Potent neutralization of Marburg virus by a vaccine-elicited antibody

Abstract

Marburg virus (MARV) is a filovirus that causes severe and often lethal haemorrhagic fever^1,2. Despite the increasing frequency of MARV outbreaks, no vaccines or therapeutics are licensed for use in humans. Here we designed mutations that improve the expression, thermostability and immunogenicity of the prefusion MARV glycoprotein (GP) ectodomain trimer, which is the sole target of neutralizing antibodies and vaccines in development^3-8. We discovered a fully human, pan-marburgvirus monoclonal antibody, MARV16, that broadly neutralizes all MARV isolates, Ravn virus and Dehong virus with 40-100-fold increased potency relative to previously described antibodies⁹. Moreover, MARV16 provided therapeutic protection in guinea pigs challenged with MARV. We determined a cryogenic electron microscopy structure of MARV16-bound MARV GP. The structure shows that MARV16 recognizes a prefusion-specific epitope spanning GP1 and GP2, which blocks receptor binding and prevents conformational changes required for viral entry. We further determined the architecture of the MARV GP glycan cap, which shields the receptor-binding site, and identified architectural similarities with distantly related filovirus GPs. MARV16 and previously identified antibodies directed against the receptor-binding site^9-11 simultaneously bound MARV GP. These antibody cocktails required multiple mutations to escape neutralization by both antibodies, a result that paves the way for the development of MARV therapeutics resistant to viral evolution. MARV GP stabilization along with the discovery of MARV16 advance prevention and treatment options for MARV disease.

Fig. 1. Design of a prefusion-stabilized MARV GP.

a, Schematic of MARV GP domain organization and mutations included in MARV GPΔMuc ectodomain constructs. CD, cytoplasmic domain; FCS, furin cleavage site; GC, glycan cap; MLD, mucin-like domain; TM, transmembrane domain. SP, signal peptide. b, Negative-stain electron micrographs of MARV GPΔMuc constructs. Scale bar, 50 nm. Five micrographs were collected for each of the four biological replicates of each ectodomain. c, MARV GPΔMuc ectodomains binding to immobilized MR191 IgG assessed by BLI. d, Ribbon diagram of the RAVV GP (Protein Data Bank (PDB) identifier: 6BP2) highlighting the residues mutated in GP2 HR1_c. GP1, purple; GP2, gold; N-linked glycans, light blue surfaces. e–g, Recombinant production yields (e), size-exclusion chromatograms (f) and differential scanning fluorimetry analysis (g) of MARV GPΔMuc_WT and MARV GPΔMuc mutants purified from Expi293 cells. The fold increase in yield relative to MARV GPΔMuc_WT is displayed above the plot (e). The bar represents the mean yield ± s.d. across four biological replicates. mAU, milli-absorbance units. Data in g are shown as the first negative derivative of the fluorescence intensity (dF) with respect to temperature (dT), and the T_m of MARV GPΔMuc_WT and MARV GPΔMuc_PV are indicated with dotted vertical lines. Data presented in c, f and g are from one biological replicate and representative of three other biological replicates. Six technical replicates were conducted and averaged for each biological replicate in g. h, Scheme of the immunogenicity study design of MARV GPΔMuc ectodomains in BALB/c mice (n = 10 per group; bleed 2 serum could not be collected for 1 animal in the 0.1 µg MARV GPΔMuc_WT group). i,j, MARV GPΔMuc_PV serum binding titre (i) and neutralizing titre against VSV pseudotyped with the MARV/Musoke GP (j) respectively, presented as average ED₅₀ (half-maximal effective dilution) or ID₅₀ (half-maximal inhibitory dilution) values obtained from two biological replicates conducted in technical duplicate and using distinct batches of protein or pseudovirus. The limit of detection (ED₅₀ of 33 or ID₅₀ of 10) is indicated with a dotted line. Black lines indicate the geometric mean titre. Statistics were assessed using the Kruskal–Wallis test with Dunn’s post-test comparing groups receiving identical doses of MARV GPΔMuc_WT or MARV GPΔMuc_PV.

Fig. 2. Discovery of a pan-marburgvirus antibody.

a, Schematic of the immunization schedule used to discover MARV GP-directed monoclonal antibodies (n = 4 mice). b, Dose–response neutralization curves for MARV16, MR78 and MR191 against VSV pseudotyped with the MARV/Musoke GP. Data are the mean ± s.e.m. from three technical replicates. Data are representative of 3–5 additional biological replicates. c, Neutralization potency of MARV16, MR78, MR191 and 4C2, a MERS-CoV antibody, against authentic MARV/Musoke. Data points reflect PRNT₅₀ values obtained from two biological replicates using distinct batches of IgGs. The black line indicates the mean PRNT₅₀ value. d, Binding affinity of the MARV16 Fab to immobilized MARV GPΔMuc_WT measured using BLI. e,f, MARV16, MR78 and MR191 IgG (e) or Fab (f) binding to immobilized MARV GPΔMuc_WT assessed by BLI. g, MARV16 IgG binding to immobilized MARV GPΔMuc_WT at variable pH values using BLI. h, Schematic highlighting GP mutations (black vertical lines) in MARV variants relative to MARV/Musoke. Residue numbers correspond to the MARV/Musoke GP. i, Neutralization potency of MARV16, MR78 and MR191 against VSV pseudotyped with the indicated MARV GP. j, Phylogenetic tree constructed using the amino acid sequences of related filovirus GPs with sequence identity relative to the MARV/Musoke GP shown to the right. k, Neutralization potency of MARV16, MR78, MR191 and EBOV-515 against VSV pseudotyped with the indicated filovirus GP. Data presented in i and k are averaged IC₅₀ values obtained from at least two biological replicates conducted in technical triplicate using distinct batches of IgG and pseudoviruses. l–o, MARV GPΔMuc_WT, EBOV GPΔMuc, SUDV GPΔMuc, thermolysin-cleaved EBOV GPΔMuc or thermolysin-cleaved SUDV GPΔMuc binding to immobilized EBOV-515 (l), MR191 (m), MR78 (n) or MARV16 (o) IgGs assessed with BLI. Data presented in d–g and l–o reflect one biological replicate and are representative of two biological replicates using distinct batches of proteins.

Fig. 3. Molecular basis of MARV16 neutralization.

a, Ribbon diagram of the cryo-EM structure of the MARV GPΔMuc_WT ectodomain in complex with three MARV16 Fab fragments. Only the Fab variable domains were modelled into the density. MARV GP1 and GP2 are shown in shades of purple and gold, respectively, MARV16 VH is shown in green and MARV16 VL is shown in light green. N-linked glycans are rendered as light blue surfaces. b, Ribbon diagram of a single MARV GP protomer in complex with one MARV16 Fab. c–e, Zoomed-in views of interactions between MARV16 CDRH1–CDRH3 and MARV GP2 (c), MARV16 CDRL3 and MARV GP2 (d) and MARV16 CDRH2 and MARV GP1 (e). Selected hydrogen bonds and salt bridges are denoted with black dashed lines. f, Binding modes of the NPC1 receptor (yellow) and MARV16 (green/light green) to MARV GP. The position of NPC1 was determined by superimposing the EBOV GP–NPC1 (PBD: 5JNX) and MARV GP–MARV16 Fab structures. The EBOV GP trimer and the region of the MARV GP glycan cap resolved in our structure (residues 191–219) are not shown for clarity. The red star denotes steric clashes. g, Ribbon diagrams of MARV GP1 and EBOV GP1 (PDB: 3CSY). The MARV GP1 core and glycan cap are shown in dark and light purple, respectively. The EBOV GP1 core and glycan cap are shown in dark and light blue, respectively. Hydrogen bonds between the glycan cap and core are indicated with black dashed lines, and glycan cap aromatic residues inserted in the RBS are shown in stick representation and labelled. The position of the experimentally determined RBS for the EBOV GP and that of the putative RBS for the MARV GP are indicated with red dashed circles.

Fig. 4. MARV16 protects guinea pigs against MVD.

a, Schematic of the MARV challenge study assessing the therapeutic efficacy of MARV16. i.p., intraperitoneal. b, Survival curves for guinea pigs (n = 6 per group) challenged with 1,000 PFU of MARV/Angola and treated with MARV16 or an isotype control antibody. Animals were monitored for 29 d.p.i. Survival curves for the groups administered with MARV16 1 d.p.i. and 4 d.p.i. excluding the animals with undetectable plasma MARV16 concentration 1 day after treatment are shown as boxes with dashed lines. Statistical differences in survival between groups were assessed using Kaplan–Meier survival analysis and excluding animals with undetectable plasma concentrations of MARV16. A two-sided log-rank test was used and Holm–Šídák correction was applied for multiple comparisons between the isotype control antibody-treated guinea pigs and MARV16-treated groups. c, MARV/Angola viral loads measured in the plasma of infected guinea pigs at 5 d.p.i. The black line indicates the geometric mean viral load for each group and the dotted line denotes the limit of detection (viral load of ≤2 × 10³ GE per ml). Statistical differences in viral loads between groups were assessed using the Kruskal–Wallis test with Dunn’s post-test comparing the isotype control antibody-treated guinea pigs to the MARV16-treated groups (excluding animals with undetectable plasma concentrations of MARV16). d–f, Daily body weights (d), body temperatures (e) and clinical scores (f) for guinea pigs for the duration of the challenge study. Animals with undetectable plasma concentrations of MARV16 1 day after treatment are denoted with triangles, which indicates that in these animals, MARV16 was sequestered at the injection site and cleared or ‘soaked up’ by the challenge virus immediately before reaching the blood stream.

Fig. 5. Formulation of MARV antibody cocktails.

a, Competitive binding assay of MARV16, MR78 and MR191 IgG to the MARV16-bound MARV GPΔMuc_WT ectodomain using BLI. Data presented are from one biological replicate and are representative of data from two biological replicates using distinct batches of protein. b,c, Representative 2D class averages and 3D reconstruction of negatively stained MARV GPΔMuc_WT ectodomain bound to MR78 and MARV16 Fab fragments (b) or MR191 and MARV16 Fab fragments (c). The position of the MR78 (pink) or MR191 (blue) Fab fragments were determined by superimposing the RAVV GP–MR78 Fab (PDB: 5UQY) or RAVV GP–MR191 Fab (PDB: 6BP2) structures with our MARV GP–MARV16 Fab structure. Scale bar, 400 Å. d–f, Escape mutations identified for MARV16 alone (d), the MARV16–MR78 antibody cocktail (e) and the MARV16–MR191 antibody cocktail (f) using replication-competent VSV encoding the MARV/Musoke GP instead of the VSV G protein. Two selection experiments were performed using the separately plaque-purified VSV-MARV/Musoke GP isolates 2B2 and 2B4. The virus was passaged in the presence of increasing concentrations of antibody until observation of obvious cytopathic effects (>20% of the field of views) in the presence of 100 µg ml^–1 of antibody. Mutations were identified by deep sequencing of the viral supernatant, and those that reached a frequency of at least 10% are displayed in the plots and coloured according to the isolate they were identified from. The V547G mutation that was also identified in 2B2 passaged without antibody is displayed in grey. g, Escape mutations (red) identified during the antibody-selection experiments mapped on the surface of MARV GPΔMuc_WT (grey). The MR78 position (pink) was determined by superimposing the RAVV GP–MR78 Fab (PDB: 5UQY) structure with our structure of MARV GP in complex with the MARV16 Fab (green). C226, F447S and L448P were not resolved in our structure and are not displayed on the MARV GP.

Extended Data Fig. 1. Analysis of serum binding and neutralization titres for MARV GPΔMuc_WT- and GPΔMuc_PV-immunized mice.

a,b, Dose-response curves for the ELISAs against the MARV GPΔMuc_PV ectodomain (a) and for the neutralization assays against VSV pseudotyped with MARV/Musoke GP (b) for the sera collected from MARV GPΔMuc_WT- or GPΔMuc_PV-immunized mice. Serum was unable to be collected at bleed 2 for mouse D5120. Two biological replicates were performed for the ELISAs using distinct batches of proteins. Two technical replicates were performed per biological replicate. Data presented are from one representative biological replicate and presented as mean ± standard error from the two technical replicates. Two to six biological replicates were performed for the neutralization assays using distinct batches of antibodies and pseudoviruses. Three technical replicates were performed per biological replicate. Data from all biological replicates are shown and presented as mean ± standard error from the three technical replicates.

Extended Data Fig. 2. Analysis of binding and neutralization titres for the monoclonal antibodies discovered from ATX-Gx mice.

a, Flow cytometry gating strategy used for sorting MARV GPΔMuc_WT-reactive memory B cells. b,c, Dose-response curves for the ELISAs against the MARV GPΔMuc_WT ectodomain (b) and neutralization assays against VSV pseudotyped with MARV/Musoke GP (c) for the 10 antibodies discovered from the immunization study using the ATX-Gx mice. Two biological replicates were performed for the ELISAs using distinct batches of proteins and antibodies. Two technical replicates were performed per biological replicate. Data presented are from one representative biological replicate and presented as mean ± standard error from the two technical replicates. Two to six biological replicates were performed for the neutralization assays using distinct batches of antibodies and pseudoviruses. Three technical replicates were performed per biological replicate. Data from all biological replicates are shown and presented as mean ± standard error from the three technical replicates. d, Dose-response curves for plaque reduction neutralization tests for MARV16, MR78, MR191, and the MERS-CoV 4C2 IgG, conducted using authentic MARV/Musoke. Two biological replicates were performed with one to three technical replicates using distinct batches of monoclonal antibodies. Data are shown as the mean ± standard error of the technical replicates.

Extended Data Fig. 3. MR78 and MR191 binding to MARV GPΔMuc_WT at different pHs.

a,b, Binding of MR78 (a) and MR191 (b) IgGs at a concentration of 100 nM at the indicated pH to immobilized MARV GPΔMuc_WT, as measured by biolayer interferometry. Data shown are one representative out of two biological replicates using distinct batches of protein and antibodies.

Extended Data Fig. 4. Analysis of neutralization breadth of MARV monoclonal antibodies.

a–m, Neutralization dose-response curves for MARV16, MR78, MR191, and EBOV515 against VSV pseudotyped with the MARV/Musoke GP (a), MARV/Angola GP (b), MARV/Ci67 GP (c), MARV/Equatorial Guinea-2023 GP (d), MARV/Kakbat-SL-2017 GP (e), MARV/Kasbat-SL-2018 GP (f), MARV/Ozolin GP (g), MARV/Ghana-2022 GP (h), RAVV GP (i), DEHV GP (j), MLAV GP (k), EBOV GP (l), or SUDV GP (m). Each of the two to six biological replicates used distinct batches of pseudoviruses and antibodies and data are shown as the mean ± standard error of technical triplicates.

Extended Data Fig. 5. Cryo-EM data processing of the MARV GPΔMuc_WT-MARV16 Fab complex.

a,b, Representative cryo-EM micrograph (a) and 2D class averages (b) obtained for MARV GPΔMuc_WT in complex with MARV16 Fabs. Scale bar: 100 nm. c, Cryo-EM data processing workflow. NUR (per-particle defocus): non-uniform refinement with per-particle defocus refinement. d, Gold-standard fourier shell correlation curves for the locally refined MARV GPΔMuc_WT-MARV16 Fab complex (using a mask comprising the GP trimer and MARV16 Fab variable domains). e, Local resolution calculated with cryoSPARC for the locally refined MARV GPΔMuc_WT-MARV16 Fab complex. f, Heat map of angular distribution for the particles contributing to the final reconstruction. g, Conical fourier shell correlation plot.

Extended Data Fig. 6. Analysis of the MARV16-bound MARV GPΔMuc_WT complex.

a, Sequence conservation of MARV GP residues comprising the MARV16 epitope across the Marburgvirus isolates (MARV variants and RAVV) assessed in this study. Conservation scores were assigned using ConSurf. b, Superimposition of the MR191 Fab-bound RAVV GP (PBD: 6BP2; orange) and MARV16 Fab-bound MARV GP (blue) comparing the N-terminus of the GP2 core domain of the two models. The arrow indicates the shift of the equivalent residues in each model. The MR191 and MARV16 Fabs are not shown for clarity. c, MARV GP2 residues 504–510 modelled into the cryo-EM map (semi-transparent grey surface). d, Comparison of the binding modes of MARV16 (green) and the EBOV antibody EBOV-515 (orange). The EBOV GP trimer from the EBOV GP-EBOV-442-EBOV-515 complex structure (PDB: 7M8L) was superimposed with the MARV GP trimer from the MARV GPΔMuc_WT-MARV16 structure to compare the EBOV-515 and MARV16 binding poses. MARV GP1 and GP2 are shown in different shades of purple and beige, respectively. N-linked glycans are rendered as light blue surfaces. e, The MARV GP1 glycan cap (residues 191–219) modelled into the cryo-EM density (semi-transparent grey surface). f, View of the putative RBS residues (shown in orange) that are shielded by the MARV glycan cap (light purple). The rest of GP1 is rendered purple.

Extended Data Fig. 7. Identification of MARV GPΔMuc_WT antigenic sites recognized by non-neutralizing monoclonal antibodies.

a,b, Percent binding (a) to the MARV GPΔMuc_WT and BLI traces (b) for antibody pairs evaluated in the epitope binning experiments. Antibody pairs showing reciprocal blocking relationships were classified as belonging to the same binding group. Data from one biological replicate are shown and representative of two biological replicates. c–e, 3D reconstructions, representative 2D class averages, and angular distribution plots obtained by single particle electron microscopy analysis of negatively stained MARV GPΔMuc_WT bound to the MARV18 (c), MARV14 (d), or MARV7 (e) Fabs. f, Representative 2D classes from an electron microscopy dataset of negatively-stained MARV GPΔMuc_WT-MARV20 Fab complex.

Extended Data Fig. 8. MARV16 does not activate FcγRIIa or FcγRIIIa.

a,b, In vitro evaluation of MARV16 mAb-mediated activation of human FcγRIIIa V158 (a) and FcγRIIa H131 (b) using a bioluminescent reporter assay as a surrogate assay for Fc-mediated effector functions. ExpiCHO cells transfected with MARV/Musoke GP and Jurkat-Fcγ cells were used as target and effector cells, respectively.