Join speaker Dan Bedinger, PhD, Head of Applications Science at Carterra as he discusses the LSA platform and the use of HT-SPR for kinetics and high-resolution, competition-based epitope binning in the rapid characterization of diverse monoclonal antibody panels. Hear how the Carterra LSA and Epitope software made Eli Lilly’s path to the clinic for two recent molecules faster than ever before.
|Dan Bedinger, PhD
Field Applications Science Manager Carterra
Posted by Daniel Bedinger, PhD
0:00:00.0 Rob Burgess: Good afternoon, good evening, good morning to you, wherever you might be logging in from around the world, I am Rob Burgess and I head business development for Sino Biological, and I would like to officially welcome everyone to the next installment in our series titled Lock and Key Immunodetection webinars. We have a wonderful and excellent speaker today from Carterra, and he's gonna talk about a very exciting technology focused on antibody discovery as well as epitope analysis. Before I get to introducing our speaker today, I have just one very brief housekeeping issue that I wanted to mention. And that is, if you have a question for the speaker, I ask that you type it into the chat box, and then at the end of the lecture, I will verbally walk through as many of those questions as possible within the chat box. So come up with some good questions for the speaker today.
0:01:54.2 RB: Also, please introduce yourselves in the chat box and tell us where you're from, it's always interesting to see where people are logging in and calling from around the world. And so without further ado, it is my pleasure to introduce to everyone today's speaker, it's Dr. Daniel Bedinger. Daniel helped launch Carterra's Loadstar Ray platform and now leads the company's global application science team. He has over two decades of experience in the generation and characterization of therapeutic monoclonal antibodies, most notably Xoma and Abgenix. Daniel earned his doctorate from UC Davis in cellular and molecular physiology, and it was there that he actually focused on and studied insulin receptor signaling. And the title of Dr. Bedinger's talk today is high throughput epitope analysis and the latest in antibody discovery from Carterra. Dan, it's a pleasure to have you today and we will now turn over the screen sharing to you.
0:02:08.3 Daniel Bedinger: Alright, thank you. Let's get this shared. Okay. Hopefully, you all can see these slides now.
0:02:17.9 RB: Yeah, sure. Thanks.
0:02:20.1 DB: Thank you... Alright, thanks for joining me today. So we're gonna talk about high throughput characterization of antibodies. The industry in general has sunk a lot of investment into the optimization of monoclonal antibody generation technologies, and a lot of organizations now have the ability to access large sequence spaces of antibody diversity, whether through phage display, B-cell screening, synthetic library design, AI. And the analytical tools to accurately measure binding affinity and specificity and epitope analysis for a large number of clones really lacks orders of magnitude behind. And so at Carterra, we're focused on trying to develop tools that bring the capability to get accurate analysis at more of a library scale or at least push the throughput as far as we can on that because we don't want you to throw a blockbuster candidate away.
0:04:03.8 DB: So typically, binding kinetics and specificity screening has been done on a pretty limited set of antibodies or clones really due to the limitation of conventional label-free bio-sensors which make those large analyses either really time-intensive or very sample-intensive. And so typically people try to compromise by either doing low resolution assays with one concentration, two concentrations of antigen, limited concentrations, or do significant down selection before they proceed with that characterization. But with automated array-based SPR, we call it HT-SPR, in the LSA platform, you can really get high quality kinetics and specificity characterization at a new scale, typically at the types of sample levels that people get out of screening campaigns. And this can also be done with supernatants and extracts, so crude samples are often applicable for high resolution kinetic data in early discovery. Also, the LSA platform can dramatically increase the scale and accessibility of competition-based epitope binning, which we'll talk about quite a bit in this talk.
0:04:37.9 DB: A little bit of agenda here, so I'm gonna start off talking about how high throughput SPR can enable and accelerate the antibody discovery by giving three examples from the literature. I'll do a brief introduction to the Carterra LSA and now LSA-XT platforms, and talk about how we do kinetic analysis with them. And then talk about our process for epitope binning and why we think it's useful to do high resolution epitope binning and our assay strategies and analytical approaches, and then going into a little bit more detail, an example of the application of that with the coVIC consortium study.
0:05:20.3 DB: High throughput, high resolution epitope binning is becoming more popular in the industry. The SARS Covid-2 pandemic coincided really with the broad adoption of the Carterra LSA platform. And the good thing about that was that most of our antibody discovery customers were able to shift when the labs were shut down and things to doing Covid discovery work and still work and there was a heavy emphasis on publication, rapid publication of SARS Covid-2 work. So we actually generated quite a body of literature around our technology. I guess the downside of that is that it's primarily SARS Covid-2 related work, which is great. It's good work and good biology and some amazing demonstrations of the technology, but it's not very broad. And so we have a lot of customers that are doing a lot of great work in other disease areas, all that you would expect, other infectious diseases, cancer, immuno oncology, neurobiology, all those things. And so I encourage people to check out our symposium presentations at our website, carterrabio-com if they're interested in some other applications because we have seminar presentations of that, although not a lot of that work has been published.
0:06:41.7 DB: Okay. So going into the first example here, this is when we love to talk about... The groups at AbCellera and Lilly collaborated very early in the pandemic. They had received some DARPA funding to build a rapid antibody infrastructure earlier. And when the pandemic came about, AbCellera collected blood from one of the earliest recovered Covid patients in Canada and did a B-cell discovery effort where they identified many punitive SARS code 2 spike binder antibodies. They expressed many of those, and in the first tranche, they moved 187 small scale expression clones into Carterra LSA analysis. So both Lilly and AbCellera had LSAs at the time, and they did this work kind of in joint. And so in seven days, they were able to characterize all of these clones for affinity and kinetics, ACE2 competition, which is neutralization of viral entry of essentially in vitro study, and full symmetrical pairwise competition epitope binning.
0:07:57.7 DB: So they were able to identify the highest affinity neutralizes from broad epitope diversity so they could select key clones from individuals or with unique epitope properties. From that, they took 24 into their manufacture ability assessment and also downstream potent C assays and then triage that rapidly into manufacturability and production. And they were able to actually file an IND for bamlanivimab 90 days after the start of getting those B-cells from the patient.
0:08:38.4 DB: And at 94 days, they were actually in the clinic with that. And bamlanivimab went on to receive emergency youth authorization in, I believe it was October of that year, making it the first biologic Covid-19 therapeutic available.
0:08:54.3 DB: So this really shows, I think, the apex, it's definitely the fastest A biologics therapy ever been developed at this point and it really shows what is achievable if companies have the right tools in place. Lilly has a very sophisticated characterization and developability paradigm that they were able to move these things into. And the LSA was a key part about accelerating that initial phase.
0:09:25.9 DB: Another example I'd like to give is a publication from a group at NIAID, this was Josh Tan's group, and they were also making SARS Covid-2 neutralizing antibodies. And they found they had about four discrete communities of neutralizing antibodies against the RBD and one discrete community against the NTD domain of that protein that showed potent neutralization. And in some early cocktail studies, they realized that they were only seeing additive potency and not synergistic, but they were more interested in something that would have a real high neutralization with synergy and maybe some different specificity that would give some variant resistance. So they looked into making biparatopics or bispecific DVDIGs, and that's this format shown here, where you have an IgG construct with two independent binding domains tethered to the end.
0:10:23.1 DB: By using this understanding of epitope binning, they were able to... In the potency infinity relationship center, they were able to design a discrete set of these biparatopic molecules to take into characterization knowing that they bound to noncompetitive epitopes with each other and should have some bispecific function. When they characterized those, they found some really interesting examples of synergy, the one shown here was the most potent, where you can see in black the potency of the pooled version of the two... Or the pooled test of the two clones, and then the biparatopic, or the other two curves, those are a replicate data showing more than 100 fold shift in potency with the biparatopic activity. They did some cryo-EM work to try and understand why this was so potent, and they found that there was a really interesting cross-linking created by the RBD. So this set of clones was actually one of the MTD binders in an RBD binder, and it would cross-link across two neighboring spike monomers and lock the SARS code 2 spike protein into an extremely inactive confirmation. This was pretty cool application of that.
0:11:41.6 DB: The last example I'll talk about here is on actually a different disease, it's still antiviral. Tom Yuan and the group at Twist was looking at ways of validating their platform and demonstrating just how efficient it could be. Twist has the ability to generate large amounts of synthetic DNA, so they went to the literature and found a paper that had published a couple 100 or I think more than 200 sequences, patient-derived ebola-recovered patient antibodies that they believed were Ebola viral protein binders.
0:12:22.7 DB: In about 14 days, they expressed all of these sequences, cloned them into expression vectors and started making small-scale 1 mil transfections of these. In another about four days, they purified those antibodies and were able to run them on the LSA for general kinetics of binding. Although, it's a trimeric antigen, so is avidity but at least characterized binding and a binning assay. So they used about a little over 50 analytes, and they included a number of clones that bound to known residues in the literature. And so this enabled them to actually map all of the sequence diversity they pooled out of the publication in just a couple of days to either known binning epitopes or identify them as binding to unique previously uncharacterized epitopes. This is a really strong validation of how in a very small amount of time can go from unknown sequence space to a fairly high resolution of understanding of the binding properties and epitope of these clones. And this could be used to design cocktails or bispecifics going forward.
0:13:41.5 DB: Okay. So now I'm gonna actually introduce the LSA platform itself. The LSA is a ray-based SPR bio-sensor, and it's been pretty widely adopted by the instrument. It launched in the beginning of 2018, and now it's in 19 of the top 20 pharma, many biotechs, numerous CROs, and multiple reagent suppliers. We actually have two instruments now, we have the original LSA and a updated version called the LSA-XT. I'll talk a little bit about the differences later on. Then we have our software suite, which is three software packages, the navigator software is our instrument control software used to design experiments and collect data. Then we have two analytical software packages, one is called kinetics and one is called epitope, it's relatively self-explanatory what they focus on. And then we have an entire line of bio-sensor chips and consumables that you can use to run the assays.
0:14:45.6 DB: So the LSA really excels in many of the core applications of biologics or antibody discovery, those being kinetics and affinity analysis, competition-based epitope binning, peptide or mutant mapping, and quantization. And so in addition to having a hardware architecture that makes these assays very high throughput scalable and sample efficient, we've put a huge amount of effort into developing analytical software packages that make processing these large data files fast and efficient but also visually rich with lots of tools to access the different types of figures and images around the data.
0:15:28.8 DB: The way the platform works has two semiindependent fluid systems within it that address the same chip, so the chip doesn't move, it sits in the center of the instrument. There is a 96-channel flow head, where you can flow 90 samples at once, and you can use that to index to four locations within the surface to create a 384-spot array. And then the single-flow cell will dock over that entire ray area and flow one sample over everything. We call this "one-on-many analysis," and it's where the really high levels of efficiency for both kinetics and epitope binning or generally multiplex studies comes in.
0:16:12.0 DB: So a little bit more resolution on this process. This is a video showing the 96 needles descending into the sample deck and drawing 96 samples at a time. This printhead device docks under the chip surface and flows the samples back and forth before returning them to the plate. And then it can go pick up additional samples, go to a new location and print 96 more, eventually creating a 384 spot array, which is actually shown in the picture now on the screen. So it's a high density array. There's some real advantages. We call this "continuous flow micro-spotting technology," and this is our core differentiator. Unlike most microarray systems where you're depositing material on the surface, this is actual flow, so it's going from running buffer to sample, then back to running buffer. So all of the types of workflows that are done on conventional bio-sensors can be done here, where you're either mobilizing from low-concentration samples using electrostatic preconcentration or you can capture from crude samples if you have an affinity surface. And because we're flowing the sample back and forth, we can maintain a high flow rate for very long contact times. So if you have a B-cell supernatant with tiny amounts of antibody in it and you wanna capture your antibodies for 45 minutes onto an antiFc surface, for example, you can do that and allow the surface to build up over time. So we get really high sensitivity that way. And it's quite flexible with chemistries as well.
0:18:57.7 DB: This is just a view of how the array works. The pink vertical rectangles are the 96 channels that are docked, the blue horizontal ones are the 48 interspot references. So those are unprinted locations used for real-time referencing. If you do four of these prints from above interlaced, you create a 384 spot array shown below. The other fluid module here is the single-flow cell, this docks over the entire area where the array can occupy and flows one sample. This can be really useful for preoperative steps if you're going to regenerate the chip surface or mobilize, capture a lawn, we call it a "lawn" when you put one sample over the entire surface. It is also used to deliver the samples for kinetic injections or competition assays like epitope binning. So that one injection is giving you simultaneous data for all, really 432 spots on the array, the 384 active plus the 48 reference.
0:19:07.0 DB: So this is one 270 microliter volume of this injection to get you data in all those parallel spots. Again, it cycles the sample back and forth, so with that relatively small volume, we're able to maintain a high dynamic flow rate to minimize things like transport effects. So again, with the single-flow cell, one sample goes over everything, collecting parallel data.
0:19:36.2 DB: Just real briefly, wanna talk about our chip types offerings. Much of the chemistries people are used to using on bio-sensors are available through us. This list is always expanding as we learn and develop new applications, but currently we have a couple different thicknesses of linear polycarboxylate chips, several different carboxymethyl-dextrans including a planer. We have nickel NTA surfaces for capture of His-tag proteins, several different thicknesses of streptavidin and protein A and protein AG chips currently available.
0:20:12.8 DB: With that, we'll start talking about high throughput kinetics and how we approach this. On the right is a diagram of a very common format for capture kinetics high throughput screening assay on the LSA. This shows an antihuman IgG surface, this would be made using the single-flow cell, as what we call a [0:20:35.6] ____ so this is everywhere, and then you would use the 96 channel side to capture supernatants or even purified antibodies diluted to the surface via their FC. Or this could be any type of antibody fragment or construct that has some affinity linkage, biotinylated peptides or aptamers, Fabs, VHHs with His-tags or HA tags, we do all those things. And then the single... And that uses tiny amounts of sample, way less than 100 nanograms per clone is necessary. Then you inject a titration series of your target with the single-flow cell/ and the example I'm gonna give on the next slide, it's PD-1, it's a 17 kilodalton protein, and to generate a titration series, starting at 1 micromolar, you use 7 micrograms of antigen. So very little material for the amount of data you generate.
0:21:37.3 DB: This assay, because this is a regeneratable surface, where you can remove the bound ligands, can be scaled all the way up to 1152 unique ligands in a run. Our instrument has positions for three 384 well plates on the 96 channel side, and so you can repeatedly load those samples to the array and strip them off. So you can even automate, say, like three antigens by 384 well plates, it's only nine steps in a run. Yeah. Oh, and you can use this to capture high-end or dilution series of ligands as well, so you can get high-quality data in one go.
0:22:23.6 DB: This is an example of a run that was done on the LSA set up in an afternoon run overnight, 384 interactions. In this case, it's actually only about 38 unique antibodies spotted in eight to 12 semi replicates each. That's why you see the repeating patterns. But all of this data... This was the... The analyte injections you're seeing are eight wells on a 96 deep well plate and use 7 micrograms of antigen to generate all of this. Some cool software features are shown here. We like people to report accurate results with the LSA, so we do some flagging and annotation of the experimental results as they come out. The ones shown in gray are nonbinders that are below a threshold you set in the software, so in your big list of kinetic rates those would just say "NA." If they have a poor fit, a high error, they're flagged, in this case in yellow.
0:23:28.1 DB: And then there's two kinetic limitations shown here as the same color, although they could be set to be different colors. If there is a slow off rate that is hitting a limit that you define in the software, it will flag that and report it as less than function. The affinity is less than 50 picomolar, off rate is less than two E times 10 to the minus five, etcetera, based on what your settings are. Also, if you haven't injected a high enough concentration of analyte to necessarily accurately inform the R max or the saturation point, it will flag that. Technically, it's flagging something, that the observed binding level is less than 50% the calculated R max. But you can think of it as highlighting to you that if you are going to hang your hat on those affinity and on-rate values, that you might wanna inject a higher concentration of antigen. It will give you those rate constants, but it flags them for you.
0:24:27.5 DB: I mentioned earlier, if you don't have 384 things to mobilize in your array, you can increase your N. So this is a really common thing that's done on the LSA, is instead of just getting one value with a goodness of fit parameter, you can actually get mean and standard deviation of equivalent analyses. This is showing one clone in 12 replicates within one experiment. It was actually captured from three slightly different dilutions, that's why you see the three densities. But if you label them with the same name, the software will automatically calculate mean and standard deviation of all the rate constants and parameters from the fit. This could be a really powerful way if you're looking at subtle changes like single amino acid mutations or maybe a deamidated form, an ion exchange peak of an antibody prep.
0:25:23.8 DB: If you spot it 12 times at multiple different densities and you run the same injections of analyte over those two preps on the array, you can look, Do my error bars overlap or is there a real difference in the kinetic behavior between those two forms? This is a pretty powerful application of the technology.
0:25:43.4 DB: In terms of visualization of this data, the software has some great tools, this is an iso-affinity plot, where you have the on rate as the Y axis and the off rate as the X axis because the affinity or KD value of an interaction is the off rate divided by the on rate. These diagonal lines represent single affinities, so you have 100 picomolar, 1 nanomolar, 10 nanomolar, 100 nanomolar here down these lines. So this is a great way to visualize kinetic distribution of the antibodies in your panel. In this case, some of the clones are highlighted to show all of the replicates that were done.
0:26:25.6 DB: One more example on the LSA of kinetics I wanna show was presented at one of our seminars by Denisa Foster from Eli Lilly. They were making covid-19 antibodies, and this is actually the bamlanivimab Fab form binding to immobilized or captured SARS Covid-2 RBD mutants. So using a shallow well 96 well plate, they were able to do small-scale transfections of many mutants, take those soups, directly capture them to the array, and in an afternoon get the full kinetics of the Fab profile to all of these 96 mutants to the receptor. So I think this is a great example of both crude kinetics and also the richness of data you can get easily on the LSA platform.
0:27:26.5 DB: So I did mention earlier, we have a new version of the LSA, we're calling that the LSA-XT. So the same story applies to it as... We like to say about the LSA, you can get 100 times the data 10% of the time using 1% of the sample compared to other leading bio-sensors. But now we have better than two times signal to noise, two times better signal uniformity in the array, and two and a half times faster data collection rate. So what this really translates to is, it's the same fluidic system as the original LSA, you can think of just enhanced optics and... Some software improvements that make processing the signal data better. The LSA was really well suited and all of the data you've seen up till now was from the original LSA for protein-protein interactions, antibody screening, affinity characterization.
0:28:26.5 DB: But it was a bit of a struggle for really small analytes or extremely rapid interactions, things that dissociate very quickly we could get an affinity but maybe not a rate constant. Or things that were very small or harder to measure, you'd have to get higher surface densities to stay above the noise structure. So the XT brings this to a new level with the faster data collection rate, a little better referencing, and much less noise. So we can do things like kinase inhibitor profiling.
0:29:01.2 DB: The next couple slides will show some assays where six kinases were coupled to an HC 30 M chip in duplicate, and we injected kinase inhibitors from three micromolar and three fold dilutions, and this is done in some 3% DMSO buffer.
0:29:19.2 DB: This is staurosporine, this is a 466 Dalton compound. So for us, very small. And you can see, we can clearly elucidate to kinetics to these different kinases. The results match the specificity and the binding rates that we'd expect to see from the literature. Another one, sunitinib, this is a 530 Dalton interaction, you can see it has a quite different specificity profile, but again, we're able to characterize both the kinetics and the affinity of these interactions. So we're very comfortable with the LSA-XT saying that analyses down into about the 500 Dalton range are totally doable. And there are some applications where using the array in this one-on-many format makes a lot of sense, like kinase inhibitor profiling.
0:30:14.2 DB: Another application that we've been working with on the LSA-XT is Fc receptor binding, and some of these... They're not small proteins, so it's not a signal limitation, but they do have rapid dissociation. So the faster data collection rate of the LSA-XT makes this nice. It's also a very high throughput, you can put down many Fc receptors into the array and inject your antibodies of interest and collect all that data in parallel. So this greatly speeds and significantly reduces the materials required for these assays based on the experiments our customers were doing in the past. Also, I will mention, Sino Biologicals has a quite robust, I think about 30 Fc receptors, many of which are biotinylated, in their catalog, so that's probably a great source to build an assay like this.
0:31:13.2 DB: This is just some data, these are 11 human [0:31:21.3] ____ Fc receptors spotted in eight replicates each, so each row is the replicates and each column contains the different receptors as you go down. So you can see we get really good agreement across the replicates, we're able to accurately measure both rapid and low affinity and high-affinity receptors. And this is... Everything you're looking at here, if you remember, is one injection per concentration of the antibody. This is trastuzumab. So very little analyte is required for these experiments to generate such robust data. If we zoom in on the data, you can see we're able to get really nice fits for both subnanomolar and micromolar interactions.
0:32:07.0 DB: Okay. And with that, I'm gonna switch to talk about epitope binning. So why do people want to do epitope binning? The function and the mechanism of action of therapeutic antibodies is linked to its epitope. The affinity of an interaction can be pretty readily assessed and engineered or optimized, but the binding epitope is innate and you can't take an antibody typically that binds to one epitope and mutate it to bind to a different one. But that's really the core part of what that antibody is. And so early epitope characterization can serve as a surrogate for functional diversity. Sequence diversity is one thing, but you can have a fair amount of sequence space that reflects a limited epitope space, so it's good to verify that in addition to sequence diversity, there is epitope diversity especially in a new discovery panel where there's interest in multiple MOAs or maybe combinations, bispecifics, cocktails, things like that.
0:33:13.8 DB: You can also use high throughput epitope binning data to inform large sequence sets. Say you did a campaign where you ended up with 3000 putative binding sequences. You can probably use bioinformatics to organize those into family trees where you know clone sequences are relatively related. So say you took 300 relatively unique sequences and ran a full-scale epitope mapping or epitope binning experiment, you could then bucket a large portion of that sequence diversity into expected epitope space. It can also be used to identify sandwiching pairs or establish IP if there's publications or patterns around certain binning profiles and epitopes. The more clones you include, the more likely you are to show a difference.
0:34:05.0 DB: So the assay setup for epitope binning on the LSA, excuse me, scales linearly, so if you add one more clone to the assay, it's one more well on your ligand plate and one more injection of analyte and sandwiching antibody in the final assay. Most other platforms, if you try to work out the logistics of scaling these pairwise competition assays, they scale exponentially in complexity and typically are limited around 20 to 30 clones maximum, where on the LSA we can scale all the way up to potentially 384 by 384 in one automated run. I'd probably recommend breaking it up into a couple where you have a 384 spot array and you only do 180 ligand or analyte injections for each one and then merge those files together, but those are details. This generates a huge amount of data.
0:35:00.9 DB: If you were to do a 96 by 96 competition binning assay, that's over 9000 interactions, if you did 384 by 384, that's almost 150,000 interactions that would show up in your heat map. So each unique interaction can be thought of as a probe. You can imagine if you have four antibodies and you've binned them together, you can tell whether they compete with each other, but you're not gonna have very good resolution about whether there's any subtle differences in that. But if you have a diverse set, so you have 96 antibodies that bind to a wide range of epitopes, you start to uncover subtle differences between them based on their shared competition profiles because each antibody that you inject or use in the assay can potentially reveal differences in the others.
0:35:56.6 DB: The LSA epitope binning workflow is outlined here, the process starts by immobilizing the clones onto the array, and then we have two assay formats that we typically use. The one on the top shown here is what we call "the premix format," where you inject either the antigen alone or the antigen in the presence of a potential competitor antibody. And if it can sandwich, you see additional binding, if not, it will inhibit the binding. This is really useful for multivalent analysis. If you have a dimer or a trimer antigen, this is the preferred format. If you have a monomer, you can use the one shown on the bottom, which is called "the classical binning assay." And this is where you inject the antigen and then subsequently inject the sandwiching antibody and look for a gain of signal if you have a sandwiching interaction.
0:36:49.1 DB: So once those experiments are run we have analytical software where you define cutoffs and thresholds, and then the software will automatically make and sort these heatmap and these network plots that show how the epitopes relate to each other of these clones.
0:37:04.4 DB: A little bit more on the binning cycle here. So this is a classical binning cycle where you have a baseline followed by an antigen injection. And the green bar is actually a normalization bar, where it scales the binding of that antigen to one. So if you were to have, say, some loss of surface activity over time or some variance there, it still allows you to set global cutoffs because it does a normalization. The second bar is a report point bar, and it's looking at the average value of the signal under that bar, and that is what is used to populate the heatmap. And it's actually the difference from the blue control injections. And so even if you have dissociating antigen, you can still set cutoffs for that. And then it allows you to position cutoffs, and everything that crosses the report point bar in the green is a sandwicher. Everything that crosses it in the orange is a blocker. And these can be adjusted on a per ligand basis if needed.
0:38:09.6 DB: The analysis software interface is shown here for the epitope binning. And this is really powerful because you have three views that are all interrelated. So there's a sensorgram view, there's the heatmap view. So immobilized antibodies are rows, injected antibodies are columns. If they're sandwiching, they're green, if they're blocking, they're red. And if they're self versus self, they have the dark outline.
0:38:33.0 DB: And then we have the network plot. In this plot, every clone is a node, a circle, or a square. If they're competitive with each other, there's a cord connecting them, and if they have the same competition profile, they're clustered into one of these colored epitope bin regions. And these three plots are interactive, so in this case, this cord was clicked, it highlights the cells and the heatmap related to that relationship and shows you the sensorgram. So this was a asymmetrical interaction. So this allows you to dig deeply into your data, explore the relationships in the heatmap and the network plot and see whether you believe all these calls or if you need to adjust your settings to properly interpret the data. This is really an enabling feature to work with these large data sets.
0:39:26.6 DB: I mentioned the bin network plot, where if the clones have the same competition profile, they are clustered into one of these groups. Another thing the software does automatically when you sort these heatmaps is it generates a dendrogram. So this shows the level of shared competition profiles across clones as you go up. There is an option to set a custom community threshold where you can define basically an arbitrary cutoff height and define more generalized bin communities. And this is really useful, we'll talk about this a lot in the CoVIC example at the end.
0:40:05.8 DB: Also, there are some great visualization tools where you can color your data by orthogonal data. In this case, that's showing whether or not the clones are mouse cross-reactive, what library they come from, or what subdomain they bind. You can also do this numerically with potencies and affinities. And this is just a toggle in the option after you load in tables of data.
0:40:31.8 DB: One more thing on epitope binning. A common application is to look for sandwiching pairs. If you're doing biomarker analysis, PK/PD studies, or looking for infectious diseases, you need to find highly efficient sandwiching pairs. So on the LSA, typically in 24 to 40 hours, you can take a panel of clones, create an array, do a kinetic analysis to the target to figure out on and off rates and then do a symmetrical pairwise competition binning, where you'll see, which clones sandwich well together and which ones don't and even identify which ones are likely to be more efficient capture or detections based on their behavior. And this typically translates really well to other plate-based or more conventional immunoassay formats.
0:41:22.0 DB: Okay. With that, I'm going to move on to talking about the CoVIC study. This was a large collaboration study run by La Jolla Institute for Immunology, and I think it's one of the richest examples of epitope binning data being overlaid with structural data. And I was a part of this and it was published in Science. The CoVIC Consortium was a global nonprofit consortium funded by the COVID-19 therapeutic accelerator, Bill and Melinda Gates and some others. As I said, run by LJI, and it involved a number of labs around the world collaborating to characterize antibodies that were contributed to the consortium that neutralized SARS Covid-2. So a number of biologies like viral neutralization, Fc receptor interactions, viral escape, were all looked at, but two of the groups, Duke and us at Carterra were using the LSA to do affinity and epitope binning characterization. So about 75% of the 400 antibodies that were contributed to this study bound to the SARS Covid-2 RBD, which is the receptor binding domain, and there was a small set beyond that that bound to either the N-terminal domain or only full-intact spike.
0:42:43.9 DB: So the binning I did, I'm going to talk about here, is for the RBD binding clones. So these clones all bound to a shared globular protein that we could express, or the LJI expressed as a separate entity. As I mentioned, there was an affinity component to this study, the group at Duke was primarily responsible for reporting this data. You can see they generated affinities to the RBD, which was a monomer and gives discreet one-to-one kinetics and quite a bit of kinetic diversity across the panel from the CoVIC. And then they also looked at spike binding, and this is a chimer with some avidity, so the kinetics are slightly less descriptive but still illustrative of the binding properties. I did the epitope binning of the entire panel, this is a representative assay from that, this is 170 ligands by 188 analytes, almost 32,000 interactions shown. This is a single experiment, this is really the way you would like epitope binning run at this scale to look. The matrix is highly symmetrical, where you're getting equivalent competition generally in both orientations, and we're able to find some really discrete, shared behaviors to make some nice communities. I'll highlight a couple of those here.
0:44:03.5 DB: This is a community that was identified, you can see that they are competitive with each other and they share many competition and sandwiching relationships, and it's generally coherent how they do it. This purple community is a closely related community, but it's different in that it has clones from these other... The yellow and blue community are sandwiching, wherein the yellow community, they were blocking. So even though these clones compete with those and they share many relationships, due to the resolution of this assay, we're able to say this is a shifted epitope with a different behavior. And you can see you can keep going on highlighting communities that have additional properties and differences. So this is how this is applied, we try to find coherent breaks in the data where there's obvious shared behavior.
0:44:57.3 DB: So this was done... This was the dendrogram and some community assignments from the CoVIC consortium. LJI did cryo-EM on selected antibodies from these various communities. Over 40 antibodies were evaluated for structure and they found that the binding interface from Cryo-EM was highly related to the competition epitope cluster bin. And so we mapped it here on the RBD, you can see where the different communities bin. The dotted line is the ACE2 binding site, so you can move around various faces of that as these epitopes shift. So this was really interesting, really great confirmation that these communities, if applied strategically, can really translate direct to discrete structural domains.
0:45:49.7 DB: It was interesting that there was some patterns with variant resistance from the different communities. You can see the ones that are dark had high levels of neutralization. And all the way through delta and mu, there were clones that seemed to have real pockets in certain communities of resistance and branches on the dendrogram. As the Omicron strain developed and the number of mutations increased, this became hard to maintain for many of these antibodies. And you can see that if we include the BA1.1 and later Omicron strains in this, that it becomes a bit more stochastic, which ones are resistant. There are still some little pockets of resistance, but this is really high sequence diversity to expect these antibodies to bind. So out of the entire panel, there were 66 neutralizers that maintain Omicron resistance.
0:46:54.6 DB: LJI published a second paper looking at the mechanism of this resistance and they found some interesting things in that, I'll talk about in the next slide. That the clones that were antibody-derived and bound to the RBD, all of the ones that had potency did so with avidity. They would bind to two monomers of spike within a trimer, much like the biparatopic antibody that was shown earlier from NIAID. There was one subdomain, RBD4B, which actually still maintained potent neutralizers with monovalent binding, and it's interesting that in that area or region where those clones bind, there's three mutations in some of the omicron strains, two are conserved and one is right on the edge. So those clones are able to bind around the mutational set and maintain potency, whereas the high avidity in the other ones seems to maintain some potency as the affinity is enhanced. And I think there are structural limitations to how the spike can move when bound bivalently by a nanobody. So this is another real interesting paper that came from this work.
0:48:13.0 DB: And with that, I'm gonna thank the contributors for these, the team at Carterra, the folks in the CoVIC consortium, especially Sharon Schendel, Catherine Hastie, Haoyang, and all the CoVIC participants.
0:48:31.2 DB: And I'll end with one final summary slide, I think high-throughput epitope binning is broadly becoming an integral part of early antibody discovery projects. Epitope competition data is crucial for selecting antibodies for cocktails and bispecifics. The LSA really allows you to run kinetics at a scale that is unrivaled. And then I want to thank Sino Biologicals for the invitation to present today.
0:48:58.5 RB: Thank you, Dan, that was an excellent presentation, we really appreciate it, wonderful technology, lots of great data. I'll go ahead and jump right into the questions as it looks like we're getting a little bit short on time here. The first question is from Rahela Kaurisakhani. And Rahela asks, "Is it possible to use cell lysate as a source of antigen for epitope binning?"
0:49:33.1 DB: It's possible, yeah, as long as there aren't other components in the media that are binding heavily to the surface or the antibodies. I've actually done quite a bit of work like with kinetics with solubilized cells to look at membrane protein binding. I think if you have a mixed micelle detergent lipid, you can have a high expressing cell line, lice it up, spin out the particulates, and then inject that as a titration series of antigen. And the antibodies should have sufficient specificity to recognize that component, so that can definitely work.
0:50:09.1 RB: Great. Great, and another question from Rahela is, "Can you use phage antibodies instead of actual purified antibodies for the epitope binning studies?"
0:50:23.7 DB: Phage display derived, like SCFEs or VHHs, the way most people make periplasmic extracts or bacterial supernatants tend not to be high enough concentration to use in epitope binning. You can absolutely get kinetics with them because if they have an expression tag, you can capture them to the array and inject the antigen. But in epitope binning, there's a component of injecting the analyte over the surface, where you really need a molar concentration of that analyte high enough to drive binding. If the clones are... If your expression's good, and the clones bind with very high affinity, then you might be able to get away with it, but typically, if you have a diversity of affinities and a diversity of expression levels, those assays would be really hard to interpret because you might see very small binding signals in that sandwiching response. And it's just due to the concentration of analyte and solution isn't high enough to drive that binding or block the binding sites in a premix assay. So typically, you're better off using either purified or high-expressing samples for those things like Expi293 transfection soups have plenty of antibody in them to use in epitope binning, or even hybridoma soups, but bacterial extracts tends to be dicey.
0:51:44.6 RB: Great, thanks for that answer. Yushan Hao is asking, "Can you perform kinetic analysis for epitope competition beta?"
0:52:00.2 DB: Not sure. Those are separate applications. Yeah, you can look at binding kinetics to the epitopes you're interested in. And the competition studies are typically not used to interpret kinetics. Usually, that's part of the same workflow but a different set of experiments. We do have some applications around Protax, the targeted protein degraders which are hetero bifunctional molecules that are binding to two components. And we measure affinity of ternary complex formation versus the two individual binary complexes, and you can calculate things like cooperativity or inhibition from those by comparing the rate constant. So maybe that's what you're thinking.
0:52:53.3 RB: Yeah, I think that probably answers the question there. Wen Gao is asking, "How is a neutralizing antibody recognized by LSA?"
0:53:03.5 DB: Yeah, so typically, when people are doing neutralization experiments on the LSA, it's that if you're looking for an antibody that will inhibit like, say, a receptor ligand interaction. If you bind to a receptor, can the ligand of that receptor still bind? Or in SARS Covid-2, it was, If you bind to that RBD or that spike molecule, can the ACE2 target still be X? So it's a pseudo neutralization study. But people do it quite a bit, and it tends to translate relatively well to cell-based confirmation assays if you're looking at those simple blocking-type mechanisms. Obviously, there's other mechanisms of neutralization that would be much harder to tease out with this type of application, like if you're dimerizing a receptor that causes it to be inactive on a cell, you might not see that if it still allows ligand binding.
0:53:56.9 RB: Great, thanks for that. My colleague Amy Sheng at Sino is asking, "Is there any buffer restriction? For example, is glycerol or trehalose a problem on the platform?"
0:54:08.7 DB: Yeah. The system is chemically compatible with those solutions, when you're running SPR, you need to be cognizant of refractive index mismatches between the sample and the analytes that you're injecting because if you get very large differences, we call them "bulk shift," bulk refractive index shifts, it can make it hard to interpret the data. But if you had samples that you couldn't remove the trehalose in, what you would want to do is have a running buffer that matches that refractive index. And you can do a simple test by doing a buffer injection to see if you get a big response or not, so you can add trehalose to your buffer or match them. Like we do a lot of experience experiments with small amounts of glycerol or DMSO or other things that do contribute refractive index, but we want to match our analytical samples to that, especially for kinetics. It gives a much cleaner result if you can match those.
0:55:11.6 RB: Right. Well, this question is from Raghu Sanganapari, and it's about really the universality of your epitope binning software. "Can that software be used with other instruments or it's only restricted to the Carterra instrument?"
0:55:24.7 DB: The software is developed to mesh directly with the LSA data and it's only available to customers that have an LSA. That being said, there is some capability to import like Excel-based tables of data in there and do some processing to make network plots and stuff that sometimes our customers take advantage of. And if you're interested in the technology in general... I didn't say earlier, if you want to see LSA data or try it out or learn more about the platform, reach out to us, you can contact us at our website and things.
0:56:18.1 RB: Ghazia Shafiq is asking, "How many times could a chip actually be used? So a chip has the same ligand on it, like with different antibodies. Can you do like multiple screens, I guess?"
0:56:28.8 DB: Yeah. So the capture surfaces, like if you make a antihuman Fc or use a protein AG or a nickel chip, they can be used quite a few times, you can strip them and reload them, even take them out of the instrument, store them and reload them a few times, not indefinitely. Assays, where you covalently immobilize an array of ligands, like you would do for an epitope binning experiment, those tend to be more single-use. You use them... You can use them extensively while they're in the instrument, but once you pull them out and make other arrays, they're done, so.
0:57:07.8 RB: Right. I have a question here I think everybody would be interested in. "I'm just wondering if you could talk about the advantages of using SPR technology as opposed to plate-based technology like ELISA, I mean, why is it so much better than an ELISA system?"
0:57:23.1 DB: Sure. I mean for things like early kinetic characterization, you're really learning a lot about the system: Rate constants, affinities, and it's easy to include things like specificity screening if you have multiple isoforms of a receptor or off-target proteins, those can all be incorporated into these initial screen runs. And so rather than an iterative process, where you find hits, select them, scale them up, and then come back to it to try, "Okay, now let's understand the affinity and then select leads." You really have all that information right at the beginning, so you can jump right ahead to putative lead selection or move smaller amounts of clones into downstream assays than you might otherwise be able to do. Also, in things like epitope binning, it scales really easily on the LSA, 96 by 96 binning is two plates, a plate of ligands, plate of analytes. On an ELISA, it'd be like 100 plates.
0:58:26.5 RB: Right. Right. Just much more high throughput and capable in that respect.
0:58:33.2 DB: Yeah.
0:58:34.2 RB: Thank you, Dan, I appreciate that. We've run out of time, I apologize to Chu Nan Chen and many others if I didn't get to your questions today. Please email those over to Sino, and we'll forward them on to Dan and others at Carterra, get answers for you. Let me just go ahead and briefly thank all the attendees. We had a huge turnout today, appreciate everybody attending. I just have to say this because it's very interesting where everybody attended from very quickly. We had folks from Egypt, Germany, of course, all across the United States, the UK, Poland, Canada, India, Portugal, Yemen, Switzerland, Iran, Nigeria, and Ukraine. So thanks for a wonderful attendance. Certainly, Dr. Bedinger, we appreciate you lecturing today under the Sino Lock and Key webinar series, that's fantastic work. Congrats to you and congrats to the entire Carterra team on the exciting technology and the wonderful success that you've achieved.
0:59:34.9 DB: Well, thank you very much, thanks for listening, everybody.
0:59:37.7 RB: Yeah, and one other person, my colleague, Max Blechman, who organized and executed this talk today. Thank you for all your hard work and your efforts, it was another successful webinar. And with that, I will say good night, good morning, good evening, wherever you are, and that completes this week's webinar, thank you.