A structural basis for antibody-mediated neutralization of Nipah virus reveals a site of vulnerability at the fusion glycoprotein apex

Significance Despite causing regular outbreaks with high case fatality rates, there are currently no licensed vaccines or therapeutics for Nipah virus (NiV) infection. Here, we sought to determine the molecular basis for how the antibody response neutralizes NiV by targeting the surface-displayed fusion glycoprotein, NiV-F. Our structural study reveals a neutralizing antibody epitope at the membrane-distal portion of the prefusion trimeric NiV-F, which is well conserved across known NiV strains. Further structure–function analyses demonstrate that additional antibodies bind this region of the NiV-F apex, suggesting that this is an immunologically accessible site of vulnerability. This work reveals the membrane-distal regions of NiV-F and the F glycoprotein from closely related Hendra virus as attractive targets for antiviral and vaccine development.

. Summary of monoclonal and polyclonal antibody features. *In blue are codon optimized constructs. 1 Binding was quantified by flow cytometry, as described in the Materials and Methods.

Polyclonal pAb835
NiV-F, G and M Anti-F pAb1187 Soluble NiV-G Anti-G pAb2489 NiV-F, G, and M Anti-F and Anti-G pAb2490 NiV-F, G, and M Anti-F and Anti-G Supplementary Table S2. Crystallographic data collection and refinement statistics.
a Values for the outer resolution shell are shown in parentheses. b Rfree is calculated as for Rwork, using only 5% of the data separated prior to refinement. c Ramachandran analysis determined with the Molprobity server (1). d RMS deviations: root mean square deviation from ideal geometry.   the DOG software (6). The Fab66-bound structure was organized into three domains, DI, DII, and DIII, followed by HRB linker and HRB. The residues comprising each domain were determined based off of the domain organization of the prefusion PIV5 fusion protein, as described previously (7). The cathepsin-L cleavage site R109L110 is noted with a black arrow. The signal sequence precedes DI and the GCNt trimerization motif follows HRB (not shown).   forms a backbone hydrogen bond with residue Ser66 on NiV-F, and also heavily contacts Ser153 and Asn155. The sidechain of CDR L2 Tyr50 forms hydrogen bonds with residues Ser66, Ser69, and Gln70, and also contacts Asn67. (B) Heavy chain CDR H2 and H3 residues contribute to the interface. CDR H3 residue Ser98 forms sidechain and backbone contacts with NiV-F residues Gln70 and Glu196, while hydrophobic Trp100 contacts the sidechain and backbone of NiV-F residue Ser69. CDR H2 residues Thr52A and Asn53 hydrogen bond with NiV-F residue Ser74.

NiV-F Fab66
Tyr52 and Thr56 from CDR H2 interact with additional residues Gly73 and Glu77 on NiV-F.   Compared to wt NiV-F, none of the mutant NiV-F or HeV-F constructs showed any statistically significant differential binding to either pAb2489 or pAb2490 (ordinary one-way ANOVA with Dunnett's correction for multiple comparisons). Note, these pAb sera were generated using codonoptimized NiV-F, -G and-M (Table S1) and contain variably cross-reactive antibodies against both HNV-F and G glycoproteins. The polyclonal response is spread over both F and G, which likely dilutes immune focusing on any particular region of F and G. In any case, the data show that these antibodies are valid for use in normalizing the binding of other Abs that may be affected by the indicated HNV-F mutations. In contrast, pAb835 was generated with codon-optimized NiV-F, but wt NiV-G and NiV-M. Our wt NiV-G is expressed at much lower levels from the pcDNA3 vector we were using at the time. Thus, our immunization scheme that elicited pAb835 resulted in an anti-NiV-F-specific polyclonal response ( Table S1 and last subsection of results section) that is likely focused on the immunologically accessible regions posited in the model presented in this study.   S13. The Fab66 epitope is highly conserved among NiV virus strains. Available NiV-F protein sequences were aligned using Multalin (4) and plotted using ESPript (5). The Fab66 epitope (calculated by the PDBePISA server (10)) was mapped onto the alignment; residues contacted by the light chain are noted by a green box below the sequence, the residues contacted by the heavy chain with a pink box, and those contacted by both chains with a grey box.

D E F
Comparison of ω estimates among buried vs exposed residues of the NiV fusion protein Comparison of ω estimates among buried vs exposed residues of the MeV fusion protein Comparison of ω estimates among buried vs exposed residues of the PIV5 fusion protein (A) Representative histogram plots of 1:1,000 dilution of pAb835 binding to HNV-F or -G glycoproteins. Binding was performed as described in the methods and analyzed as described in three biological replicates (as described in Fig. 3D). ND for mAb36 denotes 'not determined' due to ambiguous or no fit (representing no neutralization for HNVpp bearing the indicated HNV-F proteins, Fig. 6A and B). Statistical significance was determined by one sample t-Test (*, p ≤0.05; **, p ≤ 0.01; ***, p ≤ 0.001). replicates (14). Partitions were derived from the global alignments, corresponding to distinct protein domains (Group 1-4, as defined in the main text). Global and site-specific ω values were estimated under M0/M1a/M2a and the M8/M8a site models using CODEML in PAML (15), and confirmed using the FUBAR method (16). The statistical significance of nested models was evaluated using a likelihood ratio test (LRT), using the χ2 approximation to the likelihood ratio.
Only positively selected sites (PSS) scored under Bayes Empirical Bayes (BEB), or/under a posterior probability above 0.9 were considered (15,16 (17), in which ω at each branch and site of a given partition, p = 1.4, is modelled from a 2-bin general discrete distribution, with 0 ≤ ω1(p) ≤ 1 ≤ ω2(p), and a probability of (ω = ω1 (p)) = f(p). Consequently, ω can vary both across sites and branches using random effects models, while the effect of the partition is fixed. Because viral sequences isolated from individual hosts may harbour many neutral or even deleterious mutations (at the population level) due to within-host evolution dynamics (18,19), the evolution along terminal branches of the tree is modelled with separate (nuisance) distributions, and the focus of interference are the parameters governing evolution along internal tree branches. These distribution parameters, together with other model parameters (branch lengths, nucleotide substitution biases, nuisance terminal branch parameters) are estimated jointly from all partitions using maximum likelihood. We further assumed that branch lengths are shared by all partitions up to a per-partition scaling factor (the relative ratio model of evolution = f(1), and assessing significance using the likelihood ratio test. P-values are derived from the 2 distribution with 9 degrees of freedom.

Supplementary Results for dN/dS (ω) analyses.
No evidence for positive selection (ω >1) upon the complete coding region or per functional group within the F protein, including the apex of DIII and the Fab66 binding epitope, was detected for any of the paramyxoviruses analyzed ( Fig. S14 and S15). No positively selected sites were detected under the M2a and M8 site models, as were significantly scored under BEB (Table S3). For the MeV-F domain analysis, no significant differences were observed amongst the four different functional groups compared to a global ω estimate, and no significant differences were observed between the ω estimates for buried vs.
solvent exposed residues ( Fig. S14B and E, Table S3). A similar pattern was observed for NiV-F, with no significant differences amongst the four different group regions, and across the ω estimates for buried vs. solvent exposed residues ( Fig. S14A and D, Table S3). For the PIV5 fusion protein, the ω estimate for Group 1 was marginally lower compared to all other regions, whereas the ω estimate for Group 2 were marginally higher compared to all other regions, and no differences were observed between the ω estimates for buried vs. exposed residues ( Fig. S14C and 0.00000. FUBAR analysis also revealed a similar pattern at the site level (Fig. S15). A mixed effects model analysis revealed a strong and consistent degree of conservation across all codon groups for all virus datasets analyzed, confirming our previous observations. For NiV, an ω of 0 or near zero was obtained along internal branches in viral phylogeny, with no evidence for episodic positive selection anywhere in the tree ( Table S4). The evolutionary patterns observed for PIV and MeV, with Group 1 and 4 (in PIV) and Group 1 (in MeV) yielded non-zero estimates for a fraction of branches and sites evolving with an ω >1. However, there was insufficient statistical support for positive selection acting upon any of the partitions tested. The