mixed with the antibodies where the antibodies were in molar excess to ensure that all the binding sites were saturated. Then, the complex was injected over the chip surface to determine pairwise blocking of antibodies. Similar to the antibodies, the spike protein was pre-mixed with a human ACE2 receptor protein to determine if the antibodies on the chip showed an epitope with ACE2. And so, as mentioned previously, we determined the affinities of the antibodies by coupling the antibodies to the chip, then used multi-cycle kinetics with regeneration with low pH in between cycles. So, using this approach, instead of the more typical protein A or anti-human Fc capture, allowed us to conserve antibody as well as some time, because we were able to do the kinetics and binning experiment all in the one chip.

0:08:55.4 DF: But on the other hand, due to the coupling process and low pH regeneration, we did not get information for all of the antibodies. Approximately, 10% of them regenerated poorly and did not have activity once they were coupled on the surface. Nonetheless, we did get very good data. The figure on this slide shows the affinities measured on a subset of the antibodies. Shown in red are the affinities for the trimeric spike protein, and we were very pleased to see that a good number of them had very high affinity in the picomolar range. And from this experiment we were also able to identify quite a few of antibodies bind into the S2 domain of the spike protein.

0:09:54.4 DF: We weren't able to get a lot of reliable data for many of the antibodies, and where we did get the S1 binding affinities, they were pretty low, but we believed that the protein that we were using at the time was unstable. So, not long before we begin our discovery, it was shown and published that ACE2 was the receptor being engaged by the virus to gain entry to the cell. So determining whether our antibodies could inhibit binding to the spike protein by blocking this interaction was an important piece of the puzzle. To quickly screen large number of antibodies, ACE2 was pre-mixed with a spike protein and blocking was tested in the same experiment as the epitope binning.

Cartoon here at the bottom is showing two different antibodies

0:10:50.2 DF: The cartoon here at the bottom is showing two different antibodies that are depicted to bind different epitopes on the trimeric spike. When the spike is pre-mixed with ACE2 in this cartoon and shown in purple, we can see one of the antibodies can still co-bind the spike protein and ACE2 complex, whereas the other antibody at the bottom cannot, potentially, sharing the same binding epitope on the spike protein as the ACE2.

0:11:33.2 DF: And in the experimental setting, we observed blocking of the ACE2 binding of the antibody to the spike protein as lower signal than the referenced spike protein injections alone, shown here in blue, whereas the spike and ACE2 pre-mixed complex are shown here in the red signal. We set the threshold for blocking of the antibodies at 0.7, meaning that the signal of the pre-mixed complex was at least 30% less than the spike alone. So using this simple setup and only four injections, we were able to quickly visualize which one of our antibodies had a higher potential for virus neutralization by blocking the ACE2 bind into the spike protein.

0:12:43.9 DF: In addition to kinetics and the blocking of ACE2, epitope diversity was explored using pairwise competition. Each of the 192 antibodies was tested against the antibodies on the chip surface, and this was to determine blocking or no blocking, and similarly to the ACE2 setup. And due to lack of RBD and NTD domain-specific proteins, as well as poor binding to the S1 domain, we did not have data at this point to show specificity to these domains. So including benchmarks, antibodies that were known to bind RBD or NTD or S2 domain, gave us some knowledge of blocking and domain blocking.

0:13:44.0 DF: From this experiment, 171 individual bins were identified, and it's noticeable that many of the antibodies, they do have similar blocking and binding competition profiles, but they are also different enough from one another to not be grouped together. Nonetheless, when we do see some clustering, the NTD and the RBD benchmarks that we used in this experiment were clustered over here to the upper-left part of the figure, whereas the S2 benchmarks were found in this larger cluster at the bottom. And additional clustering can be applied to a binning experiment. The Carterra epitope software does have two options for clustering. A walktrap grouping in which the antibodies are grouped via the software without user input. And then, the communities grouping where the software generates a combined binary dendrogram and then the user gets to pick how many communities, what they would like to cluster their samples into.

Combined Binary Dendrogram

0:15:12.8 DF: So shown here is the combined binary dendrogram from the communities clustering option. This red bar up here, it can be moved up or down to determine clustering and form the communities. Also, at the bottom of the dendrogram, all of the antibodies are annotated. I would like to point out again how the RBD-specific benchmarks are clustered in this group. Most of the entities are in this group and the S2 are actually somewhat spread out. From this clustering option, we were able to identify eight large communities, and right here, in this arrow over here, in this cluster, as well as here on the dendrogram, I have pointed out the Bamlanivimab, Lilly Antibody 555. For more clarity, additional information can be overlaid to the communities such as domain binding. Where we had information available, I overlaid the S2 domain binding right here and the S1 domain binders were in this cluster.

0:16:48.8 DF: So at the end of the biophysical and functional screening, 24 antibodies were selected to move forward with more characterization and ultimately pick our lead antibodies. To integrate all this data and quickly select antibodies with desirable properties, we used AbCellera's Celium software. Samples that satisfy the following criteria were prioritized. Strong binding to the spike protein on cells, and/or, and bead assay were prioritized higher than 30% pseudovirus neutralizing activity was... And this assay was performed at the vaccine research center and a dose-dependent neutralization profile using the VRC's pseudo-neutralization assay was prioritized.

We determined RBD competition to the benchmarks

0:17:53.5 DF: Also, we determined RBD competition to the benchmarks. I'm making the assumption that antibodies that blocked the RBD benchmarks would also be binding the RBD domain themselves. Blocking of ACE2 in the pre-mixed binning format at 30% or higher and high affinity to the spike protein. In addition to the above mentioned characteristic, we also measured biophysical data such as melting temperature, solubility, polydispersity, as well as predicted hydrophobicity and immunogenicity of the samples. And then, once we picked a set of antibodies that satisfied all of these requirements, we went back and added individual antibodies to make sure that we had a good diverse sample set.

0:19:00.0 DF: Some of the characteristics of the selected set of antibodies are shown in this table. In addition to the RBD binders, the team selected antibodies that bind the NTD region as well as the S2 region to maximize diversity in the set. Additional characterization of the epitope was performed using hydrogen deuterium exchange or HDX, as well as cryo-EM. So, combined with the finding data in the HDX and cryo-EM, we get a full picture of the domains that were represented as the set. Affinities to the spike protein are mostly in the low picomolar range, although we did include a few antibodies in the low nanomolar. In addition to that, we measured binding to each specific domain to the full antibody, as well as binding of the FAB portion of the antibody to the spike protein, and this was done on a subset of these antibodies. As previously spoken, we determine ACE2 blocking, and as expected, the majority of the ACE2 blockers also are RBD binders. And in this table, the Bamlanivimab antibody is shown here, this 0555 antibody. This is an RBD binder, and it binds a spike protein at 24 picomolar affinity, and it is ACE2 blocker.

0:21:02.1 DF: Using the methods that I described previously, I repeated there being an experiment now with the smaller set of antibodies, and having obtained RBD and NTD, specific proteins at this point, I was able to determine the binding to each... For each one of these antibodies in addition to the S1 and S2 domain. And now, we can start seeing a much clearer picture of the diversity in the set as expected, we see a good separation between the antibodies that bind the RBD protein, shown here in blue and separated pretty well from the NTD antibodies and red and orange for S1 binder. And then, the S2 binder antibodies are shown in purple. We were also, as expected, we saw that the ACE2 blocking was mostly concentrated to the antibodies that bind the RBD portion of the spike protein. What we were very pleased to see at the end that the data that was generated using epitope binning, this bidirectional epitope binning, correlated very well with the epitope information that we later on identified by either HDX or cryo-EM.

0:22:46.6 DF: To screen the remainder of the antibodies, I began utilizing a new approach and I started using a new set of benchmarks where we have know a very well-defined epitope and desirable functional activity. So at this table at the top, I'm showing the residues involved in binding of each of this molecules to the spike protein and this were identified by crystal structures. And so here at the top are the Lilly and AbCellera as well as Lilly and Junshi antibodies. Then we have included the two Regeneron antibodies that are found in the Regeneron antibody COV2. The CR3022 antibody is from the Swiss Research Institute, and what is interesting about this antibody is that it binds both SARS and SARS-COV-2. And then in addition, we're showing the binding residues, the residues that are involved in binding of ACE2 to the spike protein. So not shown on this table, but included in our binning experiments are the two antibodies by Vir as well as an additional two antibodies from Regeneron for which they structure and epitope was published.

We were focusing on antibodies that bind to the RBD region

0:24:29.7 DF: So, for this experiment, we were focusing on antibodies that bind to the RBD region, and upon screen in the 354 and antibodies are found 92 that binds specifically to the RBD domain and also compete with the benchmark shown here, so combining epitope knowledge as well as blocking or non-blocking of the antibodies to these known benchmarks can give a very clear view of where the unknown antibody is binding as well as potential functionality. And what is most advantageous to an experiment like this is the time difference, where a typical bidirectional binning of this size can take several days of instrument time, as well as analysis and interpretation can be pretty tedious.

0:25:42.8 DF: This kind of experiment takes only a few hours of instrument time to complete, and analysis is simple, and interpretation can be straightforward. Now, as the discovery of the lead and backup antibodies began to slow down, another important aspect of the project was, and still is, the surveillance of emerging mutations. These emerging mutations could impact monoclonal antibodies. So as of April 2021, the CDC had identified five variants of concerns shown here and an additional three variants of interest. More recent data as of yesterday, which is not shown here, has identified many more variants of interest, and it shows that these variants of concern are becoming much more prevalent. So what is important to the antibody therapeutics is that these variants contain one or more mutations in the RBD domain of the spike protein where these antibodies bind.

0:27:04.6 DF: So in order to determine if our antibodies retain activity to the newly emerging variants, we express and purify the RBD portion of the spike protein as an Fc fusion and then couple it to the chip surface and measure the binding affinities using only the FAB portion of the antibodies. So the ability to couple 384 proteins at once gives us the potential to identify concerns early and quickly. In this experiment, 96 RBD variants were immobilized in duplicates, and affinities were measured against the FAB portion of Bamlanivimab.

0:28:00.5 DF: Now, we can easily get a picture of loss of binding altogether due to these different mutations or weakened binding, so just shown here in the yellow and purple. Now, we can start seeing mutations where the binding was unaffected when compared to the wild-type RBD. So being able to leverage this kinda technology even late in the antibody process has given us the tools needed to quickly and efficiently answer questions about our antibodies. And I'm extremely proud of the work that we have done to get to this point, and I'm eager to continue work with this technology on our journey to... With monoclonal antibodies against COVID-19. Thank you for your interest, and many thanks to all the different teams across multiple sites and collaborations that made all of this possible.

0:29:17.6 JC: Excellent. Thank you for your presentation, Denisa. It's now time for the Q&A. Joining Denisa for the Q&A is Dr. Daniel Bedinger, application science team lead from Carterra. Daniel helped launch Carterra's LSA platform and now leads the company's application science team in California. He has over two decades of experience in the generation and characterization of therapeutic monoclonal antibodies most notably at XOMA and Abgenix. Daniel has broad experience in utilizing multiple types of label-free biosensors from a myriad of vendors and has authorship of numerous publications regarding antibody characterization and holds patents relating to therapeutic monoclonal antibody discovery. Welcome, Daniel. To ask Denisa or Daniel a question, just type it in where it says, "Questions," and then press, "Submit." So our first question, and this one is for you, Denisa, and it asks, "And did you find antibodies with apparently potent bioactivity without ACE2 blocking? And did any of these non-blocker inhibitors show less affinity difference with the variants of concern?"

We find antibodies that block ACE2

0:30:46.1 DF: Yes. So yeah, that's a two-part question. For the first part, did we find antibodies that block ACE2... Or that do not block ACE2 that are functional, yes, we did. Although the majority of the potent neutralizers were ACE2 blockers, or least as tested in the pseudovirus neutralization assay. But we did also see a few that did not block the ACE2 interaction that could have some amount of neutralization. We have not tested any of these ACE2 non-blockers against the RBD variants of concern, and the reason for that is that we have to generate FAB for that assay, and that requires a little bit more work. And they haven't been part of our clinical or the antibodies that we are focusing that work on.

0:32:02.5 DF: But nonetheless, the several different antibodies that we did test that are ACE2 blockers, a few of them were unaffected by the mutations that are present in the variants of concern.

0:32:20.6 JC: Excellent. Thanks, Denisa. Our next question asks, you mentioned very little reagent is required to run a binding or a binning experiments, how much antibody do you need to run these types of experiments?

0:32:43.3 DF: Yeah. So this is going to be dependent on the molecular weight of the antigen that you're using, as well as things like affinity. So, in order to get pretty good SPR signal in both of those things, we need to keep in mind, nonetheless, there are multiple different things such as a smaller antigen has higher molar concentration for the same amount of protein. So without being able to exactly answer that question, I can tell you that for this type of experiment or with a spike protein which is around 400 kilodalton, the concentration that we used it at, at 20 millimolars, I end up using about 200 to 250 micrograms for a 192 by 192 experiment.

0:33:45.7 Daniel Bedinger: Yeah. And I think that the question also might have been about how much antibody is actually required. And typically, I know it varies a little bit based on the experimental design, but it's somewhere in the 15 micrograms range to do a pre-mix binning. Does that sound right, Denisa?

0:34:02.5 DF: Yes.

0:34:02.6 DB: 'Cause you need to have fairly high antibody concentrations present in the pre-mix experiment against a primer, 'cause you need to be both in stoichiometric excess of the antigen and in significantly above the KD of the antibody-antigen interaction for the competitor in that solution injection. So I think, typically, I like to run those experiments between 30 and 50 micrograms per milliliter a day which requires, I believe, it's 10 to 14 micrograms per injection at those relatively high concentrations. I'm not exactly sure about the concentrations you are using on your experiment, Denisa.

0:34:46.4 DF: Yes. So, I believe after we did this several times, for both the kinetics and the binning, I would ask for about 20 micrograms total antibody. And you're right, about 15 is what was used up.

0:35:09.0 DB: And most of that is for these analyte injections in the competition experiment, the actual array creation of the antibodies is like you would need to do a kinetics. You're more on the microgram range of protein, typically, like one microgram. Thanks.

0:35:30.5 JC: Excellent. Thanks, Daniel. Thanks, Denisa. And our next question asks, what considerations go into the selection of the panel to include in the benchmark binning approach, and can this technique be broadly applied? Denisa.

0:35:47.9 DF: Yes. So, I believe the benchmarks should represent the diversity that you're hoping to achieve, and I think of them as a bait. So when identifying benchmarks, I look for both function and lack of function, even though that may not be something that you're gonna move forward with. Being able to identify antibodies that don't have that or the potential of not having that out of a large set could be advantageous. And then, look for both antibodies that benchmarks that block or do not block each other, of course. And then, like in blocking, if that top of the information is available, and perhaps even diversity in the affinities.

0:36:51.4 DF: Again, you may want to finally choose samples that have high affinity, but, and in your benchmark sets, perhaps, use even lower affinity antibodies. And yeah, I believe this technique can be broadly applied depending on where you're applying it. And for our use and for my use, I use it as a first step to identify a set of samples in which then I would perform bi-directional binning and that way, I can actually see the true diversity within the sample set of the unknown antibodies.

0:37:42.0 JC: Thanks, Denisa. Daniel, anything further to add there?

0:37:49.6 DB: No.

[laughter]

0:37:50.0 JC: Fair enough. Excellent, no worries.

0:37:52.0 DB: Let's move to the next question.

0:37:53.9 JC: I will move on to the next question which asks, what's the pros and cons of immobilizing antibodies on the chip by coupling and capturing? Denisa, perhaps I can start with you.

0:38:10.3 DF: Yeah. So again, the pro... Well, for the binning experiment, immobilizing antibodies on the chip is a must because at the end of each injection, then you would have to remove the antibody and antigen complex that was bound. And doing that in a capture form can then defeat the purpose of using very little antibody. So that being a requirement, one of the cons is that whether it is by virtue of low pH regeneration or the coupling process. Some of the antibodies do not like those kind of conditions, so we lose activity.

0:39:03.8 DF: And again, the pros are very quick and very high-throughput and interrogation of the samples that do retain it's activity. And as for the capture, when running a kinetic experiment, having the antibodies that are captured in a uniform way presented it on, perhaps, all in the same direction, that can be very advantageous to the quality of the data, so.

0:39:42.9 DB: Yeah, I think that's right, but when you immobilize proteins covalently, there is a risk that when you mobilize primarily via lysines and the N-terminus, right, so if there are lysines in the binding domain and stuff, you can potentially impact things like binding kinetics. But with IgG specifically, or even larger proteins, there's typically enough surface lysines where you get quite a diversity and a lot of Fc-mediated linkage of the antibodies anyway, so the impact on kinetic performance of the actual mobilization step itself is usually pretty minimal. Like sometimes if you were to compare a really high quality capture kinetics experiment to an immobilized kinetics experiment, you may notice that there's an introduction of a sort of small or moderate amount of heterogeneity to the kinetic profile, but the overall kinetic picture will remain the same.

0:40:44.6 DB: It may just affect sort of the goodness that fit slightly, whereas on the capture, and you're only binding via the Fc, say, with an anti-Fc antibody, you'll get a bit more conservative kinetic. So there's a little bit of a trade-off there, but oftentimes, especially on these high-throughput applications, it makes sense to immobilize things 'cause you get so much more flexibility and simplicity, especially, if you're going to try to duplex kinetics and epitope binning experiments.

0:41:15.6 JC: Excellent. Thanks, Denisa. Thanks, Daniel. Our next question to asked, would you be able to sort of comment on the high-throughput monoclonal antibody purification step? And they're specifically looking at what was the final stop buffer suitable for pre-mixing epitope binning?

Sort of comment on the high-throughput monoclonal antibody purification step

0:41:42.4 DF: Yes. So we're very fortunate to have a pretty automated lab and usually, we do the high-throughput verifications in batches of 24 or 48. So, for that, generally, what happens is most people who are purifying antibodies will know that there's either a some tris or glycine buffer being used for neutralization purposes, so to bring the antibody from the low pH of the illusion to the higher, more neutral pH. And in previous experiments, what I have found has been that due to the high density of the surface on these chips, and generally, the good expression of our antibodies, we get a high enough concentration of anti-bodies that upon dilution prior to the immobilization on the chip, the effect of these immune containing buffers, it's kind of minimized. So personally, I just leave it, I leave it in the trays and then further dilutions can be done in either PBS or the HBS buffers that are used for this kind of experiments.

0:43:27.5 JC: Excellent. Thanks Denisa. Our next question asked, so what is the difference in classification between the community and an epitope bin? Daniel, is that something you can help answer?

0:43:47.9 DB: Sure, yeah. So, an epitope bin is the purest differentiation in the kinetic software, and it requires true identity between the clones and in their competition profile to be clustered into a bin. So if there's a single difference in the competition profile between two clones in terms of what they compete with and what they sandwich with, they will be classified as not bin-mates. That's why in Denisa's presentation, there was a large number of small bins for the overall spike competition profile, 'cause there's a lot of subtlety in the way that a diverse panel of antibodies interacts with that protein and also in the pre-mix format, there's a bit of a variance based on kinetic performance of the antibody, so it may not be entirely the same.

0:44:42.4 DB: So you tend to see very small, especially, if these antibodies that went into that, it's important to remember, they were selected based on having the sequence diversity, so they were not all actually quite different from each other. So the bin map is very granular, and it's what shows you the specifics, where even a single difference bumps them out. The community plot, on the other hand, it generates this dendrogram where you have the degree of relatedness of the antibodies in the heatmap based on their euclidean difference in terms of accumulated differences in the Sandwiching and the blocking profile.

0:45:19.3 DB: And so, you're allowed... It allows you to set a cut-off at a different point other than sort of the base line, so you can have more generalized bin clusters and include things that have very similar competition profiles, but allows some differences. And so, for these large binning maps, oftentimes, that's a very useful and effective strategy to interpret the data. We're doing a lot of this type of work, very similar to what Denisa presented here with the COVIC consortium being run by the La Jolla Institute for Immunology, where we're looking at many antibodies generated from many sources and doing this profile, and we found that we could get very interpretable maps of at least the receptor binding domain of COVID-19 antigen, coming down to about maybe the low teens, like 13 to 15 bins.

0:46:17.2 DB: It seems to be quite descriptive of where these antibodies are actually binding on the RBD and translating that to the cryo-EM structure. So I think, we actually find quite a bit more bins than that, but when we sort of triage them to related communities, we get into this low to mid digit teens number of community that seemed to make sense in terms of actually describing the general aspect of how the antibodies interact with that RBD.

0:46:56.7 JC: Excellent. Thanks, Daniel. Our next question asks, what considerations were given to avidity and multi-valency of the trimeric spike protein in the design of these experiments and the interpretation of the results? Denisa.

We use the trimeric protein for both the kinetics and the binning experiments

0:47:17.6 DF: Yes. So, yeah, we use the trimeric protein for both the kinetics and the binning experiments. So, for the kinetics, we are fully aware that we have a multivalent protein, and we kind of expect that some of that very high picomolar affinity that we saw, it was gonna be driven by this avidity. And when we later ran experiments using the monomeric RBD or NTD domains, we did see that that was indeed the case. But the rationale for actually using the trimeric protein in a kinetic setting was that that would actually best represent the true nature of the binding of the antibody to the virus.

0:48:17.5 DF: And so, yeah, that was for kinetics. For the binning experiments, so having a multivalent protein pretty much forces the user to use a pre-mix assay instead of a different type of assay where you may sequentially bind the antigen, then the second antibody. So in a pre-mix assay, the antigen and the second anti-body that's gonna be flown as an analyte, they're pre-mixed ahead of time. And as a rule of thumb, you want the antibody to be in molar excess so that all of the binding sets on antigen are saturated. So, for my experiments, I had that at 10 fold molar excess. And then, once they're pre-mixed and some amount of time where they achieve this equilibrium in saturation, then they can be bound to the antibodies on the chip, so yes.

0:49:42.4 DB: Yeah, I think that's a great explanation. Thanks, Denisa. So, as she said, we can only use typically the pre-mix format with multivalent antigens, whereas for monovalent antigens, we usually prefer to use the classical sandwich approach 'cause it's a little simpler assay, it has a bit less stringent requirements for in terms of saturating the receptor or the target like Denisa said, she'd use a 10-fold molar excess.

0:50:11.4 DB: The other thing that can be done to sort of help this, it doesn't eliminate avidity effects and the kinetics, it's to use planer chip types and low density surfaces. So if you're trying to do kinetics against a multivalent target and you spot your antibodies at either multiple densities or at very low densities, and you use, like say our CMGP, which is a planer carboxymethyl dextran surface, you can have... You have a better chance, I'll say, at revealing more of the one-to-one kinetics because you're less likely to get the bridging between different molecules on the surface than you would if you were using a more matrixed and higher density antigen surface.

0:51:00.9 DB: There are different ways that proteins interact with multi mars, for example, you can have one antibody bind to both binding things of a dimer itself at which point no amount of dilution or spacing out of the molecules on the surface can prevent that. But other times, and I think this is more common with the spike, is that when you inject the protein over the surface, it will bridge between two different antibodies that are mobilized on the surface. So, by using low captured or low mobilization densities on planer surfaces, you can reduce the impact or the frequency of that and have a better chance of teasing out kinetic differences. It's not a guarantee, but that is one strategy you'll apply.

0:51:44.4 JC: Excellent. Thanks, Daniel. Thanks Denisa. Our next question asks, what was the correlation between pseudovirus and real virus neutralization titers? Denisa.

We did not perform real virus neutralization on a lot of samples

0:52:02.7 DF: For this sample set, we did not perform real virus neutralization on a lot of samples, as you may imagine, everybody in the world was trying to run the same type of assays, so we... For the handful of samples that we tested or antibodies that we tested from this sample set, there was a good correlation between the pseudovirus and real virus assays, and there were multiple of those. And then to be fair, we did go back and perform pseudovirus assays that were more representative in a way of measuring IC50 and IC80 and so, where I showed that we had a certain cut-off, once we got into the weeds of the top 24, we were able to actually tease them more apart using the pseudovirus and then the real virus assays.

0:53:12.4 JC: Okay, thanks Denisa. Our next question asks, how broadly applicable would you consider the threshold of 30% signal difference? So from your presentation, apparently, that was used to classify whether antibody spike protein binding was blocked by acetate?

0:53:40.1 DF: Yes, so this was more in hindsight when we did compare that 30% threshold to then to pseudovirus neutralization potency, we did see that there was a fairly good correlation between the two. And that again, comes back to the question that, yes, we were actually seeing true blocking. Broadly applicable, I think the more blocking that you have, the better. I believe that in a different project, this would... This would have to be answered by biology, trying to really see if you do actually expect strong like and blocking and then you can put that threshold even lower.

0:54:54.6 DB: Yeah, I think that with the ACE2 interaction in the pre-mix format, it's a little bit of a challenge given the sort of moderate affinity of the interaction between ACE2 and spike, but I guess the good thing is that at least, if you could really load up the ACE2 in that assay to a relatively high concentration, it's not like, you know you have the same interaction between ACE2 and the protein for everywhere right, it's the same complex you're exposing to all of the antibodies, so if you're able to demonstrate that you are completely saturating it for at least some of the high affinity interactions, you can apply at least conceptually, a fairly robust cut-off where if you're not seeing very good inhibition, you know that you're probably in that middle ground between a blocker and a sandwicher.

0:56:00.0 DB: There's lots of things, and this is one thing that makes the interpretation of pre-mix binning data a bit more complicated, is many things are sort of affinity modulating or hysterically hindering. They're not really a blocker, they're just things that decrease the affinity of the interaction, which can be important and can be relevant to classifying things as inhibitors, which I think is a good, why the 30% cut-off was meaningful, 'cause if you're decreasing the binding by that much, it suggests that it's a non-favorable interaction anyway, even if it's not necessarily fully blocking, but yeah.

0:56:43.4 DB: And there's quite a range of things where you can have everything from affinity enhancers to total blockers, and to really demonstrate that, you often need to do a titration of inhibitor to see that as you add more and more of the inhibitor, do you get convergence on a certain binding rate, and if you do, then you've just proved that you have a kinetic modulator essentially, instead of an actual blocker. And that's one thing that makes both the pre-mix and the classical sandwich assays very interesting to run together or even alternately, is that in the classical sandwich assay, if you see sandwiching signal, like say you have an antibody that captures your receptor RBD, and then you inject the ACE2, if you see really any binding to that captured RBD, assuming it's monomeric, you know that it's not a blocker, but there might be quite a range for how well the ACE2 binds to that captured RBD.

0:57:44.1 DB: And so, in the classical sandwich assay, many things would be considered not competitors based on the fact that they allow some ACE2 sandwiching, but if that interaction is not very favorable and you run it in the pre-mix assay, it's gonna show inhibition of binding and be more likely to be classified as a blocker. So I think the group that Denisa works with, Lilly, and probably a collaboration of the group, and so they took a very sort of thoughtful approach to this, I think, where they were trying to figure out where the meaningful level of cut-off was to call things that should be considered to be inhibiting the interaction or not.

0:58:29.2 JC: Excellent. Thank you, Daniel and thank you, Denisa. Well, I'm afraid that is all we have time for today. I would now like to thank Doctor Denisa Foster first for her presentation and Denisa and doctor Daniel Bedinger for answering your questions, and of course you the audience for taking the time to be with us today. Remember, you can watch this webcast at any time on demand at nature.com webcasts. Thanks for watching, and I hope you'll join us again soon.

0:58:57.8 DF: Thank you.

0:58:58.0 JC: Thanks everyone.