key: cord-031060-0o9agjiq
authors: Yuan, Tom Z; Lujan Hernandez, Ana G; Keane, Erica; Liu, Qiang; Axelrod, Fumiko; Kailasan, Shweta; Noonan-Shueh, Madeleine; Aman, M Javad; Sato, Aaron K; Abdiche, Yasmina N
title: Rapid exploration of the epitope coverage produced by an Ebola survivor to guide the discovery of therapeutic antibody cocktails
date: 2020-08-01
journal: Antib Ther
DOI: 10.1093/abt/tbaa016
sha: 
doc_id: 31060
cord_uid: 0o9agjiq

BACKGROUND: Development of successful neutralizing antibodies is dependent upon broad epitope coverage to increase the likelihood of achieving therapeutic function. Recent advances in synthetic biology have allowed us to conduct an epitope binning study on a large panel of antibodies identified to bind to Ebola virus glycoprotein with only published sequences. METHODS AND RESULTS: A rapid, first-pass epitope binning experiment revealed seven distinct epitope families that overlapped with known structural epitopes from the literature. A focused set of antibodies was selected from representative clones per bin to guide a second-pass binning that revealed previously unassigned epitopes, confirmed epitopes known to be associated with neutralizing antibodies, and demonstrated asymmetric blocking of EBOV GP from allosteric effectors reported from literature. CONCLUSIONS: Critically, this workflow allows us to probe the epitope landscape of EBOV GP without any prior structural knowledge of the antigen or structural benchmark clones. Incorporating epitope binning on hundreds of antibodies during early stage antibody characterization ensures access to a library’s full epitope coverage, aids in the identification of high quality reagents within the library that recapitulate this diversity for use in other studies, and ultimately enables the rational development of therapeutic cocktails that take advantage of multiple mechanisms of action such as cooperative synergistic effects to enhance neutralization function and minimize the risk of mutagenic escape. The use of high-throughput epitope binning during new outbreaks such as the current COVID-19 pandemic is particularly useful in accelerating timelines due to the large amount of information gained in a single experiment.

. Furthermore, since viral antigens have evolved a remarkable propensity to mutate rapidly as a strategy to defy human immune responses, antibody cocktails with broad epitope coverage lower the risk of mutagenic escape which will otherwise render antibodies ineffective, as observed in non-human primates following treatment with a cocktail comprised of antibodies targeting highly similar epitopes on the EBOV GP 11 .

Although characterizing the antigenic surface of viral glycoproteins is advantageous in developing therapies, detailed information on their structure and the roles of particular binding epitopes in protection are often lacking which poses a critical bottleneck in responding to new outbreaks or viral isolates. Furthermore, discrete epitopes of the viral antigen may play distinct mechanistic roles that are unknown, cooperative 3, 7 and have varying levels of risk for mutagenic escape 12 . These factors confound the ability to design effective cocktail therapeutics.

Epitope binning is a useful empirical method for organizing antibodies into epitope families by assessing the blocking relationships that emerge from a pairwise

and combinatorial competition of antibodies against their specific target antigen 13 .

However, when studying large panels of antibodies comprising hundreds of clones, exploring an exhaustive pairwise competition matrix of the entire set by standard technologies such as FACS, ELISA, and label-free biosensors is tedious and resource intensive, so for practical reasons, the scope of these assays is often limited to the blockade against a small set of benchmark 'reagent' antibodies of known specificity or function. Antibodies that are binned in competition with a handful of controls constitute a 'few-on-many' approach and rely upon the existence of such standards. In contrast, high-throughput epitope binning assays that expand the number of pairwise permutations that can be addressed in a single experiment not only provide practical advantages of speed and minimal sample consumption, but also provide exquisite resolution revealing small differences in epitope specificity and nuanced binding modes such as allosteric modulation that may be relevant for functional activity [14] [15] [16] .

Additionally, such assays are self-referencing and do not require the use of controls, so are universally applicable to any antibody library as soon as sufficient protein is expressed and provided that the target antigen is available 17 . Improvements in the throughput of label-free biosensor technologies such as Octet-HTX (ForteBio), IBIS-MX96 (IBIS Technologies), and LSA (Carterra) enables the use of epitope binning as a high-throughput screening process rather than being reserved for small panels of antibodies 13, 16 . Previous studies have employed this method to determine fine epitope differences among antibodies to human progranulin 16 Figure 1 ). Synthetic biology advances enabled the rapid, high-throughput DNA synthesis, expression, and purification of 321 antibodies with only the variable domains' amino acid sequences as an input. The EBOV GP target was chosen as it is wellstudied, biologically relevant and available in recombinant purified form commercially.

This set of antibodies was chosen due to the availability of in-depth characterization reported in the literature, allowing us to benchmark our binning assignments to published bins assigned by FACS. Following a non-optimized, 'first pass' HT-SPR binning that revealed the epitope landscape including rare epitopes, representative antibodies were selected from each bin. These 'pathfinder' antibodies were further investigated in an independently conducted 'second pass' binning along with benchmark anti-EBOV GP antibodies supplemented with orthogonal data such as antibody sequence, in vitro cell-based neutralization of live virus and in vivo survival to lethal challenge in mouse models to provide a comprehensive analysis that can aid in the rational design of therapeutic cocktails.

Variable heavy and light domains of anti-EBOV GP antibodies were sourced from Bornholdt et al. 2 

Thirteen antibodies with defined structural epitopes from the literature were 

cells as described previously 6 . The EBOV GP sample used in all epitope binning experiments appeared to be homogeneous in analytical size exclusion chromatography while migrating at an apparent molecular mass of ~386 kDa, indicating substantial glycosylation (Supplemental Figure S1 ).

Epitope binning was performed in a premix format using a Carterra LSA SPR A premix assay format (Figure 2A ) was chosen for the epitope binning study because EBOV GP is a multivalent antigen comprising of a trimer of GP1+GP2

heterodimers 22 . Figure 2B shows an overlay plot of the sensorgrams obtained for a ligand that gave clear binding responses in the premix binning assay, due to its facile regeneration, giving reproducible antigen binding responses, and full self-block. In this rapid, non-optimized set-up, no attempt was made to optimize the concentration of the premixed antibodies and they were used in a 'batch dilution' mode as a 5-fold dilution of the supplied stock (corresponding to final antibody binding site concentrations of 53 -1893 nM, with a mean of 507 nM). Clearly, this would not have achieved a molar excess of premixed antibody to antigen for some of the low-expressing antibodies, so we limited the premixed analytes to only those clones that showed clearly detectable ligand binding and above-mean expression (>20 µg from 1.2 ml culture). Since each analyte injection consumed 300 μl, we needed around 2 μg per EBOV GP per injection and since we had a limited supply of our purified antigen (0.5 -1 mg), we would have nearly

exhausted it if we injected EBOV GP premixed individually with all 321 antibodies (>300 injections, including antigen alone injections).

shows a heat map of the results from all active ligands (rows) and fully saturated premixes (columns) in the binning assay, revealing seven distinct epitope clusters or 'communities' without the inclusion of the benchmark antibodies. Due to the low expression of some clones and the requirement for their saturation of the antigen as premixed analyte, as judged by premixes giving a full block somewhere (self or elsewhere) in the competition matrix, the number of antibodies that were successful as premixes and gave clearly interpretable responses was rather low. Despite ligand attrition and the stringent requirement for premixed antibodies to achieve antigen saturation, which reduced our heat map to 52 analytes x 233 ligands, we were able in this non-optimized 'first pass' binning to assign 234 antibodies to one of seven epitope communities without the use of benchmark antibodies (Supplemental Table S1 ). HT-SPR binning heat maps including limited embedded standards, or only bidirectional ligand/analyte pairs were also generated (Supplemental Table S2 , S3).

Antibodies representative of the epitope coverage of the full library were used as 'pathfinder' reagents to assign missing clones in a second pass, focused binning study

The results from our first pass binning led us to select a set of 15 'pathfinder'

antibodies that were representative of each epitope community and re-express them on a larger scale for use as high quality reagents that recapitulated the entire epitope diversity of the library in a few key clones. The pathfinders represented antibodies with good expression and performance in the binning assay as both analyte and ligand to facilitate their use in future binning experiments. We also scaled up the production of seven randomly selected clones from those that we had failed to assign to a community in our first pass binning, due to their poor performance as ligand, to see if we could assign these "unknowns" in a more focused 'second pass' binning experiment. Also included were a set of 13 literature controls of known specificity from a collaborator that

served as structural benchmarks to assess whether our SPR-derived epitope communities (represented by the aggregate coverage of our 'pathfinders') overlapped or extended beyond known epitopes.

The pathfinders, unknowns, and structural benchmarks were merged into a single binning experiment that was performed in a completely independent manner than the first-pass binning. In this so-called "second-pass" binning, conducted by a different operator working on a different LSA unit in a separate lab with a fresh batch of antigen and scaled-up antibodies, we successfully recapitulated the expected epitope coverage of the pathfinders (Supplemental Table S4 ) and used them to assign the 12 structural benchmarks to one of our 7 identified communities as well as assign the unknowns to a community. Even with only the pathfinder antibodies, all 7 epitopes were recapitulated (Supplemental Table S5 ). Consistent with our first pass results, most of the previously unassigned clones failed to perform as ligand, making us reliant on their performance in the role of premixed analyte to determine their epitope specificities. This was now made possible due to the larger yield available for those clones upon scale-up. The results from the second-pass binning are summarized as a heat map ( Figure 3A ) and as list of assignments ( Figure 3B ).

Next, we compared our SPR binning results obtained from first and second pass experiments to literature assignments made by FACS 2 . Up to this point, antibody production had been blinded using internal clone names rather than ADI names, to test the reliability of our methods. Figure 4A shows a dendrogram of the antibody sequences for the 241 clones assigned to an epitope community by SPR. Clones are colored by their SPR-derived epitope community (inner circle) and compared to their FACS-derived epitope bin (outer circle). Antibody sequence does not appear to be fully predictive of epitope as unrelated sequence lineages converge upon the same epitope community (Figure 4, Supplemental Figure S6 ), while highly related antibodies can

bind to distinct, non-overlapping epitopes, as shown by ADI-15731 (Comm3) and ADI-16052 (Comm6), both germline VH3-49 antibodies (Supplemental Table S1 ).

Overall, we found excellent agreement between the SPR and FACS epitope binning determinations from independent studies, with only a few (4 out of 241) discordant assignments. We were able to align our communities with those determined Table   S1 ). The high-throughput nature of the HT-SPR binning assay facilitates the identification of rare epitopes as antigen blockade is tested across so many antibodies that those showing unique blocking profiles are readily apparent. Of the 241 antibodies that we assigned to epitope communities, two or fewer were assigned to communities 6- To obtain a clean result (block or not block) in a premix assay format, two caveats must be satisfied. First, the ligands must maintain their antigen-binding activity upon multiple rounds of regeneration. Second, the premixed antibody analyte must bind up and saturate (or nearly saturate) its recognized epitopes in the antigen sample to effectively diminish the antigen's free concentration to a barely detectable level. A useful test to verify empirically that the premixed antibodies have achieved saturation of the antigen sample is a 'self-block' where the same antibody is used in the role of both analyte (premixed) and ligand (coupled to chip). Thus, a premixed antibody (analyte) that reduces antigen binding to a barely detectable level when probed by its coupled counterpart (as ligand) verifies that the premixed antibody is capable of saturating all epitopes in the antigen sample and therefore can be used reliably to assess the epitope-based competition of other ligands. Since we had no 'a priori' knowledge of the true binding affinities of any of the antibodies to our EBOV GP antigen (published kinetic data are avidity-influenced measurements on BLI 2 ) we elected to use a large molar excess of each premixed antibody relative to the antigen concentration used, while balancing the need to conserve sample (both antigen and small-scale purified antibodies).

While we intended to explore the entire 321 x 321 'analyte x ligand' competitive matrix, we observed about 25% ligand attrition due to ligands that showed poor antigen binding responses due to their low activity/affinity or being damaged (or conversely, unaffected) upon regeneration, so were excluded from our analysis. To conserve precious antigen, we elected to use as premixed analyte only those antibodies that had shown good ligand binding and good expression. Even if the antigen had not been precious, the yields of some antibodies in the small-scale expression would not have been sufficient to produce full saturation of the antigen, which reduced the number of premixed analytes that fulfilled this caveat. Conversely, epitope binning with a well behaved, monomeric antigen can be performed as a classical sandwich assay format

which requires significantly less analyte because it does not depend on fully saturating the antigen. Unfortunately, for EBOV GP such a monomeric construct with biological relevance is not available.

Benchmark antibodies can add tremendous value to binning assays in providing 'mapping' information, since cross-blocking of such standards would infer overlapping epitopes. However, the binning assay itself is not reliant on them, but enhanced by them. Four interesting observations that emerged from our high throughput binning analysis that would have likely been overlooked in a simple 'few-on-many' approach are, (1) cross-talking antibodies, (2) asymmetric blockade, (3) apparent antigen heterogeneity, and (4) rare epitopes ( Figure 5 ).

While most of the antibodies fell neatly into one of seven discrete communities, a few antibodies were able to 'crosstalk' and block members of more than one

community. An example of this behavior is ADI-15878, a structural benchmark clone, which blocked antibodies in Comm1 and Comm2 ( Figure 5A ). This contrasts mAb114, a Comm3 member that only blocks other Comm3 members and shows significant sandwiching to antibodies from other communities. Most antibodies (like mAb114) in our study only blocked antibodies in their own community. In addition to ADI-15878, its genetic sibling, ADI-15742 also showed crosstalk of communities 1 and 2 (Supplemental Figure S1) . 19, 24 . In a 'few-on-many' binning paradigm, ADI-15878 (or ADI-15742)

blocking is only known in context relative to the tested benchmarks. From our binning analysis, these two clones were clearly differentiated from the rest without examining

their sequences -see Supplemental Table S2 showing first pass binning merged with a limited set of embedded controls.

When antibody competition is tested in both orders of addition, in rare cases, asymmetric or order-dependent competition is observed. In our study FVM-09 showed markedly asymmetric blockade of Comm2 members, only blocking them when presented first to the antigen (as premixed analyte), but not blocking any other antibody (except for itself) when presented second (as ligand) ( Figure 5B) . Indeed, FVM-09 has been reported to act cooperatively in cocktails due to it possibly triggering an 'induced' epitope, an epitope that is either formed or exposed upon binding of GP by another antibody 7 . Therefore, high throughput binning can reveal possible cases of allosteric modulation that may enable nuanced mechanisms of action (Supplemental Table S2 ).

Visual inspection of the SPR sensorgrams suggests the presence of antigen heterogeneity. Comm4 and Comm5 antibodies appeared to bind a subpopulation of the antigen that was distinct from that recognized by other communities. This manifests in the sensorgrams as premixes of Comm4 and Comm5 antibodies showing 'no effect' on the binding of some ligands, while completely blocking others 25 . An example is shown in Figure 5C where ADI-15779 (Comm4) and KZ52 (Comm2) fully self-block but appear to have no effect on EboV binding to the opposing antibody when premixed in solution.

The antigen appeared conformationally intact as judged by its high degree of homogeneity when tested by standard analytical sizing methods (Supplemental Figure   S1 ), so the existence of functional subpopulations that are differentially recognized by antibodies in the panel would not have been obvious otherwise. Despite possible antigen heterogeneity, each antibody still exhibits reliable binding to EBOV GP, full selfblock and thus did not affect our ability to assign antibodies to epitope communities.

High-throughput epitope binning of large panels of antibodies is uniquely positioned to identify rare epitopes. As the blockade matrix size increases, these rare communities are repeatedly assayed and confirmed to only block within their small communities (Comm5, Comm6) or only to self-block (Comm7) ( Figure 5D) . Mapping

the epitope footprint of structural benchmarks show that epitope communities belonging to large, exposed epitopes generate large numbers of community members (Comm3) while smaller number of antibodies bind to more occluded epitopes such as the GP1 core (Comm5) (Figure 6) . Assembling antibody cocktails inclusive of these rare epitopes allows further probing of antibody combinations to uncover synergistic effects and avoid biasing therapies to an immunodominant epitope that may be cleaved as a decoy or be tolerant of high rates of mutation.

In our binning study, in addition to the trimeric EBOV GP we also tested a cleaved form of GP (GPcl), the product of cathepsin cleavage in the endosome and loss of glycan cap. We observed that antibodies in communities 1-3 bound to GPcl, while communities 4-8 did not, suggesting they targeted the sequence that was cleaved away from the viral membrane (Supplemental Table S8 ). One notable exception was FVM-09 (Comm2) which did not bind GPcl, consistent with its known epitope specificity 26 .

Antibody cocktails that comprise antibodies targeting disparate non-overlapping epitopes enhance neutralization potency by combining multiple mechanisms of action.

Cocktails that combine several antibodies targeting disparate epitopes can unlock unexpected synergistic effects that enhance neutralization activity beyond the additive contribution of each antibody, as has been demonstrated in treatments for Botulinum neurotoxins 10 , Sudan virus 27 , Ebola virus 28 and HIV-1 29 . Furthermore, antibodies that individually do not display neutralizing activity as a monotherapy may exert a synergistic neutralizing effect when combined in a cocktail 7 . Understanding the epitope landscape of antibodies generated by the human immune response to authentic viral infection is necessary to rationally assemble therapeutic antibody cocktails that condense the entire immune response to a handful of key clones that confer protection.

However, the human immune system responds differently to different pathogens.

Staphylococcus aureus, a commensal pathogen, are strongly germline-biased 30 

recent study of the human immune response to vaccination by Yellow Fever Virus 17D also reported a germline-encoded neutralization 31 . Conversely, by merging our empirically determined epitope communities with functional data from the literature, we

found that anti-EBOV GP antibodies from multiple epitope communities were able to neutralize in a cell-based assay ( Figure 4C) and confer protection in a mouse challenge model (Supplemental Table S7 ). These findings reinforce the need to evaluate the epitope landscape of the human immune repertoire to each pathogen empirically.

While antibody affinity and developability can be engineered, the binding epitope of an antibody is an innate property that cannot be engineered downstream. 

forms of their glycoproteins that subvert the immune response towards production of non-neutralizing antibodies. This escape mechanism is present in Ebola virus in the form of secreted glycoprotein isoform (sGP) 33 . Cocktails that carefully combine antibodies targeting disparate epitopes by utilizing knowledge of the epitope landscape early in the drug discovery process, as enabled by high throughput binning, may be better equipped to overcome mutagenic escape and decoy epitopes. 

Molecular mechanisms of Ebola virus pathogenesis: Focus on cell death

Isolation of potent neutralizing antibodies from a survivor of the 2014 Ebola virus outbreak. Science (80-. )

Analysis of a Therapeutic Antibody Cocktail Reveals Determinants for Cooperative and Broad Ebolavirus Article Analysis of a Therapeutic Antibody Cocktail Reveals Determinants for Cooperative and Broad Ebolavirus Neutralization

Therapeutic Monoclonal Antibodies for Ebola Virus Infection Derived from Vaccinated Humans

Development of Clinical-Stage Human Monoclonal Antibodies That Treat Advanced Ebola Virus Disease in Nonhuman Primates

Post-exposure immunotherapy for two ebolaviruses and Marburg virus in nonhuman primates

Cooperativity Enables Non-neutralizing Antibodies to Neutralize Ebolavirus

Antibody Treatment of Ebola and Sudan Virus Infection via a Uniquely Exposed Epitope within the Glycoprotein Receptor-Binding Site

Immunization-Elicited Broadly Protective Antibody Reveals Ebolavirus Fusion Loop as a Site of Vulnerability

Monoclonal antibody combinations that present synergistic neutralizing activity: A platform for next-Generation anti-toxin drugs

Emergence of Ebola Virus Escape Variants in Infected Nonhuman Primates Treated with the MB-003 Antibody Cocktail

High-throughput epitope binning of therapeutic monoclonal antibodies: Why you need to bin the fridge

Screening for GPCR ligands using surface plasmon resonance

The emerging role of biosensor-based epitope binning and mapping in antibody-based drug discovery

High-throughput epitope binning assays on label-free arraybased biosensors can yield exquisite epitope discrimination that facilitates the selection of monoclonal antibodies with functional activity

Exploring blocking assays using Octet, ProteOn, and Biacore biosensors

Assessing the binding properties of the anti-PD-1 antibody landscape using label-free biosensors

Antibodies from a Human Survivor Define Sites of Vulnerability for Broad Protection against Ebolaviruses

Structures of Ebola virus GP and sGP in complex with therapeutic antibodies

Pan-ebolavirus and Pan-filovirus Mouse Monoclonal Antibodies: Protection against Ebola and Sudan Viruses

Structure of the Ebola virus glycoprotein bound to a human survivor antibody

The Importance of Epitope Binning for Biological Drug Discovery

Structural basis of Pan-Ebolavirus neutralization by a human antibody against a conserved, yet cryptic epitope

Label-free epitope binning assays of monoclonal antibodies enable the identification of antigen heterogeneity

Macaque Monoclonal Antibodies Targeting Novel Conserved Epitopes within Filovirus Glycoprotein

Development of an antibody cocktail for treatment of Sudan virus infection

Systematic Analysis of Monoclonal Antibodies against Ebola Virus GP Defines Features that Contribute to Protection

Super Potent Bispecific Llama VHH Antibodies Neutralize HIV via a Combination of gp41 and gp120 Epitopes

Germline-encoded neutralization of a Staphylococcus aureus virulence factor by the human antibody repertoire

Longitudinal dynamics of the human B cell response to the yellow fever 17D vaccine

Expanding the watch list for potential Ebola virus antibody escape mutations

Antigenic Subversion: A Novel Mechanism of Host Immune Evasion by Ebola Virus

Sym015: A highly efficacious antibody mixture against metamplified tumors

The EMBL-EBI search and sequence analysis tools APIs in 2019

InterPro in 2019: Improving coverage, classification and access to protein sequence annotations

Structure of the Ebola virus glycoprotein bound to an antibody from a human survivor