Posted by Daniel Bedinger PhD

Erica Ollmann Saphire, PhD, Professor, La Jolla Institute for Immunology

Presentation at the National Institutes of Health (NIH) by the Coronavirus Immunotherapy Consortium (COVIC)

Dr. Erica Ollmann Saphire of the La Jolla Institute for Immunology presents how the Gates-funded COVID-19 consortium has fully characterized 350 therapeutic candidate antibodies from all over the world in record time. The revolutionary Carterra LSA has been integral in identifying the antibodies’ mechanism of action so potential therapies can be pursued and how the best therapeutic cocktail can be developed.

Dr. Erica Ollmann Saphire from the La Jolla Institute for Immunology

0:00:03.3 Speaker 1: This session will begin with the presentation by Dr. Erica Ollmann Saphire from the La Jolla Institute for Immunology, followed by a panel discussion moderated by Dr. Adrian McDermott from the Vaccine Research Center and NIAID. Dr. Saphire, the stage is yours. You can start sharing your screen.

0:00:22.6 Erica Ollmann Saphire: Thank you. So, thank you to the organizers for the invitation to speak today. I'm here on behalf of a very large consortium, CoVIC, the Coronavirus Immunotherapeutic Consortium, which was launched with dual funding from the Bill & Melinda Gates Foundation and a supplement to our Center for Excellence in Translational Research, U19 of NIAID. The goal of the study is to figure out which antibodies are best, what features that are that makes them best, how we might make combinations better than any of those any lab could come up by themselves and how to make these therapeutics available to everybody at low cost of goods.

Idea for this consortium came from Ebola

0:00:57.6 ES: The idea for this consortium came from Ebola, a work we had been doing in the field at the time. In 2013, right before the large Ebola virus pandemic, the research field had antibodies that would neutralize in vitro, but not protect non-human primate models in vivo, and other molecules that would protect non-human primates in the absence of any good neutralization. And so when faced with molecules that neutralized, it didn't protect and other molecules protected but didn't neutralize. We wondered what we were missing.

0:01:29.6 ES: Was this data telling us that neutralization wasn't the most important thing or that we hadn't been measuring it in the right way? Was it telling us there was something else in addition to neutralization? Did we need to learn something about the Fc and its protective activities, or was this data telling us that we needed a cocktail instead of a monotherapy?

0:01:48.6 ES: We reasoned as a research field that we needed to do a broader study. We needed to understand what it was that leads to protection, and we needed a statistically significant pool of antibodies in order to do so. Perhaps, we just found the exceptions and not the rule so far. So, we recognized, as a body of scientists, that what we needed to do was build a bigger study. We needed better tools. We needed more samples, and the issue was that everybody was working in a silo. Everyone was working on antibodies from their own lab that they had selected in a certain way and using their own assays to analyze them, and different assays across different laboratories give different results.

We needed to build was a research framework

0:02:25.8 ES: The other issue was that there was a pandemic, which was... Had cases doubling every two days. And so, what we needed to build was a research framework that would enable us to do a broader study and work together, a framework that would enable collaboration and the sample on the data sharing that we all knew we needed, but that would function within an environment built on competition. The research enterprise is built on competition, because competition drives innovation. Competition drives you to get results faster. Competition drives you to do a better job. And so how do we build an environment whereby we can share samples, but intellectual property is protected and people's ability to publish is protected?

0:03:07.2 ES: And so we built a study whereby all the samples were blinded and codenamed, so nobody knew whose was whose, but we could look at aggregate data of how antibodies performed. Intellectual property would be protected, and we could run parallel experimental tracks with a fast track to get molecules to our clinic as quickly as we could and a parallel comprehensive track, whereby we looked at more samples with more different kinds of assays to understand what we, together as a field, were missing, and that program was supported by an NIAID Center of Excellence in Translational Research or CETR.

0:03:40.2 ES: So that CETR launched the global collaboration called the VIC, the Viral Immunotherapeutic Consortium, which ultimately united 44 previously competing labs across five continents, whether they were academic, industry, or government laboratories, into a single comprehensive effort to understand which antibodies are best against Ebola and how we find them, what makes them best. We looked at 200 different monoclonal antibodies across three years, so in this plot, the antibodies that we looked at are listed in the vertical axis and the vertical rows and everything we measured about them in the horizontal axis, the features.

0:04:14.8 ES: We look to see which variables correlated with other variables, which features or functions track together, does a function happen at a certain epitope? We did a logistic regression to figure out which high throughput assays would best forecast in vivo success so that we could figure out which were the best ones to do, which ones would be more predictive, and we found that we were... There were better ways to measure neutralization and we're missing activities of the Fc. So together in that program over a couple of years, we published 78 papers, conferred 10 PhDs, revised assays used in the research pipeline and forwarded six new therapies to the clinic where none were before.

0:04:54.3 ES: So it was that framework that we were asked to pivot to SARS-CoV-2 when the new epidemic emerged. And the funding for this program came from the Bill & Melinda Gates Foundation and a supplement to our CETR U19. And the goal of this program was to compare antibodies against SARS-CoV-2 side by side, figure out which features led to the most effective protection, and if we could make better or more potent, more protective cocktails from a global pool of samples than any one laboratory could alone.

Different about this pandemic

0:05:25.8 ES: Now, there's something different about this pandemic as well. Before, for other emerging viruses, we would select things in vitro, in plastic, then put them through rodent and then non-human primate models, and very few would head toward clinical use. In this epidemic, we have clinical studies ongoing while other things are being discovered and evaluated in vitro, and so we have an opportunity to use that clinical data to address and inform whether or not the in vitro assays and animal models we are using are predictive in what we can learn in all these different studies, so we have an opportunity to get more complete information and human information at the same time.

0:06:08.2 ES: We wondered if companies advancing these products for therapeutic use would be willing to participate and they did. We now have over 350 different monoclonal antibodies in the study and they've all been discovered in different ways. Some were found in survivors of SARS-CoV-2, some were re-engineered from the original SARS in 2003, survivors of that disease. Some were modeled or built in silico. Some came from immunized mice, maybe populated with human antibody genes.

0:06:37.4 ES: Some are multivalent constructions and some are nanobodies. And for each one of these antibodies, we are looking at its ability to bind to different, to spike different sequences and domains of spike, its competition with other antibodies, its ability to confer neutralization in three different assays, its ability to confer Fc protection measured at a systems level and in cellular assays, its structure, its location and propensity of escape, its in vivo protection and more.

0:07:02.9 ES: For each one of these antibodies, we are looking at their ability to bind spike... Spikes domains, variance of spikes, their competition with each other, their ability to neutralize pseudovirus and live virus systems, their ability to bring about Fc effector functions on the systems level and in cellular assays, their structures, their location and propensity of escape and their ability to deliver in vivo protection and more.

0:07:26.9 ES: The antibodies come from laboratories in four different continents. About two-thirds from industry, one-fourth from academia, 8% from government laboratories. And a lot of the molecules come from small startups and might not get attention on their molecule a number of way. About 8% come from the major multinational corporations and we are also analyzing the therapeutics that are being considered by the active clinical trials to provide that independent analysis in standardized platforms.

0:07:55.6 ES: Now, the array of company sizes here probably reflects the array of companies in the United States and abroad, that there's a lot of small companies. But this is what it looks like from over here. Every antibody is codenamed, blinded and they're all given the same fair shake. What happens when the antibodies come in is that they're given that codename. Now the donor knows which antibody is theirs, and they get all of our data back and they can use it for their IND Filing.

0:08:22.6 ES: To me, the data are anonymized, and so I can look at the aggregated data of what do antibodies do and what do assays tell us. So, we aliquot and distribute these antibodies. And we send them out to different experts in the field, each doing a different kind of assay to evaluate the function of the antibodies. The data is then aggregated into a database called CoVIC-DB. There are a number of contributors.

[automated voice]

0:08:49.5 ES: And a number of people running the assays. And I'd very much like to thank these groups of people that are doing a lot of the heavy lifting to build this publicly available database. In that database, you can see data for your own antibody and you can see the unblinded data for everybody else's. So if you go to our website,, and you click on the database, it looks something like this. Where for example, if you were interested in antibody number two, CoVIC 2, you could look and see how... Where it ranks in neutralization in the whole pool, where it ranks in neutralization among others of its epitope group. So CoVIC 2 is among the most potent in group two. And you can look at its footprint by structural biology.

Kind of information on Antibody 96

0:09:28.0 ES: If you're interested in Antibody 96, you could find that kind of information here. Now Antibody 96 is kind of interesting. It's not the most potent neutralizer. In fact, it's kind of toward the bottom of its epitope bin five, but it does deliver very potent protection in vivo. Now, that's a study that we were running in this broad database. And six months ago, the variants of concern emerged and we were asked to pivot the studies to also find out which CoVIC antibodies would be resistant to these variants of concern.

0:09:57.9 ES: So the first thing we did toward that effort is the competition analysis. And this was done by Dan Bedinger at Carterra, using his high throughput surface plasmon resonance platform. So, we have antibodies in CoVIC against the receptor-binding domain, the N-terminal domain, S2, everything else. The majority are against the receptor-binding domain because this is where one achieves very potent neutralization and this is where people looked first to find a candidate, therapeutics. So if we look at the first 250 in the receptor-binding domain reactive antibodies, we find that they divide themselves into seven major groups. We call these communities. Seven major kind of competition bins or footprints of where they bind.

0:10:38.5 ES: Now other studies divided therapeutic antibodies into classes or groups, looked at them by structure or looked at them by germline. This is a slightly different and complementary approach. This is by competition. Which ones interfere with each other. We chose this method because we knew we were going to look for a cocktail. And so the separation here is by degree of physical competition to bind the RBD.

0:11:03.6 ES: The first major division here removes the green and cyan group from the others, four and five. The second major division over here picks groups one, two and three, and separates them from groups six and seven. Now you can draw as many divisions as you want. Antibodies bind across the spike in a complete spectrum. And the cutoff, this red bar across the center, you can raise it or lower it as you wish. You could have few groupings, you could have all the way down to 350 different individual groupings, but we place this cutoff bar where it makes practical sense in divisions of different antibody behavior.

0:11:42.6 ES: So, we have seven major communities that map to places in space, but within them, we can name 17 smaller clusters. So, group two has an A and a B, and the B can be divided into one, two and three. And so you might think of these things in continuum like a city. So for example, if the RBD is New York City, it's divided into boroughs. The boroughs are divided into neighborhoods, and then there's individual streets and individual houses on those streets. So, you know that the line between one neighborhood and another blurs, and ones on the edge has some characteristics of one neighborhood and some characteristics of the other.

How we've divided them based on their behavior in biology

0:12:16.3 ES: This is how we've divided them based on their behavior in biology. Now we have solved structures of a couple members of each one of these groups and subdivisions to map where they are. So, for example, these three are directed against the RBM, the receptor-binding motif, exactly where ACE2 binds. And so ACE2, the structure that is shown here in red. And we have three different footprints within that RBM. RBD-1, which overlaps precisely with the ACE2 binding site. RBD-2 shifted a little bit one way, RBD-3 shifted a little bit in the other way. So, we map the footprints and we could map the angles. We have two classes against the outer face of the receptor-binding domain. This is the domain that's always exposed, whether the RBD is down or up.

0:13:02.5 ES: Ad then we have two classes against the inner face. These are ones that only bind the RBD when it's up and not when it's down. And so, this division is backed up by a lot of structure, so we're building this database here and this will be available in the EM Database and also in bioRxiv very soon, mapping lots of structures for each one of these domains and we're showing you example images, 2D classes, from looking at the Fab fragment of the antibody and also the complete IgG, where you see these Y's coming out. So, this is based on a lot of structure and a lot of biophysical analysis.

0:13:36.5 ES: So, we're looking at each one of these antibody groups by their footprint because their footprint determines which mutations they may or may not be susceptible to. So, for example, in each one of these structures, the footprint of ACE2 itself is outlined in the black dotted line, the footprint of that group of antibodies, and they're remarkably consistent, 2B1s always hit the same spot. They overlap to different degrees with the ACE2 binding footprint.

RBD1 is nearly precisely the ACE2 binding site

0:14:02.4 ES: So, RBD1 is nearly precisely the ACE2 binding site, you see it overlaps that black dotted line very closely, and on the right, EM structures of spike in a side view and a top view of the three Fabs of, for example, COVIC259, Antibody 259, which falls into group RBD1 overlapping that ACE2 binding site. RBD-2 shifted a little over to one side, you're looking at Antibody 252 and its Fab fragments.

0:14:31.5 ES: RBD3 is shifted down a little bit onto the lower, what we call the mesa, that's Antibody number 80. And then of course, the inner face and outer face antibodies are shifted, sometimes entirely outside that ACE2 binding footprint. Now if you look at these in more detail, so here is a model of spike, oriented in the side view, so the viral membrane is toward the bottom and the target cell will be toward the top. And this has one receptor-binding domain, one RBD lifted into the up position, as if it were going to interact with the ACE2 receptor.

0:15:04.4 ES: Now, the next view will be from the perspective of the target cell looking down onto spike. So if you were ACE2 looking on to spike, this is what you would see, we'll zoom into that single receptor-binding domain and the residues I'm coloring are those of high frequency mutations that have emerged alone or in variants of concerns, so we can map where they are. Now, the competition binning makes perfect sense with the structural biology. You can roughly divide the receptor-binding domain in half between the ones that have the inner face, that they're only exposed when the RBD lifts up and the one on the outer face, that's always exposed whether the RBD is down or the RBD is up.

0:15:46.7 ES: Here is RBD1. This is also the ACE2 binding site, and there are molecules in CoVIC which are ACE2 linked to an Fc. RBD-2 shifted a little bit more toward the inner face, RBD-3 shifted a little bit downward, RBD-4 and 5 on the outer face, shifted up and shifted down, RBD-6 and 7 overlap, but behave very differently on the inner face. Now how does this compare to what you have seen before? So, Pamela Bjorkman's Lab, Chris Barnes, originally defined four classes. Class 1, those against receptor-binding motif are RBDs 1, 2, and 3. What we see about these antibodies in CoVIC is that they nearly completely or completely block binding of ACE2.

0:16:29.1 ES: When we look at the IgG instead of the Fab, it binds with a one-to-one stoichiometry. One IgG to one trimeric spike. They frequently bind bivalently with both Fabs parallel like goalposts and the Fc sticking up from there. They require the RBD to be up in order to bind. And if you look at the structure of what an IgG looks like, down to that spike, you can see that goalpost thing here, where you see spike at the bottom, you see the two clear Fabs with the little holes in the center and a little blur at the top. The blur at the top is the flexible Fc attached to those Fabs. Ian Wilson's group recently published another definition of spike receptor-binding sites and they divided these into RBS-AB and RBS-D.

Pamela Bjorkman's group also defined Class 2

0:17:19.1 ES: Pamela Bjorkman's group also defined Class 2, now this looks to us like RBD-4, Ian Wilson has called it RBS-C. Most of the ones in this group of outer domain are ACE2 blocking and they can bind spike, whether the RBD is up or down. RBD-5 forms Class 3 also where S309 binds. These do not block ACE2. They combine the RBD when it's up or down, and the interesting thing about these IgGs, I'll show you one later, they can crosslink different spikes together. So, you can see that they can either, the IgGs, in solution, can either bind two spikes facing each other, and this wouldn't happen on the same virus, but I suppose two viruses could oppose each other, but we can also see is that neighboring spikes within the range that is observed, one IgG can bridge these two. RBD-6 and 7 together form Class 4 and they require two RBDs to be up. We've divided these because they have different behavior in biology, one blocks ACE2, one only sometimes does.

0:18:20.7 ES: So, across the entire CoVIC, we can create a competition grid of which antibodies compete with each other. So, a dark blue are antibodies that compete, light blue are antibodies that don't compete. It's hard to get a sense across hundreds, and so I've zoomed into 50 key examples. I'm showing you here a competition grid of 50 example antibodies divided into their groups by rainbow coloring one to seven across the RBD. We can see a couple of things. So, for example, this is why we have divided this class into groups six and seven. Six competes with the potent neutralizers of 2A, 7 do not. If we look at groups four and five, they both bind the outer domain, but four competes entirely with 2B and five does not. It exhibits some different behavior.

0:19:08.0 ES: And so, this grid gives us a way by which we would choose cocktails. So for example, if you wanted to take advantage of the very potent neutralization observed in 2A and a lot of the clinical candidates come from this orange group too, you could pair it with one from groups four or five or seven. But four and five compete, so you'd have to pick one, so for example, a 2A, a four and a seven would work. If you wanted to choose one of the therapeutic candidates in 2B, you could pair it with a group of five or six or seven. 2A can't pair with six, only 2B and so you might say something binds the receptor-binding motif, but you need to know exactly where in order to choose what cocktail you would pair it with.

0:19:46.7 ES: Six and seven compete with each other. You can't choose both. You have to pick one. So, for example, the 2B with the five and the six. This competition matters because the different footprints typically have commonalities in which are and are not susceptible to different mutations in the variants of concern. So, 2A with a lot of therapeutic candidates is quite sensitive to the K417 mutations. 2B is quite sensitive to Eek E484K. On the far right, the cyan, blue and purple, groups 5, 6 and 7, these are variant resistant, so these are ones that are very attractive for making durable cocktails moving forward.

0:20:30.5 ES: And so, some pairings that you might consider with the larger pool that we have, and keep in mind that we have antibodies from 50 different groups raised in different ways and selected in different ways, these are combinations that you wouldn't have put together any other way. Let's say you wanted to focus on Antibody 249, this extremely potent neutralization or COVIC2 for its potent neutralization, you could pair them with these molecules, for example, and that would give you the advantage of the potent neutralization with some variant resistance.

We've also looked at neutralization escape

0:21:05.4 ES: We've also looked at neutralization escape by each community and I'll show you how we presented these results. So we've looked at mutations individually and as assembled variants and we... There's some terminology here, four by M and five by M, those are mink mutations in a cluster of five and then minus one, because there's a particular point mutation that is interesting in that it enhances antibody neutralization, and we're looking at the full change in pseudovirus neutralization for VSD, where those antibodies that are gray have no effect by this mutation, those that are turquoise neutralized that mutant or that variant better, than wild type, those that are brown are knockouts.

0:21:47.6 ES: And so, what you can see are... For example, this one here, CoVIC 69, that's ACE2 linked to an Fc. Its neutralization only gets better with those variants, as you might expect. Now it's not incredibly potent, but it's an interesting proof of concept. Those in groups 2A and 2B, the receptor-binding motif, now, there are exceptions, but they are kinda universally knocked out by B1351 and P1. The same thing with RBS or group four, but groups five, six and seven tend to be variant resistant. Now, four and five both bind the outer face, but because they have a different footprint, they have a different susceptibility to emerging variants.

0:22:29.9 ES: And so something of interest that you might look at here, for example, CoVIC 268 is mutation-resistant. It might be more durable in time and give you an option for a therapy moving forward. Same thing in group RBD-5, there are interesting molecules there that crosslink spike. Antibodies 251, 170, and 166 are also variant resistant. And with an RBD-6 CoVIC 250 is very variant resistant. And we've done the structural biology two different ways.

0:22:58.6 ES: We've looked at it with an Fab to get resolution and we've looked at it with an IgG to get relevance because the thing that's actually being mobilized in vivo is an IgG. Now, we've all grown up looking at structures of Fabs because this was the only way we could get high resolution and when crystals were our only option, that was the only way you're gonna build a crystal lattice. The flexibility of the Fc or the crosslinking of the IgG would've prevented structure determination by that way. Now that we have tools of cryo-EM, we can do both. And we can think about what happens when the IgG binds and we can learn a lot that way.

First thing to understand is that spike on the surface of the virus

0:23:33.1 ES: So, the first thing to understand is that spike on the surface of the virus is not rigidly upright. And so for example, this work here, this beautiful tomography shows that spike more often leans one way or another, maybe 40 degrees this way or 40 degrees that way. So, they tilt on the surface of the virus. And the behavior of how the IgG interacts with spike has to do with its footprint on spike, its community. So, for example, RBD-1, and this is a human antibody that winds up overlapping nearly precisely with the ACE2 binding site, frequently binds three separate Fabs onto the trimer, and they're angled outward such that it's probably three different IgGs. And if you look at the 2D class from the microscope, you can see that blurry halo across the top, that's the rest of the IgG in space moving around.

0:24:23.3 ES: And you can see the Fabs in red down here. That group gives direct ACE2 competition. Now, group RBD-2 and these are a lot of therapeutic candidates for their potency of neutralization, these tend to bind one IgG for spike, upright like goalposts, you can see the two Fabs very clearly and rigidly, and one single small blur above, that's the Fc. So, they bind onto two RBDs, they leave the third one unoccupied, but they might sterically block access of ACE2 with that bulk of the IgG.

0:25:00.9 ES: So, these bind bivalently, intra-spike, one IgG to one spike. Group five is interesting. Groups five always link spikes together, we rarely ever, ever... Never see a single IgG bound to a single spike. So what you're... Let me show you the structure models on the right first. You're looking at two spikes facing each other. This won't happen on a single virus, it happens when the spikes are soluble ectodomains and can tumble around in any degree in solution or maybe two viruses come together. But the way that these Fabs are organized, an IgG is anchoring two different spikes together.

0:25:37.6 ES: And a lot of classes that we see are the picture on the bottom right, with two tilted spikes linked with one IgG. Now, this antibiotic here, number 96 and group 5 is interesting. You would never have chosen it by its neutralization alone, but its protection is among the best, so it's punching above its weight through some mechanism. It's possible that what it is doing is achieving linkage of spikes, maybe greater ACE2 steric hindrance by crosslinking on the variant's surface. We don't know yet, but it's intriguing. And we should think about how the different types of antibodies behave as IgGs.

0:26:17.2 ES: Those against RBD1 tend to bind three per spike, sometimes they crosslink RBD-2 bivalently in one and every behavior around the circle through different categories. We also have an antibodies against the NTD, and NTDs bind in every orientation around the clock face. I'm showing you three extremities here. And unfortunately, those antibodies against the NTD are very easily scaped by all the deletions that emerge in the variants.

0:26:46.4 ES: That is what we have done in our first year of running, to build this consortium, contract all the molecules, organize ourselves with the contributors and the people standing at the assays, and develop this database body and pivot to looking at the variants of concern. What you can expect from this consortium to the next year is to complete the variant analysis on the entire CoVIC panel. We've done an in-depth analysis of about 50, we're gonna go through all of them now.

0:27:08.9 ES: We're going to expand to more S2 reactive and additional variant-resistant candidates, as people discover these out in the community. We'll complete our in vivo studies and then using that data and the ones in the study for which there's human clinical data available, use that body of data to understand the capacity of our different in vitro assays to forecast protection. Often for emerging viruses, we have to evaluate things in plastic than in rodents and in primates, and very few ever see clinical utility.

0:27:39.9 ES: In this effort, we have the opportunity for the first clinical data to actually re-inform the next generations of therapies. What we'll also be doing is using this very broad array of antibodies, 350 that come from four different continents to make more variant-resistant potent cocktails that would not have come together any other way. Also what we are building is a database, a very broad and deep analysis of therapeutic candidates across the landscape of spike, chosen a different way, selected a different way. This database is being built with fair ideology, so that the data in there is findable, accessible, interoperable and reusable.

0:28:16.6 ES: You can go in there and because it's all anonymized, you can download yourself an Excel spreadsheet of what you wanna learn about antibodies in general. If you have an antibody in CoVIC, you can see how it's faring. In this database, we will list residues that comprise each of these footprints and links to the EM-structures and how-tos and how to make this broadly applicable for the antibodies that you are discovering now and other ones to come, and we'll be using the information from this database to advance variant resistant cocktails.

0:28:45.9 ES: And to come up with the information that will inform the next generation of vaccines, because looking at monoclonals not only gives us that therapeutic or prophylactic option, it also tells us what kinds of antibodies we would really like a vaccine to elicit. And so by studying where these bind and what sort of processes give rise to them, we can make next generation vaccines. This is a very large consortium, I'd like to thank the Gates Foundation, NIH for the funding that launched it, GHR and the Overton Family for emergency support when we needed it to continue the variant analysis. I'd like to thank our program officers for shepherding the program along, and all of the contributors of samples and contributors of data.

Data that was led by the efforts of our program manager Dr. Sharon Schindel

0:29:29.1 ES: The data that I've shown you today was led by the efforts of our program manager Dr. Sharon Schindel, scientists Dr. Kathryn Hastie, Haoyang Li, Dan Bedinger, Tim Germann, Georgia Tomaras, Bjoern Peters and our database manager, Brendan Ha and Mari Kojima. And I'd like to thank you for your attention.

0:29:51.6 Speaker 1: Thank you so much, Dr. Saphire. We're going to continue with the panel. Dr. McDermott, are you ready?

0:30:00.1 Adrian McDermott: Yeah. Now, can the panel light their cameras up so we can start our conversation? Obviously, Chris is gonna get the first questions because Erica pointed out all of his brilliant work that he's been doing. But as I said in our pre-meeting, I come to this very much as an acolyte and not a guru, so willing to learn in the next 20 minutes or so. The assembled expertise here is really quite impressive. I'm going to go and ask Chris just to introduce himself briefly. Chris Barnes?

0:30:41.1 Christopher Barnes: Yeah, thank you. Hi. I'm Christopher Barnes, currently a postdoc in Pamela Bjorkman's lab here at Caltech, soon to be assistant professor of biology at Stanford.

0:30:49.4 AM: Congratulations on that. Mike Diamond, you want to introduce yourself briefly?

0:30:56.6 Mike Diamond: Yeah, I cannot access the video, it's still locked out for me, but hopefully they'll be able to correct it. I'm Mike Diamond, I'm a professor at Washington University School of Medicine.

0:31:11.0 AM: David Montefiori.

0:31:12.0 David Montefiori: David Montefiori, I'm a professor in the department of surgery at Duke University.

0:31:18.0 AM: Penny?

0:31:18.4 Penny Moore: Hi everyone. Penny Moore from the National Institute for Communicable Diseases and University of Witwatersrand in South Africa.

0:31:27.2 AM: Rachel?

0:31:29.5 Rachel Liberatore: Hi. I'm Rachel Liberatore, I'm the Director of Research and Development at RenBio, a startup biotech focusing on antibody therapeutics.

0:31:38.1 AM: And last but not least, Laura.

0:31:40.8 Laura Walker: Hi, I'm Laura Walker, I'm the Director of the Antibody Sciences department at Adimab and the Chief Scientific Officer at Adagio Therapeutics.

0:31:48.1 AM: Fantastic. Clearly, Erica gave such a fantastic talk that she's raised a lot of questions off the bat. So initially, I think we should talk about, what is the prospect for combination therapy and in the different sites that Chris identified, maybe he can expand a little bit on what his predictions or insight would be moving forward into which kind of antibodies we would need.

0:32:21.5 CB: Yeah, so Dr. Saphire presented a great talk, recapitulating our work and really, you think about the variants of concern right now and where these mutations reside, and I think that's really the focus moving forward, trying to understand exactly which class of antibodies can we combine based off of where these variants, where these commonly reoccurring mutations are occurring in the RBD. So, if you go back to her classification, my classification, the Class 1 and Class 2 are those that bind to these, towards the apex of the RBD that overlap with the ACE2 binding sites. And of course, the E4A4K mutation, which is common among a lot of the circulating variants, knocks off the number of these antibodies or mixed these antibodies sensitive to the virus.

0:33:12.4 CB: So, in this sense, you would definitely wanna add antibodies that bind to these other regions outside the ACE2 binding sites, and so this is typically a Class 3 or Class 4 in my definition. I think in Erica's definition, it was Class 5 or Group 5, 6, and 7. The group, Class 3 or Group 5 antibodies that she showed, which can crosslink, enter crosslinked spikes, that's a really interesting class because they can be very potent antibodies. And so when you think about combining those types of antibodies with those from other classes that don't overlap, you can definitely do that based off of what we understand structurally from these antibodies.

What about antibodies that are outside the RBD?

0:33:53.6 AM: So, I don't know if anyone else got any questions for Chris, but Chris, would you combine... What about antibodies that are outside the RBD? Would you combine those as well, so for example, like the NTD?

0:34:08.4 CB: Yeah, no, definitely, you could look at combining NTD antibodies as two antibodies with those that bind into the RBD. Obviously, David Veesler's done a lot of work, as well as group at VRC on defining the neutralizing antibodies within the N-terminal domain, and there seems to be one supersite which people have pointed out, are easily escaped by a lot of the variants, through either indels or mutations in this one epitope.

0:34:37.9 CB: But yeah, definitely the NTD provides additional region in which you can bind maybe two or three antibodies, so maybe two RBD virus plus a NTD binder and that would definitely give you a lot more coverage of the spike and allow for better potent therapeutic cocktails. I think one thing you have to be careful with, with NTD antibodies is can you combine them with some of these RBD antibodies that bind with different orientations that may now or sterically be hindered by NTD binder. So, teasing out exactly how the NTD antibodies target and orient on the NTD is really important if you think about combining them, in that the supersite are those that could potentially conflict with the RBD.

0:35:24.7 AM: Right, so from a antibody development, more industrial point of view, I'll pose a question to Rachel and Laura and say, the Group 5 antibodies, do they represent a very good target to, say, bispecific antibodies or trispecific antibodies and are you working on any of those, if you can tell us?

0:35:56.2 RL: Sure. So, I guess I'll start. So, I think the short answer is yes, that they represent very interesting targets, and I think that's really a key point of a lot of what's been discussed already today, that a real focus on trying to think about conserved epitopes and therapeutics that can get at those epitopes is quite important with the concerns of variants, and to get at one of the points that Chris was making with respect to bispecifics, just the same sort of principles apply when thinking about pairing the different arms.

0:36:34.3 RL: So in a bispecific, each arm is binding a distinct epitope on the SARS-CoV-2 spike, in the same way that a cocktail would have two antibodies binding different sites. And so, not all of them work together. And with a bispecific, there are perhaps even more structural constraints because now, these two different binding sites are physically linked, and so there are pros to that, but there are also cons to that. And so when designing bispecifics, it's very much not the case that you just are thinking about that two sites are compatible in terms of these sort of binding classes, but they also have to work at some level when they're physically linked together, so there are some additional concerns there.

0:37:22.2 LW: And on our end, Adagio, we are not developing bispecific antibodies. As everybody probably knows, there are some manufacturing challenges associated with that. Rather we're focusing on development of broadly neutralizing antibodies in our cocktails of broadly neutralizing antibodies because we think that it might be a more optimal long-term solution to the problem. I think some of the antibodies that are being discovered now, like these so-called Class... I use Chris's nomenclature, the Class 3 type antibodies that seem to be pretty resistant to escape, might not be true in the future as Betty very nicely showed, and we're seeing the emergence of new variants every week now.

Start developing these cocktails

0:37:56.5 LW: And so, as you start developing these cocktails, you might need to develop new cocktails the following year, but I think broadly neutralizing antibodies overcome some of that because they generally target epitopes that aren't readily targeted by endogenous main responses. There isn't much pressure on those sites and they also target conserved residues that might be more difficult for the virus to mutate without suffering a fitness cost. So, that's what we're focusing on.

0:38:20.2 AM: So Laura, while I've got variants of concern, how concerned are you about the variant of AY.1 or even 617.2? Does that really create a challenge if you're trying to put these into clinical trials before you actually get a therapeutic?

0:38:49.6 LW: Sorry, do you mean concerned from a monoclonal perspective or more generally in terms of vaccines...

0:38:54.1 AM: A monoclonal perspective, of course.

0:39:01.8 LW: From the perspective of people that are developing broader antibodies, like our company Adagio and then also at Vir, we haven't seen evidence to date, or we haven't identified any variants to date that are resistant to our antibody or the Vir antibody for that matter, while when we look at other antibodies that are SARS-2 specific, we're seeing that there are many mutations that are rising at relatively high frequency that escaped some of those antibodies. And I'm not concerned yet, but having a cocktail of two that are broadly neutralizing would really increase the barrier.

0:39:37.1 LW: And that's sort of the direction that we're moving now, because of course, you can never be 100%. And as we get closer and closer, get higher levels of immunity in the population, it's unclear where things are going to go. Maybe as there's more pressure on the virus and these sort of common sites, maybe responses will shift to more conserved sites and now there'll be pressure on the Vir antibody site and our site, it's hard to predict that. But I think cocktails might overcome some of that.

0:40:03.3 AM: How do you down-select? I don't know, I may ask Mike Diamond this. In the down selection of some of these monoclonals, what kind of... I know that we've got hamsters, and we've got mice, and we've gotten non-human primates, where do you start with this Mike?

0:40:19.6 MD: Well, the first thing is what Laura and Rachel both echoed, and this relates to Betty's comment is, in the setting of all of these variants and variants that will emerge, I've come to the conclusion that one of the key components is trying to identify antibodies that can really probably bind very much at conserved sites and have great potency. And in Erica's classes, she has all of these great potency data and she's got competition data, and she has structural data on many, but not all.

0:40:54.3 MD: But identifying in those classes those that are actually binding to the most conserved epitopes and figuring out assays that you can do that, either you can look for cross to SARS-1, but it doesn't have to go to SARS-1, it can still bind to conserved epitope for SARS-2, without necessarily binding SARS-1. That's one aspect going forward, maybe I wouldn't have said that at the beginning of the pandemic, but certainly now, in the setting of our understanding of variation, and recombination and change that the virus probably is gonna have under immune selection pressure.

0:41:25.1 MD: The second issue for me, what we have learned over time, is that even though monoclonal antibodies don't appear to generate escape easily, I have a feeling, in the same way that was echoed, is that we are gonna be more comfortable with having... Even with a very broadly neutralizing antibody, a second antibody, to potentially control resistance. And we've seen in vivo that resistance doesn't happen as Regeneron did early on against variants or against historical strains, as long as they have activity in a combination, whereas you do get resistance in ways that are unpredictable as monotherapy.

0:42:05.4 MD: For example, you will get resistance, even though they may not be resistant to a given variant at time A, you may see resistance to that variant but not to another variant, even though there's the same selection pressure. In other words, the sequences and how you get there may actually change the way the mutations occur. In other words, how many nucleotides or which amino acids have to change to become resistant in that particular variant. So, I do think that being cognizant of the evolution of the virus in different backgrounds is important.

Other question relates to in vivo assays

0:42:38.0 MD: And then your other question relates to in vivo assays, which in vivo assay is most predictive? Well, the smaller animal models are models, and we use them all the time, and they need to be corroborated as you go up, but I do think there's utility in the hamster model, it seems to have some predictive ability, and certainly some of the mouse models are very stringent, especially the therapeutic models, and my feeling is if you can show efficacy in those models, that may be one way to move up to the next model and to down-select.

0:43:12.9 MD: There's utility in it, but it needs to be corroborated going up. And part of the problem is that what we are learning is that at least in the animal models, effector function matters in the context of therapeutic activity. I think Erica alluded to the studies that are being done through the consortium in terms of understanding this. Galit Alter's part of it, there's in vivo data. How does it correlate with your nude activity? And that's not always predictable from animal to animal because of the differences in isotype, because of the differences in Fc receptors. And we're still learning how to make those predictions based on activity. The long and short is highly conserved multiple antibodies, probably multiple models, and doing them in combinations that are rationally designed, as Erica said. I can't hear...

0:44:03.3 LW: You're muted, Adrian.

0:44:04.6 AM: Picking up on something else that Erica said was the... Maybe you'll see protection in an animal model that doesn't echo what is found in in vitro neutralization and that would indicate some sort of Fc function. I know that Galit has done a lot of beautiful work on this but has anybody else been looking at this? Has David or Penny been looking at Fc functions in any of their assays, in humans?

0:44:41.2m MD: You're asking me the question?

0:44:42.6 AM: Well, I'm asking... Really, I want Penny and David to lead off on that and then Mike, I'm sure you've actually got data on it, so.

0:44:51.3 MD: I'll wait to hear.

0:44:52.5 DM: So Adrian, I'll start out. My lab in particular is not looking at Fc effector functions, but other labs are. And I think that they're very important. And I'm really glad to hear that they're part of the program that Erica put together.

0:45:11.2 PM: Yeah, we are looking at Fc effector function, but mostly in the context of infection rather than monoclonal antibodies. But I think it might be interesting. One of the things that we don't really understand is what the targets of the most potent Fc effector function antibodies might be, and they may not necessarily overlap with the antibodies that we focus on isolation, which are more, those with neutralizing activity.

0:45:37.0 PM: So, I do think that there's some unknowns in the sense that we probably don't know what the targets are of the best Fc effector antibodies. And nor do we, I think we're only beginning to understand how relevant that might be. So, I think if it's to isolate antibodies with the Fc effector function may give a very different answer to those that are fixed on the isolation neutralizing antibodies.

0:46:02.6 AM: Alright, Mike, do you have any data that would reflect any of this?

Fc effector functions need to be broken down

0:46:06.8 MD: Yeah, we do. I would say also the Fc effector functions need to be broken down, and certainly Galit has a toolkit of assays to break these down and as do others. One of them are gonna be the antibodies that bind to the virion and yet engage Fc effector functions perhaps for complement-mediated augmentation of neutralization via complement deposition or opsonization into cells and then targeted destruction. And then there's of course, the antibodies that are recognizing spike on the surface of infected cells and engaging immune cells directly.

0:46:41.6 MD: And those are different events and may have different qualities of the antibodies that you need to do that to engage the particular components. One of the things that we've noticed is that all antibodies don't bind spike on the surface equivalently. Some antibodies bind with very high avidity and also at higher levels. And so the EC50 binding is different and then the magnitude, the absolute amount that you would see by flow cytometry or some other visualizing assay is different.

0:47:14.3 MD: And this may be a correlate of protection. In fact, we've seen it as in limited data that some of this information correlates with the ability to have effector function mediated protection. The second issue is, and it's different for different regions, for example, NTD antibodies may bind differently than RBD antibodies, than RBM antibodies, in terms of the amount of spike that's actually being seen. The second is that we've seen, at least in vivo, in the mouse and the hamster, and those are the only two animals that we've looked, that therapeutic activity of neutralizing antibodies is augmented substantially by Fc-mediated functions. For example, if you take a LALA-PG variant of antibodies, they don't do nearly as well as the intact antibody.

0:48:00.8 MD: And we've measured PKPD, it's not due to that. It's, they just don't have this added activity, and we've been able to show, at least in the mouse, that certain cell types are important, for example, monocytes are important for modulating inflammation and antibodies also are helping priming functions. And this was seen earlier by Jeff Ravetch and Skip Virgin, I think on the line some place, they published a paper earlier last year, I believe on flu, where they also saw Fc receptor-mediated priming of CD8 T cells, which resulted in a clearance at a more rapid pace.

Data for it in the smaller animal models

0:48:37.7 MD: So, I think that there is data for it in the smaller animal models. Where we don't have really good data for this yet is in monkeys and we don't have, at least not that I've seen, maybe it exists, but I haven't seen data on this, and certainly not in humans yet, because we haven't done paired analysis, let's say ones that are modulated. But I do think there's an opportunity, one, to learn a lot about the biology, a lot about how antibodies engage, different antibodies engage spike, whether on the virion or on the surface, and then potentially optimizing those functions in a way where you could actually get enhanced activity over time. So, I think there's still a lot to learn.

0:49:16.0 LW: Mike, I was wondering if you could speak to whether you have any data to suggest, but antibodies that are able to... That do shedding of the S1 subunits, so some of these ACE2 targeting antibodies shed S1, if those antibodies have reduced effector function?

0:49:35.6 MD: Yeah, we do not have direct data to address that. I believe there are people in the field that have looked at it more carefully than we have. We have not looked at shedding per se, there is this concept that certain antibodies will bind and then, the S will get... The S1 will drop off, and of course, if it drops off the cell surface and then it's not gonna have effector functions. I think some groups do have this data, we don't have it, but I believe that if it is occurring, certainly will impact this, yes. And other...

0:50:07.4 LW: I've seen in vitro data, but not in vivo data.

0:50:08.6 MD: Correct.

0:50:08.8 LW: We might have a lot of uncleaved S on the surface of infected cells, for example, so.

0:50:12.9 MD: Right.

0:50:13.2 LW: A lot of these in vitro assays are performed with stabilized S, and so they might be missing some of this as well.

0:50:19.4 MD: Correct. Yeah, and it is different also, if you transduce S on the surface, as opposed if you have a viral infection with S. I don't think they... We have data to suggest that the S's are not displayed exactly the same way in terms of the level of antibody that might bind for a given one. So there's certainly, there's utility, but there's gonna have to be more validation comparing to, let's just say, spike that's transduced versus actual spike that's coming from an infected cell.

0:50:54.8 AM: So we've got five minutes left, I'm told. So, I would also like to know is, has anybody taken antibody and trying to improve it, either for Fc function or trying to improve it by modification of the CDR H3. Chris, have you tried to do that yet? Or Laura, Rachel, anybody else?

0:51:21.9 LW: Our antibody... I should have mentioned this earlier, that one of the problems with broadly neutralizing antibodies that come straight out of humans, at least the ones that we've identified and we've looked very hard for these, is that there does seem to be a trade-off between breadth and potency, some of these antibodies, these so-called Class 4 antibodies that are very broad across the house are base of subgenus, they're definitely not as potent.

0:51:45.7 LW: We've overcome that by engineering our antibodies, so that they have that breadth combined with the potency seen with other clinical antibodies, but I'm not sure if those antibodies actually arise naturally, or at least we haven't been able to find any. So yeah, we've overcome it with engineering. We have not engineered the Fc region beyond a half-life extension mutation though.

0:52:04.3 MD: We found one mouse antibody, that is a sort of Class 4 I guess, which neutralizes in the single digit nanogram per mil of all viral strains that we've tested, all VOCs that we've tested. And it engages with its light chain, not its heavy chain as the dominant mode of engagement structurally, but I agree that what we've noticed like you, is that many of these ones with great breadth don't quite have the same potency, but there are some exceptions, which leads me to believe that either through engineering or through just brute selection campaigns, the way Erica has... I wouldn't call it brute, because I think it's elegant, but large numbers of these, you may be able to find the needles in the haystack, which actually engage in a certain way, which give you both qualities. So, I think they can exist, it's just that they're rare.

0:52:55.2 LW: Well, and you might be identifying antibodies in mice that wouldn't arise in humans due to the different germline repertoire.

0:53:00.7 MD: Totally, totally possible, yes.

0:53:04.0 RL: We've not been engineering in a sequence-based way, but through the engineering of the bispecific, one thing I didn't mention before is that although there are some sort of cons to limitations in the targeting domains that you can pair successfully, one of the benefits that's seen with some bispecifics, not all, is that you actually get a synergistic effect of the binding, and so the bispecific will have a better potency than the cocktail of monoclonals, and so in some cases, you can sort of... If you're targeting conserved domains with both side of... Both arms of your bispecific, you can get a boost in potency just by physically linking them together and sometimes getting this synergistic effect.

0:53:56.5 AM: Chris, you had something?

We've been exploring libraries of antibodies to improve them

0:53:58.7 CB: I was just gonna add, we've been exploring libraries of antibodies to improve them, but I guess the trade-off also was, for the ones that are usually broad, they aren't as potent, so can we improve those? But those that are potent, can we improve that breadth... That's very hard to do. And a lot of the takeaway was that a lot of these antibodies coming out of humans have high affinities towards our target, towards the RBD. So really, when you look at metrics to improve antibodies, when it comes to display mechanisms and affinity maturation, that's something that the human body is doing better, and so now we're going back and exploring clonal variants found at six months and now one year, that have acquired more cementite mutations, and those are being pulled out by our collaborator Michel Nussenzweig at Rockefeller.

0:54:48.3 CB: And they're showing that these antibodies have better affinities, and are becoming more resistant to some of the variants, which are... Which is really good from natural infections. So, I think that's gonna be something to explore now also in the vaccinated people moving forward, is that looking at time points now or a year down the road, are antibodies that were expended upon vaccination, are they also requiring these mutations to become better over time? Which is going to be important for protection moving forward.

0:55:17.2 AM: Okay, well, I think Cesar will appreciate it if we started to wrap up right now. I apologize really to David and Penny, for not asking all my great questions, I was gonna ask about neutralization, but we didn't have time for that, but thank you very much to the panel. And thank you, Erica, for outlining her paper. And over to Cesar. Thank you very much.