Anti-Ebola virus mAb 3A6 with unprecedented potency protects highly viremic animals from fatal outcome and physically lifts its glycoprotein target from the virion membrane

Monoclonal antibodies (mAbs) against Ebola virus (EBOV) glycoprotein (GP1,2) are the standard of care for Ebola virus disease (EVD). Anti-GP1,2 mAbs targeting the stalk and membrane proximal external region (MPER) potently neutralize EBOV in vitro. However, their neutralization mechanism is poorly understood because they target a GP1,2 epitope that has evaded structural characterization. Moreover, their in vivo efficacy has only been evaluated in the mouse model of EVD. Using x-ray crystallography and cryo-electron tomography of 3A6 complexed with its stalk– GP1,2 MPER epitope we reveal a novel mechanism in which 3A6 elevates the stalk or stabilizes a conformation of GP1,2 that is lifted from the virion membrane. In domestic guinea pig and rhesus monkey EVD models, 3A6 provides therapeutic benefit at high viremia levels, advanced disease stages, and at the lowest dose yet demonstrated for any anti-EBOV mAb-based monotherapy. These findings can guide design of next-generation, highly potent anti-EBOV mAbs.


INTRODUCTION
Ebola virus (EBOV; family Filoviridae: species Orthoebolavirus zairense) causes severe and frequently fatal acute human disease in outbreaks that can cause thousands of deaths.Complicating containment efforts, EBOV may persist subclinically in survivors for years and reignite outbreaks.Ebola virus disease (EVD) can be prevented with two licensed vaccines and treated with approved monoclonal antibody (mAb)-based therapeutics 1 .However, even with approved mAbs, acute outcomes remain poor in EVD patients with high viral loads and/or advanced disease, and the impact of on viral persistence is unknown.As such, identi cation and optimization of novel mAbs is needed to address these gaps.
All advanced anti-EBOV mAb therapeutics and vaccines target protein spikes protruding from virion envelopes 2 .Each spike comprises a single EBOV-encoded glycoprotein (GP 1,2 ), synthesized by translation of a preproprotein that is cleaved in the Golgi apparatus into GP 1 and GP 2 subunits.A disul de bond links these two subunits to form heterodimers (Fig. 1A top) that assemble into GP 1,2 trimers 3,4 .GP 1 contains a heavily glycosylated mucin-like domain (MLD) that obscures the upper and outer portions of GP 1,2 and a glycan cap domain that shields the virion receptor-binding site in the GP 1 core from the host immune response (Fig. 1A top).Upon virion entry, host-cell cathepsins proteolytically process GP 1,2 in the endosome to remove the MLD and glycan cap domain to expose the GP 1 core and binding site for the virion receptor, NPC intracellular cholesterol transporter 1 (NPC1) [5][6][7] .GP 2 , a typical class I fusogen, mediates fusion of virion and endosomal membranes to release viral ribonucleocomplexes into the target cell 8 .GP 2 contains an internal fusion loop (IFL), two consecutive heptad repeat regions (HR1 and HR2), a membrane proximal external region (MPER), and a C-terminal transmembrane (TM) domain (Fig. 1A top).HR2, also called the stalk, is largely alpha-helical and connects the GP 1 core to the MPER and TM domain 7 .
Major recognition sites for anti-EBOV mAbs are the GP 1 MLD and glycan cap domain; the GP 1,2 trimer base; and the GP 2 IFL, stalk, and MPER [9][10][11][12][13][14][15][16][17] .The stalk-MPER is of special interest for therapeutic and vaccine design.The mAbs that bind the stalk-MPER have potent neutralization activity and target a region with high amino acid sequence conservation across orthoebolaviruses.Indeed, the region has 70% sequence conservation among all six orthoebolaviruses and for the three viruses that can cause fatal disease, Bundibugyo virus (BDBV), EBOV, and Sudan virus (SUDV), the sequence conservation increases to ~ 90% 9,11,13 .Despite these important features, mAbs against the stalk-MPER are the least wellcharacterized of anti-orthoebolavirus mAbs in part because of a comparative lack of structural information for this site: to increase solubility and stability, MPER was deleted from GP 1,2 constructs used for all high-resolution GP 1,2 structures determined to date.Only one structure, of a stalk-MPER targeted mAb, the BDBV223, against BDBV, is available 18,19 .Curiously, BDBV223 binds to a GP 2 epitope that in current models of the EBOV GP 1,2 structure is predicted to be occluded by the virion membrane 19 .As such, the mechanism by which mAbs that target the stalk-MPER access their epitope, and whether accessibility is associated with therapeutic e cacy for patients infected with EBOV or related viruses, remains unclear.

RESULTS
Crystal structures of unbound human mAb 3A6 Fab bound to the Ebola virus glycoprotein stalk-MPER mAb 9.6.3A6(henceforth abbreviated as "3A6 IgG") was isolated from a human survivor of the 2013-2016 Western African EVD outbreak 6 months after hospital discharge 11 .The predicted linear epitope of 3A6 IgG encompasses GP 1,2 residues 626-640 and extends from the C-terminal end of the stalk to the start of the MPER 11 (Fig. 1A bottom/inset, orange box, outlined in purple in the sequence alignment).To determine the mode of molecular recognition and neutralization of EBOV by 3A6 IgG, we crystallized the 3A6 Fab fragment alone (Supplementary Fig. 1) and in complex with a 14-amino acid peptide having a sequence corresponding to the EBOV GP 1,2 stalk-MPER epitope (aa 626-640; Fig. 1B-C).Crystals of 3A6 Fab diffracted to 2.5 Å and had an asymmetric unit containing four Fab fragments.Meanwhile, crystals of the 3A6 Fab-stalk-MPER peptide complex diffracted to 1.27 Å and had one Fab fragment in the asymmetric unit (Table S1).The 3A6 Fab structure was essentially identical in the unbound and peptide-bound states as evidenced by the 0.46Å root-mean-square deviation (RMSD) (Fig. 1C, Supplementary Fig. 1).Residues I627-G639 of the 3A6 IgG-peptide epitope (Fig. 1A bottom/inset) are visible (Supplementary Fig. 1).The EBOV stalk-MPER peptide is -helical from its N terminus (I627) to residue T634 and then slightly unravels through the visible terminus at residue G639 (Supplementary Fig. 1).
Ebola virus glycoprotein MPER residues D632 and P636 are critical to mAb 3A6 binding 3A6 IgG binds to and neutralizes EBOV but not SUDV in vitro 11 .In the 3A6 IgG-peptide epitope that includes stalk-MPER residues I627-G639, four residues differ substantially between EBOV and SUDV GP 2 : K633 vs. N, T634 vs. P, D637 vs. N, and G639 vs. D (Fig. 1A bottom/inset).Using a cell-based antibody-binding assay we next compared binding of 3A6 IgG to full-length (MLD-containing) EBOV GP 1,2 with each of these four residues changed individually or in combination with the corresponding SUDV residues (Fig. 2A).We also measured binding with an ELISA using puri ed EBOV GP 1,2ΔTM/ΔMLD containing the same amino acid changes (Fig. 2B).None of the individual mutations affected 3A6 IgG binding to cell-surface GP 1,2 or GP 1,2ΔTM/ΔMLD , but binding was completely inhibited when all four residues were changed to the SUDV counterparts (Fig. 2A, 2B; Supplementary Fig. 2).
Next, we used alanine scanning mutagenesis of GP 1,2 to identify individual residues throughout the epitope that are critical for 3A6 binding.We made alanine point mutations (wild type alanines were mutated to serines) at each amino acid residue between positions 627 and 639 of EBOV GP 1,2ΔMLD (GP 1,2 lacking the mucin-like domain; Fig. 1A) and analyzed each resulting EBOV GP 1,2ΔMLD for 3A6 reactivity by ow cytometry (Supplementary Fig. 2 and Supplementary Table 2).Notably, D632A and P636A mutations produced a ≤ 20% reduction in 3A6 Fab binding relative to wild-type (WT) GP 1,2ΔMLD (Fig. 2C).Both residues are identical in EBOV and SUDV GP 2 (Fig. 1A bottom/inset), and changes at these sites do not substantially affect binding of control mAbs KZ52, 1H3, or 4G7 that target conformational epitopes on the base (KZ52, 4G7) and glycan cap (1H3) of EBOV GP 1 7,20,21 (Fig. 2C).These results are consistent with those of previous studies in which morphologically authentic "biologically-contained" EbolaΔVP30 virions 22 passaged in the presence of 3A6 IgG led to the emergence of glycoproteins bearing P636S and P636Q mutations 11 .Therefore, we next evaluated neutralization of P636S-bearing EbolaΔVP30-eGFP virions by multiple mAbs using a plaque-reduction assay.The P636S mutation abolished neutralization activity of both 3A6 and 1E6, another stalk-binding mAb, but did not affect neutralization of mAbs targeting the glycoprotein core (Supplementary Table 3).

Binding of mAb 3A6 lifts Ebola virus glycoprotein relative to the membrane surface
We superimposed the 3A6 Fab-stalk-MPER structure onto trimeric GP 1,2ΔTM/ΔMLD using the overlapping portion of the epitope as a guide (Protein Data Bank [PDB] identi cation number [ID]: 5JQ7; Fig. 3A).This superimposition revealed steric clashes of the bound 3A6 Fab with the other two GP 2 monomers of stalk-MPER such that three Fabs could not simultaneously bind to the tightly bundled conformation of GP 2 observed in previous crystal structures (Fig. 3B).The EBOV GP in the previously published highresolution crystal structure (resides 32-312 fused to 464-632; compare to Fig. 1A, top) was stabilized by fusion of a britin trimerization motif to the C terminus 23 .A structure of GP 2 of the related Marburg virus that was not fused to any trimerization motif also demonstrated the same close bundling of the three GP 2 monomers in the trimer 24 .Hence, these structures may represent the native bundled conformation and binding of three copies of 3A6 to GP 1,2ΔTM/ΔMLD , were observed (Supplementary Fig. 3).These results indicate that the GP 1,2ΔTM/ΔMLD -3A6 complex indeed contains three copies of 3A6 Fab bound to stalk-MPER.Composition gradient multiangle light scattering (CG-MALS) supported this result, revealing that three 3A6 Fabs consistently bind to one GP 1,2ΔTM/ΔMLD trimer with equal a nities (K D =52.15 (± 1.3) nM; Supplementary Fig. 3).Furthermore, negative stain EM (nsEM) analysis of the GP-3A6 complexes also showed binding of three 3A6 Fabs to the stalk-MPER (Fig. 3C).
The GP 2 subunits in the natural membrane-anchored form have less freedom to open and separate from each other than would the free GP 2 C-termini of the ectodomain.To image 3A6 Fab bound to GP 2 in its natural transmembrane form, we produced lamentous EBOV-like particles consisting of EBOV matrix protein (VP40) and full-length GP 1,2 .These virion-like particles (VLPs) were incubated with 3A6 Fab for cryogenic electron tomography (cryo-ET) and subtomogram averaging analysis (Supplementary Table 4).
Tomogram reconstructions showed extra densities anchored to stalk-MPER that indicated the presence of 3A6 Fabs (Fig. 4A).We also complexed VLPs with the Fab of a well-characterized anti-EBOV control mAb, KZ52 7 , to represent the state of GP 1,2 in the absence of 3A6 Fab (Fig. 3).We used this core-binding mAb instead of unbound GP 1,2 because antibody binding generates a larger structure that can be more accurately aligned during subtomogram averaging.The KZ52 epitope at the base of GP 1 is su ciently separated from the stalk-MPER 3A6 binding site and its binding does not disrupt the native structure of GP 1,2 7 .
Comparison of the structures of VLP-GP 1,2 -3A6 (18 Å) and VLP-GP 1,2 -KZ52 (8.7 Å) revealed that in the presence of the 3A6 Fab the GP 1,2 body was displaced vertically away from the VLP membrane by approximately 3 nm (Fig. 4B).We conclude that 3A6 binding induces this vertical displacement since VLP-GP 1,2 complexed with both 3A6 and KZ52 Fabs together resulted in a similarly lifted GP 1,2 (Fig. 4C) with densities that overlapped with VLP-GP 1,2 -3A6 (Fig. 4D).Steric hindrance of 3A6 Fab with the lipid membrane likely necessitates the vertical lift of GP 1,2 from its natural position upon 3A6 Fab binding.
This lift could inhibit conformational changes that stalk-MPER must undergo for GP 1,2 to mediate Ebola virion entry into target cells.
Together, these data unveil a novel mechanism-of-action for antibodies and identify 3A6 IgG as a rst-inclass antibody that appears to perform physical work.
Low-dose mAb 3A6 monotherapy is e cacious in domesticated guinea pigs and rhesus monkeys exposed to Ebola virus In vitro, 3A6 IgG neutralized EBOV at a concentration of 0.33 nM (50% plaque reduction neutralization test In vivo, prophylactic administration of 3A6 IgG protected laboratory mice from fatal outcome after exposure to a typically lethal dose of mouse-adapted EBOV (100% protection after a 100 µg dose [~ 5 mg/kg] and 50% protection at a 25 µg dose [~ 1.25 mg/kg]) 11 .To increase stringency, in this study we evaluated the e cacy of 3A6 IgG in a post-exposure domesticated guinea pig model.Groups of six (three male and three female) guinea pigs were exposed intraperitoneally (IP) to 1,000 plaque-forming units (PFU) of domesticated guinea-pig-adapted EBOV (Day 0).On Day 3, the guinea pigs were either left untreated or treated IP with a single 5 mg dose of the EBOV antibodies 3A6 IgG, 1A2 IgG (targets the EBOV GP 2 fusion loop), 7G7 IgG (targeting an unknown epitope on EBOV GP 1,2 11 ), or the anti-in uenza A virus (FLUAV) antibody 42-2D2.All animals in the untreated group and those treated with 1A2 or 7G7 succumbed to EBOV infection.All but two of the anti-FLUAV 42-2D2 treated control animals succumbed to EBOV infection.In contrast, all guinea pigs treated with 3A6 survived and exhibited few or no clinical signs of disease (Fig. 5A, Supplementary Fig. 4).
The rhesus monkey model recapitulates key features of EVD and is generally preferred over rodent models for development of EVD medical countermeasures 1 .We randomized four rhesus monkeys into treatment (n = 3, rhesus monkeys 1-3) and no treatment (n = 1) groups.All monkeys were given an intramuscular (IM) injection of a typically lethal 1,000-PFU dose of EBOV (day 0).On Day 4 and 7 after infection, the treatment group monkeys received 25 mg/kg of 3A6 in PBS intravenously, whereas the control monkey received intravenous PBS only.EBOV replication was con rmed in all monkeys on Day 4 by plaque assay titration and quantitative real-time reverse transcription polymerase chain reaction (RT-qPCR), with 10 4 -10 6 EBOV PFU per mL and 10 8 -10 10 EBOV glycoprotein gene equivalents per mL of serum (Supplementary Fig. 4).Notably, these high levels of viremia could still be reversed by 3A6 administration, as evidenced by a decrease in viral load after the rst dose on Day 4 and continued reduction to below the limit of detection by Day 21 (Supplementary Fig. 4).Clinical signs consistent with EVD were observed in monkeys 1 and 3 as early as Day 4 but resolved by Day 13 (Supplementary Fig. 4).Notably, monkey 3 had signi cantly elevated AST activity and rapidly rising serum creatinine levels suggesting a marked reduction in the glomerular ltration rate (the initial rise was similar to the control animal) with pathologic evidence of EBOV-induced liver injury.Nonetheless, this animal still recovered after 3A6 IgG treatment (Supplementary Fig. 5).All three monkeys in the treatment group survived, whereas the control monkey was found dead on Day 8 (Fig. 5B).These results demonstrate that postexposure dosing of 3A6 IgG alone reverses the course of EBOV infection and protects animals of different species from fatal outcome.

Discussion
Vaccines are currently only approved for prevention of EBOV infection 25 , but not for infections caused by other loviruses.Antibody therapeutics that can be used at a low dose to reverse advanced disease are urgently needed to treat people with lovirus infections who live in countries with limited resources.The EBOV GP 1,2 stalk-MPER is of interest for therapeutic/vaccine design due to its relatively high amino acid sequence conservation among all orthoebolaviruses, indicating that a single mAb targeting this region could have therapeutic activity against infections by any of these viruses.Moreover, known antiorthoebolavirus stalk and/or MPER mAbs are highly potent neutralizers in vitro 11,18,19 , suggesting that they may be applied in much lower doses compared to mAbs that are currently used in the clinic.
Here we built on previous in vitro and prophylactic laboratory mouse e cacy studies of the EBOV GP  10,26 .Our data therefore pave the way for development of novel therapeutics that potentially expand the treatment window for effective intervention in highly viremic patients and later in the EVD course.Such therapeutics could increase the likelihood of survival for this group of patients seen relative to currently approved mAb therapeutics.
We previously hypothesized that binding of BDBV223, an anti-stalk antibody targeting a similarly occluded epitope in the EBOV-related Bundibugyo virion 19 , requires either bending or lifting of GP 1,2 .In this study, we experimentally addressed this hypothesis using 3A6, which binds an epitope that is closer to the C terminus (i.e., even more occluded) than that bound by BDBV223.The linear epitope of 3A6 spans residues I627-G639 in the lower region of the EBOV GP 1,2 stalk and our structural studies suggest that a portion of the 3A6 epitope is embedded within the membrane prior to antibody binding.Our structural data further suggest that 3A6 Fab rst binds to the exposed stalk polypeptide above the membrane and then displaces and separates the GP 1,2 monomer stalk bundles.The rst and second 3A6 Fab likely promote gradual stabilization of intermediate conformational states to tilt the GP 1,2 relative to the membrane surface and increase stalk-MPER exposure as it partially elevates above the membrane surface.Due to the steric hindrance between the 3A6 Fab and the membrane, binding of the Fab at the third stalk-MPER site on the GP 1,2 trimer vertically lifts GP 1,2 relative to the membrane surface.We hypothesize that following binding 3A6 IgG achieves potent neutralization activity by blocking conformational changes needed to drive fusion of virion and cell membranes.Human immunode ciency virus 1 glycoprotein and FLUAV hemagglutinin can also be tilted by binding of anti-MPER antibodies 27,28 , indicating that positional exibility is a common property of class I fusogens.
In conclusion, our studies establish 3A6 IgG as the founding member of a new group of immunotherapeutics against Ebola virus that achieves complete protection against advanced disease at the lowest dose yet observed for a monotherapy via a novel mechanism of action.The next desired feature of this new group is breadth: 3A6-like antibodies against stalk-MPER epitopes that have panorthoebolavirus activity likely exist.Such antibodies could be used at even lower concentrations and at more advanced stages across the lovirus disease spectrum.
Drosophila Schneider 2 (S2) cells (Thermo Fisher Scienti c) were cultured with Schneider's Drosophila medium (Thermo Fisher Scienti c) in stationary asks at 27°C.Stable cell lines were adapted to serumfree conditions and maintained on orbital shakers at 27°C.

Antibody and antibody fragment expression, puri cation, crystallization, and visualization
Protein fragment generation, protein and protein fragment puri cation, crystallization, X-ray structure determination, and negative-stain electron microscopy were performed following standard protocols.

Protein expression and puri cation
The Expi293 Expression System (Thermo Fisher Scienti c) was used for expression of immunoglobulins.
Light and heavy chain-encoding plasmids were prepared using an endotoxin free kit (Takara Bio, NucleoBond Xtra Midi Plus EF) and used to transfect Expi293 cells at a 2:1 ratio of light chain to heavy chain using Expifectamine 293 transfection reagent (Thermo Fisher Scienti c) according to the manufacturer's instructions.Monoclonal antibodies (mAbs) containing supernatants from transfected cells were clari ed by centrifugation and then incubated with protein A agarose resin (GenScript, Piscataway, NJ, USA) in batch format overnight, followed by washing, elution, and buffer exchange into Dulbecco's phosphate-buffered saline (DPBS; Thermo Fisher Scienti c) as previously described 29 .
Antibodies used in vivo were veri ed to be endotoxin-free using a commercial kit (Thermo Fisher Scienti c).All antibodies produced in this study were expressed as human IgG1.
Fragment antigen binding (Fab) fragments were generated from puri ed IgG1s through digestion with 3% immobilized papain (Thermo Fisher Scienti c) for 2 h, followed by puri cation with a Mono Q anion exchange chromatography column (GE Healthcare, Chicago, Illinois, USA) and size-exclusion chromatography with a Healthcare Superdex 75 Increase 10/300 GL column (GE Healthcare) in 1X trisbuffered saline (TBS, Thermo Fisher Scienti c).Fractions with pure Fab were concentrated using Ultra Centrifugal Filter Units (Amicon, Miami, Florida, USA).Epitope peptide representing glycoprotein (GP 1,2 ) residues 626-640 was chemically custom-synthesized by Thermo Fisher Scienti c and puri ed via highperformance liquid chromatography (HPLC).
X-ray data collection and protein structure determination X-ray diffraction data of Fab-peptide complexes were collected on beamline 12 − 2 at the Stanford Synchrotron Radiation Lightsource, and Fab diffraction data were collected on beamline 23ID-B at the Advanced Photon Source 30,31 .One dataset for the Fab crystal was used, and two datasets from separate Fab-peptide complex crystals were merged for processing using AutoPROC with XDS 32,33 for indexing and integration, followed by POINTLESS 34 and AIMLESS 35 , programs for data reduction, scaling, merging, and calculation of structure factor amplitudes and intensity statistics.One Fab-peptide complex per asymmetric unit was present in space group P1 21 1 (a = 52.

Size-exclusion chromatography coupled to multi-angle light scattering
Size-exclusion chromatography coupled to multi-angle light scattering (SEC-MALS) experiments were performed using a Superdex 200 Increase 10/300 column (Cytiva), and an ÄKTA FPLC puri er in line with a Wyatt miniDAWN MALS detector and a Wyatt Optilab digital refractive index (dRI) detector (Amersham Biosciences).All experiments were performed in 1X TBS.ASTRA VI software was used to combine these measurements and enable the absolute molar mass and extinction coe cient of the eluting glycoprotein, Fab, or glycoprotein-Fab complex to be determined 41,42 .

Composition gradient multi-angle light scattering
Composition gradient multi-angle light scattering (CG-MALS) experiments were performed with a Calypso II composition gradient system (Wyatt) to prepare different compositions of buffer, glycoprotein, and antibody and deliver to the miniDAWN detector and an online ultraviolet (UV) detector (Cytiva).The extinction coe cient obtained from the SEC-MALS experiment was used to measure the concentration of the glycoprotein during CG-MALS experiments.Polycarbonate lter membranes with 0.1-µM pore size (Millipore Sigma, Burlington, MA, USA) were installed in the Calypso system for sample and buffer ltration.Glycoprotein was diluted to a stock concentration of 40-60 µg/mL in TBS.Fab was diluted to a stock concentration of 50-60 µg/mL in TBS.The automated Calypso method consisted of a dualcomponent "crossover" gradient to assess hetero-association between the glycoprotein and Fab.For each composition, 0.7 mL of protein solution were injected into the UV and MALS detectors until an equilibrium was reached within the MALS ow cell and the ow stopped for 300-800 s.Data were collected, and analyses were performed with CALYPSO software.GP 2 -3A6 Fab association was measured in triplicate with 2 different preparations of glycoprotein and Fab.

Protein assays
Enzyme-linked immunosorbent and cell-based antibody binding assay were performed with wild-type virus glycoproteins or variants created via alanine scanning following standard protocols.

Cell-based antibody binding assay
To evaluate binding of mAbs to glycoprotein variants, HEK 293T cells expressing full-length GP 1,2 or variants thereof were incubated with unlabeled mAbs at 10 µg/mL, followed by staining with DyLight 488 anti-human IgG and detection of uorescence by microscopy.The binding of a control conformational mAb (ADI-15878 Ig) 9,43,44 was used as a control for glycoprotein expression levels.Secondary antibody binding only was used as a negative control to assess background binding.In detail, HEK 293T cells were plated at ≈ 1×10 5 cells per well in 24-well plates treated with Poly-L-lysine (Millipore Sigma) 1 d prior to transfection.Cells were transiently transfected with 0.5 µg DNA per well using TransIT-LT1 transfection reagent (Mirus Bio, Madison, WI, USA).At 48 h post-transfection, cells were xed with 4% paraformaldehyde (PFA, Electron Microscopy Sciences) in DPBS for 20 min.Cells were then incubated for 1 h at room temperature with 10 µg/mL primary mAbs in DPBS supplemented with 1% BSA (Millipore Sigma).Cells were subsequently incubated at room temperature for 1 h with DyLight 488 anti-human IgG secondary antibody (SA5-10126; Thermo Fisher Scienti c) and Hoechst 33342 (Invitrogen) in DPBS supplemented with 1% BSA.Finally, cells were imaged on a wide eld uorescence Axiovert 200M Marianas microscope with a 10x/0.3dry objective (ZEISS, Feasterville, PA, USA) or a confocal LSM780 microscope with a 10x/0.3dry objective (ZEISS).Images were analyzed in QuPath 45 .Nuclei were segmented using the Hoechst image, and the objects were expanded by 5 µm to locate approximate cell boundaries.DyLight 488-positive and DyLight 488-negative cells were counted using a trained object classi er.The classi er was optimized for the wide eld and confocal images separately.Then, all data from 3 biological replicates were combined, and the 3A6-positive cell percentage was normalized against that obtained with ADI-15878.

Enzyme-linked immunosorbent assay
Microtiter plate wells were coated with puri ed recombinant WT or mutant GP 1,2ΔTM/ΔMLD and incubated at room temperature for 1 h before blocking with 3% bovine serum albumin (BSA; Millipore Sigma) in DPBS containing 0.05% TWEEN-20 (Fisher Scienti c) for 1 h.Serial dilutions of mAb were applied to the wells and incubated for 1 h at room temperature.The bound antibodies were detected using Jackson Immuno Research Labs peroxidase-conjugated goat anti-human IgG (Thermo Fisher Scienti c; #109036006) with horseradish peroxidase (diluted 1:4,000) and 3,3',5,5"-tetramethylbenzidine (TMB) substrate (Thermo Fisher Scienti c) before 50 µl of 1 N sulfuric acid (Fisher Scienti c) was added to stop the reaction.Absorbance at 450 nm was then measured using a Spark microplate reader (Tecan, Männedorf, Switzerland).Half-maximal response (EC 50 ) values for mAb binding were determined using Prism 7 (GraphPad Software, Boston, Massachusetts, USA) after log-transformation of antibody concentrations using EC 50 shift nonlinear regression analysis.

Plaque reduction assay using biologically contained EBOV
A biologically contained EBOV, EbolaΔVP30 virus (Halfmann et al., 2008), was used to assess the impact of a P636S mutation on 3A6-mediated neutralization as previously described (Davis et al., 2019).Brie y, Ebola-GP-P636SΔVP30-eGFP virus was incubated with 10 µg/mL of monoclonal antibody (mAb) at 37°C for 60 min.The virus/mAb mixture was then inoculated onto Vero VP30 cells, seeded the previous day in 12-well plates.After a 60 min incubation, cells were washed to remove any unbound virus, and overlaid with 1.25% methylcellulose media to allow for plaque formation.Seven days after infection, the number of plaques was quanti ed after immunochemistry staining with an antibody against the VP40 protein.
Negative-stain electron microscopy GP 2 -3A6 complexes were obtained by incubating GP 1,2ΔTM/ΔMLD with a three-fold molar excess of 3A6 Fab overnight followed by puri cation using a Superdex 6 Increase 10/300 GL SEC column.The complexes were diluted to 0.01 mg/mL, and 4 µL of the complex solutions were each applied to freshly plasma-cleaned carbon-coated 400-mesh copper grids (Electron Microscopy Sciences, Hat eld, PA, USA) for 1 min.The solutions were blotted from the grids, followed by staining with 1% uranyl formate (Electron Microscopy Sciences, Hat eld, PA, USA) for 1 min.The stain was then blotted from the grids, which were then air-dried before imaging.Images were collected on an FEI Titan Halo 300 kV electron microscope (Thermo Fisher, Waltham, MA, USA) at a magni cation of ×57,000 with a Falcon II camera.Contrast transfer function (CTF) correction, particle picking, 2-dimensional class averaging, and 3dimensional reconstruction and re nement were all performed using cryoSPARC v3.1.0 46.

Virion-like particle preparation, puri cation, and visualization
Ebola virus virion-like particles (VLPs) were prepared from HEK 293T cells by co-expression of full-length GP 1,2 and matrix protein (VP40) essentially as described previously 47 except that phosphate-buffer saline (Gibco PBS; Thermo Fischer Scienti c) instead of TNE buffer was used.Clari cation of supernatants from 4 x 150 mm dishes was performed at 3,000 xg for 15 min at 4°C.Final pellets after density gradient puri cation were resuspended in 200 µL Gibco PBS.

Sample preparation, data collection, and tomogram reconstruction
Fabs (1 mg/mL) were mixed with puri ed VLPs and 10-nm colloidal gold and incubated for 30-60 min at 4°C.Different combinations of Fabs were prepared, imaged, and processed in parallel.The different mixtures (4 µL) were added to C-Flat 2/2 EM grids (Protochips) and vitri ed by back-side blotting (4-s blotting time) using a LeicaGP cryo plunger (Leica; Deer eld, IL, USA) and stored in liquid nitrogen until imaging.
Cryogenic electron tomography data collection was performed essentially as described previously 48 on a Titan Krios electron microscope equipped with Gatan Bioquantum energy lter and K3 direct electron camera (Thermo Fisher).The nominal magni cation was 64,000×, giving a pixel size of 1.39 Å on the specimen.The defocus range was − 2.0 to -4.5 µm, with a 0.25-µm step size (Supplementary Table 2).
To generate an initial starting model for each structure, 50-100 copies of glycoprotein were manually identi ed from VLP laments that were down-scaled by 6× binning of the voxels and subjected to reference-free subtomogram alignment.To identify the positions of all the particles on the viral surface of viral laments, markers were placed manually along the lament of the tube using the Volume Tracer function in UCSF Chimera (v.1.13.1) 52 and the radius of the lament was determined centered at the membrane using the Pick Particle Chimera Plugin 53 .An oversampled cylindrical grid of points was generated on the lament surface with ≈ 8 nm spacing, and subtomograms were extracted for all grid points with a box size of 64 pixels (approximately 50 nm) centered at a radius 10 nm above these grid positions.Initial Euler angles were assigned to each subtomogram based on the orientation of the normal vectors relative to the cylinder surface.

Alanine scanning and antibody binding test
Alanine scanning was performed by introducing alanyls (alanyls were changed to seryls) into GP 1,2ΔMLD region 627-639 via site-directed mutagenesis of GP 1,2ΔMLD -encoding plasmid.The plasmid clones were individually arrayed into 384-well plates and transfected into HEK 293T cells.Protein variants were cell surface-expressed for 22 h 21 .The indicated mAbs were incubated with the cells for 1 h before an Alexa Fluor 488-conjugated secondary antibody (Thermo Fischer Scienti c) was added.Antibody binding was assessed by detection of cellular uorescence with a high throughput ow cytometer (Intellicyt, Albuquerque, NM, USA).Background uorescence was measured in vector-transfected control cells and mAb reactivity against the variants was calculated with respect to reactivity with GP 1,2ΔMLD by subtracting the signal from mock-transfected controls and normalized to signals from wild-type GP 1,2ΔMLDtransfected controls.Residues predicted to be involved in the epitope were identi ed when mAb and variant did not react, but when reactivity of other control mAbs was observed, which excludes glycoproteins variants that were misfolded or were expressed at low levels.

Domesticated guinea pigs
Hartley strain domesticated guinea pigs (Cavia porcellus (Linnaeus, 1758)) of both sexes, aged 6-8 weeks, were acquired from Charles River Laboratories and six animals (three males and three females each) assigned to ve groups.All animals were exposed intraperitoneally (IP) to 1,000 plaque forming units (PFU) of domesticated guinea pig-adapted Ebola virus/UTMB/C.porcellus-lab/COD/1976/Yambuku-Mayinga-GPAon Day 0. Animals in each group were injected IP on Day 3 with 5 mg of i) 3A6, ii) 1A2, iii) 7G7, or iv) 42-2D2 anti-in uenza A virus (FLUAV) IgG in 3 mL of DPBS or received no treatment.Animals were observed daily for clinical signs of disease and were assigned a clinical score of 0-3 (0 = none; 1 = mild; 2 = moderate; 3 = severe).Animals reaching endpoint criteria (score of 3) were euthanized.Weight was recorded daily starting 1 d before exposure until all animals recovered from disease (Day 15), then twice weekly until the study endpoint on Day 28.Blood was collected twice, on Day 6 after exposure and at the time of euthanasia.

Rhesus monkeys
Four rhesus monkeys (Macaca mulatta (Zimmermann, 1780)) of both sexes (WorldWide Primates, Miami, FL, USA) were single-housed and acclimated to BSL-4 conditions prior to virus exposure.On Day 0, monkeys were sedated using intramuscular (IM) injection of 15 mg/kg of Ketamine HCl (KetaThesia, Henry Schein, USA), and injected IM with a target dose of 1,000 PFU of Ebola virus/H.sapienstc/COD/1995/Kikwit-9510621(NR-50306, Lot 9510621, ≥ 95% 7U abundance at the GP editing site, BEI Resources, USA; the same dose and lot of this virus previously resulted in death at days 5-8 postexposure in 12 out of 12 untreated rhesus monkeys ("historical controls" 55 ).On Day 4 and Day 7 after exposure, monkeys 1-3 received 25 mg/kg of 3A6 IgG in DPBS (kindly provided by Chakravarthy Reddyvia) by intravenous infusion, and the control monkey received an equivalent volume of DPBS.All monkeys were observed for the development of clinicals signs of EBOV infection and scored daily according to a four-point scoring scale.Physical examination and blood collections were conducted on the monkeys once prior to exposure (Day − 1) and at 4, 7, 9, 12, 21, and 28 d after exposure.Complete blood counts with reticulocytes and differential were analyzed on a Sysmex XT-2000iV hematology instrument (Sysmex America, New York, NY, USA).Sera were obtained after separation at room temperature and centrifugation for 15 min at 1,500 xg followed by analysis using the Piccolo general chemistry 13 panel on a Piccolo Xpress analyzer (Abaxis, NJ, USA).Prothrombin and activated partial thromboplastin times were measured on a CS-2500 system automated coagulation analyzer (Sysmex America).Infectious titers were determined in sera using an Avicel-based crystal violet stain plaque assay on Vero E6 cell culture monolayers (BEI Resources) as previously described 56 .Sera were inactivated in TRIzol LS according to the manufacturer's instructions (Thermo Fisher Scienti c), and nucleic acid extracted using the QIAamp Viral RNA Mini Kit (Qiagen, Germantown, MD, USA).The BEI Resources Critical Reagents Program (CRP) EZ1 RT-PCR kit assay was used in accordance with manufacturer's instructions 31 on an Applied Biosystems 7500 FastDx Real-Time PCR instrument (Thermo Fisher Scienti c) to quantify EBOV nucleic acids in sera and to transform results into log 10 genome equivalents (GE) per mL of sample.The control monkey died on Day 8 whereas the treated monkeys underwent elective euthanasia approximately 3 mo after virus exposure.

Statistical analysis
details of experiments, including numbers of replicates and measures of precision (standard deviation, SD), can be found in the gure legends, gures, results, and methods.Dose-response ELISA curves were t to a EC 50 shift by nonlinear regression analysis.All analyses were performed with Prism 7.

Figures
3 Å, b = 66.4 Å, c = 68 Å, α = γ = 90°, β = 104.2°),and four Fabs were present in the asymmetric unit of the Fab structure in space group P1 (a = 53.7 Å, b = 65.7 Å, c = 125.6Å, α = 98.7°, β = 91.4°,γ = 96.0°).Crystal structures were determined by molecular replacement using Phaser 36 within the CCP4 package37 , with a homology model predicted with SWISS-MODEL 38 as a starting model.Iterative manual model rebuilding was performed using Coot39 and re ned with Phenix40 .The peptide was built into different Fourier maps and calculated prior to inclusion of the respective structural elements.Final atomic coordinates and structure factors of the Fab-peptide complex and Fab were deposited in the Protein Data Bank (PDB) under identi cation numbers (IDs) 7RPU and 7RPT, respectively.Figures were created in PyMOL (http://www.pymol.org/).
Animal exposure and treatment experiments using infectious EBOV were performed in the biosafety level 4 (BSL-4) laboratory at the Integrated Research Facility at Fort Detrick (IRF-Frederick), Division of Clinical Research (DCR), National Institutes for Allergy and Infectious Diseases (NIAID), National Institutes of Health (NIH) under accreditation (000777) by the Association for Assessment and Accreditation of Laboratory Animal Care (AAALAC), Laboratory Animal Welfare approval (D16-00602) by the Public Health Service (PHS), and United States Department of Agriculture (USDA) registration (51-F-0016).Animal experiments were approved by the NIAID DCR Animal Care and Use Committee (ACUC) and followed the recommendations provided in the Guide for the Care and Use of Laboratory Animals.

Figure 1 Crystal
Figure 1

Figure 4 Binding
Figure 4