Design and evaluation of bi- and trispecific antibodies targeting multiple filovirus glycoproteins

Filoviruses (family Filoviridae) include five ebolaviruses and Marburg virus. These pathogens cause a rapidly progressing and severe viral disease with high mortality rates (generally 30–90%). Outbreaks of filovirus disease are sporadic and, until recently, were limited to less than 500 cases. However, the 2013–2016 epidemic in western Africa, caused by Ebola virus (EBOV), illustrated the potential of filovirus outbreaks to escalate to a much larger scale (over 28,000 suspected cases). mAbs against the envelope glycoprotein represent a promising therapeutic platform for managing filovirus infections. However, mAbs that exhibit neutralization or protective properties against multiple filoviruses are rare. Here we examined a panel of engineered bi- and trispecific antibodies, in which variable domains of mAbs that target epitopes from multiple filoviruses were combined, for their capacity to neutralize viral infection across filovirus species. We found that bispecific combinations targeting EBOV and Sudan virus (another ebolavirus), provide potent cross-neutralization and protection in mice. Furthermore, trispecific combinations, targeting EBOV, Sudan virus, and Marburg virus, exhibited strong neutralization potential against all three viruses. These results provide important insights into multispecific antibody engineering against filoviruses and will inform future immunotherapeutic discoveries.

Filoviruses are negative-strand RNA viruses that cause severe hemorrhagic fever with case fatality rates of up to 90% in humans and nonhuman primates (1,2). There are five known ebolaviruses: Ebola virus (formerly "Ebola Zaire," EBOV 3 ), Res-ton virus, Taï Forest virus, Sudan virus (SUDV), and Bundibugyo virus. EBOV has been the most prevalent in terms of human infections, but both SUDV and Bundibugyo virus have caused large (Ͼ300 cases) outbreaks in the past (3).
With more than 28,000 suspected cases of infection, the 2014 -2016 EBOV disease epidemic in western Africa far exceeded the scale of any previous filovirus outbreak and underscored the need for filovirus pre-and post-exposure treatments. (4) To date, no therapeutic drug is approved, although several mAb mixtures are undergoing clinical trials. A number of studies have shown that mAb or mAb mixtures provide protective efficacy in nonhuman primates (NHPs) (5)(6)(7)(8). One mAb mixture, ZMapp TM (Mapp Biopharmaceutical), reversed the course of advanced Ebola virus disease in NHPs when provided 5 days after infection (5). A recent clinical study (PREVAIL II) indicated that although patients receiving ZMapp TM fared better than those receiving the previous standard of care, these differences within the small sampling of patients were below the threshold of statistical significance (9). Nonetheless, based on the strong trend of efficacy, the NIAID, National Institutes of Health and the Food and Drug Administration now consider ZMapp TM to be the standard of care. These results and more recent data showing that mAbs can protect NHPs from MARV challenge indicate that mAbs or mAb mixtures have strong therapeutic potential (10).
Until recently, there were few filovirus mAbs with demonstrated cross-neutralizing or cross-protective activity. This is likely due to the high degree of sequence variability across ebolaviruses in the envelope glycoprotein (GP), the primary target of neutralizing antibodies (11)(12)(13). Nonetheless, several mAbs from macaque immunizations or natural human infections have recently been shown to confer in vivo protection of rodents from multiple ebolaviruses (14 -19). The prefusion GP spike consists of three copies each of the surface subunit (GP1) and the transmembrane subunit (GP2) (20 -22). Regions of GP1, including its heavily glycosylated mucin-like domain (MLD), constitute the majority of the solvent-exposed surface. Cross-neutralizing GP1 mAbs typically target the glycan cap region (23,24). At the exposed region of the GP1-GP2 interface (the "base"), the GP2 fusion loop and surrounding residues provide other targets of cross-neutralizing mAbs (23). Given the sporadic nature of filovirus outbreaks and the inability to predict which viral species will be the causative agent, there is a strong potential benefit to cross-protective therapies.
We and others have been exploring multispecific antibody engineering strategies as a complementary approach to generation of cross-protective antibodies (25,26). We showed recently that combining two species-specific variable domains into a bispecific antibody (bsAb) confers cross-protection in mice for EBOV and SUDV (27). More recently, the specific combination of variable domains targeting broadly reactive (but non-neutralizing) epitopes on the viral surface with variable domains that disrupt critical interactions between the host receptor (Niemann Pick C1) and the endosomally exposed receptor-binding site (RBS) of GP1 resulted in pan-ebolavirusneutralizing activity (28 -30). Here we examine an extended panel of bsAbs targeting EBOV and SUDV and explore trispecific combinations (tsAbs) designed to provide activity against EBOV, SUDV, and MARV. The results illustrate that crossneutralization across diverse filoviruses is possible with appropriately engineered antibodies. Our results further reveal sites of vulnerability in the glycoprotein and add to the repertoire of cross-neutralizing antibodies.

Design, expression, and purification of multispecific antibodies
Previously we reported the construction and evaluation of bsAbs targeting the GP base epitope of EBOV and SUDV utilizing the scFv-Ig format (27). For these bsAbs, we employed the variable domains of two humanized variants of SUDV mAb 16F6 (F4 and E10) and the human EBOV mAb KZ52. We found that the most efficacious combination was fusion of the scFv from KZ52 to the C terminus of the F4 heavy chain (scKZ52-F4 HCC). This molecule provided potent cross-neutralization of both surrogate viral particles (vesicular stomatitis virus bearing the GP of EBOV or SUDV, VSV-GP(EBOV) or VSV-GP(SUDV), respectively) as well as authentic EBOV (Mayinga) and SUDV (Boniface). Furthermore, in two separate murine models of disease, scKZ52-F4 HCC afforded complete protection with post-exposure dosing.
We sought to explore alternative variable domains and bsAb formats to determine the effects on binding and inhibitory properties. Specifically, although the KZ52 mAb provided protection in mice and guinea pigs, and the variable domains, when integrated into the scFv-Ig fusions, provided protection in mice, in a single study, KZ52 was not protective in NHPs. To explore alternative EBOV-specific variable domains targeting the GP base epitope, we designed and generated scFv-IgG constructs utilizing the F4 IgG with scFv from c2G4, one of the neutralizing ZMapp TM components, as an alternative (Fig. 1A) (5). The c2G4 scFv was fused at the heavy chain C terminus (sc2G4-F4 HCC), light chain N terminus (sc2G4-F4 LCN), and light chain C terminus (sc2G4-F4 LCC). Previously we had determined that fusion of scFvs to the N terminus of the F4 IgG light chain abrogates activity, likely by sequestration of required interactions of the F4 heavy chain with SUDV GP.
To expand the potential scope of filovirus multispecific antibodies, we sought to integrate a third specificity toward MARV. The domain organization of MARV GP differs significantly from those of the ebolaviruses, including in sites of susceptibility for neutralization. The MARV GP mucin-like domain is proposed to sit lower on the trimer, likely occluding binding at the base region. However, the RBS is more exposed, and thus mAbs that bind the RBS have been shown to neutralize MARV GP (31)(32)(33). Such RBS-binding mAbs can cross-react with the corresponding regions on ebolavirus GP because the RBS is highly conserved across the family, but do not confer strong neutralization of ebolaviruses because the RBS is generally sequestered in the native form. Thus, targeting of the ebolavirus RBS requires intracellular delivery of RBS mAbs to the endosome, where viral membrane fusion occurs (28). Nonetheless, MARV mAb MR78 has been shown previously to inhibit MARV entry (32). Thus, we created a series of trispecific antibodies (tsAbs) in an scFv-scFv-IgG fusion containing the variable domains of MR78, KZ52, and F4 (Fig. 1B). Initial tsAb designs containing the F4 IgG as the central molecule with scFvs of KZ52 and MR78 appended as fusions to the N and/or C termini of the light and heavy chains were generated but did not exhibit neutralizing properties against VSV-GP(MARV) (see below). Thus, subsequent tsAb designs utilized an MR78 IgG core but with KZ52 and F4 scFvs added to the C termini of the light and heavy chains.

Engineered multispecific antibodies against filoviruses
Although the scFv-IgG format provides a straightforward design and expression strategy for generation of bsAbs, the long polypeptide linkers between variable domains on the scFv or between the scFv and IgG represent a potential liability for proteolytic degradation or immunogenic presentation. Furthermore, scFvs are known in some instances to oligomerize or be prone to aggregation because of inter-scFv association of the variable domains that can still dissociate from one another because they are connected by a long flexible linker. For EBOVspecific mAbs, our previous studies have explored mostly variable domains that target a single neutralizing epitope at the base of the GP, but other regions (such as the glycan cap) are also known sites of susceptibility (24). We therefore also generated bsAbs comprising the variable domains of F4 or E10 (SUDV-specific) fused to those of c13C6 (EBOV-specific, and a ZMapp TM component) in the dual variable domain (DVD-Ig) format ( Fig. 1C) (34). The DVD-Ig utilizes short human-derived peptide linkers to directly append one set of variable domains on top of another set within the IgG backbone. Presumably, the shorter linker length is less susceptible to proteolysis, and there is less inherent flexibility and, thus, aggregation because variable domain dissociation is less likely.
All bsAbs and tsAbs were expressed readily in human embryonic kidney 293 suspension cells and were purified by protein A chromatography. bsAbs and tsAbs migrated with the expected molecular weight by SDS-PAGE analysis and fragmented into the expected heavy and light chains upon reduction ( Fig. 1 and Fig. S2).

Binding studies and biochemical properties
Binding activities for bsAb and tsAbs were assessed initially by ELISA (Fig. 2) and then more quantitatively by biolayer interferometry (Fig. 3). For sc2G4-F4 bsAbs, the binding activity against both EBOV and SUDV GP was retained in all three of the tested formats (HCC, LCC, and LCN). ELISA with either EBOV or SUDV GP immobilized onto the well provided nanomolar EC 50 values. BLI studies, in which the bsAb was immobilized onto the sensor surface and then dipped into an analyte solution containing EBOV or SUDV GP, indicated that association with either GPs occurred in the 10 4 -10 5 /Ms range, and dissociation in the 10 2-10 3 /s range. The kinetic binding data could be described reasonably with a 1:1 model. However, because the bsAbs are bivalent in their ability to engage the antigen, and the GPs themselves are trimeric, it's likely that the binding stoichiometries are more complicated than a simple 1:1 model and thus the kinetic data represent an ensemble of interaction stoichiometries. Nonetheless, the k off /k on ratios ("apparent K D s") were in the range of 10 Ϫ7 -10 Ϫ8 M. These values were consistent with the approximate affinities that we have determined previously for F4 (2.5 ϫ 10 Ϫ7 M) and c2G4 (2.2 ϫ 10 Ϫ7 M), indicating that the affinity of the variable domains within the bsAb was not affected (35).
A tsAb containing F4 as the core IgG with scFvs of KZ52 and MR78 as fusions to the heavy chain C terminus and light chain N terminus, respectively, ("scKZ52(LCN)-scKZ52(HCC)-F4") retained binding activity toward EBOV, SUDV, and MARV ( Figs. 1 and 4). However, this tsAb and other tsAbs bearing F4 as the central IgG and scFvs for MR78 and KZ52 failed to neutralize VSV-GP(MARV) at 270 nm (Fig. 4). These results suggest that conversion of the MR78 variable domains into the scFv format results in loss of proper assembly of the variable domains to faithfully recapitulate the MR78 activity, despite the ability to bind MARV GP.

Engineered multispecific antibodies against filoviruses
In contrast, a tsAb bearing MR78 as the core IgG with scFvs of F4 and KZ52 appended as C-terminal fusions to the heavy and light chains, respectively, ("scF4(HCC)-scKZ52(LCC)-MR78") was able to bind and neutralize all three GPs. BLI measurements provided apparent K D measurements in the doubledigit nanomolar range against MARV and EBOV and ϳ650 nM
To examine whether all three specificities on scF4(HCC)-scKZ52(LCC)-MR78 could be engaged simultaneously, a threestep sequential binding experiment was performed in which immobilized tsAb was sequentially dipped in solutions containing GP(EBOV), GP(MARV), and GP(SUDV) (Fig. 5D). All three antigens could be bound simultaneously.
The c13C6ϳE10 DVD-Ig bound both GP(EBOV) and GP(SUDV), but, surprisingly, the c13C6ϳF4 DVD-Ig did not bind to the GP(SUDV) (Figs. 2 and 3). The basis for this difference in activity is not clear, and is unanticipated because F4 and E10 are highly related to one another in terms of sequence. Nonetheless, kinetic binding data indicate that affinity for both components was again retained for the c13C6ϳE10 DVD, with 5.3 ϫ 10 Ϫ8 M for GP(SUDV) and 3.8 ϫ 10 Ϫ9 M for GP(EBOV) (previously reported to be 2.1 nM by Doranz and co-workers (36).

Neutralization profiles
We next sought to explore the capacity of bsAbs and tsAbs to inhibit cell entry mediated by multiple filovirus glycoproteins. For this purpose, we utilized recombinant vesicular stomatitis virus particles containing filovirus GP in place of the native glycoprotein G (rVSV-GPs). All rVSV-GPs contained an enhanced GFP (eGFP) in the genome to facilitate quantification of infection events by fluorescence. The rVSV-GPs have been shown previously to provide faithful recapitulation of GP-mediated cell attachment and viral fusion (37). Furthermore, neutralization of rVSV-GPs has been shown previously to be a highly predictive measure of antibody neutralization of authentic pathogens but can be manipulated under routine (BSL2) laboratory conditions (18,27,28).
All three variants of the scFv-IgG fusions of 2G4 and F4 exhibited strong neutralization potential against rVSV-GP-(SUDV) and rVSV-GP(EBOV), similar to a mixture consisting of a stoichiometric 1:1 mixture of the monospecific parental antibodies (Fig. 6, A and B). The sc2G4-F4 HCC construct had the strongest neutralization potential against both rVSV-GPs, whereas activity of sc2G4-F4 LCC was lower for rVSV-GP(EBOV) despite comparable binding affinity relative to the two other sc2G4-F4 scFv-IgG variants. A possible explanation for this discrepancy is that the location of the scFv on the light chain C terminus somewhat occludes the potential for intraspike cross-linking by the c2G4 scFv. Notably, we and others have found that Fab fragments of base-targeting filovirus mAbs have a much lower neutralization potential than the corresponding IgGs, suggesting that the bivalent cross-linking contributes to the neutralization activity (35). sc2G4-F4 HCC had comparable IC 50 values against both rVSV-GPs relative to individual parental antibodies as well as to the respective 1:1 mixture. However, the activity of sc2G4-F4 LCN and sc2G4-F4 LCC was somewhat diminished against rVSV-GP(EBOV) relative to the c2G4 parental control. This diminution in activity was not observed for rVSV-GP(SUDV) for any of the bsAbs compared with the F4 monospecific control. The two monospecific controls were orthogonal in their activities, with F4 having no activity against rVSV-GP(EBOV) and c2G4 likewise being inactive against rVSV-GP(SUDV). The c13C6ϳE10 DVD-Ig was potently neutralizing against rVSV-GP(SUDV), albeit with a reduction in activity relative to the WT parental E10 IgG as well as the respective comparator mixture (Fig. 6D). Neither c13C6 nor the DVD-Ig showed any significant neutralization activity against rVSV-GP(EBOV) (Fig. 6C). However,

Engineered multispecific antibodies against filoviruses
this result was expected and consistent with previous studies of c13C6. c13C6 engages the glycan cap epitope and is non-neutralizing but indispensable for ZMapp TM in vivo protection. The specific mechanism of c13C6-mediated protection is unclear but may involve Fc-mediate function for infected cell or virus clearing.
The scF4(HCC)-scKZ52(LCC)-MR78 tsAb neutralized all three rVSV-GPs, albeit with somewhat reduced potency against rVSV-GP(EBOV) relative to the parental KZ52 mAb (Fig. 7, A-D). IC 50 values against all three viruses were in the nanomolar range, similar to those observed for the mixture consisting of the monospecific parental antibodies, despite the fact that SUDV binding had ϳ10-fold lower affinity. When the tsAb was tested in a microneutralization assay with authentic EBOV, SUDV, and MARV, high levels of neutralization against all three pathogens were observed at concentrations of 500 nM, with diminishing levels of neutralization at lower tsAb concentrations (Fig. 7, E and F). However, the parental F4 and KZ52 mAbs were strongly neutralizing (Ͼ70%) against their respective authentic pathogens at concentrations as low as 2 nM. Levels of neutralization against authentic MARV were similar among the parental MR78 and the tsAb.

In vivo activity of E10ϳ13C6 DVD Ig
The protective capacity of the E10ϳ13C6 DVD was tested in two separate post-exposure mouse models. For EBOV, WT C57BL/6 mice were challenged with a lethal dose of mouseadapted EBOV (maEBOV) and then treated 1 day post-infec-tion with 100 g of bsAb. For the SUDV challenge, we utilized a previously reported model in which type I ␣/␤ interferon receptor knockout (IFNAR Ϫ/Ϫ ) mice were challenged with human lethal Sudan virus (Boniface). bsAb treatment (500 g/dose) was provided on days 1 and 5 post-exposure. In each case, the monospecific controls E10 and 13C6 were included, as was a 1:1 mixture of these two mAbs for comparison. For the SUDV challenge, because 13C6 exhibits some SUDV activity, the monospecific EBOV mAb 6D8 was included as a further control. As shown in Fig. 8, A and B, E10ϳ13C6 DVD-Ig provided 80% protection against maEBOV and 70% protection against SUDV; however, protection against SUDV was not statistically significant relative to the control mAb 6D8 (30%), with a group size of n ϭ 10. In both models, weight loss in E10ϳ13C6 DVD-Ig-treated mice was observed, a clinical sign of illness. However, in both cases, the aggregate weight loss was followed by weight gain after day ϳ10, indicating recovery from disease. For SUDV, this weight loss trend was again indistinguishable from the negative control 6D8. For both viruses, the mixture of E10 and 13C6 provided significant protective efficacy relative to controls (100% and 90% protection in maEBOV and SUDV, respectively), but mice receiving the combination therapy experienced no (maEBOV) or little weight loss (SUDV). The monospecific SUDV control E10 afforded no activity against maEBOV, with all mice losing weight until succumbing to the disease. However, E10 was 90% protective against SUDV, with only moderate weight loss, as reported previously. 13C6 was

Engineered multispecific antibodies against filoviruses
100% protective against maEBOV (as reported previously), and 60% protective against SUDV, but again indistinguishable from the negative control mAb 6D8 in terms of survival and weight loss. Together, these results suggest that E10ϳ13C6 DVD Ig is strongly protective against EBOV but does not confer a statistically significant survival advantage for SUDV relative to the negative control mAb 6D8.

Discussion
Here we have explored newly designed bsAb and tsAbs for neutralization of multiple filoviruses. These reagents are complementary to the existing set of cross-neutralizing mAbs from NHP immunization and human infection (14 -19, 27, 28). Furthermore, the structure-activity relationship studies reported here with different bsAb and tsAb formats provide novel insights into requirements for cross-neutralization and crossprotection from filoviruses that can be utilized for future mAb engineering efforts.
We found here and previously that many bsAb combinations employing combinations of SUDV mAb F4 and base-binding EBOV mAbs (KZ52 and 2G4) afford potent cross-neutralization of these two filoviruses (27). These studies further confirm the base structural subdomain as an important site of vulnerability for at least SUDV and EBOV. Recently, human mAbs that bind slightly higher on the prefusion spike (e.g. ADI-15878, ADI-15742, and ADI-15946) have been shown to afford broad or even panebolavirus-neutralizing activity (18). Mechanistically, binding in this region likely sequesters critical transformations of GP during viral entry. For example, ADI-15946 stabilizes EBOV GP to proteolysis by CatL, a critical process that exposes the RBS for binding to NPC1. Furthermore, KZ52 may sequester critical membrane fusion events within endosomes. The work described here, along with structural data on ADI-15878, ADI-15742, and ADI-15946, suggest that binding of GP in this region is enough to confer neutralization and protection potential, at least in rodents.

Engineered multispecific antibodies against filoviruses
We further explored the potential for a bsAb combination employing an SUDV GP base binder (E10) linked to an EBOV GP glycan cap binder (13C6) in the DVD-Ig format. Although this bis-mAb reflected aspects of both parental mAbs in binding and neutralization profiles, it provided in vivo efficacy only against EBOV. Therefore, the precise combination of mAbs and the bispecific format into which they are integrated can have nonobvious effects for in vivo efficacy. Some degree of trial and error, as with all bsAb design strategies, is required to identify optimal combinations.
Results with tsAbs indicate that, in principle, neutralization against disparate members of filoviruses is possible with a single agent. However, it seems likely that the parameters required for neutralization of MARV differ significantly from any of the ebolaviruses. Only a single non-neutralizing mAb, FVM02p, has been described to cross-react across multiple ebolaviruses and MARV (17). Overall, MARV GP is only ϳ30% identical to EBOV (whereas the GP of SUDV, the most distant ebolavirus from EBOV, is ϳ55% identical). Furthermore, there are distinct differences in the domain organization and spatial location of the MLD between the prefusion GP of EBOV and MARV. It has been reported that the MLD sits lower on the viral spike in MARV, likely sequestering base epitopes but leading to a more exposed RBS (33). Thus, true pan-filovirus activity may require combination of multiple mAbs, some specific to Ebola viruses

Engineered multispecific antibodies against filoviruses
but others specific for Marburg viruses, either as a mixture or as engineered multispecific Abs. Our results further indicate that engineering of tsAbs with three distinct functional specificities is, in fact, possible with the scFv-scFv-IgG format. However, we found that only particular combinations yielded productive tsAbs.
The bsAbs and tsAbs reported here may have immunotherapeutic potential, especially when included in mixtures with other cross-neutralizing mAbs. Overall, for viral immunotherapy, it seems advantageous to target multiple epitopes to mitigate viral escape by a single point mutation and to leverage potential synergism in mechanisms of action. In this context, it seems that a potentially tractable approach to multiepitope targeting with broad mixtures would be to combine multiple appropriately engineered bsAbs or tsAbs. Furthermore, for cases where bsAbs provide equivalent activity to a mixture consisting of the individual mAbs, inclusion of those mAbs as part of bsAb allows the use of both specificities but without the manufacturing burden of having to produce both molecules. In theory, such a manufacturing burden could be tractable for combination of two mAbs, but combination of four or more mAbs may be too onerous to be feasible on a large scale. Therefore, multispecific antibody engineering approaches described here may be implemented to reduce the manufacturing complexity of producing a mixture.
Multispecific mAbs have been utilized extensively in oncology, but their applications in viral immunotherapy are only now emerging (25). At present, the true immunotherapeutic potential of multispecific, engineered mAbs in relation to combinations of traditional mAbs remains to be determined. Multispecific mAbs contain engineered, non-natural segments that could potentially affect critical parameters such as stability and in vivo half-life and could elicit an anti-therapeutic antibody response by the host. However, the advantages of multispecific mAbs include the ability to target multiple epitopes within a single agent and the capacity to engender properties within an engineered bi-or trifunctional molecule that is not possible with either component alone. A good example of this latter principle is the development of "Trojan horse" bsAbs that engage the ebolavirus GP RBS deep within the endosomal pathway (18). The results with bsAbs and tsAbs presented here, and in other reports for filoviruses and other viral pathogens (25,26), indicate that multispecific antibodies may exhibit enhanced properties relative to canonical monospecific mAbs that are particularly beneficial for immunotherapy.

Antibody expression and purification
To facilitate the expression of multispecific antibody molecules, the respective single-chain variable fragment (scFv) as well as DVD fusions were subcloned into the respective pMAZ-IgH and pMAZ-IgL vectors developed by Mazor et al. (38) (Fig.  S1). Subsequently, the respective pMAZ-IgH and pMAZ-IgL vectors encoding each antibody were co-transfected into Freestyle TM -293F suspension-adapted human embryonic kidney 293 cells using linear polyethyleneimine (Polysciences, Warrington, PA). Cell cultures were incubated at 37°C and 8% CO 2 for 6 days post-transfection. Cleared cell supernatants were then applied to a protein A affinity column (ϳ1 ml of packed beads per 600 ml of culture) (Thermo Scientific). Antibodies were purified using the Gentle Antibody Elution System (Thermo Scientific) according to the manufacturer's instructions and subsequently exchanged into 150 mM HEPES (pH 7.4) and 200 nM NaCl.

BLI
The OctetRed TM system (ForteBio, Pall LLC) was used to determine the binding properties of IgGs and DVD-Igs. Antihuman Fc capture sensors were used for initial Ab loading. For single-phase binding experiments, global data fitting to a 1:1 binding model was used to estimate values for the k on (association rate constant), k off (dissociation rate constant), and K D (equilibrium dissociation constant). For double-phase binding experiment, the Ab was first immobilized on an Fc sensor and then allowed to equilibrate in a solution containing the first antigen. The sensor was then transferred to a second solution containing the second antigen.

Virus neutralization assays
Recombinant vesicular stomatitis Indiana viruses (rVSVs) expressing eGFP and EBOV GP in place of VSV G have been described previously (28,37). Similar rVSVs expressing eGFP and representative GP proteins from SUDV and MARV were generated as above. The infectivities of rVSVs were measured by automated enumeration of eGFPϩ cells (infectious units) using a CellInsight CX5 imager (Thermo Fisher) at 12-14 h post-infection. For Ab neutralization experiments, pretitrated amounts of rVSV-GP particles (multiplicity of infection Ϸ 1 infectious unit/cell) were incubated with increasing concentrations of test Ab at room temp for 1 h prior to addition to cell monolayers in 96-well plates. Viral infectivities were measured as above.

Mouse challenge experiments
Research was conducted under an IACUC-approved protocol in compliance with the Animal Welfare Act, Public Health Service Policy, and other federal statutes and regulations relating to animals and experiments involving animals. The facility where this research was conducted is accredited by the Association for Assessment and Accreditation of Laboratory Animal Care, International and adheres to principles stated in the Guide for the Care and Use of Laboratory Animals, National Research Council, 2011.
Female C57BL/6 mice (6 -14 weeks old) (The Jackson Laboratory) were challenged via the intraperitoneal route (I.P.) with 1000 pfu of maEBOV. Mice were treated 1 day post-challenge I.P. with 100 g of antibody or PBS.
Male and female type 1 IFN ␣/␤ receptor knockout mice (type 1 IFN␣/␤ R Ϫ / Ϫ ) purchased from The Jackson Laboratory (4 -14 weeks of age) were utilized in these experiments. Mice were challenged I.P. with a target dose of 1000 pfu of WT SUDV and treated I.P. with 500 g of the indicated mAb at day ϩ1 and day ϩ5.
Following all challenges, mice were observed daily for clinical signs of disease and lethality. Daily observations were increased Engineered multispecific antibodies against filoviruses to a minimum of twice daily while mice were exhibiting signs of disease. Moribund mice were humanely euthanized on the basis of IACUC-approved criteria.