Broad neuraminidase antibodies confer protection against seasonal and avian influenza viruses

Abstract

Neuraminidase (NA) is a critical target for universal influenza vaccines and therapeutic antibodies, yet its antigenic landscape remains incompletely understood. Here we identify two broadly cross-protective monoclonal antibodies, CAV-F6 and CAV-F34, from influenza-infected individuals. These antibodies inhibit NA enzymatic activity across multiple subtypes and confer protection against seasonal influenza in female mouse models. Importantly, the two antibodies also neutralize emerging avian strains, including recent bovine H5N1 and H7N9 strains, both with pandemic potential. Structural studies reveal that both antibodies target conserved regions of the NA active site via HCDR3, blocking sialic acid interaction. Furthermore, we observe distinct occupancy for the two antibodies on N2 tetramer, which is likely due to differences in binding affinity. Our findings provide molecular insights into NA-targeted immunity and offer a foundation for developing broadly protective influenza vaccines and therapeutics.

Influenza A virus neuraminidase (NA) is an important target for universal influenza vaccines. Here, the authors identify two monoclonal antibodies, CAV-F6 and CAV-F34, that inhibit NA activity across multiple subtypes, offering protection against seasonal and emerging avian influenza strains, thus advancing the development of broadly protective vaccines and therapeutics.

Introduction

Influenza viruses pose a significant and persistent threat to global public health, causing a spectrum of acute respiratory illnesses and incurring substantial economic burdens. Influenza A viruses (IAV) are further classified into subtypes based on surface membrane proteins hemagglutinin (HA) and neuraminidase (NA), with 19 HA and 11 NA subtypes identified to date^1,2. Influenza B viruses (IBV) predominantly infect humans and represent a significant health concern, particularly for young children, the elderly and immunocompromised individuals³. Avian influenza viruses like H5N1, H5N6 and H7N9 have demonstrated the capacity to cross species barriers and infect humans^{4, 5, 6, 7–8}, with extremely high case fatality rates. In March 2024, an outbreak of H5N1 highly pathogenic avian influenza (HPAI) virus was reported in dairy cattle, with subsequent infections identified among dairy farm workers⁹. Although the bovine H5N1 virus retains specificity for avian-type (α2,3-linked sialic acid) receptor, a mutation in the HA may allow it to switch to human-type (α2,6-linked sialic acid) receptor^10,11. Research demonstrates that HPAI H5N1 viruses isolated from infected farmers are lethal to mice and ferrets, and can be transmitted between mammals through respiratory droplets¹². These characteristics suggest that the future HPAI H5N1 may spread more readily among human populations, posing profound risks to public health.

NA, the second most abundant glycoprotein on the surface of influenza viruses¹, has been demonstrated to be a promising target for combating influenza. As a sialidase, NA cleaves sialic acid linkages on host cells¹³, facilitating the release of new virions and aiding viral spread. Furthermore, NA catalyzes the hydrolysis of sialic acid receptors in respiratory mucus^14,15, aiding the migration of viral particles in the respiratory tract. Notably, cross-reactive NA antibodies generated from prior infection can effectively reduce morbidity and mortality during pandemics. Compared to HA, NA is more conserved, making it an attractive target for broadly protective therapeutics and vaccines^{16, 17–18}. In recent years, several broadly reactive NA epitopes have been identified^{19, 20, 21, 22, 23, 24, 25–26}. Antibodies targeting the NA enzymatic site, like 1G01²⁰ and FNI9²¹, directly inhibit enzyme activity across various influenza subtypes. Antibodies binding to the underside of the NA head domain, such as 3C08, 3A10, 1F04²³, and NDS.1, NDS.3²⁴, show cross-reactivity with H3N2 NAs by sterically hindering NA-sialic acid interactions. Moreover, NA-reactive antibodies can enhance virus clearance through antibody-dependent cell-mediated cytotoxicity (ADCC) and complement-dependent cytotoxicity (CDC)^{27, 28–29}.

Current influenza vaccines predominantly utilize HA as the primary antigen for immunization, with serum conversion of HA antibodies commonly employed to assess the immunogenicity of these vaccines^30,31. However, accumulating evidence highlights the critical role of NA in eliciting cross-protective immunity. Previous studies have shown that incorporating both HA and NA proteins into influenza vaccines can significantly boost the breadth of protective coverage and improve immune responses^{32, 33–34}. A broadly protective influenza vaccine that elicits a robust NA antibody response may enhance the capacity to withstand a pandemic of avian influenza. For instance, antibodies targeting the N1 protein of H1N1 virus have demonstrated effective protection in mice infected with the H5N1 virus³⁵. Anti-NA antibodies have also been shown to prevent airborne transmission of the influenza virus after postexposure in animal models³⁶. Therefore, it is imperative to elicit a broader repertoire of NA-neutralizing antibodies that synergize with HA-mediated immunity. Additionally, integrating critical epitope information from these broadly neutralizing antibodies holds promise for advancing the design of next-generation influenza vaccines.

Here, we report the isolation and characterization of two broadly protective antibodies, CAV-F6 and CAV-F34, from an influenza-infected individual. These two antibodies demonstrated strong binding to NA proteins of Group 1 and Group 2 IAVs, as well as IBVs. Both exhibit potent neuraminidase inhibition (NAI) activity, particularly against the newly emerged HPAI H5N1 in the USA, with promising prophylactic and therapeutic effects in mouse model challenged with various seasonal and avian influenza viruses. Structural analysis revealed that both antibodies blocked sialic acid interacting with the NA active site, effectively inhibiting NA enzymatic function. Importantly, our research revealed a correlation between the occupancy of bound antibodies on the NA tetramer enzymatic active sites and their NAI activity. Mechanistically, we identified that variations in antibody binding affinity drive the differential occupancy observed on the NA tetramer. These findings offer valuable insights into the neutralization mechanisms of NA antibodies, informing the development of influenza therapeutics and universal vaccine design.

Results

Isolation and characterization of anti-neuraminidase antibodies

To isolate the NA-specific antibodies, plasmablasts (PBs) were sorted from peripheral blood mononuclear cells (PBMC) derived from three influenza-infected individuals in January 2020 (Fig. 1A). ScRNA-seq and single B cell cloning techniques were used separately to screen and obtain monoclonal antibodies specific to NA³⁷. Among these antibodies, CAV-F52 (single cell sorting-derived) was specific to IBV NA, CAV-F108 (single cell sorting-derived) exhibited specificity for the N2 subtype, whereas CAV-F6 (from scRNA-seq) and CAV-F34 (from single cell sorting) displayed broad reactivity to NAs of Group 1, Group 2 IAVs as well as IBVs (Fig. 1B). Both CAV-F6 and CAV-F34 were isolated from the donor 2. We then assessed the binding specificities of CAV-F6 and CAV-F34 towards NAs of IAVs and IBVs via ELISA (Fig. 1B). Simultaneously, biolayer interferometry (BLI) was used to further verify the affinity between the antibodies and various NA antigens (Fig. S1). CAV-F34 displayed binding to N1, N2, N3, and IBV NAs, while CAV-F6 exhibited a little broader cross-reactivity, encompassing N1, N2, N3, N4 and IBV NAs (Fig. 1B, C). Both of them showed weak binding to N5, N8 and N9 subtypes (Fig. 1B, C). Genetic lineage tracing revealed that CAV-F34 and CAV-F6 originate from the IGHV3-48 and IGKV3-20 germline genes (Fig. 1D; and Table S1). To further characterize their evolutionary relationships, we performed phylogenetic analysis on IGHV3-48 sequences derived from the same donor (donor 2) antibody repertoires. Both antibodies localized the terminal positions in the evolutionary tree (Fig. S2), suggesting their acquisition of somatic hypermutations during affinity maturation.

Fig. 1 Identification and characterization of broadly cross-reactive NA-targeting antibodies. [Images not available. See PDF.]

A Schematic representation of the antibodies screening process. B Binding heatmap of mAbs to recombination NAs representative of IAV and IBV strains by ELISA. Colored boxes represent the value of the optical density at 405 nm (OD405). C Phylogenetic tree of IAV and IBV NAs constructed using maximum likelihood analysis of amino acid sequences. Pentagrams indicate antibody-binding subtypes (OD405 over twice the negative control value; red: CAV-F6; yellow: CAV-F34). D Immunoglobulin variable heavy-chain and light-chain gene usage, HCDR3 and LCDR3 sequences of NA mAbs. E Half maximal inhibitory concentration (IC50) values of CAV-F34 and CAV-F6 against NAs via ELLA (orange) and MUNANA (red) assay. F–I Inhibitory activity of CAV-F34 and CAV-F6 against drug-resistant NA mutants in MUNANA assays. Data represents one of three independent experiments, shown as mean ± SD of three technical replicates.

We next employed an ELLA assay with recombinant NA proteins to evaluate the NAI activity of CAV-F6 and CAV-F34, using CAV-F52 and CAV-F108 as reference antibodies. Both CAV-F6 and CAV-F34 displayed robust NAI activity against seasonal influenza N1 and N2 subtypes, including strains with a historical range exceeding a century (A/Brevig Mission/1/1918 (H1N1)) and a recent strain of HPAI that can be transmitted among mammals (A/dairy cow/Minnesota/24_016288-003/2024 (H5N1)) (Fig. 1E; and Fig. S3A). Moreover, CAV-F6 displayed stronger inhibition compared to CAV-F34 against avian influenza N1, N3, and N4 subtypes from H5N1, H2N3, and H10N4 viruses (Fig. 1E; and Fig. S3A). To address the potential interference caused by the substantial size of the substrate used in the ELLA assay, we also conducted a MUNANA (2′-(4-methylumbelliferyl)-α-D-N-acetylneuraminic acid) assay using a small molecule substrate to further evaluate the NAI activity (Fig. 1E; and Fig. S3B). CAV-F6 demonstrated broad and potent inhibition against seasonal influenza N1 and N2 as well as avian influenza N1 and N4, albeit exhibiting weak inhibition towards IBV NA. CAV-F34 displayed enzyme inhibition against N1 from A/California/04/2009 and A/India/Pun151245/2015, but its inhibitory activity against other NA types was considerably weaker. Overall speaking, CAV-F6 showed considerably stronger inhibitory activity against H1N1, H3N2 and H5N1 strains compared to CAV-F34 in the MUNANA assay (Fig. 1E).

Given the rise of oseltamivir-resistant influenza strains in clinical specimens, we further investigated the effectiveness of these two antibodies towards resistant strains using MUNANA assay (Fig. 1F–I; and Table S2). The primary mutations associated with reduced sensitivity to oseltamivir include H274Y, I222R, and S246N on Group 1 NA (introduced into the N1 protein derived from A/California/04/2009 (H1N1)) and N294S on Group 2 NA (introduced into the N2 protein derived from A/Switzerland/9715293/2013 (H3N2))^38,39. The results revealed that both CAV-F6 and CAV-F34 demonstrated robust enzyme inhibitory activity against the N1 carrying these mutations. As for the N294S mutation on N2, CAV-F6 maintained its inhibitory effect, whereas CAV-F34 lost its effectiveness.

NA-reactive mAbs confer protection effects against seasonal influenza virus challenge in vivo

To evaluate the protective effects of CAV-F34 and CAV-F6 in vivo, a series of lethal challenge studies with seasonal influenza viruses was conducted using BALB/c mouse model. Mice were administered either CAV-F34 or CAV-F6 at a dosage of 10 mg/kg via intraperitoneal injection, 24 h before (prophylactic) or 24 h (IBVs)/48 h (IAVs) after (therapeutic) exposure to a lethal dose of seasonal influenza virus (A/California/04/2009(H1N1), A/Guizhou/54/1989(H3N2), B/Colorado/06/2017(V) or MA-B/Florida/4/2006(Y)) (Fig. 2A, B). Both CAV-F34 and CAV-F6 provided complete protection in mice challenged with H1N1 or H3N2 virus in both the prophylactic and therapeutic groups (Fig. 2C–F). Mice exposed to H1N1 and H3N2 exhibited stable body weights, with only a slight initial decrease followed by recovery. For lethal strains of influenza B virus, both CAV-F34 and CAV-F6 showed protective efficacy against both Victoria and Yamagata lineage strains (Fig. 2G–J). CAV-F34 conferred complete protection against both subtypes in prophylactic and therapeutic settings. In the prophylactic group, the CAV-F6 offered 40% protection against the Victoria lineage and 100% protection against the Yamagata lineage. In the therapeutic group, mice receiving CAV-F6 antibodies exhibited body weight loss, yet they still achieved complete protection against the Victoria lineage. For the Yamagata lineage, the CAV-F6 antibodies provided 60% protection. These findings suggest that CAV-F34 and CAV-F6 hold promise for use in the prophylaxis and early therapeutic intervention of seasonal influenza virus infections.

Fig. 2 In vivo protection of CAV-F34 and CAV-F6 against seasonal influenza viruses. [Images not available. See PDF.]

A, B Experimental design for pre-exposure prophylaxis (A) and post-exposure therapy (B). BALB/c mice (6-8 weeks old, female) were intraperitoneally administered 10 mg/kg mAbs either 24 h before challenge (prophylaxis) or 24 h (IBV)/48 h (IAV) post-challenge (therapy). Virus dose was 5 × LD₅₀. Survival was monitored for 14 days. C, D Protection against A/California/04/2009(H1N1): Prophylaxis (C) and therapy (D) with CAV-F34 (green) and CAV-F6 (blue). Yellow: PBS control. Left panels: mean percentage change in body weight relative to baseline. Right panels: survival curves. Five animals were used per group (n = 5), and error bars represent SD. E, F Protection against A/Guizhou/54/1989 (H3N2): Prophylaxis (E) and therapy (F). G, H Protection against B/Colorado/06/2017 (Victoria): Prophylaxis (G) and therapy (H). I, J Protection against MA-B/Florida/4/2006 (Yamagata): Prophylaxis (I) and therapy (J). For (E–J) Data presentation identical to (C–D).

NA-reactive mAbs provide effective protection against avian influenza virus challenge in vivo

For avian influenza viruses, the effectiveness of these two antibodies was assessed in prophylactic and therapeutic settings against A/dairy cattle/Texas/24-008749-003-original/2024 (H5N1) and A/Anhui/1/2013 (H7N9) NIBRG-268 (Fig. 3A, B). As for H7N9 virus, CAV-F34 and CAV-F6 provided partial protection when 10 mg/kg of mAbs were administered either 24 h before or after infection (Fig. 3C, D). CAV-F6 provided 80% protection in the prophylactic group, effectively preventing both weight loss and mortality, whereas CAV-F34 displayed 40% protection in the prophylactic group. Both antibodies offered weaker protection in the therapeutic assay (40% for CAV-F6, 20% for CAV-F34) compared to the prophylactic group. Notably, CAV-F6 provided complete protection against A/dairy cattle/Texas/24-008749-003-original/2024 (H5N1) (2×LD₅₀) when mice were treated prophylactically with 20 mg/kg of mAb via intraperitoneal injection 24 h prior to infection (Fig. 3E). CAV-F34 achieved 60% protection under the same prophylactic conditions. Additionally, the mAbs were administered 24 h post-infection to evaluate therapeutic treatment, and all mice experienced weight loss in this setting (Fig. 3F). Both antibodies provided partial protection (60% for CAV-F6 and 20% for CAV-F34). The results indicated that both mAbs provided a measurable degree of protection against avian influenza viruses. Notably, CAV-F6 showed significant prophylactic efficacy against the recently emerging bovine H5N1 strain when administered at a higher dose.

Fig. 3 In vivo protection of CAV-F34 and CAV-F6 against avian influenza viruses. [Images not available. See PDF.]

A, B Experimental design for pre-exposure prophylaxis (A) and post-exposure therapy (B). BALB/c mice (6-8 weeks old, female) were intraperitoneally administered 10 mg/kg (H7N9) or 20 mg/kg (H5N1) mAbs either 24 h before challenge (prophylaxis) or 24 h post-challenge (therapy). Survival was monitored for 14 days. C, D Protection against A/Anhui/1/2013 (H7N9) NIBGR-268: Prophylaxis (C) and therapy (D) with CAV-F34 (green) and CAV-F6 (blue). Yellow: PBS control. Left panels: mean percentage change in body weight relative to baseline. Right panels: survival curves. Five animals were used per group (n = 5), and error bars represent SD. E, F Protection against A/dairy cattle/Texas/24-008749-003-original/2024 (H5N1): Prophylaxis (E) and therapy (F). For (E–F) Data presentation identical to (C–D).

The broad NAI activity of CAV-F34 and CAV-F6 is achieved by blocking the NA active site

To elucidate the mechanism by which CAV-F34 and CAV-F6 achieve broad NAI activity, we conducted cryo-electron microscopy (cryo-EM) studies to determine the structures of Fabs in complex with NAs from four different strains, including N1 protein from H1N1 virus A/California/04/2009 (CA09 N1), N2 protein from H3N2 virus A/Switzerland/9715293/2013 (SL32 N2), N4 protein from A/Red knot/Delaware Bay/310/2016 (rkDB16 N4) and N1 protein from H5N1 virus A/dairy cow/Minnesota/24_016288-003/2024 (dcMN24 N1). Five structures of NA-Fab complexes were determined (Fig. 4A). All structures were resolved at high resolutions (2.5-3.1 Å) (Fig. 4; and Table S3; Fig. S4–S8). Among these five NA-Fab complexes, three NA-Fab complexes adopted a C4 symmetry, with each NA protomer bound with a Fab, including SL32 N2/CAV-F6, rkDB16 N4/CAV-F6, and CA09 N1/CAV-F34 complexes (Fig. 4A). Interestingly, the SL32 N2/CAV-F34 complex exhibited only two Fabs bound to the NA tetramer, potentially explaining the lower NAI activity of CAV-F34 against SL32 N2 observed in the MUNANA assay (Fig. 1E). These two CAV-F34 Fabs were positioned either adjacent or opposite each other. Due to poor density for one CAV-F34 Fab in the SL32 N2/CAV-F34 complex, only the Fab with well-defined density was modelled for further analysis (Fig. 4A).

Fig. 4 Cryo-EM structures of the NA-antibody complexes. [Images not available. See PDF.]

A Overall structures of the complexes of NA tetramers and Fabs. Calcium ions (red spheres) and glycans (green) are shown. B Overall structures of Fab-NA protomers. Both CAV-F34 and CAV-F6 Fab utilize HCDR3 to approach into the enzymatic site. C Binding epitopes on the NAs. Footprints of CAV-F34 and CAV-F6 Fabs on the NAs are represented by red surface. The interacted CDRs and FR loops are shown. For all the panels, A/California/04-CIP100_RGCM_1868/2009 (H1N1) NA tetramer, A/Switzerland/9715293/2013 (H3N2) NA tetramer, A/Red knot/Delaware Bay/310/2016 (H10N4) NA tetramer, and A/dairy cow/Minnesota/24_016288-003/2024 (H5N1) NA tetramer are colored by light purple, cyan, cornflower blue and dark violet, respectively. CAV-F34 and CAV-F6 are colored as orange and pink, respectively. Abbreviations: variable domain of heavy chain (VH); variable domain of kappa chain (VK); complementarity-determining region (CDR); framework region (FR). D–H Detailed interactions between the NA enzymatic site and HCDRs. Salt bridges and hydrogen bonds are depicted by solid lines. I–M Detailed interactions between sialic acid (SA) and the NA enzymatic site of A/Tanzania/205/2010 (H3N2) (PDB ID: 4GZQ) (I), A/Tanzania/205/2010 (H3N2) NA and oseltamivir (OSE) (PDB ID: 4GZP) (J), A/Tanzania/205/2010 (H3N2) NA and FNI9 (PDB ID: 4G3N) (K), A/California/04/2009 (H1N1) NA and 1G01 (PDB ID: 6Q23) (L) and A/Kansas/14/2017 (H3N2) NA and 4N2C402 (PDB ID: 8ZR4) (M). Salt bridges and hydrogen bonds are depicted by solid lines, and other interactions within 4 Å are represented by dashed lines. For all the panels, A/Tanzania/205/2010 N2 NA, A/California/04/2009 (H1N1) NA and A/Kansas/14/2017 (H3N2) NA are colored by dark blue, light purple and salmon, respectively. SA, OSE, FNI9, 1G01 and 4N2C402 are colored as yellow, lime, dark green, grey and brown, respectively. N, Sequence logo plot illustrating amino acid conservation of a portion of NA sequences from H1N1 and H5N1 IAVs. O, Sequence logo plot illustrating amino acid conservation of a portion of NA sequences from human seasonal H3N2 IAVs.

We then analyzed one protomer of the NA-Fab tetramers and found that the binding orientation of CAV-F34 on CA09 N1 and SL32 N2 is nearly identical (Fig. S9A). The similar binding orientation results in a comparable binding epitope on the NAs, with buried surface area (BSA) of 1116.9 and 1173.2 Ų, respectively (Fig. 4B, C; and Fig. S9A). CAV-F6 also exhibited similar binding angles in the complexes of NA/CAV-F6, including SL32 N2, rkDB16 N4 and dcMN24 N1, with BSAs of 1167.1, 1035.8 and 1140.9 Ų, respectively (Fig. 4B, C; and Fig. S9B). When the CAV-F34/NA complex was superimposed onto that of CAV-F6 based on NA tetramer, a slight but obvious twist of antibody binding angles was observed (Fig. S9C). Although CAV-F6 and CAV-F34 share the same germline genes of variable heavy (IGHV3-48) and light chain (IGKV3-20), differences are observed at the three complementarity-determining regions (CDRs) of the heavy chain (Fig. S10). These differences might correlate with the distinct angles at which CAV-F6 and CAV-F34 approach their NAs (Fig. 4B; and Fig. S9). CAV-F6 engages with SL32 N2, rkDB16 N4 and dcMN24 N1 NAs by utilizing all of its CDRs (H1, H2 and H3) in the heavy chain, as well as CDR1 (K1) and CDR2 (K2) of the kappa chain, similar to the interaction between CAV-F34 and SL32 N2 (Fig. 4C). Interestingly, CAV-F34 additionally utilizes framework 3 of the kappa chain (KFR3) to interact with CA09 N1 NA (Fig. 4C).

We further analyzed the local interactions between the HCDR3 region and the NA enzymatic pockets (Fig. 4D-4M). Both CAV-F34 and CAV-F6 employ a similar binding mode, with their HCDR3 regions inserting into the enzymatic pocket of NA tetramers. Within this pocket, a complex network of hydrogen bonds and salt bridge interactions forms, particularly between residues R103 and D104 of the HCDR3 region, and five residues of NA active pocket (R117/118, D150/151, R292/293, R368/369/371 and E227/228) (Fig. 4D–H). Notably, these residues also play crucial roles in sialic acid (SA) interacting with NA proteins, underscoring the NA active site blocking mechanism of CAV-F34 and CAV-F6, which contributes to their broad neutralizing activity (Fig. 4I). Compared to previous studies, the binding mode of CAV-F34 and CAV-F6 within the enzymatic pocket resembles that of oseltamivir (OSE) and FNI9 but differs from 1G01 (Fig. 4J–L).

We subsequently analyzed the conservation of binding epitopes for CAV-F6 and CAV-F34 on NAs. The majority of residues that interact with CAV-F34 and CAV-F6 on N1 NA have remained highly conserved across nearly all H1N1 and H5N1 NA sequences from 1968 to 2025, with approximately 84% of these residues conserved for CAV-F34 and 88% for CAV-F6 (Fig. 4N). Similarly, the amino acid residues interacting with N2 NA in CAV-F34 and CAV-F6 show minimal variation in seasonal influenza H3N2 sequences (Fig. 4O). Overall, the binding epitopes of CAV-F6 and CAV-F34 are notably conserved among HxN1 (including both H1N1 and H5N1) and human seasonal H3N2 viruses, which underscores their broad cross-reactivity.

CAV-F6 R54S mutation reduces binding affinity and consequentially diminishes inhibition potency

Cryo-EM analysis revealed an intriguing difference of the binding number between CAV-F6 and CAV-F34 when bound to the SL32 N2 NA. Four CAV-F6 Fabs were found to bind to SL32 N2 NA, whereas only two CAV-F34 Fabs bound to N2 NA (Fig. 5A, B). To explore the reasons behind this difference in binding, we focused on the dominant interactions of CAV-F6 compared to CAV-F34 when binding to SL32 N2 NA. Upon examining these interactions, especially hydrogen bonds and salt bridges interactions, CAV-F6 demonstrated a notably stronger interaction, particularly at the R54, D102 and W105 sites of heavy chain (Fig. 5C). Thus, we generated six single-point mutated antibodies by swapping residues between CAV-F6 and CAV-F34 (CAV-F6 R54S, D102N and W105L; CAV-F34 S54R, N102D and L105W). We used ELISA assay to confirm their binding abilities to SL32 N2 NA. All the mutants showed similar binding affinities to that of wild-type antibodies (Fig. S11A). Consistent with binding, ELLA assays demonstrated comparable NAI activities for all single-point mutants (Fig. 5D; and Fig. S11B). However, the MUNANA assay revealed a dramatical decrease of NAI activity for CAV-F6 harboring the R54S or D102N mutations, while no significant change was observed upon introducing corresponding mutations in CAV-F34 (Fig. 5D; and Fig. S11C).

Fig. 5 CAV-F6 R54S mutation reduces binding affinity and consequentially diminishes inhibition potency. [Images not available. See PDF.]

A, B Two-dimensional classification and three-dimensional models of the A/Switzerland/9715293/2013 (H3N2) in complex with CAV-F6 (A) and CAV-F34 (B). C Comparison of key amino acid sites between CAV-F34 and CAV-F6 binding to the A/Switzerland/9715293/2013 (H3N2) NA tetramer. Hydrogen bonds and salt bridge interactions formed between NA and CAV-F6 are marked with blue triangles, while those with CAV-F34 are marked with orange inverted triangles. Amino acid residues that form plenty of contacts with NA are marked with black circles. D The IC₅₀ value of enzyme inhibitory ability of CAV-F6 and its single point mutant including R54S, D102N and W105L against A/Switzerland/9715293/2013 (H3N2) NA protein via ELLA and MUNANA assay. E–G Binding affinity of CAV-F34, CAV-F6 and CAV-F6 R54S with A/Switzerland/9715293/2013 (H3N2) NA via BLI assay. H, I Detailed interactions between the wild-type CAV-F6-R54 residue (H) and mutated CAV-F6-R54S residue (I) with the A/Switzerland/9715293/2013 (H3N2) NA protomer. Salt bridges and hydrogen bonds are depicted by solid lines. J Superimposition of the CAV-F6-R54S fab and the unbounded NA protomer of the A/Switzerland/9715293/2013 (H3N2) NA/CAV-F6-R54S complex displayed an obvious steric hindrance. K Superimposition of the CAV-F6-R54S fab of the A/Switzerland/9715293/2013 (H3N2) NA/CAV-F6-R54S complex and the protomer of apo A/Switzerland/9715293/2013 (H3N2) NA tetramer also displayed a similar steric hindrance. L Fold change in IC50 values for CAV-F6-R54S and CAV-F6 in NAI of A/Switzerland/9715293/2013 (H3N2) NA with different mutations. M–Q Inhibitory activity of CAV-F6-R54S and CAV-F6 against N2 NA carrying K199R, K199A, K199D, K199E mutations in MUNANA assay. Data represents one of three independent experiments, shown as mean ± SD of three technical replicates.

To understand the impact of the R54S and D102N mutations on CAV-F6 binding to the N2 NA tetramer, we prepared the corresponding complexes and collected cryo-EM data. One CAV-F6 R54S Fab and four CAV-F6 D102N Fabs were observed in the reconstructed map (Fig. S11D and S11E). This indicated that R54 is a key residue for CAV-F6 to fully bind to N2 NA tetramers. We further measured the binding affinity of CAV-F6 R54S Fab to SL32 N2 and found that, while its association rate constant (ka) was comparable to that of the wild-type CAV-F6 Fab, its dissociation rate constant (kd) was nearly 13-fold higher—similar to that observed for CAV-F34 Fab and SL32 N2 (Fig. 5E–G). These results suggest that the reduced binding stoichiometry of both CAV-F6 R54S and CAV-F34 on N2 observed in cryo-EM likely arises from their increased dissociation rates, highlighting residue 54 as a key modulator of antibody binding stability to N2. To further figure out the local interactions surrounding residue 54, we determined the SL32 N2/CAV-F6 R54S structure at a high resolution of 2.3 Å, and an unambiguous difference was observed compared to WT (Fig. S12; and Table S4). In the wild-type complex, R54 formed a network of hydrogen bonds and salt bridge interactions with D198, K199 and D221 of the SL32 N2 NA (Fig. 5H). In contrast, S54 in the mutated complex exhibited notably weaker interactions, with only one hydrogen bond (Fig. 5I). Although S54 formed fewer interactions with the NA tetramer, the K199 side chain on the bound NA tetramer remained positioned away from S54, similar to that in the wild-type complex. We analyzed the conformation of K199 in unbound NA protomers, including the apo NA form and unbound protomers from N2 NA/CAV-F6 R54S complex. All these side chains of K199 in unbound NA protomers approached S54 of CAV-F6 face to face, leading to a potential steric hindrance when CAV-F6 R54S bound (Fig. 5J, K; and Fig. S13).

These observations suggest that in the apo form, the K199 on the SL32 N2 NA tetramer prefers to adopt a conformation incompatible with CAV-F6 binding. Upon CAV-F6 approach, the presence of R54 in CAV-F6 repels K199 due to similar charges, thus facilitating hydrogen bond and salt bridge formation with D221 on NA. These critical interactions might be disrupted when R54 was substituted with serine, resulting in a weaker repulsion of K199 and a lower binding ratio to SL32 N2 NA tetramers.

Moreover, the K199 of seasonal influenza virus H3N2 NA from all strains remains highly conserved over approximately the past 60 years (Fig. 4O). To validate the role of K199, we introduced K199R, K199A, K199D, and K199E mutations into the influenza virus SL32 N2 NA protein and evaluated their impact in the MUNANA assay (Fig. 5L–Q). The results revealed significant differences in NAI activity between CAV-F6 and CAV-F6 R54S against these mutants. For the wild-type N2, CAV-F6 R54S exhibited an IC₅₀ value over two orders of magnitude higher than CAV-F6 (Fig. 5L, M). After the mutation of K199R, the IC₅₀ value of CAV-F6 R54S remained over 60 times higher than that of CAV-F6 (Fig. 5L, N). However, this difference was reduced to within a factor of 10 following K199A, K199D, and K199E mutations (Fig. 5O–Q). These findings underscore K199 as a critical site influencing antibody-mediated N2 inhibition, consistent with our structural observations.

Discussion

The isolation of broadly neutralizing antibodies (bnAbs) against neuraminidase from human populations has positioned NA as a promising target for universal influenza vaccine development^{19, 20, 21, 22, 23, 24–25,40, 41–42}. In this study, we successfully isolated two NA-specific monoclonal antibodies, CAV-F6 and CAV-F34, from an influenza-infected individual. They displayed broad cross-reactivity against various NA subtypes in Group 1 IAVs, Group 2 IAVs, and IBVs. Both CAV-F6 and CAV-F34 potently inhibited NA enzymatic activity in vitro and provided remarkable protection against seasonal influenza challenge in vivo. Notably, these antibodies showed partial protective efficacy against H7N9 virus and bovine-derived H5N1 virus. Additionally, our antibodies CAV-F6 and CAV-F34 target the enzymatic active site of NA, similar to the broadly reactive antibodies FNI9²¹ and 4N2C402⁴³. Notably, their HCDR3 loops penetrate the enzymatic pocket to engage highly conserved residues, highlighting the critical role of a shared “RD” motif (or “DR” in 4N2C402) on HCDR3 in mimicking sialic acid-NA interactions^44,45 (Fig. 4D-4M). Despite these shared mechanistic features, differences in germline gene usage among these antibodies can lead to variations in binding orientation and buried surface areas (BSA) (Tables S5 and S6).

Although several NA-specific antibodies have been isolated by us and other groups, the underlying mechanisms of neutralization remain partially understood. Our findings highlight the significance of antibody occupancy for optimal NA inhibition. The tetrameric structure of NA, with each subunit containing an active site, requires full occupancy for complete enzymatic blockade^{46, 47, 48–49}. This phenomenon is likely influenced by the antibody affinity and the distribution of epitopes on the NA molecule. Antibodies targeting enzymatic sites may face kinetic challenges in binding all monomers simultaneously. The R54S mutation in CAV-F6 results in reduced binding affinity due to an increased dissociation constant (kd). This weaker interaction allows CAV-F6 R54S to dissociate more readily from the N2 tetramer (A/Switzerland/9715293/2013(H3N2)), enabling NA to maintain its enzymatic activity (Fig. 5E–G). Due to the incomplete elucidation of the catalytic mechanism of NA, we were previously uncertain about the impact of partially blocking enzymatic active sites on its enzymatic function. Previous studies on viral surface oligomeric proteins have shown that the activity of each subunit in the oligomer is closely related to the function exerted by the oligomer. For example, in the case of the trimeric envelope glycoprotein of HIV-1, at least two protomers need to remain active for virus entry⁵⁰.

In addition, we identified the key antibody-antigen interactions that underlie the differential affinities of CAV-F6 Fab and CAV-F34 Fab in binding to N2 NA. CAV-F6, which strongly inhibits N2 enzymatic activity, showed significantly reduced inhibition following the R54S mutation (Fig. 5D). Based on analysis of the interactions surrounding residue 54 of CAV-F6, we proposed that K199 on N2 protein might utilize its large positive side chain as a gate for antibody binding. When lysine was replaced with shorter side chain residue (K199A) or reverse electricity residues (K199D, K199E), the CAV-F6-R54S regained strong enzyme activity inhibition against N2 (Fig. 5O–Q). This suggests that the K199 residue likely acts as a gatekeeper for certain antibodies that bind to enzyme activity sites. It’s likely that only the basic amino acids on antibodies, such as arginine, can exert sufficient repulsive force to facilitate unlocking, allowing the HCDR3 to insert the enzyme active pocket more effectively. Our structural studies highlight the importance of both the enzymatic active site and the surrounding epitopes of NA. Furthermore, previous studies have demonstrated that the influenza virus is inclined to mutations at position 199 under the selection of N2 antibodies^{51, 52–53}, indicating that although this position serves as a prominent interaction site for many broadly reactive antibodies²⁰, it is a crucial site where the virus is likely to escape. Our study suggests that K199 influences the ability of antibodies, such as CAV-F34, to access the enzymatic active region of N2. Therefore, we propose that modifying K199 on N2 could be a strategy for developing comprehensive N2 immunogen-based vaccines.

Furthermore, we resolved the apo structure of the NA head from the avian influenza H5N1 clade 2.3.4.4b virus at a resolution of 3.0 Å (Fig. S14A, and Table S4). The NA head exhibits a typical tetrameric fold, with each protomer forming a six-bladed propeller structure composed of four anti-parallel β-strands per blade. Glycans were identified at Asn88, Asn146, and Asn235. The 150-loop in the H5N1 NA is positioned away from the enzymatic site, creating an additional 150-cavity, consistent with other N1 NAs^54,55 (Fig. S14B, C). In the cattle H5N1 NA-CAV-F6 complex, we observed a significant shift of the 150-loop toward CAV-F6, resulting in a closed conformation. These high-resolution structures of the H5N1 clade 2.3.4.4b NA and the flexibility of 150-loop offer valuable insights for future drug design.

In conclusion, our study expands the repertoire of antibodies against NA, providing evidence for cross-conserved epitopes on NA and offering insights into NA antibody neutralization mechanisms. The conserved antigenic sites are key factors for the development of universal vaccines. The conservation of the HA stem region has been validated, and universal vaccine candidates based on this region have progressed to the clinical trial stage^56,57. As the functionality of NA antibodies continues to be elucidated, future influenza vaccine designs may need to incorporate both HA and NA to achieve broader protection, inducing protective polyclonal responses against different antigens to counter the variability of influenza viruses.

Methods

Patient

Human peripheral blood mononuclear cells (PBMC) were obtained from donors who had seasonal influenza virus infection in January 2020. The study was approved by the Biomedical Research Ethics Committee of School of Public Health of Sun Yat-sen University in China (Ethics Code: School of Public Health of SYSU Medical Ethics [2019] No. 118). Written consent was obtained from all enrolled donors, who participated in the study. Donor 1 was a 19-year-old male, infected H3N2 subtype of IAV; donor 2 was a 25-year-old male also infected with H3N2; donor 3 was a 32-years-old male, infected with influenza B virus. PBMC were collected on day 7 post-confirmed infection.

Cell lines

HEK293T cells were cultured at 37 °C with 5% CO₂ in Dulbecco’s Modified Eagle medium (DMEM, Gibco), supplemented with 10% fetal bovine serum (FBS, Gibco) and 1% Penicillin-Streptomycin Solution (Gibco). Sf9 (Spodoptera frugiperda 9) cells were maintained in SF-SFM medium at 27 °C.

Virus and Biosafety Statements

A/California/04/2009 (H1N1), A/Guizhou/54/1989 (H3N2), B/Colorado/06/2017 (Victoria), MA-B/Florida/4/2006 (Yamagata) and A/Anhui/1/2013 (H7N9) NIBRG-268 passaged in 10-day old specific pathogen free (SPF) eggs at 33 °C (IBV) for 3 days or at 37 °C (IAV) for 2 days, respectively. All viruses were manipulated under Biosafety Level 2 (BSL-2) conditions.

Recombinant cattle_H5N1 virus⁵⁸ was kindly provided by Professor Honglin Chen (The University of Hong Kong). In vivo experiments involving infectious live cattle_H5N1 animal experiments were performed in a Biosafety Level 3 (BSL-3) laboratory at the Department of Microbiology, The University of Hong Kong, and strictly followed the approved standard operating procedures as previously described⁵⁹.

A full list of viruses used in this study can be found in Table S7.

Animal ethics approval

The animal experiments were performed in accordance with protocols approved by Sun Yat-sen University Institutional Animal Care and Use Committee (IACUC). (Ethics File Approval No.: SYSU-IACUC-2024-002734).

The BALB/c mice (6–8 weeks old) used in H5N1 virus challenge were obtained from the Centre for Comparative Medicine Research (CCMR) at the University of Hong Kong. The use of animal and experimental operating procedures has complied with the approved operation protocol by the Committee on the Use of Live Animals in Teaching and Research of The University of Hong Kong under CULATR 24-281.

PBMC isolation and cell sorting

Blood was collected in ethylenediaminetetraacetic acid (EDTA)-anticoagulated sample tubes using standard phlebotomy techniques. PBMCs were layered over human lymphocyte separation medium and centrifuging at 900×g for 20 minutes. The PBMC layer at the Ficoll interface was collected, washed twice with PBS/0.2%BSA. Cell counts were obtained, and cells were stained for 30 min at 4 °C with CD19-PB (Biolegend, clone HIB19, 302232), CD38-PE (Biolegend, clone HIT2, 303506), and CD27-APC (Biolegend, clone O323, 302810). Then cells were washed twice and single plasmablasts (CD19⁺ CD27⁺CD38^high) were single cell-sorted using a FACSAria III into 96-well plates and immediately frozen on dry ice.

10X genomics scRNA-seq library preparation and sequencing

CD19⁺ CD27⁺ memory B cells and CD19⁺ CD27⁺ CD38^high plasma B cells were sorted by FACS using a BD FACSAria III Flow Cytometer and pooled together. Single cell suspensions were loaded onto the 10X Genomics Chromium Controller microfluidic device and processed according to manufacturer’s protocol for a target capture of 12,000 B cells. ScRNA-seq and scVDJ-seq libraries were prepared as previously described³⁷. The cDNA-enriched libraries were size-selected using SPRI beads and sequenced on an Illumina NovaSeq 6000 instrument using 150 paired-end reads.

Sequence analysis of antibody

Single-cell V(D)J sequencing data were processed as previously described analysis pipeline³⁷. Paired-end fastq files were initially processed using Cell Ranger (v.5.0.0) vdj pipeline against the GRCh38 reference genome. V(D)J sequences were consequently re-annotated with “igblastn” using the IMGT database as a reference. Single paired BCR chains were continued for clonotypes analysis. The Hvdj sequences of VH3-48 clonotypes were selected to build B cell trees using Dowser and IgPhyML which takes B cell somatic hypermutation into consideration⁶⁰.

Monoclonal antibody generation

Antibodies were generated as previously described⁶¹. Briefly, Immunoglobulin heavy and light chain variable genes from single cell sorted plasmablasts were amplified by reverse transcriptase polymerase chain reaction (RT-PCR) and nested PCR reactions then sequenced. The amplified heavy and light genes were cloned into IgG1 expression vectors and co-transfected into HEK293T cells. After 5 days, secreted mAbs were purified from the supernatant using protein A agarose (Cytiva).

Expression and purification of NA proteins

The NA ectodomain (residues 82-469) was fused to an N-terminal gp67 signal peptide, 6×His-tag, vasodilator-stimulated phosphoprotein (VASP) tetramerization domain, and thrombin cleavage site to generate recombinant NA. Bacmid DNA carrying the recombinant NA was generated using the Bac-to-Bac system (Thermo Fisher Scientific). Baculovirus was produced by transfecting purified bacmid DNA into Sf9 cells with Cellfectin Reagent (Thermo Fisher Scientific), followed by amplification through serial passage in Sf9 cells. For protein expression, suspension Sf9 cells were infected at an MOI of 5. At 72 h post-infection, cells were pelleted by centrifugation (1000 × g, 30 min). Soluble NA was purified using Ni-NTA Sepharose High-Performance chromatography (Cytiva).

Oseltamivir-resistant variants were generated by introducing specific mutations: H274Y/I222R/S246N in N1 (A/California/04/2009(H1N1)) and N294S in N2 (A/Switzerland/9715293/2013(H3N2)). Mutant proteins were expressed and purified using the aforementioned protocols.

Enzyme-linked immunosorbent assay (ELISA)

Flat-bottom 96-well plates (Thermo Scientific) were coated with 50 μL of NA protein (2 μg/mL in PBS; Biosharp) and incubated at 4 °C overnight. After blocking with PBS containing 3% BSA (GBCBIO) for 1 h at 37 °C, diluted antibodies (10 μg/mL in PBS) were incubated for 1 h at 37 °C. Following six washes with PBS-T (PBS with 0.05% Tween-20), HRP-conjugated goat anti-human IgG antibody (Jackson ImmunoResearch; 1:2000) was added for 1 h at 37 °C. After washing with PBS-T, Super AquaBlue ELISA substrate (eBioscience) was added. Absorbance was measured at 405 nm using a SpectraMax ABS Plus microplate reader (Molecular Devices). Antibodies showing OD values ≥2-fold higher than blank controls were considered positive.

Enzyme-linked lectin assay (ELLA)

Flat-bottom 96-well plates (Thermo Scientific) were coated with 100 μL/well of fetuin (25 μg/mL; Sigma-Aldrich) in PBS containing Ca²⁺/Mg²⁺ and incubated at 4 °C for 24 h. After three washes with PBS-T, a separate plate was used to prepare mixtures containing serially diluted mAbs and fixed concentrations of NA proteins for 2 h at 37 °C. These mixtures were transferred to pre-washed fetuin-coated plates and incubated for 20 h at 37 °C. Following six washes with PBS-T, 100 μL of HRP-conjugated peanut agglutinin (PNA; 1:400 dilution) was added. Plates were incubated in the dark for 2 h before adding Super AquaBlue ELISA substrate (eBioscience). Absorbance was measured at 405 nm using a SpectraMax ABS Plus microplate reader (Molecular Devices). Data were analyzed in GraphPad Prism 9.0, with the 50% inhibition concentration (IC₅₀) defined as the antibody concentration reducing NA activity by 50% relative to negative controls.

MUNANA assay

Serial dilutions of antibodies were mixed with fixed concentrations of NA protein in 96-well black plates (Greiner Bio-One) and incubated at 37 °C for 1 h. Subsequently, 4-MUNANA substrate (GlpBio) was added to a final concentration of 300 µM. After 2 h of incubation at 37 °C, reactions were terminated with stop solution (NaOH-EtOH solution). Neuraminidase inhibition activity was measured by fluorescence spectroscopy using a BioTek Synergy microplate reader. Fluorescence excitation was set at 360 nanometers and emission detection at 455 nanometers. Data were analyzed in GraphPad Prism 9.0, with the 50% inhibition concentration (IC₅₀) defined as the antibody concentration reducing NA activity by 50% relative to negative controls.

Biolayer interferometry binding assay (BLI)

Binding affinity was measured by biolayer interferometry (BLI) using an Octet R8 instrument (Sartorius). For antibody-NA affinity measurements, antibodies diluted to 10 μg/mL in PBS (pH 7.4, 0.02% Tween-20) were loaded onto anti-human IgG Fc biosensors. The sensors were then exposed to serially diluted NA proteins (200, 100, 50, 25, and 12.5 nM) for association phases of 120 or 300 s. Following 600 s dissociation in assay buffer, regeneration was performed using 10 mM glycine (pH 1.5) for 30 s per cycle. For Fab-NA affinity measurements, biotinylated NA proteins were coupled to streptavidin biosensors. Serially diluted Fabs (five 2-fold dilutions starting from 200 or 500 nM) were flowed over the sensors for 300 s association, followed by 500 s dissociation in assay buffer. For estimating the KD, a 1:1 binding model was used. The fitting data was analyzed by GraphPad Prism 9.0. “No binding” was defined as maximum response of less than 0.05 nm.

Preparation of Fab fragments

CAV-F34 and CAV-F6 Fab fragments were prepared by digesting CAV-F34 and CAV-F6 IgG with papain (Sigma), respectively. And then CAV-F34 and CAV-F6 Fab fragments were separated from their Fc fragments by using Protein A beads (GenScript). Finally, gel-filtration chromatography was conducted to further purify the Fab fragments, using a Superdex 200 column (GE Healthcare) pre-equilibrated with HBS buffer (10 mM HEPES, pH 7.2, 150 mM NaCl).

Cryo-electron microscopy sample preparation and data collection

To prepare the SL32 N2 NA/CAV-F34, SL32 N2 NA/CAV-F6, CA09 N1 NA/CAV-F34 and rkDB16 N4 NA/CAV-F6 complexes, the purified NA tetramers were mixed with Fabs with a molar ratio of 1:4 and the complexes were incubated for 2 hours, followed by gel-filtration chromatography using a Superdex 200 column pre-equilibrated with HBS buffer. The final concentrations of the four NA-Fab complexes were about 1.5 mg/mL in HBS buffer. To prepare the dcMN24 N1 NA/CAV-F6 complexes, the purified NA tetramers were mixed with Fabs with a molar ratio of 1:4.4 and the complexes were incubated for 4 hours with a concentration of 1.43 mg/mL in HBS buffer. Then, 4.5 μL of NA-Fab complexes were applied to the pre-glow-discharged holey carbon grids (Quantifoil grid, Cu 300 mesh or Au 300 mesh, R1.2/1.3). After waiting 2 seconds, the grid was then blotted for 2 seconds with filter paper in 100% relative humidity, 8 °C, and plunged into the liquid ethane to freeze samples using FEI Vitrobot system (FEI).

Cryo-EM data of the CA09 N1 NA/CAV-F34, SL32 N2 NA/CAV-F34, SL32 N2 NA/CAV-F6, and dcMN24 N1 NA/CAV-F6 complexes were collected using FEI Titan Krios (Thermo Fisher Scientific) electron microscope operating at 300 kV with a Gatan K3 Summit direct electron detector (Gatan Inc.) at Tsinghua University. 1,754 movies were collected for the CA09 N1 NA/CAV-F34 complex, 1,962 movies were collected for the SL32 N2 NA/CAV-F34 complex, and 415 movies were collected for the SL32 N2 NA/CAV-F6 complex using the AutoEMation software⁶². These data were collected at a magnification of 81,000 with a pixel size of 1.0742 Å and at a defocus range between 1.3–1.8 μm. Each movie had a total accumulate exposure of 50 e⁻/Ų fractionated in 32 frames of 66 ms exposure. 3076 movies were collected for the dcMN24 N1 NA/CAV-F6 complex using the SerialEM software⁶³. The data was collected at a magnification of 29,000 with a pixel size of 0.97 Å and at a defocus range between 1.3–1.8 μm. Each movie had a total accumulate exposure of 50 e⁻/Ų fractionated in 32 frames of 66 ms exposure. Cryo-EM data of the rkDB16 N4 NA/CAV-F6 complex was collected using FEI Talos Arctica (Thermo Fisher Scientific) electron microscope operating at 200 kV with a Gatan K2 Summit direct electron detector (Gatan Inc.) at Tsinghua University. 320 movies were collected for the rkDB16 N4 NA/CAV-F6 complex using the SerialEM software⁶³. The data was collected at a magnification of 45,000 with a pixel size of 0.94 Å and at a defocus range between 1.3–1.8 μm. Each movie had a total accumulate exposure of 50 e⁻/Ų fractionated in 32 frames of 125 ms exposure.

The SL32 N2 NA/CAV-F6-R54S, and SL32 N2 NA/CAV-F6-D102N complexes were prepared as mentioned above. Cryo-EM data of these complexes were collected using FEI Tecnai Arctica (Thermo Fisher Scientific) electron microscope operating at 200 kV with a Gatan K2 Summit direct electron detector (Gatan Inc.) at Tsinghua University. 529 movies were collected for the SL32 N2 NA/CAV-F6-R54S complex and 1,440 movies were collected for the SL32 N2 NA/CAV-F6-D102N complex using the SerialEM software⁶³. These data were collected at a magnification of 39,000 with a pixel size of 0.836 Å and at a defocus range between 1.3-1.8 μm. Each movie had a total accumulate exposure of 50 e⁻/Ų fractionated in 32 frames of 100 ms exposure. Besides, cryo-EM data of the SL32 N2 NA/CAV-F6-R54S complex was also collected using FEI Titan Krios G3i (Thermo Fisher Scientific) electron microscope operating at 300 kV with a Gatan K3 Summit direct electron detector (Gatan Inc.) at Shuimu BioSciences Co., Ltd. to obtain a high-resolution structure. 3153 movies were collected for this complex using the SerialEM software⁶³. These data were collected at a magnification of 81,000 with a pixel size of 0.856 Å and at a defocus range between 1.3–1.8 μm. Each movie had a total accumulate exposure of 50 e⁻/Ų fractionated in 32 frames of 52 ms exposure.

Cryo-electron microscopy sample preparation, data collection and processing

MotionCor2 v.1.2.6⁶⁴ was used for beam-induced motion correction of whole frames in each movie, and GCTF v.1.18⁶⁵ was used to estimate the parameters of contrast transfer function (CTF) for each micrograph. Particles were automatically picked using cryoSPARC⁶⁶. And 1,611,425 particles for CA09 N1 NA/CAV-F34 complex, 4,228,701 particles for SL32 N2 NA/CAV-F34 complex, 639,868 particles for SL32 N2 NA/CAV-F6 complex, 152,517 particles for rkDB16 N4 NA/CAV-F6 complex, 3,937,205 particles for dcMN24 N1 NA/CAV-F6 complex, 998,044 particles for SL32 N2 NA/CAV-F6-R54S complex, and 692,709 particles for SL32 N2 NA/CAV-F6-D102N complex were extracted using cryoSPARC⁶⁶, which were used for subsequent 2D classification. These complexes also used cryoSPARC for subsequent data processing. After two or three rounds of 2D classification, the better classes were selected, and the selected particles were used to create 3D initial model and perform heterogeneous refinement. The proportion of different conformations of the SL32 N2 NA/CAV-F6-R54S, and SL32 N2 NA/CAV-F6-D102N complexes was calculated. Finally, a total of 242,813 particles for CA09 N1 NA/CAV-F34 complex, 113,270 particles for SL32 N2 NA/CAV-F34 complex, 46,554 particles for SL32 N2 NA/CAV-F6 complex, 17,312 particles for rkDB16 N4 NA/CAV-F6 complex, 545,144 particles for dcMN24 N1 NA/CAV-F6 complex, 170,595 particles for SL32 N2 NA/CAV-F6-R54S complex and 154,601 particles for SL32 N2 NA tetramer were applied to non-uniform refinement to generate density map and C4 symmetry was imposed for CA09 N1 NA/CAV-F34 complex, SL32 N2 NA/CAV-F6 complex and rkDB16 N4 NA/CAV-F6 complex. Based on the gold-standard Fourier shell correlation (FSC) cut-off of 0.143 criterion, the resolutions were 2.5 Å for CA09 N1 NA/CAV-F34 complex, 3.0 Å for SL32 N2 NA/CAV-F34 complex, 2.8 Å for SL32 N2 NA/CAV-F6 complex, 3.1 Å for rkDB16 N4 NA/CAV-F6 complex, 3.0 Å for dcMN24 N1 NA/CAV-F6 complex, 2.3 Å for SL32 N2 NA/CAV-F6 complex and 2.4 Å for SL32 N2 NA tetramer. Data collection and processing statistics of CA09 N1 NA/CAV-F34 complex, SL32 N2 NA/CAV-F34 complex, SL32 N2 NA/CAV-F6 complex, rkDB16 N4 NA/CAV-F6 complex, dcMN24 N1 NA/CAV-F6 complex, SL32 N2 NA/CAV-F6-R54S complex and SL32 N2 NA tetramer were listed in Table S2 and Table S3.

Model building and refinement

The CryoNet (https://aaai.org/ojs/index.php/AAAI/article/view/3918) was used to generate the initial atomic models for the CA09 N1 NA/CAV-F34 complex, SL32 N2 NA/CAV-F34 complex, SL32 N2 NA/CAV-F6 complex, rkDB16 N4 NA/CAV-F6 complex and dcMN24 N1 NA/CAV-F6 complex structures. The initial atomic model of SL32 N2 NA/CAV-F6-R54S complex and SL32 N2 NA tetramer is based on the structure of SL32 N2 NA/CAV-F6 complex. These atomic models fit into the final density maps well. Coot v.0.9.2⁶⁷ was subsequently used for manual adjustment and correction according to the protein sequences, map densities, Ramachandran plot, rotamers, bond geometry restraints and so on. The Real Space Refinement of PHENIX v.1.18.2⁶⁸ was also used to refine these structures. The quality of the final models was evaluated by PHENIX v.1.18.2⁶⁸. The validation statistics of these structural models were listed in Table S2 and Table S3. Figures were generated using UCSF Chimera v.1.16⁶⁹ and UCSF ChimeraX v.1.3⁷⁰.

Sequence conservation analysis

A total of 51117 full-length non-repeating H1N1 and H5N1 (1968-2025) and 42133 human H3N2 (1968-2025) NA protein sequences were downloaded from Global Initiative for Sharing Avian Influenza Data (GISAID; https://gisaid.org). N2-A/Switzerland/9715293/2013(H3N2) and N1-A/California/04/2009(H1N1) amino acid sequences were chosen as references in multiple sequences alignment using mafft tool⁷¹, respectively. Sequences (<0.05%) with insertions/deletions ratio greater than 10% were removed from alignment. Conservation of filtered sequences were showed with seqlogo plot using ggseqlogo package.

Prophylactic and therapeutic protection experiments

For seasonal influenza virus and H7N9 vaccine virus, BALB/c female mice at 6-8 weeks old (n = 5 mice/group) were anesthetized and intranasally infected with 5× median lethal dose (LD₅₀) of A/California/04/2009(H1N1), A/Guizhou/54/1989(H3N2), B/Colorado/06/2017(Victoria), MA-B/Florida/4/2006(Yamagata) or A/Anhui/1/2013 (H7N9) NIBRG-268. Mice were given the indicated antibody at a dose of 10 mg/kg intraperitoneally at 24 h before infection (prophylaxis) or 24 h(IBVs)/48 h(IAVs) after infection (therapeutics). Weight loss was monitored daily for 14 days. The humane endpoint was defined as a weight loss of 25% from initial weight on day 0.

In the prophylactic study against cattle_H5N1 virus, groups of five BALB/c female mice at 6–8 weeks of age were received intraperitoneally (i.p.) with a dose of 20 mg/kg of indicated mAb 24 h prior to intranasal challenge with 2× LD50 cattle_H5N1. For the therapeutic setting, the mice were given 20 mg/kg indicated mAb 24 h after infection with cattle_H5N1 virus. An equivalent volume of PBS was i.p. administered as a control. Body weights and survival rates of all mice were monitored and recorded for 14 days. Mice that displayed 25% or more weight loss were subsequently euthanized. The mice were kept in cages with individual ventilation under 65% humidity and an ambient temperature of 21–23 °C and a 12–12 h day-night cycle for housing and husbandry.

Declaration of the use of Artificial Intelligence (AI) assisted technology in the manuscript writing

The authors utilized ChatGPT 3.5 to enhance grammar and readability during the manuscript writing. Following this, the authors thoroughly reviewed and edited the article, assuming full responsibility for its content.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Acknowledgements

This work was supported in whole or in part by the National Key R&D Program of China (Grant number: 2021YFC2300100 to Y.-L.S. and Y.-Q.C., 2022YFC2304204 to Y.-Q.C.), Shenzhen Medical Research Fund: B2302044 to Y.-Q.C., Shenzhen Science and Technology Program (Grant number: RCJC20210706092009004, KQTD20200820145822023, ZDSYS20230626091203007 to Y.-Q.C. and JCYJ2020010914243811 to Y.-L.S.), CAMS Innovation Fund for Medical Sciences grant 2022-I2M-1-021 to Y.-L.S., Non-profit Central Research Institute Fund of Chinese Academy of Medical Sciences (2023-RC180-01 to J.G. and 2023-PT330-01 to Y.-L.S.), Research Grants Council of HKSAR Government (grants 27113524 and 17108223 to S.Y.), the National Natural Science Foundation of China (32171202 to X.W.) and Shenzhen Medical Research Fund: A2302039 to K.X.L. We thank the Experimental Teaching Center, School of Public Health (Shenzhen) of Sun Yat-sen University, the Tsinghua University Branch of China National Center for Protein Sciences (Beijing), and Shuimu BioSciences Co., Ltd. (Beijing) for technical support (Cryo-EM, Protein Preparation, Characterization and Biocomputing).

Author contributions

K.X.L., X.M.L., X.C.Z., S.N.L., and J.W.G. performed experiments and contributed to the experimental design. J.P.C. and S.F.Y. contributed to the experiment in vivo on HPAI H5N1. Y.Z.S., L.L., R.C., X.Y.C., X.Y.Y., Y.M.Z., and W.Y.L. constructed antigens and amplified flu strains. M.X.X., R.X.Z., H.J.L., R.Y.L., and H.T.M. participated in the preliminary sample collection. F.Y., S.F.C., L.D., H.L.L., and M.S. provided samples. Y.-Q.C., Y.L.S., and J.W.G. conceived the study design and supervised laboratory teams. K.X.L., X.M.L., and J.W.G. wrote the manuscript. Y.-Q.C., Y.-L.S., and X.Q.W. revised the manuscript. All authors read, edited, and approved the manuscripts.

Peer review

Peer review information

Nature Communications thanks Yohei Watanabe and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Data availability

The coordinates of CA09 N1 NA/CAV-F34 complex, SL32 N2 NA/CAV-F34 complex, SL32 N2 NA/CAV-F6 complex, rkDB16 N4 NA/CAV-F6 complex, dcMN24 N1 NA/CAV-F6 complex, SL32 N2 NA/CAV-F6-R54S complex, SL32 N2 NA tetramer and dcMN24 N1 NA tetramer have been deposited in the Protein Data Bank (PDB) with the accession numbers 8YVL, 8YVM, 8YVN, 8YVK, 9KOD, 8ZYX, 9IJV and 9L44, respectively; their corresponding maps have been deposited in the Electron Microscopy Data Bank (EMDB) with the accession numbers EMD-39606, EMD-39608, EMD-39609, EMD-39605, EMD-62473, EMD-60582, EMD-60644 and EMD-62805, respectively. Other structures for analysis, including 4GZQ, 4GZP, 8G3N, 6Q23 and 8ZR4, were obtained from the PDB.  are provided with this paper.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s41467-025-62040-1.

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

References

1. Krammer, F et al. Influenza. Nat. Rev. Dis. Prim.; 2018; 4, 3.

2. Fereidouni, S et al. Genetic characterization of a new candidate hemagglutinin subtype of influenza A viruses. Emerg. Microbes Infect.; 2023; 12, 2225645.

3. Zaraket, H; Hurt, AC; Clinch, B; Barr, I; Lee, N. Burden of influenza B virus infection and considerations for clinical management. Antivir. Res; 2021; 185, 104970.

4. Tian, H et al. Spatial, temporal and genetic dynamics of highly pathogenic avian influenza A (H5N1) virus in China. BMC Infect. Dis.; 2015; 15, 54.

5. Li, XL et al. Risk distribution of human infections with avian influenza H7N9 and H5N1 virus in China. Sci. Rep.; 2015; 5, 2015NatSR..518610L 18610.

6. Zhu, W et al. Epidemiologic, Clinical, and Genetic Characteristics of Human Infections with Influenza A(H5N6) Viruses, China. Emerg. Infect. Dis.; 2022; 28, pp. 1332-1344.

7. Peacock, T. et al. The global H5N1 influenza panzootic in mammals. Nature637, 304–313 (2024).

8. Morens, D. M., Taubenberger, J. K. & Fauci, A. S. H7N9 avian influenza A virus and the perpetual challenge of potential human pandemicity. mBio4, e00445-13 (2013).

9. Uyeki, TM et al. Highly pathogenic avian influenza A(H5N1) virus infection in a dairy farm worker. N. Engl. J. Med; 2024; 390, pp. 2028-2029.

10. Lin, TH et al. A single mutation in bovine influenza H5N1 hemagglutinin switches specificity to human receptors. Science; 2024; 386, pp. 1128-1134.

11. Good, MR et al. A single mutation in dairy cow-associated H5N1 viruses increases receptor binding breadth. Nat. Commun.; 2024; 15, 10768.

12. Gu, C. et al. A human isolate of bovine H5N1 is transmissible and lethal in animal models. Nature 636, 711–718 (2024).

13. Seto, JT; Rott, R. Functional significance of sialidose during influenza virus multiplication. Virology; 1966; 30, pp. 731-737.

14. Cohen, M et al. Influenza A penetrates host mucus by cleaving sialic acids with neuraminidase. Virol. J.; 2013; 10, 321.

15. Yang, X et al. A beneficiary role for neuraminidase in influenza virus penetration through the respiratory mucus. PLoS One; 2014; 9, e110026.2014PLoSO..9k0026Y

16. Wu, NC; Ellebedy, AH. Targeting neuraminidase: the next frontier for broadly protective influenza vaccines. Trends Immunol.; 2024; 45, pp. 11-19.

17. Daulagala, P et al. Avian Influenza A(H5N1) Neuraminidase Inhibition Antibodies in Healthy Adults after Exposure to Influenza A(H1N1)pdm09. Emerg. Infect. Dis.; 2024; 30, pp. 168-171.

18. Liang, Z et al. A(H2N2) and A(H3N2) influenza pandemics elicited durable cross-reactive and protective antibodies against avian N2 neuraminidases. Nat. Commun.; 2024; 15, 5593.

19. Chen, YQ et al. Influenza Infection in Humans Induces Broadly Cross-Reactive and Protective Neuraminidase-Reactive Antibodies. Cell; 2018; 173, pp. 417-429 e410.

20. Stadlbauer, D et al. Broadly protective human antibodies that target the active site of influenza virus neuraminidase. Science; 2019; 366, pp. 499-504.2019Sci..366.499S

21. Momont, C et al. A pan-influenza antibody inhibiting neuraminidase via receptor mimicry. Nature; 2023; 618, pp. 590-597.2023Natur.618.590M

22. Hansen, L et al. Human anti-N1 monoclonal antibodies elicited by pandemic H1N1 virus infection broadly inhibit HxN1 viruses in vitro and in vivo. Immunity; 2023; 56, pp. 1927-1938.e1928.

23. Lei, R et al. Leveraging vaccination-induced protective antibodies to define conserved epitopes on influenza N2 neuraminidase. Immunity; 2023; 56, pp. 2621-2634.e2626.

24. Lederhofer, J et al. Protective human monoclonal antibodies target conserved sites of vulnerability on the underside of influenza virus neuraminidase. Immunity; 2024; 57, pp. 574-586.e577.

25. Wolters, RM et al. Isolation of human antibodies against influenza B neuraminidase and mechanisms of protection at the airway interface. Immunity; 2024; 57, pp. 1413-1427.e1419.

26. Yasuhara, A et al. A broadly protective human monoclonal antibody targeting the sialidase activity of influenza A and B virus neuraminidases. Nat. Commun.; 2022; 13, 2022NatCo.13.6602Y 6602.

27. Wohlbold, TJ et al. Broadly protective murine monoclonal antibodies against influenza B virus target highly conserved neuraminidase epitopes. Nat. Microbiol; 2017; 2, pp. 1415-1424.

28. Broecker, F. et al. Extending the Stalk Enhances Immunogenicity of the Influenza Virus Neuraminidase. J Virol93 (2019).

29. Wang, M et al. A human monoclonal antibody targeting the monomeric N6 neuraminidase confers protection against avian H5N6 influenza virus infection. Nat. Commun.; 2024; 15, 8871.

30. Dormitzer, PR et al. Influenza vaccine immunology. Immunol. Rev.; 2011; 239, pp. 167-177.

31. Couch, RB. Seasonal inactivated influenza virus vaccines. Vaccine; 2008; 26, pp. D5-D9.

32. Eichelberger, MC; Monto, AS. Neuraminidase, the Forgotten Surface Antigen, Emerges as an Influenza Vaccine Target for Broadened Protection. J. Infect. Dis.; 2019; 219, pp. S75-S80.

33. Guzman Ruiz, L et al. Increased efficacy of influenza virus vaccine candidate through display of recombinant neuraminidase on virus like particles. Front Immunol.; 2024; 15, 1425842.

34. Kang, H. J. et al. Multiple neuraminidase containing influenza virus-like particle vaccines protect mice from avian and human influenza virus infection. Viruses14, 429 (2022).

35. Wan, H et al. Molecular basis for broad neuraminidase immunity: conserved epitopes in seasonal and pandemic H1N1 as well as H5N1 influenza viruses. J. Virol.; 2013; 87, pp. 9290-9300.

36. Tan, J et al. Human anti-neuraminidase antibodies reduce airborne transmission of clinical influenza virus isolates in the guinea pig model. J. Virol.; 2022; 96, e0142121.

37. He, B et al. Rapid isolation and immune profiling of SARS-CoV-2 specific memory B cell in convalescent COVID-19 patients via LIBRA-seq. Signal Transduct. Target Ther.; 2021; 6, 195.

38. Hurt, AC et al. Increased detection in Australia and Singapore of a novel influenza A(H1N1)2009 variant with reduced oseltamivir and zanamivir sensitivity due to a S247N neuraminidase mutation. Eur. Surveill.; 2011; 16, 19884.

39. Samson, M; Pizzorno, A; Abed, Y; Boivin, G. Influenza virus resistance to neuraminidase inhibitors. Antivir. Res; 2013; 98, pp. 174-185.

40. Kirkpatrick Roubidoux, E et al. Novel epitopes of the influenza virus N1 neuraminidase targeted by human monoclonal antibodies. J. Virol.; 2022; 96, e0033222.

41. Zhu, X et al. Structural basis of protection against H7N9 influenza virus by human anti-N9 neuraminidase antibodies. Cell Host Microbe; 2019; 26, pp. 729-738.e724.

42. Tulip, WR; Varghese, JN; Laver, WG; Webster, RG; Colman, PM. Refined crystal structure of the influenza virus N9 neuraminidase-NC41 Fab complex. J. Mol. Biol.; 1992; 227, pp. 122-148.

43. Wang, X et al. Identification of a broad-inhibition influenza neuraminidase antibody from pre-existing memory B cells. Cell Host Microbe; 2025; 33, pp. 151-166.e158.

44. Jo, G. et al. Structural basis of broad protection against influenza virus by a human antibody targeting the neuraminidase active site via a recurring motif in CDR H3. bioRxiv, https://doi.org/10.1101/2024.11.26.625467 (2024).

45. Lin, C., Yang, Q., Huang, L., Wang, X. & Zhang, Z. Elucidating the characteristics and clonal evolutionary trajectory of influenza neuraminidase broadly reactive B cell. bioRxiv, https://doi.org/10.1101/2025.02.23.639775 (2025).

46. da Silva, DV; Nordholm, J; Madjo, U; Pfeiffer, A; Daniels, R. Assembly of subtype 1 influenza neuraminidase is driven by both the transmembrane and head domains. J. Biol. Chem.; 2013; 288, pp. 644-653.

47. Bucher, DJ; Kilbourne, ED. A 2 (N2) neuraminidase of the X-7 influenza virus recombinant: determination of molecular size and subunit composition of the active unit. J. Virol.; 1972; 10, pp. 60-66.

48. Colman, PM; Varghese, JN; Laver, WG. Structure of the catalytic and antigenic sites in influenza virus neuraminidase. Nature; 1983; 303, pp. 41-44.1983Natur.303..41C

49. Ellis, D et al. Structure-based design of stabilized recombinant influenza neuraminidase tetramers. Nat. Commun.; 2022; 13, 2022NatCo.13.1825E 1825.

50. Yang, X; Kurteva, S; Ren, X; Lee, S; Sodroski, J. Subunit stoichiometry of human immunodeficiency virus type 1 envelope glycoprotein trimers during virus entry into host cells. J. Virol.; 2006; 80, pp. 4388-4395.

51. Gulati, U et al. Antibody epitopes on the neuraminidase of a recent H3N2 influenza virus (A/Memphis/31/98). J. Virol.; 2002; 76, pp. 12274-12280.

52. Kirkpatrick Roubidoux, E et al. Identification and characterization of novel antibody epitopes on the N2 neuraminidase. mSphere; 2021; 6, pp. e00958-20.

53. Wan, Z et al. Identification of amino acids in H9N2 influenza virus neuraminidase that are critical for the binding of two mouse monoclonal antibodies. Vet. Microbiol; 2016; 187, pp. 58-63.

54. Amaro, RE et al. Mechanism of 150-cavity formation in influenza neuraminidase. Nat. Commun.; 2011; 2, 2011NatCo..2.388A 388.

55. Li, Q et al. The 2009 pandemic H1N1 neuraminidase N1 lacks the 150-cavity in its active site. Nat. Struct. Mol. Biol.; 2010; 17, pp. 1266-1268.

56. Nachbagauer, R et al. A chimeric hemagglutinin-based universal influenza virus vaccine approach induces broad and long-lasting immunity in a randomized, placebo-controlled phase I trial. Nat. Med; 2021; 27, pp. 106-114.

57. Widge, AT et al. An influenza hemagglutinin stem nanoparticle vaccine induces cross-group 1 neutralizing antibodies in healthy adults. Sci. Transl. Med; 2023; 15, eade4790.

58. Liu, Y et al. Intranasal influenza virus-vectored vaccine offers protection against clade 2.3.4.4b H5N1 infection in small animal models. Nat. Commun.; 2025; 16, 3133.

59. Wen, K et al. Broad-spectrum humanized monoclonal neutralizing antibody against SARS-CoV-2 variants, including the Omicron variant. Front Cell Infect. Microbiol; 2023; 13, 1213806.

60. Hoehn, KB; Pybus, OG; Kleinstein, SH. Phylogenetic analysis of migration, differentiation, and class switching in B cells. PLoS Comput Biol.; 2022; 18, 2022PLSCB.18E9885H e1009885.

61. Chen, YQ; Lan, LY; Huang, M; Henry, C; Wilson, PC. Hemagglutinin Stalk-Reactive Antibodies Interfere with Influenza Virus Neuraminidase Activity by Steric Hindrance. J. Virol.; 2019; 93, pp. e01526-18.

62. Lei, J; Frank, J. Automated acquisition of cryo-electron micrographs for single particle reconstruction on an FEI Tecnai electron microscope. J. Struct. Biol.; 2005; 150, pp. 69-80.

63. Mastronarde, DN. Automated electron microscope tomography using robust prediction of specimen movements. J. Struct. Biol.; 2005; 152, pp. 36-51.

64. Zheng, SQ et al. MotionCor2: anisotropic correction of beam-induced motion for improved cryo-electron microscopy. Nat. Methods; 2017; 14, pp. 331-332.

65. Zhang, K. Gctf: Real-time CTF determination and correction. J. Struct. Biol.; 2016; 193, pp. 1-12.2016JSG..84..1Z

66. Punjani, A; Zhang, H; Fleet, DJ. Non-uniform refinement: adaptive regularization improves single-particle cryo-EM reconstruction. Nat. Methods; 2020; 17, pp. 1214-1221.

67. Emsley, P; Cowtan, K. Coot: model-building tools for molecular graphics. Acta Crystallogr D. Biol. Crystallogr; 2004; 60, pp. 2126-2132.2004AcCrD.60.2126E

68. Adams, PD et al. PHENIX: a comprehensive Python-based system for macromolecular structure solution. Acta Crystallogr D. Biol. Crystallogr; 2010; 66, pp. 213-221.2010AcCrD.66.213A

69. Pettersen, EF et al. UCSF Chimera-a visualization system for exploratory research and analysis. J. Comput Chem.; 2004; 25, pp. 1605-1612.

70. Pettersen, EF et al. UCSF ChimeraX: structure visualization for researchers, educators, and developers. Protein Sci.; 2021; 30, pp. 70-82.

71. Nakamura, T; Yamada, KD; Tomii, K; Katoh, K. Parallelization of MAFFT for large-scale multiple sequence alignments. Bioinformatics; 2018; 34, pp. 2490-2492.