Structures of Foot-and-mouth Disease Virus with neutralizing antibodies derived from recovered natural host reveal a mechanism for cross-serotype neutralization

The development of a universal vaccine against foot-and-mouth disease virus (FMDV) is hindered by cross-serotype antigenic diversity and by a lack of knowledge regarding neutralization of the virus in natural hosts. In this study, we isolated serotype O-specific neutralizing antibodies (NAbs) (F145 and B77) from recovered natural bovine hosts by using the single B cell antibody isolation technique. We also identified a serotype O/A cross-reacting NAb (R50) and determined virus-NAb complex structures by cryo-electron microscopy at near-atomic resolution. F145 and B77 were shown to engage the capsid of FMDV-O near the icosahedral threefold axis, binding to the BC/HI-loop of VP2. In contrast, R50 engages the capsids of both FMDV-O and FMDV-A between the 2- and 5-fold axes and binds to the BC/EF/GH-loop of VP1 and to the GH-loop of VP3 from two adjacent protomers, revealing a previously unknown antigenic site. The cross-serotype neutralizing epitope recognized by R50 is highly conserved among serotype O/A. These findings help to elucidate FMDV neutralization by natural hosts and provide epitope information for the development of a universal vaccine for cross-serotype protection against FMDV.

Introduction Foot-and-mouth disease virus (FMDV), a small nonenveloped single-stranded positive-sense RNA virus (genus Aphthovirus within the family Picornaviridae), is the causative agent of footand-mouth disease (FMD). FMD is one of the most contagious and economically devastating diseases of cloven-hoofed animals in many developing regions of Asia, Africa and South America [1][2][3]. FMDV exists as seven immunologically distinct serotypes (O, A, C, SAT1, SAT2, SAT3 and Asia1) [4]. Among them, serotypes O and A are the most common causative agents of FMDV outbreaks globally [5,6]. Vaccination is believed to play a predominant role in the control and prevention of FMDV [7]. However, the antigenic diversities of the currently known epitopes throughout FMDV serotypes and the lack of understanding of FMDV neutralization in natural hosts limit the development of a vaccine that is able to provide cross-serotype protection [8][9][10][11].
As the major antigen, the virus capsid of FMDV comprises 60 copies each of four virusencoded proteins, VP1, VP2, VP3 and VP4; VP1 to VP3 form most of the capsid shell, with VP4 lining the interior surface [12,13]. At present, there is limited structural information regarding how FDMV engages with neutralizing antibody (NAb). The cryo-electron microscopy (cryo-EM) structure of the FMDV-C-S8c1-SD6 Fab complex shows that the murine Fab fragment SD6 binds to the VP1 GH-loop (residues 136-147), and the GH-loop plays a key role in the interaction with integrin receptors [14,15].
As a natural host of FMDV, bovines have a different immunoglobulin (Ig) repertoire than other vertebrates, which display restricted lengths of the third heavy chain complement determining region (HCDR3), averaging 12-16 amino acids. Bovines, however, produce antibodies with HCDR3s that average~26 amino acids, with an ultralong subset (10-15% of the repertoire) that contains greater than 70 amino acids [16][17][18]. These characteristic Ig sequences point to bovines as a promising host for producing high-affinity and broadly neutralizing antibodies. This is exemplified by the rapid elicitation of broadly neutralizing antibodies to HIV after bovine immunization; these broadly neutralizing antibodies contain ultralong HCDR3s that are responsible for their serological breadth and potency [19].
As yet, it is unclear exactly how FMDV is neutralized by natural bovine host NAbs and whether an epitope exists that can elicit cross-serotype NAbs. In a previous study, we reported the isolation of two serotype O-specific NAbs (F145 and B77) from recovered natural bovine hosts through the single B cell antibody isolation technique [20]. In the current study, we isolated a NAb (R50) that neutralizes FMDV serotypes O and A. We determined the cryo-EM structures of FMDV serotype O (O/Tibet/99) in complex with B77 and F145 scFv, as well as R50 scFv in complex with FMDV serotype O (O/Tibet/99) and with serotype A (A/WH/CHA/ 09). The near atomic level details reveal a key epitope for cross-serotype neutralization and inform a structure-based rationale to design a universal FMDV vaccine that can elicit intertypic cross-reacting NAbs.

Isolation of antibodies from bovine plasmablasts
Upon isolating two FMDV serotype O-specific antibodies (B77 and F145) from a natural bovine host [20], we aimed to isolate a cross-serotype NAb that can neutralize both serotype O and A. FMDV (O/Tibet/99) (hereafter named FMDV-OTi) and FMDV (A/WH/CHA/09) (hereafter named FMDV-AWH) were applied as baits to isolate specific memory B cells from natural bovine host blood, which displayed strong cross-reactivity with both serotype O and A. As revealed in fluorescence-activated cell sorting (FACS), CD21 + IgM -O -FMDV + A -FMDV + B cells are extremely rare, displaying 21 per million PBMCs. Via FACS, a total of 96 single CD21 + IgM -O -FMDV + A -FMDV + B cells were sorted for the amplification of the IgG BCR variable gene and the development of a FMDV serotype O and A cross-neutralizing scFv antibody ( Fig 1A). Finally, 60 paired IgG clones from these cross-reactivity B cells were successfully amplified. Of these, 14 scFv antibodies displayed cross-reactivity with both FMDV serotype O and A. But, only R50 scFv antibody showed cross-neutralizing activity with both FMDV serotype O and A.
To further investigate the neutralization breadth and potency of R50 scFv antibody, we propagated, purified three representative strains of FMDV serotype O (O/Mya/98; O/HN/ CHA/93; O/Tibet/99) obtained from three currently epidemic topotypes (SEA, Cathay, and ME-SA, respectively) and two representative strains of FMDV serotype A (A/WH/CHA/09; A/GDMM/CHA/2013) from the Asia topotypes SEA-97/G1 and G2 in China [21], and separately examined their binding and neutralization abilities by flow cytometry and microneutralization assay. The flow cytometry showed that R50 scFv binds to the three topotypes of FMDV serotype O and the two distinctive strains of FMDV serotype A (Fig 1B). In line with the binding results, microneutralization assay also revealed that R50 scFv cross-neutralizes these strains of FMDV serotype O and A (Fig 1C). In contrast, B77 and F145 scFv only neutralize the three representative strains of FMDV serotype O (Fig 1C), as previously reported [20].

Overall architectures of the FMDV-NAb complexes
To clarify the mechanism of cross-serotype neutralization, we determined the structures of B77 and F145 scFv in complex with FMDV-OTi, as well as R50 scFv in complex with FMDV-OTi and FMDV-AWH (S1 Fig). The cryo-EM micrographs indicated the presence of scFv on the FMDV capsids at distinct locations (Fig 2). The final resolution of the cryo-EM reconstruction was estimated by the FSC 0.143 cutoff to be 3.90 Å for the FMDV-OTi-B77 complex, 3.68 Å for the FMDV-OTi-F145 complex, 3.94 Å for the FMDV-OTi-R50 complex and 3.49 Å for the FMDV-AWH-R50 complex (S1 Fig). Three B77 or F145 scFv molecules bind to the FMDV-OTi capsid around an icosahedral threefold axis (Fig 2A, 2B, 2E and 2F). By contrast, R50 scFv binds to a depression between the 2-and 5-fold axes of FMDV-OTi and FMDV-AWH (Fig 2C, 2D, 2G and 2H). In all cases, the cryo-EM densities were of sufficient quality to allow the atomic modeling of FMDV capsid proteins and the NAb scFv variable loops that are responsible for virus recognition. However, the densities of the bound NAbs away from the contact surface were not sufficiently clear to trace all main chains (S2 Fig). The footprints of B77, F145 and R50 on the capsid surface were also mapped (Fig 2I-2L).
The lengths of bovine HCDR3 feature a trimodal distribution: group 1 comprises very short HCDR3s (�10 amino acids); group 2 comprises intermediate lengths (11 to 47 amino acids); and group 3 comprises ultralong HCDR3s (�48 amino acids) [22,23]. The structures show that B77 contains an intermediate length HCDR3 loop with 18 amino acids, but F145 and R50 contain ultralong HCDR3 loops (50 and 61 amino acids, respectively) (S3 Table). The ultralong HCDR3s form an elongated "stalk" and a globular "knob" at the distal ends of the stalks, and the stalk region is essentially a display scaffold to deliver the knob domain, which functions as an effector domain that mediates the molecular recognition of antigen. The globular knob of F145 interacts with VP2, but the globular knob of R50 binds to a shallow depression on the capsid, contacting the two major capsid proteins VP1 and VP3 from two adjacent protomers (S3 Fig).

Interaction of B77 and F145 with FMDV-OTi
The complex structures show that B77 and F145 make similar contacts with the BC-loop and HI-loop of FMDV-OTi VP2. Residues in the VP2 BC-loop ( OTi-VP2 T71 and OTi-VP2 S72) and in the VP2 HI-loop ( OTi-VP2 N190 and OTi-VP2 Q196) interact with residues B77-VH M56, B77-VH T101, B77-VH S105, B77-VH L108 and B77-VH W114 in the B77 heavy chain (S4 Table). The OTi-VP2 T71 and OTi-VP2 S72 side chains form hydrogen bond contacts with side chain atoms of B77-VH T101 and B77-VH W114 ( Fig 3A and S4 Table). The side chains of OTi-VP2 T190 and OTi-VP2 Q196 form hydrogen bond contacts with the side chain atoms of B77-VH M56 and B77-VH S105 (Fig 3A and S4 Table). In addition, OTi-VP2 H65 and OTi-VP2 D68 in βB and the BCloop interact with residues B77-VH W93 and B77-VH S95 in the light chain of B77 (S4 Table). The side chain of OTi-VP2 H65 forms hydrogen bonds or van der Waals contacts with the side chain atoms of B77-VL S95 (Fig 3A and S4 Table). To further confirm the epitope, neutralization escape mutants were screened under B77 immune pressure. The B77-resistant mutants were isolated and characterized. An Asn!Ser mutation occurred at position 190 of VP2; this position is located in the virus-antibody binding interface and is strictly conserved among serotype O (S4 Table and Fig 3C).
F145 interacts with the VP2 BC-loop ( OTi-VP2 T71 and OTi-VP2 S72) and the HI-loop ( OTi-VP2 V189, OTi-VP2 N190, OTi-VP2 T191) (S5 Table). The side chains of OTi-VP2 S72 and OTi-VP2 T191 form hydrogen bond contacts with the side chain atoms of F145-VH S119 and F145-VH T128( Fig  3B and S5 Table). The amide nitrogen of OTi-VP2 N190 provides additional stabilization via hydrogen bonds with the hydroxyl group of F145-VH D129 (Fig 3B and S5 Table). To further confirm the epitope, neutralization escape mutants were generated under F145 immune pressure. The F145-resistant mutants were isolated and characterized. A Ser!Ala mutation was found at position 72 in VP2, which is also located in the virus-antibody binding interface. FMDV employs heparin sulfate (HS) and integrin (generally avβ6) as receptors [24,25]. In the cellular entry of FMDV-O, HS acts as a primary receptor that allows for rapid cell binding, thereby providing more time for virus binding to integrins [26]. FMDV binding to the integrin receptor is facilitated by a conserved arginine-glycine-aspartic (RGD) motif in the exposed GH-loop of VP1 [25]. Structure comparisons of FMDV-integrin and FMDV-NAbs show obvious clashes between NAbs (B77 and F145) and the integrin receptor, suggesting that FMDV neutralization by B77 and F145 is facilitated by blocking virus-receptor interaction via steric hindrance (S7 Fig). Meanwhile, in vivo the antibodies are IgG molecules or some other multi-site entity, a situation that could result in aggregation of virus particles if one virus particle becomes crosslinked with another. Aggregation of virus particles can cause a reduction in the number of infectious virions available to interact with receptor and may result in the neutralization.

Determinant for the serotype O-specificity of F145 and B77
F145 and B77 neutralized FMDV-OTi but did not have a neutralizing effect against FMDV-AWH ( Fig 1C). The structural alignment of FMDV-OTi and FMDV-AWH capsid proteins showed no substantial conformational changes in the VP2 BC-loop and HI-loop, which are the epitopes of B77 and F145 (S4 Fig). The sequence alignment of FMDV-OTi and FMDV-AWH exhibited distinct amino acid differences on the epitopes of B77 and F145, indicating that the variation of critical amino acid residues may be the determinant for serotypespecific recognition ( Fig 3C).
For B77, we identified that the neutralization escape mutation OTi-VP2 N190! AWH-VP2 S190 enables the escape of antibody-mediated neutralization (virus neutralization [VN] titer >400 μg/ml) ( Fig 3C and S8 Table). For F145, we mutated FMDV-OTi amino acid by FMDV-AWH's sequence at the corresponding positions (S72D, N190S and T191S), rescued virions by reverse genetics approach and evaluated the neutralization capacity using virus neutralization test (VNT). All mutated viruses could be rescued with an acceptable attenuated proliferation and cytopathic effect (CPE) (Fig 3D and 3E). The VNT results showed that FMDV-OTi with the OTi-VP2 N190S mutation exhibited a 2.9-fold reduction in VN titer, and the OTi-VP2 S72D mutation resulted in a significant reduction (~30-fold) in VN titer (Fig 3F). These results suggest that VP2 S72 and N190 play the most essential roles in the serotype Ospecific neutralization activity of F145 and B77.
Notably, the R50 footprint on the FMDV capsid does not overlap with identified receptor binding sites (S7 and S8 Figs), indicating that R50 may not directly compete with the binding of receptor. Previous studies have reported that A9-Fab destabilized EV71 virions and as much as 70% of EV71 F-particles had been destroyed when incubated with A9-Fab at 37˚C [27]. When we incubated FMDV-OTi or FMDV-AWH with R50 at a molar ratio of 1:240 for 30 s or 60 s at 4˚C, the negative stain electron micrographs showed that R50 destabilized the virus particle and induced dissociation of virus particles into capsid pentamers, and very few particles remained intact after 60 s (S9 Fig). Meanwhile, the cryo-electron microscopy (cryo-EM) micrographs also showed significantly lower particle density when incubated for 30s (S1 Fig). In contrast to this remarkable effect, B77 did not induce significant decrease of particle density even after one hour incubation at 4˚C (S9 Fig). These results indicated a possible neutralization mechanism of R50 by promoting the dissociation of FMDV virions before infection.

Discussion
Foot-and-mouth disease virus (FMDV) causes a highly contagious disease in bovines and other cloven-hoofed animals. Difficulties facing the prevention and control of FMDV include a high morbidity, a complex host range, and broad antigenic diversity [28]. Control of the disease in endemic regions has been mainly based on large-scale vaccinations. It is known that vaccination with one serotype cannot confer adequate protection against strains of different serotypes or heterologous strains of the same serotype based on antigenic variation; furthermore, the protection response is not immediate [29]. It is important to develop universal protective vaccines against the seven FMDV serotypes. However, the cross-serotype effectiveness of the available FMDV vaccine is limited. A previous work constructed chimeric virus-like particles displaying tandem repeats of B cell epitopes (VP1 residues 134-161 and 200-213) derived from FMDV serotypes O and A using a truncated HBc carrier; the chimeric VLPs elicited moderate neutralization antibodies against FMDV serotypes O and A in mice [30].
At present, five functionally independent neutralizing antigen sites recognized by murine monoclonal antibodies (mAbs) have been identified on the capsid surface of FMDV serotype O by neutralization escape mutants [31][32][33][34]. Site 1 is linear and trypsin-sensitive; it encompasses the GH-loop and the carboxy terminus of VP1, with critical amino acid substitutions at residues 144, 148, 154 and 208. However, sites 2-5 are all conformational and trypsin-resistant. Site 2 (isolated after pressure from C6 or C9 NAbs) is defined in the BC-loop and EF-loop of VP2 and involves critical amino acid substitutions at residues 70-73, 75, 77, and 131. Site 3 (isolated after pressure from C8 NAb) is defined in the BC-loop of VP1, with critical residues at positions 43 and 44. Critical residues at position 58 in the VP3 B-B "knob" contribute to site 4 (isolated after pressure from 14EH9 NAb). Site 5 (isolated after pressure from C3 NAb) involves an amino acid change at position 149 in the GH-loop of VP1, which is distinct from site 1 even though part of the GH-loop is encompassed. Murine antibodies usually have a short HCDR3 and generally flat binding surfaces. Antigenic sites recognized by murine monoclonal antibodies are all surface-exposed and intra-protomer sites (S10 Fig). The structures of FMDV-OTi in complex with serotype O-specific NAbs F145 and B77 show that the binding sites (BC-loop and HI-loop of VP2) are also surface-exposed and intra-protomer sites. For F145, neutralization escape mutation and reverse genetics reveal that VP2 S72 within antigenic site 2 plays the most essential roles in the serotype O-specific neutralization activity (Fig 6A). For B77, the key epitope residue on VP2 N190 is the major determinant for serotype O-specific recognition. VP2-190 around the threefold axis is a new neutralizing epitope, which is close to antigenic site 2( Fig 6A).
Bovine antibodies with ultralong HCDR3 structures have the advantage to bind to epitopes that are not easily accessible to common antibodies with flatter paratope features. The structures of FMDV-OTi and FMDV-AWH with a cross-serotype NAb R50 exhibit a novel neutralizing antigenic site located in a shallow depression on the capsid, contacting two major capsid proteins VP1 and VP3 from two adjacent protomers. The novel neutralizing antigenic site formed by OTi-VP1 V50, OTi-VP1 D52, OTi-VP1 P94, OTi-VP1 E95, OTi-VP1 L159(or AWH-VP1 L157), OTi-VP1 P160 (or AWH-VP1 P158) and OTi-VP3 D173 (or AWH-VP3 D174) is highly conserved among FMDV serotypes O and A (Fig 6B and 6C). Meanwhile, NAb R50 targeting the novel antigenic site can induce a broad-spectrum cross-neutralizing response against diverse isolates of serotype O and A. Therefore, this could be a target for development of universal vaccine.

Ethics statement
All animal experiments in the study were approved by the Review Board of Lanzhou Veterinary Research Institute, Chinese Academy of Agricultural Sciences (permit no. LVRIAEC

Virus production and purification
FMDV (O/Tibet/99) or FMDV (A/WH/CHA/09) was grown in baby hamster kidney BHK-21 cells. Inactivated O/Tibet/99 and A/WH/CHA/09 whole virus antigens were kindly provided by an FMD inactivated vaccine manufacturer. For further purification, virus antigens in the supernatant were precipitated by incubating at 4˚C overnight in 8% (w/v) PEG 6,000. The precipitated virus antigens were harvested by centrifugation at 3,500 g for 1 h at 4˚C, followed by resuspension in pH 7.4 PBS (phosphate-buffered saline) (137 mM NaCl, 2.7 mM KCl, 50 mM Na 2 HPO 4 and 10 mM KH 2 PO 4 ). The viruses were purified through a sucrose cushion that transferred the samples to an ultracentrifuge tube. A 10 ml aliquot of 30% (w/v) sucrose was gently added to the bottom of the suspension, and the sample was centrifuged at 35,000 g for 1.5 h. The sucrose was removed from the pellet, and 500 μl of PBS was added to cover the pellet. The supernatant was purified over a 20-60% sucrose gradient and fractionated by centrifugation at 35,000 g for 4 h at 4˚C. The fractions were analyzed by negative stain electron microscopy, and the virus fraction was transferred to a 100 kDa MWCO centrifugal filter for buffer exchange with PBS to remove the sucrose.

Bovine infection
One 1-year-old healthy Qinchuan bovine (Bos Taurus), a Chinese breed of beef bovine, was raised in an animal biosafety level 3 laboratory for the sorting of antigen-specific antibodysecreting B cells after FMDV infection. The bovine, designated as #2334, was first challenged subcutaneously at two sites on the tongue with 10,000 BID 50 (50% bovine infective dose) of bovine-adapted O/Mya/98 and then received a boost vaccination with O/HN/CHA/93 and O/ Tibet/99 respectively on days 35 and 132, as described in our previous report [20]. After day 225, the animal was further challenged subcutaneously at two sites on the tongue with 10,000 BID 50 (50% bovine infective dose) of FMDV (A/WH/CHA/09). Serum samples were collected at days 114 after this infection, and heparinized peripheral blood was taken from the jugular veins of the bovine for PBMC isolation.

Single-Cell sorting of FMDV serotype O and serotype A cross-neutralizing plasmablasts from bovine PBMCs using flow cytometry
The bovine PBMCs stained as above were sorted by flow cytometry (BD FACS Aria II, USA) using a 100 μm nozzle. Single-cell mode was set to sort the FMDV CD21 + IgM -O -FMDV + A -FMDV + cells using flow rate of 10,000 events/s. The subsequent single-cell lysis and cDNA synthesis procedures were performed in accordance with our previous report [20]. The obtained cDNA templates were stored at −20˚C for subsequent PCR amplification.

Cloning of scFv gene and construction of pcDNA3.4-scFv expression vector
Amplification of the antibody variable region genes (VH and VL) was performed by nested PCR using primers for bovine IgG gamma and lambda chains (Ig γ and Ig λ; S1 Table) [20]. Because the λ chain accounts for 95% of the bovine light chains, only the λ light chain was amplified by PCR. The final PCR products of paired VH and VL were Sanger sequenced. Single-chain fragment variable (scFv) was designed by splicing the VH and VL genes using a flexible linker (GGGGSGGGGSGGGGS). The N-and C-termini of the scFv included a signal peptide (MNPLWTLLFVLSAPRGVLS) of bovine Ig γ and a His tag, respectively. The scFv genes were synthesized by GenScript Inc. (www.genscript.com) with codon optimization and were expressed in CHO cells, followed by cloning into the pcDNA3.4 vector. The cloned scFvexpressing plasmids were amplified in E. coli and extracted and purified using an Endo Free Maxi plasmid kit (TIANGEN, China).

Expression and purification of NAb scFv
Single-chain fragment variable (scFv) antibodies were expressed in ExpiCHO-S cells (Invitrogen, USA) following the manufacturer's instructions. Briefly, the CHO cells were routinely cultured at 37˚C with �80% relative humidity and 8% CO2 in 125 mL flasks that were fixed in an orbital shaker platform at a speed of 125 rpm with a 25 mm shaking diameter. After the cell density reached 6 × 10 6 cells/ml, the cells were collected and resuspended in fresh media for transfection. The pcDNA3.4-scFv plasmid was transfected into ExpiCHO-S cells with an Expi-Fectamine CHO transfection kit (Invitrogen, USA). An enhancer and feed component were added at 18 h post transfection according to a standard protocol. ScFv-containing supernatants were harvested 10 days after transfection. The expressed scFv was initially purified with a HisTrap excel column. The obtained eluate was concentrated using a 100 kDa ultrafiltration tube and then further purified by size exclusion chromatography using a Superdex-200 increase 10/300 column and an AKTA plus protein purification system (GE Life Sciences, USA). The purity and size of each single-chain fragment variable (scFv) antibody was assessed by SDS-PAGE. The concentrations of the final obtained scFvs were determined by measuring their corresponding absorption values at a wavelength of 280 nm.

Binding affinity measurement of R50 scFv
Baby hamster kidney-21 (BHK-21) cells in 24-well plate were respectively infected with different FMDV strains at a multiplicity of infection (MOI) of 0.01 and incubated for 4 h at 37˚C. Subsequently, these infected cells were fixed and permeated for 20 min using a fixation and permeabilization solution (BD Biosciences, USA). After washing with PBS buffer, 1 × 10 6 infected cells/sample was stained with diluted R50-ScFv and subsequently incubated with anti-HIS FITC antibody (Genescript, China) and 3B4B1-Alexa647 (Fluorescence-labeled antibody (clone:3B4B1) against the nonstructural protein 3B of FMDV) for 30 min at room temperature (20-25˚C). After two washes, the stained cells were loaded for FC (BD LSRFortessa, USA), and 10000 Alexa647 positive cells were recorded. The percent of FITC positive cells in Alexa647 positive cells was used to assess the binding activity of R50 scFv.

Virus neutralizing test
The NAb scFvs were titrated for viral neutralizing activity against three representative strains from the three topotypes of FMDV serotype O and the Asia topotype of FMDV serotype A by using a microneutralization assay as previously described [21]. The five strains of FMDV were O/Mya/98 (SEA topotype), O/HN/CHA/93 (Cathay topotype), O/Tibet/99 (ME-SA topotype), A/GDMM/CHA/2013 (genotype 2 in SEA-97 lineage of Asia topotype), and A/WH/CHA/09 (genotype 1 in SEA-97 lineage of Asia topotype). Briefly, scFv antibody samples were 2-fold serially diluted in 96-well cell culture plates in a total volume of 50 μl, and 100 TCID50 of FMDV in 50 μl of culture media was added to each well. After incubation for 1 h at 37˚C, 5×10 4 BHK-21 cells in 100 μl of media were added to each well as an indicator of residual infectivity. Normal cell wells, and 10, 100, and 1,000 TCID50 virus control wells were used in each plate. The plates were incubated at 37˚C with 5% CO2 for 48 h and then fixed in acetonemethanol (volume ratio = 1:1) and stained with a 0.2% crystal violet solution. The endpoint titer was determined as the reciprocal of the lowest antibody dilution to fully prevent a cytopathic effect (CPE) by 100 TCID 50 FMDV in each well. The neutralizing activity was expressed as the virus neutralization (VN) titer, which was calculated as the initial antibody concentration divided by the endpoint titer.

Rescue and titration of mutated viruses
Full-length cDNAs were generated by the exchange-cassette strategy to replace the whole P1 coding region of an existing pOFS plasmid with the respective genes of O/Tibet/99 and A/ WH/CHA/09. Site-directed mutagenesis was used to introduce nucleic acid mutations to produce full-length cDNAs with single amino acid substitutions [35]. All mutant constructs were confirmed by nucleotide sequencing. The mutant viruses were rescued as previously described [36]. Briefly, Not I-linearized mutant plasmids were transfected into BSR/T7 cells using Lipofectamine 2000 following the manufacturer's instructions. The transfected cells were monitored daily for the appearance of CPE. At 72 h post transfection, culture supernatants were harvested and passaged on BHK-21 cells. The mutant virus titers were measured using the plaque forming unit (PFU) assay [37]. Confluent monolayers of BHK-21 cells were infected with 10-fold serially diluted FMDV samples in 6-well plates. The 0.6% Tragacanth gum was added after 1 h of incubation. Plaques were visualized at 48 h postinfection (p.i.) by fixing with acetone-methanol and staining with crystal violet. The amount of plaque was observed and statistically analyzed.

Generation of neutralization escape mutants
Neutralization escape mutants were generated through consecutive passages of FMDV in BHK-21 cells under the selective pressure of neutralizing scFv antibodies, according to a previous report with minor modifications [31]. Briefly, tenfold serial dilutions of FMDV (O/Tibet/ 99) in 50 μl were incubated with 50 μl of varying concentrations of scFv sample (20 μg/ml to 50 μg/ml) in 96-well microplates. The mixtures were used to infect 100 μl of BHK-21 cells (10 6 cells/ml), which were incubated at 37˚C for 48 h to allow virus propagation to occur. The first passage virus was harvested from the wells seeded with the highest dilution of virus that produced an approximately 80-100% cytopathic effect (CPE). Subsequent rounds of pressure selection were performed in 24-well plates in which the passaged virus (200 μl) was incubated with an equal volume of twofold concentration of antibodies in each well containing 400 μl of BHK-21 cells. The harvested virus was subjected to several additional rounds of selection until it completely escaped from neutralization upon the addition of scFv at concentrations greater than 1 mg/ml. The P1 region of the obtained neutralization escape mutants was amplified by one-step RT-PCR as described previously [38] using the primers Pan204+: ACCTCCAACGGGTGGTACGC/NK61: GACATGTCCTCTTGCATCTG) and were subsequently verified by sequencing. Mutated amino acids were determined by aligning the entire mutant P1 region to its parent virus.

Cryo-EM sample preparation and data collection
FMDV and antibodies were incubated at a molar ratio of 1:240 in a volume of 50 μl for 30 min (B77), 10 min (F145) and 30 s (R50) at 4˚C. A 3 μl aliquot of the mixture was applied to a glow-discharged carbon-coated gold grid (GIG, Au 1.2/1.3, 200 mesh; Jiangsu Lantuo Biotechnology Co. Ltd., China). The grid was blotted for 5 s in 100% relative humidity and plunge-frozen in liquid ethane using a Vitrobot mark IV (Thermo Fisher, USA). The cryo-EM data were collected at 200 kV with a FEI Arctica (Thermo Fisher, USA) and a direct electron detector (Falcon II, Thermo Fisher) at Tsinghua University. Micrograph images were collected as movies (19 frames, 1.2 s) and recorded at −2.4 to −1.4 μm underfocus at a calibrated magnification of ×110 kX, resulting in a pixel size of 0.932 Å per pixel. The data collection and refinement statistics are summarized in S2 Table. Image processing and three-dimensional reconstruction Similar image processing procedures were employed for all data sets. Individual frames from each micrograph movie were aligned and averaged using MotionCor2 [39] to produce driftcorrected images. Particles were picked and selected in Relion-2.1 [40], and contrast transfer function (CTF) parameters were estimated using CTFFIND4 [41] and integrated in Relion-2.1. Subsequent steps in three-dimensional (3D) reconstruction used Relion-2.1 in accordance with recommended gold-standard refinement procedures [40]. The final selected particles were further processed in THUNDER [42], which is a particle filter-based cryo-EM image processing software. The local defocus of each particle was determined by refining the defocus parameter during expectation maximization in THUNDER. For all reconstructions, the final resolution was assessed using the standard FSC = 0.143 criterion.

Model building and refinement
The X-ray structure of native FMDV O1BFS (PDB:1BBT) [43] was manually placed into the cryo-EM map for FMDV particles and rigid-body fitted with UCSF Chimera [44]. The Xray structure of native BOV-7 (PDB: 6E9U) [23] was manually placed into the cryo-EM map for scFv and rigid-body fitted with UCSF Chimera [44]. The fitting was further improved with real-space refinement using Phenix [45]. Manual model building was performed using Coot [46] in combination with real-space refinement with Phenix [45] to adjust the mismatches between the model and the target protein. The density maps were kept constant during the entire fitting process, and the atomic coordinates were subjected to refinement. The additional structures reported in this work were built and refined by using FMDV (O/Tibet/99) particles as a starting model and rigid-body fitted and refined. Validation was conducted using the MolProbity function integrated within Phenix. The refinement statistics are presented in S2 Table. Supporting information S1 Fig. Cryo-EM analysis of virus-NAb complex. Typical electron micrographs were collected with a defocus of 2.0 μm (FMDV-OTi-B77), 1.9 μm (FMDV-OTi-F145), 1.5 μm (FMDV-OTi-R50) and 1.7 μm (FMDV-AWH-R50) (Scale bar, 1000 Å). Selected 2D class averages both show prominent spikes on the outer surface of viral particles (Scale bar, 480 Å). Fourier shell correlation (FSC) of the final 3D reconstruction after gold-standard refinement using RELION and THUNDER. The resolution corresponding to an FSC of 0.143 is shown for these virus-antibody complexes. FSC curves are plotted before (gray) and after (yellow) masking in addition to post correction (orange), accounting for the effect of the mask using phase randomization.