Bispecific Antibodies Potently Neutralize SARS-CoV-2 Variants of Concern – Presentation by NIH Scientists

0:00:03.3 Speaker 1: Good morning. I'd like to thank the organizers for the opportunity to speak at this conference. Today I'm going to talk about some recent work on bispecific antibodies that target SARS-CoV-2 variants of concern. The COVID-19 pandemic is the worst pandemic that we have faced in the last 100 years, and cases were first observed around the end of 2019. Since then, there have been more than 200 million cases, and about 4.7 million deaths due to this disease. The research community in partnership with industry has launched an unprecedented effort to target this disease. And one of the results of this work is that we now have several authorized vaccines in the US from Pfizer, Moderna and J&J. And there are other vaccines that are also available in other countries such as the AstraZeneca vaccine, Sinovac, Sinopharm and so on. In terms of monoclonal antibodies, there are several that have been authorized for use in the US.

0:01:07.2 Speaker 1: And these antibodies are primarily aimed at mild to moderate COVID-19 patients who are at risk of progressing to more severe disease. These antibodies have been shown to be effective in reducing the risk of hospitalization and death. And overall, the vaccines and the antibodies have saved a lot of lives. They have reduced infection, they have prevented severe disease. However, in the last year, we have witnessed the continuous emergence of a series of new SARS-CoV-2 variants that have mutations in different parts of the spike protein. These variants are a potential threat to the efficacy of these vaccines and antibodies. And therefore, at least in the antibody space, it is important to continue developing a broad panel of antibodies that target different regions of the spike protein. And this was the aim of this study, to isolate potent antibodies that target different parts of the spike, and specifically we focused on the Receptor-Binding Domain or RBD and the N-terminal domain or the NTD.

SARS-CoV-2

0:02:22.2 S1: Our second aim was to combine these different antibody specificities in order to develop tools that could potently neutralize SARS-CoV-2 variants of concern. I'm going to start this talk with a pipeline and overall pipeline to show how we isolated the antibodies. And then I will describe each step in more detail. So first, we select the donors by plasma neutralization of authentic SARS-CoV-2. And once we identify donors of interest, we screened for antibodies from plasmablasts and memory B cells. The plasmablasts was screened directly for production of antibodies that could bind to the spike protein, using a device called the Beacon, that I'll come back to later. The memory B cells were first activated, and then it was screened in the same manner. Once we had identified B cell clones of interest, we did PCR to obtain the heavy and light changing sequences. And then we produced the antibodies as recombinant proteins and this was done by GenScript.

0:03:25.0 S1: And then once we had a panel of antibodies, they were screened for potency, affinity as well as for the part of the spike protein that these antibodies bound to. So the first step, donor selection, we... In order to increase our chances of finding good antibodies, we wanted to screen as many donors as possible at the start. And when we started this study in April 2020, the cohort that had the largest collection of samples that was available to us came from the New York Blood Center, and we screened samples from 126 donors who were confirmed to be COVID-19 positive by PCR. So first, we screened plasma from these donors for binding to Coronavirus spike proteins as well as for neutralization to authentic SARS-CoV-2, neutralization of authentic SARS-CoV-2. Right. So, this is the result of that screen. We found, as expected, that a lot of the plasma had reactivity to COVID-2 spike protein. We also found that many samples had reactivity to spike protein from other coronaviruses, indicating prior exposure to some of these seasonal coronaviruses.

0:04:45.5 S1: If you look at neutralization, which is the column on the right, there was a wide range of neutralization between the different donors, and so we chose the top neutralizing donors for monoclonal antibody isolation. Early on, we made a decision to isolate monoclonal antibodies from both memory B cells and plasmablasts with a special focus on plasmablasts. And the reason we decided to do so is that many of the studies that have been done in this field have focused on memory B cells. And so we were interested in finding out the type of antibodies that plasmablasts make, and whether they can also make potent antibodies against SARS-CoV-2. We were able to screen the plasmablasts directly for production of antibodies of interest using a device called the Beacon. And the heart of this device is a chip that's roughly the size of a teabag that contains 11,000 nanopens. And so during a run, the Beacon identifies B cells, it then draws cages of light around the B cells, and then it pushes them into these nanopens.

Beacon

0:06:00.5 S1: So this video here just shows a single cage around a single B cell. But what happens in the actual run is that the Beacon will identify many different B cells. It will draw cages around them, and then it'll push them all into their respective pens. And once they're all in their pens, then the Beacon will count a number of B cells in each pen, so that we know whether it's zero or one or more than one B cell per pen. Once the B cells are in their pens, we can rapidly screen them for production of antibodies of interest, due to the small volume of each pen just 0.25 nanoliters. And so what we did is we flooded in Bs coated with spike protein along with secondary antibodies that were fluorescently labeled. And so if a B cell producing antibody of interest, it would bind to the Bs, the secondary will bind, and this would result in an increase in the fluorescence signal over time.

0:07:02.8 S1: And so on the right, this is an actual run where we had Bs coated with spike protein on the channels above the pans. And we can see that over time, this was over half an hour, the Bs above specific pans had an increase in signal. Once this was done, we were able to identify the pans of interest and then export the B cells one by one into a 96-well plate where we could do PCR to obtain the antibody sequences. So using this approach, we obtained a total of 216 recombinant antibodies that combined to spike protein, 169 came from plasmablast and 47 from memory B cells. One of the first things we wanted to do was to compare the potency of antibodies from plasmablast and memory B cells. And so we screened all 216 for neutralization of authentic SARS-CoV-2. We found that the average potency of antibodies from both cell types was similar. And when we looked at the top five antibodies, regardless of cell type, four of them came from plasmablast, and one from a memory B cell. And therefore we concluded that plasmablasts are capable of producing potent antibodies against SARS-CoV-2, and that we should not ignore the cell population when trying to screen for potent antibodies against this virus.

What is Potent

0:08:31.5 S1: So now at this point, we knew that we had some potent antibodies. And the next question we asked was, how potent? And to answer this question, we compared our top antibodies to some of the best antibodies. From the literature, we've published sequences as shown here. So here, the antibodies in red are the ones from the study, and the ones in the other colors are from different studies. We compared them in three different neutralization assays, two with authentic SARS-CoV-2 and one with a pseudovirus. And we found overall that there was a lot of variability between the different assays. The absolute IC50 values that we obtained was quite different, but the trends between antibodies were quite similar, and our top antibodies had comparable potency overall to some of these top benchmark antibodies. And of course, the other lesson here is that if we are interested in comparing antibodies from different groups, it would be good to test them all in the same assay due to this assay-to-assay variability.

Spike Protein that our Panel of Antibodies Bound

0:09:44.2 S1: Next, we wanted to identify the part of the spike protein that our panel of antibodies bound to. At this stage, if you just go back, we had already classified them as RBD or NTD binders, but we wanted to go a bit further to find out the part of the RBD or the NTD that these antibodies targeted. And to do so, we used a device called the LSA, which can perform high-throughput SPR.

0:10:13.2 S1: For example, if we focus on the NTD antibodies, we performed a competition assay of all the antibodies versus each other in this checkerboard format. And here, red means that there is competition and green means there is not. And so, based on the pattern of competition, we were able to subdivide these antibodies into different bins, which correspond to different parts of the NTD. And if we spike in antibodies with known binding sites on the NTD, we are then able to map these bins to specific parts of the NTD. So for the NTD specific antibodies, they could be divided into four bins. And when we looked at which bin contained the most neutralizing antibodies, we found that most of the neutralizing antibodies fell in this orange bin, which corresponds to this site here, which is located at tip of the NTD.

0:11:15.2 S1: And this is consistent with work done by other groups, identifying this site as an antigenic super site for this domain. When we did the same analysis with the RBD specific antibodies, the picture was slightly different. The antibodies were also divided into four bins, but the neutralizing antibodies were quite spread out between the different bins. And interestingly, our top three neutralizing antibodies, 503, 664 and 993, they all bound to different bins. For example, 503 targeted this region at the top of the RBD, which is also where ACE2 binds.

0:11:56.2 S1: And this binding site was confirmed when our colleagues at the Scripps solved the crystal structure of this antibody bound to the RBD. In contrast, 664 and 993 bind to different sites of the RBD, and we confirmed that all three antibodies do not compete with each other for binding to this domain. So now that we knew that we had antibodies that could bind to different sites of the RBD, and some of them bound to the NTD, we were able to test for synergy between these antibodies, by mixing them in different pair-wise combinations.

0:12:32.2 S1: Unfortunately, we found that there was no clear synergy between any of the pairs that we tested, which is consistent with some findings by other groups as well. So instead of just combining the antibodies, we thought that we could mix them in this bispecific antibody format where the bindings at one antibody is linked to the bindings of the second antibody by this linker, and this format is called DVD-Ig. And we chose it because of this quadrivalent format and also due to ease of design expression. Right now, hypothesis here is that this covalent linkage between the two antibody binding sites could result in a function or a phenotype that is not observable by just mixing the antibodies. We successfully expressed 10 by specific antibodies. And if we look at the names on the right, 1-10, the name... The start on the left is the first antibody, so the one forming the outer binding site and then the second name is the one that forms the inner binding site.

0:13:44.9 S1: And we ran an SDS-PAGE gel in comparison with number 11, which is the regular IgG, and we found that nine out of 10 had the expected molecular weights. This was confirmed by a size exclusion chromatography where we found that, again, nine out of 10 had this single species, which was slightly larger than the standard antibody. Next, we tested the bispecific antibodies using SPR and ELISA and this was done to test if the outer and the inner binding sites of the bispecifics could still bind to the expected targets. And I'm not going to go through details of experiments here, but basically we found that the outer and inner binding sites of all the bispecific antibodies could still bind to their expected targets.

0:14:43.0 S1: Most importantly, we were interested in the function of these bispecific antibodies, and so we tested them for neutralization of authentic SARS-CoV-2 as well as the pseudovirus. And we found that a few of the antibodies had very high potency, less than a nanogram per mil IC50 and the most potent bispecific antibody in our panel was 1206 521 GS as shown here, and this is a bispecific antibody that combines an RBD binder, 1206 and an NTD binder, 521.

0:15:16.9 S1: This was the most potent bispecific in our panel, and when we compared it to the monoclonals, it was also the most potent antibody overall. And this was quite surprising to us as CV1206, and if you look at IC50 value here, was not one of our best RBD binders, it was a bit worse. It was actually quite a bit worse than some of the others that we tested. And this can be seen clearly in this figure, where in blue, the bispecific antibody is much, much better at neutralization than either of its parent or component antibodies. At first, we thought that this was just a synergistic interaction that we missed, but when we tested the bispecific in comparison to the cocktail, a cocktail of the two antibodies, we found that the bispecific once again, performed a lot better.

0:16:14.3 S1: And so, this was an interesting result for us, and we wanted to find out why this happened and why this was the case, and first, we thought maybe the bispecific just binds better to spike protein compared to highest component antibodies, but we found that this was not the case. While the bispecific binds similarly to its parents, and yeah, in fact it binds a bit worse to the NTD compared to its parent, NTD binder CV521. So this was a puzzling result, and so we were quite unsure about why this antibody was better, but then our colleagues at the Scripps obtained negative stain EM images of the bispecific bound to spike protein.

0:17:01.9 S1: And in this figure, the spike protein's in green and this in orange is one arm of the bispecific and this in purple is another arm, either of the same bispecific or a different one. And what we observed is that the bispecific antibody was able to crosslink adjacent spike proteins through the covalent linkage between the outer and inner binding sites. So this outer binding site 1206 binds to RBD of one spike, the inner binding site, 521 binds to the NTD of a different spike, and because these two binding sites are covalently linked, they can pull the two spike proteins together and this is also true for this pair of binding sites here in purple.

Reason Why Bispecific antibody is better than Cocktail

0:17:50.8 S1: Alright, so this is the reason that we think... We think this is the reason why this bispecific antibody is better than even a cocktail of the two parents which cannot have this phenotype. We tested this bispecific antibody in a hamster model of SARS-CoV-2 infection, and in this experiment, the bispecific antibody was administered one day before charged for authentic SARS-CoV-2, and we found that the hamsters that were given two different doses of the bispecific, they all showed no loss in body weight over a week, and they also had negligible clinical score similar to the hamsters, which were not infected at all. So these results suggested that this antibody is effective in preventing clinical symptoms in this model. And since then we have tested two other bispecific antibodies in the same model, and we have seen similar results.

0:18:51.1 S1: So this is one potential advantage of bispecific antibody. Through its covalent linkage between two different antibody-binding sites, it can have some phenotypes that may not be possible in a cocktail or just with normal, regular IgGs. The second potential advantage is that the bispecific antibodies may be more resistant to mutations in SARS-CoV-2 variants of concern due to the ability to target two different sites of the spike protein. And at the time that we were conducting this study, the alpha and the beta variants of concern were the most prevalent or the most concerning, and so we tested the ability, first, of our regular IgGs to bind to spike proteins carrying mutations in these two variants.

Mutation in Alpha & Beta

0:19:43.9 S1: Some of them are individual mutations and then some of them... Some of the spikes had all the mutations in alpha or beta. And in this heat map, red means binding and white means no binding. And so we found that the beta variant was more of a problem, compared to alpha. This is consistent with what others have found, where all the NTD antibodies did not bind to beta, and two out of five of the RBD antibodies also could not bind to beta presumably due to this E44K mutation. When we tested binding of the bispecific antibodies to the same panel of mutants, we found that only the bispecific antibodies that had both partners, both components that lost binding to the beta spike were unable to bind to this spike as well. Unfortunately, this included 1206521, which is one of the antibodies that I described earlier.

0:20:45.0 S1: In contrast, the rest of the bispecifics that... Even those that had one partner that had lost binding to the Beta spike, or the Alpha spike, all of them, could still bind to spike proteins from both variants. We went on to test the bispecific antibodies for neutralization of Alpha and Beta and then we added Gamma and Delta as well, and we tested some of the antibodies in two different pseudovirus assays. Here are the numbers in the brackets show the ratio of IC50 of the variants versus wild-type. So numbers larger than one indicate reduced potency against the variant, and numbers smaller than one indicate a greater potency against the variant compared to wild-type. There was a wide range of neutralization potencies across the different variants, but overall, we found that two bispecific antibodies are retained close to wild-type potency against the Alpha, Beta, Gamma and Delta variants.

0:21:54.9 S1: So, to conclude, in this study, we started by screening samples from COVID-19 convalescent patients, and we isolated monoclonal antibodies that bound to different parts of the spike protein. We combined monoclonal antibodies with nonoverlapping specificities to form a DVD-Ig bispecific antibodies. And we found that in one antibody this combination resulted in a function of being able to crosslink adjacent spikes, and this bispecific was potent in preventing clinical symptoms in the Hamster model of SARS-CoV-2 infection. Two of our bispecific antibodies were also effective in neutralizing the Alpha, Beta, Gamma and Delta variants of concern, with neutralization potencies close to that against the wild-type virus. So, overall, these findings support the further exploration of bispecific antibodies as an antibody... A new or a different antibody modality against COVID-19.

Thanks to the Collaborators

0:23:07.2 S1: So with that, I'd like to thank the various collaborators who participated in this project. This was work done by many different groups. I'd like to thank especially Hyeseon and Mary from Peter Crompton's group. And Christina from my group, who did a lot of the antibody cloning and characterization. And I'd like to thank also Wilson's group and Scripps for the crystal structure. Andrew Ward's lab that did the EM work, David Nemazee and Thomas Rogers, as well as our colleagues at the Vaccine Research Center and the Integrated Research Facility who did the in vitro neutralization and Hamster experiments. I'd like to thank also the sample donors, without whom none of this would be possible and NIAID for funding, and I thank you for your attention. I'll be happy to take any question.