Elimination of receptor binding by influenza hemagglutinin improves vaccine-induced immunity

The binding of influenza hemagglutinin (HA) to sialic acid (SA) receptors plays a well-defined role in shaping infection but the impact of such binding on vaccine responses has not yet been explored. We generated a virus-like particle (VLP) vaccine bearing the HA of H1N1 A/California/07/09 that is unable to bind to its α(2,6)-linked SA receptor (H1Y98F-VLP) and compared its immunogenicity and efficacy to a wild-type H1-VLP (H1WT-VLP) in mice. The H1Y98F-VLP elicited significantly stronger and more durable antibody responses (hemagglutination inhibition and microneutralization titers) and greater avidity maturation, likely attributable to improved germinal center formation. H1Y98F-VLP also resulted in a robust population of IL-2+TNFα+IFNγ− CD4+ T cells that correlated with antibody responses. Compared to H1WT-VLP vaccination, mice immunized with H1Y98F-VLP had 2.3-log lower lung viral loads and significantly lower pulmonary inflammatory cytokine levels 5 days post-challenge. These findings suggest that abrogation of HA-SA interactions may be a promising strategy to improve the quality and durability of influenza vaccine-induced humoral responses.


INTRODUCTION
Vaccination is widely recognized as the most effective way to prevent influenza infection and to reduce the societal and economic burden of seasonal influenza epidemics. However, influenza vaccines have remained largely unchanged since their introduction in the mid-1940s 1,2 despite significant year-to-year and strain-to-strain variation in effectiveness 3,4 . This inconsistency raises doubts about whether current vaccines would be suitably effective in the context of a pandemic and highlights the need for novel and more reliable approaches to influenza vaccination.
Influenza hemagglutinin (HA) is a trimeric glycoprotein on the surface of all influenza viruses that initiates infection by binding to sialic acid (SA) receptors on the surface of respiratory epithelial cells 5 . Antibodies that bind to the receptor-binding domain (RBD) of HA can block this initial interaction with the host cell and are the basis for the most widely used correlate of protection for influenza infection 6,7 . HA proteins are therefore the major and, in some cases, the only antigens in all commercial influenza vaccines 8 . The binding properties of influenza HA are strainspecific and have been extensively studied in the context of disease severity and transmissibility 5,[9][10][11] . However, the implications of HA binding properties on influenza vaccine immunogenicity and efficacy have not been investigated despite the presence of SA receptors on the surface of cells throughout the body [12][13][14] . We previously demonstrated that the strain-specific binding properties of influenza HA influence the pattern of interaction of HA-bearing virus-like particles (VLP) with human peripheral blood mononuclear cells (PBMC) in vitro 15 . These interactions were driven by differential expression of α2,3and α2,6-linked SA on human PBMC and strongly influenced downstream innate immune activation 15 , raising the possibility that the binding properties of HA may be important modulators of influenza vaccine immunogenicity.
In the current work, we sought to determine whether HA-SA interactions influence vaccine responses in vivo. We generated a plant-based HA-VLP vaccine bearing H1 (A/California/07/2009) that was unable to bind SA and compared its immunogenicity and efficacy to the wild-type (WT) H1-VLP using a murine model. Mutation of H1 binding was achieved by a single amino acid substitution from tyrosine to phenylalanine at residue 98 (Y98F, H3 numbering), which prevents HA-SA interactions by eliminating the hydroxyl group required for hydrogen bonding with SA. This mutation was originally described by Martín et al. using H3 (A/ Aichi/2/68) 16 but has subsequently been shown to prevent SA binding of many influenza A HAs, including H1 (A/California/07/ 2009), due to its position within a conserved region of the RBD 17 . This mutation does not affect the integrity of the RBD or HA folding 17,18 making the H1 Y98F -VLP a good candidate for studying the possible impact of HA-SA interactions on vaccine responses. Herein, we demonstrate that elimination of HA-SA interactions significantly improves both the immunogenicity and efficacy of a plant-based H1-VLP vaccine.

RESULTS
Generation and validation of H1 Y98F -VLP VLP composed of a plant-lipid bilayer studded with the WT H1 (H1 WT -VLP) or Y98F H1 (H1 Y98F -VLP) were expressed in Nicotiana benthamiana as previously described using a 2X35S/CPMV160/ NOS expression system [19][20][21] (Fig. 1a). Expression of HA in crude leaf digests was confirmed by western blot (Fig. 1b). Following purification, VLP preparations were separated by gel electrophoresis and visualized using Coomassie G-250 staining to evaluate the protein composition and purity (Fig. 1c). HA was predominantly expressed in its uncleaved form (HA0), however, faint bands corresponding to cleaved HA (HA1 and HA2) and HA dimers were observed. The purity of the HA-VLP products was determined by analysis of the densitometry profile of each protein band and was comparable between formulations (~95%). VLP size (~100 nm) and morphology were consistent with previous reports 20, 22 and were unaffected by the Y98F mutation (Fig. 1d).
With multiple active HA trimers on their surface, H1 WT -VLPs are capable of mediating hemagglutination by binding to SA on the surface of red blood cells (RBC). Introduction of the Y98F mutation prevented hemagglutination, indicating that SA binding was greatly reduced or abolished (Fig. 1e). Furthermore, H1 Y98F -VLP failed to agglutinate human PBMC ( Supplementary Fig. 1a). Incubation of PBMC with H1 Y98F -VLP dramatically reduced polyclonal B cell activation that occurs following binding of the H1 WT -VLP 15,23 and increased T cell activation ( Supplementary Fig.  1b-d), indicating that reduced binding can modulate downstream immune responses. To confirm that the Y98F mutation inhibits HA-SA interactions on a molecular level, we measured binding of H1 WT -and H1 Y98F -VLP to immobilized SA by surface plasmon resonance (SPR). The slope of the association curve was 1.66 resonance units per second (RU/s) for H1 WT -VLP and 0.3 RU/s for H1 Y98F -VLP (Fig. 1f). Importantly, H1 Y98F -VLP was tested at a much higher concentration than H1 WT -VLP (44x). When the slopes were adjusted for HA content, a 99.6% reduction in binding was attributable to the Y98F mutation (H1 0.59 RU/s; Y98F H1 0.002 RU/s) (Fig. 1g). Taken together, these data suggested that H1 Y98F -VLP was an appropriate tool to evaluate the role of HA-SA interactions in the context of influenza vaccine responses. H1 Y98F -VLP elicits stronger and more durable humoral responses To establish whether HA-SA interactions influence the humoral immune response to vaccination in mice, we measured the development of H1-specific antibodies in sera following vaccination with 1 dose of H1 WT -VLP or H1 Y98F -VLP (3 µg HA/dose) or an equivalent volume of PBS (placebo). Total H1-specifc IgG was measured by ELISA and functional antibodies were quantified using the hemagglutination inhibition (HI) assay to measure antibodies that block the binding of live virus to avian RBCs 24 and the microneutralization (MN) assay to measure antibodies that prevent infection of Madin-Darby canine kidney (MDCK) cells 25 . Antibodies were quantified at 21 days post-vaccination (dpv) to characterize pre-challenge humoral responses and on a monthly basis (1-3 m, 7 m) to evaluate the kinetics and durability of the antibody responses.
Immunization with H1 WT -VLP or H1 Y98F -VLP resulted in comparable H1-specific IgG titers by ELISA at all time points (Fig. 2a) but there were marked differences in antibody functionality. Most notably, vaccination with H1 Y98F -VLP resulted in significantly higher HI and MN titers at 21 dpv (p = 0.002 and p < 0.001, respectively) (Fig. 2b-c). Titers increased until 3 months post- Fig. 1 Y98F mutation abrogates SA binding without affecting HA expression or VLP structure. Wild-type (WT) H1 (H1 WT ) and Y98F H1 (H1 Y98F ) were expressed in Nicotiana benthamiana. a Representation of the H1 WT (top) and H1 Y98F (bottom) expression cassettes. 2X35S/ CPMV160 promoter, double 35S promoter fused to the 5′ UTR of a cowpea mosaic virus (CPMV) expression enhancer; SpPDI signal peptide from alfalfa protein disulfide isomerase, NOS nopaline synthase terminator signal. b Expression of HA was confirmed by SDS-PAGE of crude leaf extracts followed by immunoblot analysis. Commercially available recombinant H1 expressed in HEK-293 cells (recH1, Immune Technologies) was included as a positive control (1 µg). 1°ab: rabbit polyclonal anti-H1 1:500 (Cat. No. IT-003-SW, Immune Technology); 2°ab: horseradish peroxidase-conjugated goat anti-rabbit IgG 1:20000 (Cat. No. IT-200-01, Immune Technology). c VLP composition and purity were evaluated by SDS-PAGE of purified leaf digests followed by Coomassie G-250 staining. d Representative TEM images show the similar size and morphology of H1 WT -and H1 Y98F -VLP. Images were acquired using a Tecnai G2 Spirit transmission electron microscope. Scale bar = 100 nm. e Sialic acid (SA) binding was evaluated based on hemagglutination of turkey red blood cells following incubation (30 min) with serial twofold dilutions of H1 WT -and H1 Y98F -VLP (starting at 1:150 and 1:10, respectively). SA binding was further quantified by SPR. f SPR sensorgram showing the binding response of H1 WT -(red) and H1 Y98F -VLP (blue) to α-2,6 SA captured on a streptavidin-coated chip surface. g Relative binding of H1 Y98F -VLP when adjusted for HA content.
vaccination (mpv) in both groups but were consistently higher among mice vaccinated with H1 Y98F -VLP. Furthermore, H1 Y98F -VLP resulted in improved durability of HI titers, which declined dramatically between 3 and 7 mpv in mice vaccinated with H1 WT -VLP. As a result, HI titers were >4-fold higher among mice vaccinated with H1 Y98F -VLP at 7 mpv (p = 0.029) (Fig. 2b). MN titers were better maintained in the H1 WT -VLP group at 7 mpv but still declined in 4/8 animals and remained significantly lower than in mice vaccinated with H1 Y98F -VLP (p = 0.029) (Fig. 2c). H1-specific antibodies were not detected in the placebo group.
Sustained production of high avidity IgG is thought to be mediated by plasma cells (PC) residing in the bone marrow (BM) 26 . To evaluate whether these cells contribute to the improved durability of antibody responses in mice vaccinated with H1 Y98F -VLP, we quantified H1-specific IgG-producing BM PC at 7 mpv by enzyme-linked immune absorbent spot (ELISpot) assay. Few H1specific BM PC were detected in the placebo group (2 ± 1 PC/1 × 10 6 cells). H1-specific BM PC were detected in both vaccine groups and the frequency was higher among mice vaccinated with H1 Y98F -VLP, although this difference did not reach statistical significance (Fig. 2e). Interestingly, the frequency of H1-specific BM PC seemed to correspond with the durability of antibody responses: MN titers declined between 3 and 7 mpv in all mice with <10 PC/1 × 10 6 BM cells and were maintained in animals with ≥10 PC/1 × 10 6 BM cells regardless of vaccine group (Fig. 2e). Furthermore, the frequency of H1-specific BM PC significantly correlated with IgG (r s = 0.704, p = 0.003), HI (r s = 0.604, p = 0.015) and MN titers (r s = 0.657, p = 0.007) at 7 mpv (Fig. 2f). Sera were collected at day 21 (left panel) or on a monthly basis (right panel) to measure (a) total H1-specific IgG by ELISA, (b) hemagglutination inhibition titers and (c) microneutralization titers. d Avidity indices of sera obtained 2-7 mpv following incubation with 8 M urea. e H1-specific IgG-producing plasma cells (PC) in the bone marrow measured by ELISpot. Representative wells from each group are shown on the right. f Spearman's rank correlation technique was applied to evaluate the relationship between the frequency of PC and IgG titers (left), HI titers (middle) and MN titers (right). Error bars represent the mean ± SEM. At day 21, N = 40-70/group and data are pooled from six independent experiments. For long-term studies, N = 7-8/group. Statistical significance between vaccine groups was determined by Mann-Whitney test (*p < 0.033, **p < 0.01, ***p < 0.001). Statistical significance between time points within the same group was determined by two-way ANOVA with the Geisser-Greenhouse correction and Sidak's multiple comparisons ( ♦ p < 0.033).
Although adults typically receive a single dose of inactivated influenza vaccine each year, children who are immunologically naïve to influenza require two doses for an adequate immune response 27 . Because laboratory mice are immunologically naive, we evaluated the humoral responses in mice given two doses of H1 WT -VLP or H1 Y98F -VLP administered 21 days apart. By 28 days post-boost, HI titers were significantly higher among mice vaccinated with H1 Y98F -VLP (p = 0.02) ( Supplementary Fig. 2a) and a similar trend was observed in MN titers ( Supplementary Fig.  2b). While no significant differences were observed regarding H1specific IgG titers ( Supplementary Fig. 2c) immunization with two doses of H1 Y98F -VLP resulted in significantly higher IgG avidity ( Supplementary Fig. 2d, p = 0.03 at 4 M and 6 M urea) and H1specific IgG-producing PC in the BM by 28 days post-boost (p = 0.02) ( Supplementary Fig. 2e). Once again, the frequency of BM PC was strongly correlated with IgG titers, HI titers, MN titers, and IgG avidity in both vaccine groups ( Supplementary Fig. 2f). Improved humoral responses were also observed among mice vaccinated with two doses of a plant-based non-binding H1 Y98F -VLP based on the H1 A/Idaho/07/2018 sequence ( Supplementary Fig. 3a-c) and human cell culture-produced H1 Y98F trimers based on the H1 A/ Brisbane/02/2018 sequence ( Supplementary Fig. 4), suggesting that HA-SA interactions broadly impact humoral responses to H1.  (Fig. 3a). However, striking differences in T FH frequencies and kinetics were observed (Fig. 3b). H1 Y98F -VLP resulted in more rapid induction of T FH cells that were maintained until 19 dpv. In contrast, the H1 WT -VLP resulted in a gradual expansion of T FH cells until 13 dpv followed by a rapid decline. As a result, the frequency of T FH cells was significantly higher among H1 Y98F -VLP-vaccinated mice in early and late GCs (7 dpv p = 0.017; 19 dpv p = 0.03). No increase in GC B cells or T FH cells was observed at any time point among control mice injected with PBS ( Supplementary Fig. 6).
Given the importance of T FH cells in avidity maturation of GC B cells, we next sought to determine whether vaccination with H1 Y98F -VLP results in improved recognition of the H1 antigen among GC B cells. Antigen-specific GC B cells were distinguished based on their ability to bind H1 Y98F -VLP in vitro and were detected using fluorescently labeled anti-H1. This strategy allows for reliable detection of cognate B cells while avoiding SAmediated interactions with non-cognate cells. H1-specific GC B cells were quantified prior to vaccination and at intervals corresponding to early (7 dpv), peak (13 dpv) and late (19 dpv) GC evolution (Fig. 3c, d). Prior to vaccination GC B cells were rare in all groups (<1% of B cells) and did not bind the H1 Y98F -VLP probe (Fig. 3c). Following vaccination, H1-specific GC B cells were readily detected in both vaccine groups, however, H1 Y98F -VLP resulted in an increased frequency of these cells within the GC at all time points. This effect was most pronounced at 19 dpv when the frequency of H1-specific B cells in the GC was~30% higher among mice vaccinated with H1 Y98F -VLP compared to H1 WT -VLP (p = 0.011) (Fig. 3d). Among non-GC B cells (CD19 + GL7 − Fas − ), the frequency of H1-specific cells was similar between vaccine groups and was comparable to pre-vaccination levels ( Supplementary Fig.  7). Taken together, H1 Y98F -VLP favors a more antigen-specific GC response characterized by improved maintenance of T FH cells and expansion of H1-specific GC B cells.
CD4 + T cell response to H1 Y98F -VLP correlates with antibody responses Antigen-specific CD4 + T cells in the spleen and BM were quantified by flow cytometry following vaccination with H1 WT -VLP or H1 Y98F -VLP. Freshly isolated cells were stimulated with H1 WT -VLP (18 h) or a pool of 131 overlapping peptides (15aa) spanning the H1 sequence (6 h) and responding cells were characterized as antigen-experienced (CD44 + ) CD4 + T cells expressing IL-2, IFNγ and/or TNFα (see Supplementary Fig. 8 for full gating strategy). The frequencies of H1-specific CD4 + T cells observed following stimulation with either the H1 WT -VLP or the H1 peptide pool were highly correlated (r s = 0.5344, p < 0.0001) and had similar cytokine signatures ( Supplementary Fig. 9). However, the magnitude of the response was greater upon stimulation with the H1 WT -VLP, allowing for more reliable identification of rare populations. Thus, functional signatures were evaluated in cells stimulated with H1 WT -VLP.
A single 3 µg dose of H1 WT -VLP or H1 Y98F -VLP elicited significant populations of H1-specific CD4 + T cells in the spleen compared to placebo at 28 days post-vaccination (p = 0.01 and p = 0.009, respectively). There were no differences in the magnitude or functional signatures between vaccine groups (Fig. 4a). Both vaccines elicited significant populations of IL-2 + TNFα + IFNγ − CD4 + T cells (H1 WT p = 0.009; H1 Y98F p = 0.01) and IFNγ single-positive cells compared to the placebo group (H1 WT p = 0.02; H1 Y98F p = 0.002). In contrast, vaccination with two 0.5 µg doses 21 days apart resulted in marked differences in CD4 + T cell signatures between vaccine groups (Fig. 4b). In the spleen, the IL-2 + TNFα + IFNγ − population continued to account for the majority of polyfunctional cells. However, this population was significantly larger in mice vaccinated with H1 Y98F -VLP (p = 0.01). Conversely, IFNγ + cells were more prevalent among H1-specific CD4 + T cells elicited by H1 WT -VLP. It should be noted that the frequency of IFNγ + cells was low in both groups despite substantial IFNγ + populations elicited by a single 3 µg dose of either vaccine. This discrepancy likely reflects the lower antigen dose in the two-dose regimen, as the magnitude of the CD4 + T cell response to a plant-based H1 WT -VLP is dose dependent in humans 28 . As expected, IFNγ + and polyfunctional populations were markedly increased in mice that received two 3 µg doses, but functional signatures remained unchanged ( Supplementary Fig. 10). With the exception of the frequency and relative proportion of single-positive IL-2 CD4 + T cells, similar signatures were also observed following vaccination with H1 WT -and H1 Y98F -VLPs targeting H1 A/Idaho/07/2018 (Fig.  S3d, e).
Antigen-specific CD4 + T cells were also measured in the BM, which serves as a major reservoir for long-term maintenance of memory CD4 + T cells following vaccination 29 . Immune cells were isolated from bilateral femurs 28 days post-boost and evaluated in parallel with splenocytes. Overall, cytokine signatures in the BM resembled those observed in the spleen. However, only the H1 Y98F -VLP resulted in a significant increase in the frequency of H1-specific CD4 + T cells compared to the placebo group (p = 0.03) (Fig. 4c). The frequency of H1-specific CD4 + T cells in the BM strongly correlated with the frequency of responding cells in the spleen in mice vaccinated with H1 Y98F -VLP (r s = 1.00, p = 0.0167) but not H1 WT -VLP (r s = −0.1, p = 0.95), suggesting that a defect in recruitment contributes to the low frequency of H1-specific CD4 + T cells in the BM of H1 WT -VLP-vaccinated mice (Fig. 4d). Similar to the spleen, the response to H1 Y98F -VLP was dominated by IL-2 + TNFα + IFNγ − CD4 + T cells and IFNγ-expressing cells were virtually absent. Conversely, nearly 30% of BM CD4 + T cells elicited by H1 WT -VLP were IFNγ + , and IL-2 + TNFα + IFNγ − cells were underrepresented in the BM compared to the spleen (24 vs. 32% of responding cells) (Fig. 4c).
Given that CD4 + T cells with the IL-2 + TNFα + IFNγ − phenotype have been shown to correlate with antibody responses to a number of vaccine antigens 30,31 , we sought to determine whether expansion of this population correlated with improved avidity maturation and functional antibody titers among mice vaccinated with H1 Y98F -VLP. In the spleen, the frequency of IL-2 + TNFα + IFNγ − cells correlated with IgG avidity maturation in mice vaccinated with H1 Y98F -VLP (r s = 0.6809, p = 0.0355) but not H1 WT -VLP (r s = 0.07976, p = 0.8355) (Fig. 4e). However, in the BM, this T cell population was strongly correlated with HI titers in both vaccine groups (r s = 0.7957, p = 0.008) (Fig. 4f). This population also significantly correlated with MN titers (r s = 0.7431, p = 0.0169) and IgG avidity (r s = 0.8293, p = 0.0045), but not total IgG (r s = 0.3598, p = 0.3054). Taken together, these findings suggest that enhanced recruitment of IL-2 + TNFα + IFNγ − CD4 + T cells to the BM may contribute to improved functional antibody responses in mice vaccinated with H1 Y98F -VLP (Fig. 4f).
H1 Y98F -VLP results in reduced viral load and pulmonary inflammation following homologous challenge In a study to compare protection following vaccination with H1 WTand H1 Y98F -VLP, mice were challenged with 1.6 × 10 3 times the median tissue culture infectious dose (TCID 50 ) of H1N1 (A/ California/07/09) 28 d post-vaccination with a single dose of 3 µg H1 WT -or H1 Y98F -VLP or an equivalent volume of PBS. Infection resulted in substantial weight loss (17.3 ± 1.3% at day 5) and 64.3% mortality in the placebo group. Consistent with previous reports 32 , vaccination with H1 WT -VLP resulted in complete protection from lethal infection. Similarly, all mice vaccinated with H1 Y98F -VLP survived and there was no significant difference in post-infection weight loss between the vaccinated groups (4-6% at day 5) (Fig. 5a). However, significant differences in the rate of viral clearance and pulmonary inflammation were observed.
To evaluate viral clearance, lungs were collected at 3 days postinfection (dpi) or 5 dpi and the TCID 50 of lung homogenates was determined (Fig. 5b). Vaccination with H1 Y98F -VLP, but not H1 WT -VLP, resulted in a decrease in the viral titer compared to placebo at 3 dpi (p < 0.001). In both groups, viral titers were inversely correlated with HI and MN titers (HI r s = −0.695, p = 0.001; MN r s = −0.7067, p = 0.001). By 5 dpi, viral titers in mice vaccinated with H1 Y98F -VLP were nearly 2-log lower than mice vaccinated with H1 WT -VLP (p = 0.03) and there was no significant difference between the placebo and H1 WT -VLP-vaccinated mice (p = 0.31).
Lung homogenates were evaluated by multiplex ELISA (Quansys) to quantify several cytokines (TNFα, IFNγ, IL-1α, IL-1β and IL-6) and chemokines (MCP-1, MIP-1α and RANTES) implicated in influenza-mediated acute lung injury (Fig. 5c, d) 33,34 . Baseline levels were established using lung homogenates from mockinfected mice. Overall, vaccination with either H1 WT -VLP or H1 Y98F -VLP resulted in reduced cytokines and chemokines compared to the placebo group at 3 dpi, but there were no significant differences between the vaccinated groups. However, by 5 dpi there was a marked divergence in pulmonary inflammation and all cytokine/chemokine levels except for TNFα were significantly lower in the H1 Y98F -VLP vaccinated mice compared to the H1 WT -VLP group. Strikingly, mice vaccinated with H1 WT -VLP exhibited a significant increase in IFNγ compared to 3 dpi (p = 0.005) with levels nearing the placebo group (1570 ± 315 and 1887 ± 367 pg/ ml, respectively). IFNγ levels correlated with the viral titer (r s = −0.7050, p = 0.001) in both vaccinated groups and remained near the baseline in mice that received H1 Y98F -VLP (172 ± 46 pg/ml). IL-1β, MIP-1α and RANTES levels followed similar trends. Consistent with these findings, histopathological evaluation of lungs collected at 4 dpi revealed that mice vaccinated with H1 Y98F -VLP had less pulmonary inflammation compared to H1 WT -VLP-vaccinated mice and more closely resembled mock-infected animals (Fig. 5e). Together, these findings suggest that while both vaccines protected from lethal influenza infection, mice vaccinated with H1 Y98F -VLP were better able to control and clear the virus and exhibited reduced influenza-associated lung inflammation.

DISCUSSION
Vaccination is the most effective strategy to prevent influenzaassociated morbidity and mortality, however, vaccine effectiveness is often limited by variable immunogenicity and rapidly Color-matched pie charts depict relative distributions of cytokine-producing CD4 + T cell populations and IFNγ + populations are highlighted by a green arc. Statistical significance was determined by Kruskal-Wallis test with Dunn's multiple comparisons (total response) or two-way ANOVA with Tukey's multiple comparisons (cytokine signatures) (*p < 0.033, **p < 0.01, ***p < 0.001). Spearman's rank correlation technique was applied to evaluate the relationship between (d) the frequency of H1-specific CD4 + T cells in the spleen and BM, e the frequency of IL-2 + TNFα + IFNγ -CD4 + T cells in the spleen and IgG avidity index (4 M urea) and f the frequency of IL-2 + TNFα + IFNγ − CD4 + T cells in the BM and HI titer (*p < 0.033, **p < 0.01).
waning protection 35,36 . We have demonstrated that SA binding properties of influenza HA influence vaccine responses and that ablation of HA-SA interactions can improve the quality and durability of immune responses in mice. Non-binding H1 mutants were generated using the previously described Y98F mutation, which prevents SA binding without affecting antibody recognition of the globular head or RBD 17 . Although historically used as probes to identify HA-specific B cells 17,23,37 and more recently in studies targeting the generation of stem-specific antibodies 38 Antibodies that mediate HI have long been shown to provide protection and are widely considered to be a good predictor of vaccine efficacy. Thus, the induction of high HI titers is important for licensure of influenza vaccines in many jurisdictions [40][41][42] . MN titers also correlate with protection and provide useful insight into the functional neutralizing capacity of vaccine-induced antibodies 43 . In this study we demonstrated that ablation of HA-SA binding has broad impact on the humoral response to vaccination. Consistent with previous reports, we observed a strong correlation between HI and MN titers 43 and pre-infection antibody were inversely correlated with pulmonary peak viral load (3 dpi) in both vaccine groups, suggesting that these antibodies play a direct role in controlling infection. While viral titers in the lungs do not always correlate with morbidity 44,45 , the strong inverse correlations between viral titers and pulmonary inflammatory cytokines and infiltrates suggest that the viral load is a good reflection of disease severity in this model. The improved functional antibody response elicited by H1 Y98F -VLP is therefore likely to have played an important role in reducing the severity of infection.
Our data suggest that eliminating HA-SA binding may also improve the durability of the antibody response to influenza vaccines. Vaccine-induced HI titers typically decline rapidly postvaccination [46][47][48][49] , consistent with our observations in mice that received H1 WT -VLP. In sharp contrast, HI and MN titers in mice vaccinated with H1 Y98F -VLP were maintained at peak levels until 7 mpv. While the mechanisms underlying this observation are under further investigation, improved specificity and longevity of the GC reaction and an increase in HA-specific PC in the BM likely play an important role. BM PC can persist for decades following antigen exposure in humans and are thought to mediate long-term maintenance of serum antibody titers following vaccination 18,50 . Our finding that increased frequencies of BM PC correlate with HI, MN, and total IgG titers at 7 mpv suggests that this population is central to the durability of humoral response elicited by H1 Y98F -VLP. Weisel et al. have shown that long-lived PC in the BM arise from late GC B cells (18-20 days) after a prolonged period of somatic hypermutation and avidity maturation 51 . The increased HA-specific BM PC we found in the H1 Y98F -VLP group is therefore likely a reflection of the increased HA-specific B cells and T FH cells in the GC of these animals at 19 dpv.
Although influenza vaccine developers have focused primarily on humoral responses, antigen-specific T cells undoubtedly play important roles in protection and often generate more broad and cross-reactive responses than antibodies 52 . CD4 + T cells support both B cell affinity maturation and CD8 + T cell responses and can mediate protection in a murine vaccination/challenge model in the absence of neutralizing antibodies 53 . Previous studies have demonstrated that plant-based H1 WT -VLPs elicit strong CD4 + T cell responses in both mice and humans with increased CD4 + T cell polyfunctionality compared to inactivated vaccines 32,54-57 . The overall pattern of CD4 + T cell responses in this study was consistent with previous reports and was not compromised in the absence of HA-SA interactions. However, there were no differences between vaccine groups after a single dose, suggesting that enhanced viral clearance in this model was driven primarily by the better humoral response to the H1 Y98F -VLP vaccine. Interestingly, two doses of H1 Y98F -VLP resulted in significant expansion of the IL-2 + TNFα + IFNγ − population, which was strongly correlated with humoral responses. Others have demonstrated that this population is comprised of Th1 cells and a population of primed but uncommitted T helper cells (Thpp) with high proliferative and differentiation potential 58 . Thpp cells are frequently generated in primary responses to antigens including influenza 58 , thus, it is not surprising that both vaccines elicited similar frequencies of IL-2 + TNFα + IFNγ − CD4 + T cells after a single dose. However, subsequent exposure to influenza HA typically results in expression of IFNγ 58,59 . Expansion of the IFNγ − population in the spleen upon boosting with H1 Y98F -VLP but not the H1 WT -VLP is reminiscent of the CD4 + T cell responses to protein vaccines such as hepatitis B, diphtheria and tetanus, that are dominated by Thpp cells and elicit robust and durable antibody responses 30,31,60 . The Thpp bias induced by the "non-binding" HA and concomitantly higher HI titers were also observed with H1 A/Idaho/07/ 2018, suggesting that the impact on CD4 + T cells could broadly influence H1 humoral responses.
To our knowledge, Thpp cells have not been described in the BM and it is unknown whether this population is a common feature of other protein vaccines. However, BM CD4 + T cells are a major reservoir for long-lasting immunity and are known to provide efficient help to B cells 29,61 . Thus, enhanced homing of CD4 + T cells to the BM may contribute to the enhanced durability of the responses we observed following H1 Y98F -VLP vaccination. Interestingly, IL-2 + TNFα + IFNγ − CD4 + T cells in the BM significantly correlated with HI titers, MN titers and IgG avidity, but not total IgG titers. One possible explanation for this discrepancy is that the IL-2 + TNFα + IFNγ − CD4 + T cells play a role in B cell selection, giving rise to antibodies with increased avidity and functional capacity.
Beyond the possibility that Y98F HAs behave similarly to traditional protein vaccines, the mechanisms underlying their improved immunogenicity are not yet clear. However, a likely contributor is better trafficking to the draining LN in the absence of SA binding. Terminal α2,3 and α2,6 SAs are widely distributed throughout the body, including on skeletal muscle cells 12,62,63 . Thus, the H1 WT -VLP and recombinant WT HAs may be more likely to be sequestered at the site of injection or en route to the draining LN due to HA-SA interactions. In support this hypothesis, previous work has demonstrated that while eGFP-labeled H5-VLP can be detected in the popliteal LN 10 min after footpad injection 64 , a substantial eGFP signal remains at the site of injection for up to 24 h. Soluble eGFP was not retained in the footpad, suggesting that SA binding may contribute to injection site retention 65 . Although depot-type adjuvants that "trickle" antigen into the draining LN can enhance affinity maturation through prolonged antigen availability in the GC 66 , the rapid decline in T FH cells suggests that this is not occurring in mice vaccinated with H1 WT -VLP. It is possible that H1 WT -VLPs retained at the injection site are degraded prior to delivery to the draining LN. Fig. 5 H1 Y98F -VLP improves viral clearance and reduces pulmonary inflammation upon lethal challenge. Female Balb/c mice were challenged with 1.6 × 10 3 TCID 50 of H1N1 (A/California/07/09) 28 days post-vaccination with 3 µg H1 WT -or H1 Y98F -VLP or an equivalent volume of PBS. a Mean weight loss (left) and survival (right) were monitored for 12 dpi (n = 12-14/group, error bars represent SEM). Mice weighing <80% of their pre-challenge weight were euthanized. A subset of challenged mice (n = 9/group) were euthanized at 3 dpi and 5 dpi for evaluation of the viral load and pulmonary inflammation. b Viral titers in the supernatant of lung homogenates calculated using the Karber method and reported as TCID 50 /100 µl supernatant (GMT ± 95% CI). c Concentrations of cytokines and chemokines in the supernatant of lung homogenates measured by multiplex ELISA (mean ± SEM). The horizontal line represents the mean of mock-infected mice as a baseline. d Radar plots showing the cytokine profiles of mock-infected and infected lungs at 3 dpi and 5 dpi. e H&E stains of lungs collected 4 dpi (×10 magnification, scale bar = 100 µm). Statistical significance for b and c were determined by Kruskal-Wallis test with Dunn's multiple comparisons. Comparisons between groups at the same time point are represented by * (*p < 0.033, **p < 0.01, ***p < 0.001) and comparisons within the same group over time are represented by ♦( ♦♦ p < 0.01, ♦♦♦ p < 0.001).
Thus, retention of H1 WT antigen at the site of injection may reduce the concentration of antigen delivered to draining LN without providing sustained antigen presentation. In addition, HA-SA interactions may limit immune recognition of important neutralizing epitopes on the globular head of HA through steric hindrance. Ongoing investigations aim to characterize the impact of SA binding on trafficking of the H1 antigen to the draining LN and epitope specificity of the immune response.
In addition to these mechanistic studies, we are also evaluating the generalizability of these findings with respect to different influenza strain HAs and different production platforms. The fact that a H1 Y98F -VLP targeting an antigenically distinct H1 (A/Idaho/ 07/2018) has similar immune effects suggests that eliminating HA-SA interactions may be a promising strategy to improve immunogenicity of vaccines against current and emerging strains of influenza. Confirmatory studies with additional non-binding VLPs (e.g.: H3N2 and B strains) will be required to evaluate the broad generalizability of this approach. The fact that improved humoral responses were observed with soluble H1 Y98F trimers (A/ Brisbane/02/2018) suggests that this strategy may confer similar benefits across recombinant vaccine platforms. Although eggbased propagation of influenza A strains bearing Y98F HA has been described 16 , mutations that restore HA binding may limit the viability of this approach in platforms requiring the growth of live virus 67 . Finally, since receptor-binding proteins are often targets of choice for viral vaccines and many of the receptors for these viruses have wide tissue distribution, our observations raise the possibility that manipulation of non-cognate binding interactions may have application beyond influenza vaccines. For example, SA binding is a common feature of many viruses and vaccine antigens including the hemagglutinin-neuraminidase protein of mumps virus 68 , the VP8* domain of bovine-human reassortant rotavirus vaccine strains 69 , and SARS-CoV-2 spike protein 70 .
Taken together, we have demonstrated that altering the binding of HA to its SA receptor can have a significant impact on influenza vaccine-mediated immunity and viral clearance in mice. Elimination of the HA-SA interaction may be a simple, effective and readily implemented strategy to improve both the quality and durability of influenza vaccine responses. Although there is still much to learn mechanistically, if these findings are confirmed in human studies, the use of non-binding HAs may make an important contribution to the development of nextgeneration influenza vaccines.

Gel electrophoresis and immunoblot analysis
To confirm HA expression, 2 g of biomass were homogenized in 4 ml extraction buffer with 1% phenylmethylsulfonyl fluoride. Homogenates were clarified by centrifugation (10,000 × g, 10 min) and the crude extracts (25 µl/sample) were separated a Criterion XT 4-12% Bis-Tris gel (Bio-Rad) under reducing conditions and then transferred onto a PVDF membrane. Successful transfer was confirmed using ponceau red staining followed by de-staining with water. Membranes were blocked overnight (4°C) with 5% skim milk in TBST (tris-buffered saline, 0.1% Tween 20) and then incubated with rabbit polyclonal anti-H1 (Cat. No. IT-003-SW, Immune Technology) diluted 1:500 in TBST + 2% skim milk for 1 h at room temperature (RT). Membranes were then incubated for 1 h (RT) with horseradish peroxidase (HRP)-conjugated goat anti-rabbit IgG (Cat. No. IT-200-01, Immune Technology) diluted 1:20,000 in TBST + 2% skim milk. Bands were developed using Super Signal West Pico chemiluminescent substrate (Thermo Fisher) and detected on X-ray films. For VLP composition and purity analysis, purified VLP products (5 µg/sample) were separated on a 4-12% Bis-Tris gel as described above followed by staining with biosafe Coomassie G-250 (Bio-Rad). Gels were imaged using ChemiDoc TM XRS + system (Bio-Rad). See Supplementary Fig. 11 for unmodified gel and blot images.

Transmission electron microscopy (TEM)
VLP samples were diluted to 100 µg/ml in PBS and 5 µl of each were placed on 200 copper grids (Agar Scientific) for 45 s. Grids were washed 2x (1 min each) with 5 µl distilled water followed by two incubations with 1.5% uranyl acetate. Excess fluid was removed and samples were left to air dry. Grids were imaged on a Tecnai G2 Spirit Twin 120 kV Cryo-TEM (FEI) equipped with a Gatan Ultrascan 4000 CCD camera model 895 (Gatan).

Surface plasmon resonance
Binding of VLP to biotinylated ɑ-2,6 sialic acid glycans (6′-sialyl(LacNAc)-PEG-biotin) was quantified by SPR using a Biacore TM 8 K system (Cytiva, formerly GE Healthcare Life Sciences). Biotinylated synthetic glycan (Sussex Research Laboratories Inc.) was immobilized to a Series S sensor chip SA at a minimum target of 400 resonance units (RU) in the test flow cells. VLPs were diluted in HBS-EP + Buffer (assay running buffer) and injected at a flow rate of 50 μl/min (120 s contact time) at 4°C. H1 WT -VLP was diluted 150x and 100x (2.5 and 4 µg/ml) and H1 Y98F -VLP was diluted 10x and 5x (88 and 176 µg/ml). The standard curve included 8 dilutions ranging from 10 to 0.08 μg/ml. The flow was initially directed over a mock surface to which no protein is bound, followed by the biotinylated glycan. Response from the protein surface is corrected for the response from the mock surface (surface reference). Data analysis was performed using Biacore TM

Vaccination and sample collection
All animal procedures were carried out in accordance with guidelines of the Canadian Council on Animal Care, as approved by the Animal Care Committee of McGill University. Female Balb/c mice (8-10 weeks old, Charles River Laboratories) were immunized by injection into the quadricep muscle with VLP formulations containing 0.5-3 µg HA (50 µl total in PBS). Mice were vaccinated on day 0 and boosted on day 21 (when indicated). To evaluate humoral and cell-mediated immune responses mice were euthanized on day 28 (one-dose) or day 49 (28 days post-boost) by CO 2 asphyxiation. A small group of mice were maintained for 7 months to evaluate long-term immunity and sera were collected monthly.
Blood was collected from the left lateral saphenous vein before each vaccination and by cardiac puncture at study endpoints. Sera were obtained by centrifugation of blood in microtainer serum separator tubes (Beckton Dickinson) (8000 × g, 10 min) and stored at −20°C until further analysis. Spleens were collected in Hank's balanced salt solution (HBSS) without calcium and magnesium (Wisent). Single cell suspensions were prepared by passing spleens through a 70 µm strainer (BD Biosciences) and RBC were lysed using ACK buffer (0.15 M NH 4 Cl, 1 mM KHCO 3 , 0.1 mM NA 2 EDTA, pH 7.4). Cells were washed in HBSS and then resuspended in RPMI (Wisent) supplemented with 10% fetal bovine serum (Wisent), 1 mM penicillin/streptomycin (Wisent) and 0.5 mM β-mercaptoethanol (Sigma) (complete RPMI, cRPMI). Femurs were collected in cRPMI, sterilized in 70% ethanol (2 min) and rinsed with HBSS. Femurs were cut at one end and BM was extracted by centrifugation (8000 RPM, 30 s). Single cell suspensions were obtained as described for spleens.
To evaluate GC responses mice were vaccinated in the right hind limb footpad with 0.5 µg VLP (30 µl total in PBS). Draining popliteal LN were collected at indicated time points and digested with collagenase D (1 mg/ ml; Sigma) and DNaseI (10 µg/ml; Sigma) (40 min, 37°C, shaking at 220 RPM) prior to mechanical dissociation as described for spleens.
To measure antibodies that mediate HI sera were diluted 1:4 in receptordestroying enzyme (RDE; Denka Seiken) and incubated 18-20 h at 37°C. Sera were inactivated for 30 min at 56°C then further diluted to 1:10 with PBS. Two-fold serial dilutions of RDE-inactivated sera were added to a V-well microtiter plate (25 µl/well) and incubated for 30 min with 4 HA units of virus (25 µl/well) at RT. 50 µl of 0.5% turkey RBCs was added to each well and incubated 30 min at RT. Wells were examined visually for RBC teardrop formation upon plate tilting, indicating inhibition of hemagglutination 25 . Titers are reported as the reciprocal of the highest dilution to inhibit hemagglutination. Titers below the limit of detection (<10) were assigned a value of 5 for statistical analysis.
To measure antibodies that mediate MN MDCK cells (ATCC CCL-34) were seeded into flat-bottom 96-well plates (50,000/well) in HyClone SFM4Me-gaVir medium (Cytiva) supplemented with 10 µg/ml gentamicin (Gibco Life Technologies), 100,000 U/ml penicillin G (Gibco Life Technologies) and 20 µg/ml glutamine (Wisent). Two-fold serial dilutions of heat-inactivated sera (56°C, 30 min) were incubated with 100 TCID 50 of virus for 2 h (37°C, 5% CO 2 ) and then added to the MDCK monolayers with 1x TPCK (tolylsulfonyl phenylalanyl chloromethyl ketone)-treated trypsin. Plates were incubated for 3 h (37°C, 5% CO 2 ) and then medium was removed from each well and replaced with fresh MegaVir with 1x TPCK-treated trypsin. Cells were observed for cytopathic effects (CPE) for 3-5 days and the MN titer was defined as the reciprocal of the highest dilution to prevent CPE. Titers below the limit of detection (<10) were assigned a value of 5 for statistical analysis.
ELISpot HA-specific IgG antibody secreting cells (ASC) were quantified by ELISpot (Mouse IgG ELISpot BASIC , Mabtech). Sterile PVDF membrane plates (Millipore) were coated with Anti-IgG capture antibody and blocked according to the manufacturer's guidelines. To quantify in vivo activated ASCs, wells were seeded with 250,000 (BM) or 500,000 (splenocyte) freshly isolated cells and incubated at 37°C, 5% CO 2 for 16-24 h. HA-specific ASCs were detected according to the manufacturer's guidelines using 1 µg/ml biotinylated HA (immune tech, biotinylated using Sulfo-NHS-LC-Biotin). To evaluate memory ASCs, freshly isolated cells were polyclonally activated with 0.5 µg/ml R848 and 2.5 ng/ml recombinant mouse IL-2 (1.5 × 10 6 cells/ ml in 24-well plates) for 72 h (37°C, 5% CO 2 ). Activated cells were recounted and the assay was carried out as described above.

Flow cytometry
To identify total GC B cells and T FH cells, freshly isolated cells from the popliteal LN (1 × 10 6  To identify antigen-specific CD4 + T cells, freshly isolated splenocyte or BM immune cell suspensions (1 × 10 6 /200 µl in a 96-well U-bottom plate) were stimulated with 10% cRPMI (negative control), 2-2.5 µg/ml homologous HA WT -VLP (18 h), or a pool of overlapping peptides (15aa) spanning the HA sequence (6 h, BEI resources) (37°C, 5% CO 2 ). Golgi Stop and Golgi Plug (BD Biosciences) were added 5 h before the end of the stimulation according to the manufacturer's instructions. Cells were washed 2x with PBS (320 × g, 8 min, 4°C) and labeled with Fixable Viability Dye eFluor 780 (eBioscience) (20 min, 4°C). Cells were washed 3x followed by incubation with Fc Block (BD Biosciences) for 15 min at 4°C. Samples were incubated for an additional 30 min upon addition of the surface cocktail containing the following antibodies: anti-CD3 FITC (1 µl/sample; Thermo Fisher Cat. All flow cytometry was conducted using a BD LSRFortessa or BD LSRFortessa X20 cell analyzer. Data was analyzed using FlowJo software (Treestar, Ashland).