Sixty healthy adult volunteers (20-49 years old) were enrolled in an open label phase I dose escalating clinical trial early 2009 at Haukeland University Hospital, Bergen, Norway (www.clinicaltrials.gov, NCT00868218) (16). The study was approved by the Regional Committee for Medical Research Ethics, Northern Norway and the Norwegian Medicines Agency. All participants provided written informed consent before inclusion into the study.

Participants were randomized into 4 groups, and intramuscularly vaccinated twice with the H5N1 virosomal vaccine (Crucell Berna Biotech) containing 30μg HA alone, or 1.5, 7.5 or 30μg HA with 50μg Matrix M adjuvant (Novavax) at 3-week interval (16). None of the participants had previously received an H5N1 vaccine.

A monovalent inactivated virosomal H5N1 vaccine, containing vaccine strain NIBRG-14, a virus derived from A/Vietnam/1194/2004 (H5N1) and A/Puerto Rico/8/1934 (H1N1) using reverse genetics, was used in the clinical trial. The influenza surface antigens HA and NA were purified from beta-propiolactone (BPL) inactivated egg grown viruses, mixed with lecithin and incorporated into the phospholipid bilayer by spontaneous formation of the virosomes. The HA content of the vaccine was quantified by single radial diffusion, and the presence of NA was confirmed.

The adjuvant Matrix M used in the trial was the 3^rd generation immune stimulating complex, which contains Matrix-A and Matrix-C fractions produced from purified Quillaja saponin fractions A and C, at the proportion of 91:9. The vaccine was formulated as 30μg HA alone, or 1.5, 7.5 or 30μg HA with 50μg Matrix M adjuvant, filled into single use syringes, and stored at 4°C until use. All participants (n=60) received 2 doses of the vaccine at a 3-week interval, except one withdrawal from the 7.5µg HA adjuvanted vaccine group.

Blood samples were collected before, and up to 42 days after vaccination. Sera were separated, aliquoted and stored at -80°C until use.

All sera were treated with receptor-destroying enzyme (RDE, Seiken) at a ratio of 1 in 4 at 37°C for 18h, and heat treated at 56°C for 1h. The treated sera were analysed in duplicate (2-fold serial dilution, starting from 1:10) with 4 hemagglutinating units of viruses and 0.8% horse red blood cells, as previously described (16, 17). A panel of reassortant H5N1 viruses were used, including A/Vietnam/1194/2004 (NIBRG-14, vaccine strain, clade 1), A/Indonesia/5/2005 (IBCDC-RG2, subclade 2.1.3.2), A/turkey/Turkey/1/2005 (NIBRG-23, subclade 2.2.1), and A/Cambodia/R0405050/2007 (NIBRG-88, subclade 1.1). The hemagglutination inhibition (HI) titer was determined as the reciprocal of the highest sera dilution giving 50% inhibition of hemagglutination. Non-detected samples were assigned a value of 4 for calculation purpose.

H5N1 pseudotypes were generated by co-transfecting lentiviral vectors pHR’CMV-Luc, pCMVRΔ8.2, pCMVR-H5HA, and pCMVR-N1NA into HEK293T cells as previously described (19, 20). A panel of pCMVR-HA constructs encoding H5HA from A/Vietnam/1203/2004 (Vietnam, clade 1), A/Indonesia/5/2005 (Indonesia, subclade 2.1.3.2), A/Turkey/65596/2006 (Turkey, subclade 2.2.1), A/common magpie/Hong Kong/5052/2007 (HK5052, subclade 2.3.2.1), A/Shenzhen/406H/2006 (Shenzhen, subclade 2.3.4), A/Shanxi/2/2006 (Shanxi, clade 7), and pCMVR-N1NA construct encoding NA from A/Thailand/(KAN-1)/2004 were used to prepare H5N1 pseudotypes.

All sera were heat inactivated and analysed in duplicate (4-fold serial dilution, starting from 1:10) with H5N1 pseudotypes corresponding to 20,000 to 200,000 relative luciferase activity (RLA), as previously described (20). The pseudotype-based neutralization (PN) titer (IC₈₀) was determined as the reciprocal of the sera dilution giving 80% reduction of RLA. PN titers were normalized based on HIV-1 gag p24 quantities for different pseudotypes. Non-detected samples were assigned a value of 2 for calculation purpose.

H5HA stalk and N1NA specific immunoglobulin G (IgG) were quantified in sera using the enzyme-linked immunosorbent assay (ELISA) developed in-house, as previously described (21). Serially diluted sera were analysed in Maxi Sorp 96-well plates coated with 1 μg/ml recombinant chimeric HA (cH9/5HA) that combines the H5HA stalk domain from NIBRG-14 strain with an HA globular head domain from A/guinea fowl/Hong Kong/WF10/1999 H9 influenza A virus, or 1 μg/ml recombinant N1 neuraminidase. Immunoglobulin concentrations were interpolated from standard human IgG curves. For calculation purposes, non-detected samples were assigned as 0.04 μg/ml against cH9/5HA and 0.005 μg/ml against N1NA.

All sera were heat inactivated and analysed in duplicate (2-fold serial dilution, starting from 1:10) with 100 TCID₅₀ reassortant cH9/1N3 virus in MDCK cells, as previously described (22). The cH9/1N3 virus contains the HA stalk domain from A/California/07/2009 H1 strain, an HA globular head domain from A/guinea fowl/Hong Kong/WF10/1999 H9 strain, and N3NA from A/swine/Missouri/4296424/2006 virus. The virus neutralization (VN) titer was measured with 0.7% turkey red blood cells and determined as the highest serum dilution giving complete hemagglutination. Non-detected samples were assigned a value of 5 for calculation purpose.

All sera were treated at 56°C for 45min and analysed in duplicate (3-fold serial dilution, starting from 1:5) with a reassortant NIBRG-73 virus, as previously described (22). The NIBRG-73 virus has N1NA from the NIBRG-14 strain and H7HA from A/equine/Prague/1956 virus. The neuraminidase enzymatic activity was measured with fetuin, horseradish peroxidase-conjugated peanut agglutinin and o-phenylenediamine dihydrochloride (Sigma-Aldrich), and read as optical density (OD) value at 490nm. The neuraminidase inhibition (NI) titer (IC₅₀) was calculated as the reciprocal dilution of sera giving 50% reduction in enzymatic activity. Non-detected samples were assigned a value of 2 for calculation purpose.

All sera were heat inactivated, serially diluted and analysed using ADCC Reporter Bioassay kit (Promega), as previously described (22). MDCK cells infected with NIBRG-14 virus at multiplicities of infection (MOI) 0.34 were used as target cells, and Jurkat/NFAT-luc cells were used as effect cells. The antibody-dependent cellular cytotoxicity (ADCC) reporter activity was measured with Bio-Glo Luciferase Assay Reagent (Promega) as relative luciferase activity (RLA). ADCC titer (EC₅₀) was calculated as the reciprocal dilution of sera giving 50% of maximum RLA. Non-detected samples were assigned a value of 2 for calculation purpose.

Six to eight weeks old female BALB/c mice (Taconic, Denmark) were housed under specific-pathogen free conditions at Rikshospitalet, Oslo University Hospital. All animal experiments were approved by the Norwegian Food Safety Authority.

Pooled human sera or saline (400µl/mouse) were injected intraperitoneally (i.p.) into mice (n=12/group) 1 day prior to viral challenge. Mice were anaesthetized by subcutaneous (s.c.) injection of Hypnorm/Dormicum (0.05ml working solution/10g) and infected intranasally (i.n.) with 5 MLD₅₀ of NIBRG-14 virus in 20 µl/mouse (10µl/nostril). Mice were monitored for survival and weight loss for 2 weeks after challenge, with an endpoint of 20% weight reduction, as required by the Norwegian Food Safety Authority. Mice that lost more than 20% body weight were euthanized by cervical dislocation. For mice that did not lose weight during the 2-week monitoring, the maximum body weight loss was assigned as 0.01% for calculation purpose. At days 3 and 5 post challenge, both lobes of lungs were harvested (n=3 mice/group), snap frozen and stored at -80°C until use.

Frozen lung tissues were weighted and homogenized in DMEM with 1% antibiotics. NIBRG-14 virus was quantified in TCID₅₀ on MDCK cells (23). Hemagglutination assay with 0.7% human red blood cells was used to measure the viral load. The limit of detection was 22.49 TCID₅₀/ml of homogenate.

Full-length HA protein amino acid sequences from influenza type A viruses were downloaded from NCBI Influenza Virus Database. Phylogenetic analyses were performed at ngPhylogeny.fr using MAFFT (Multiple Alignment using Fast Fourier Transform, default settings), BMGE (Block Mapping and Gathering with Entropy, default settings), and PhyML (Phylogeny software based on the Maximum-likelihood, default settings) (24, 25). See Supplementary Table 1 for the accession no. of all HA amino acid sequences used.

Biological replicates were used in all experiments, unless otherwise stated. Antibody quantification results including HI, PN, VN, NI, ADCC titers, and IgG concentrations, and lung viral load results were Ln transformed prior to statistical tests. Turkey’s multiple comparisons and Fisher’s LSD test were performed in two-way analysis of variance (ANOVA). False Discovery Rate controlled multiple comparisons were performed in Nonparametric Kruskal-Wallis test. Nonparametric Spearman correlations and Pearson correlations were tested, linear fitting curves were plotted when Spearman or Pearson P <0.10. In multiple linear regression analyses, all independent variables were centered prior to test. * P<0.05, ** P<0.01, *** P<0.001, all P values are two-tailed. All statistical analyses were performed with GraphPad Prism 7 and SPSS 25.

Sixty adults were divided into 4 groups and vaccinated with the H5N1 virosomal vaccine alone or escalating doses of adjuvanted vaccine: Group 30μg - (30μg HA alone, 15 subjects), Group 1.5μg + (1.5μg HA with adjuvant Matrix M, 15 subjects), Group 7.5μg + (7.5μg HA with adjuvant Matrix M, 15 subjects), and Group 30μg + (30μg HA with adjuvant Matrix M, 15 subjects). All subjects received a boosting dose 3 weeks after the first priming dose, except for one withdrawal after the first dose in 7.5μg + group ( Figure 1A ). Twenty-five subjects had earlier received seasonal influenza vaccine(s) or reported influenza virus infection ( Table 1 ). Sequential pre- and post-vaccination serum samples up to 42 days were collected from all vaccinees.

We assessed HA specific antibodies using the HI assay. None of the vaccinees had detectable antibodies before vaccination. After priming, only the 30μg + group showed potent increases of vaccine specific antibodies. After boosting, all 3 adjuvanted groups had antibodies above the protective level (HI titer ≥32), while the non-adjuvanted 30μg - group remained below the protective level ( Figure 1B ). The 1.5μg + group had significantly lower antibody fold-induction after priming compared to the 30μg + group, while after boosting the difference diminished. On the contrary, the 30μg - group showed significant antibody induction after priming but limited further increase after boosting ( Figure 1C ).

We further analysed antibody responses using pseudotype-based neutralization (PN) assay. In 24 subjects, low levels of pre-existing vaccine specific neutralizing antibodies were detected. Of note, the vaccine elicited potent homologous antibody responses 7 days after priming in all 4 groups, which were further significantly elevated to higher level in the 1.5μg + and 7.5μg + groups after boosting. The 1.5μg + group had lower antibody fold-induction after priming compared to the 30μg + group but reached similar neutralizing antibody titer after boosting; whilst the 30μg - group had rapid antibody increase 7 days after priming but no significant boost after the second dose ( Figures 1E, F ).

To explore the breadth of the vaccine elicited cross-neutralizing antibodies, we developed a panel of H5N1 pseudotypes expressing divergent H5 hemagglutinins covering multiple (sub)clades of H5N1 viruses isolated from humans ( Figure 1D and Supplementary Table 1 ). Remarkably, cross-neutralizing antibodies were elicited to equivalent titers as compared to vaccine specific antibodies in all 4 groups when tested against the 3 closely related pseudotypes, namely HK5052 (subclade 2.3.2.1), Turkey (subclade 2.2.1) and Shenzhen (subclade 2.3.4). Meanwhile, although significantly boosted by vaccination, lower levels of cross-neutralizing antibodies were detected to the 2 more distant pseudotypes: Indonesia (subclade 2.1.3.2) and Shanxi (clade 7). Nevertheless, broadly neutralizing antibodies were detected in all the subjects in the 30μg + group across all the (sub)clades ( Figure 1G ).

Total HA stalk specific antibodies were measured in ELISA against the recombinant chimeric protein cH9/5HA, which consists of the HA stalk domain from the vaccine strain and an irrelevant H9HA head. All subjects were found to have pre-existing HA stalk specific antibodies. As early as 7 days after priming, we observed significant antibody increases in all 4 groups. The 30μg + group had significantly higher antibody fold-induction as compared to the 1.5μg + and 30μg - groups ( Figures 2A, B ).

We next assessed neuraminidase (NA) specific antibodies by ELISA ( Figures 2C, D ). In 51 subjects, NA specific antibodies were detected prior to vaccination. The H5N1 vaccines elicited potent antibody increases in all 4 groups at 7 days after priming. Further significant antibody increases were found after boosting in 1.5μg + and 7.5μg + H5N1 groups ( Figure 2C ). Overall, the 7.5μg + and 30μg + groups showed equivalently potent responses, while the 1.5μg + and 30μg - groups had lower fold change of NA specific antibodies ( Figure 2D ).

To quantify the antibodies that inhibit NA enzymatic activity, we performed enzyme-linked lectin assay (ELLA). Before vaccination, detectable NA inhibition (NI) titers were observed in 36 of 60 subjects. Vaccinees in the groups 30μg -, 7.5μg + and 30μg + had significant antibody increases 7 days after priming. While vaccinees in the 1.5μg+ group showed antibody induction 21 days after priming, which increased significantly after boosting ( Figure 2E ). The kinetics of NI antibody fold change among 4 vaccine groups was similar to the kinetics of NA specific binding antibodies by ELISA ( Figures 2D, F ).

Lastly, we measured ADCC inducing antibodies. Thirty-seven out of 60 subjects had pre-existing ADCC inducing antibodies. The H5N1 vaccines further elevated the antibody levels, which remained high after boosting in all groups, especially the 30μg+ group ( Figures 2G, H ).

In summary, the virosomal H5N1 vaccines elicited potent and multifaceted antibody responses. The adjuvanted intermediate (7.5μg +) and high (30μg +) dose vaccines potently induced antibody increases 7 days after priming, which were further elevated or maintained after boosting. By comparison, the adjuvanted low (1.5μg +) dose vaccine showed delayed antibody kinetics; whilst the non-adjuvanted (30μg -) vaccine elicited antibodies of an overall lower magnitude.

As the majority of our subjects had some detectable antibodies against the H5N1 vaccine components prior to vaccination, we investigated whether pre-existing immunity hampers the H5N1 vaccine elicited antibody responses. Firstly, the age of the subjects enrolled in this study ranged from 20 to 49 years old ( Table 1 ). We saw significant correlations between the age and pre-existing level of neutralizing antibodies, HA stalk and NA specific binding antibodies ( Figure 3A ). In contrast, the antibody levels after H5N1 vaccination had inverse or no correlation with age ( Figure 3B ). Next, subjects who had had seasonal influenza vaccination or infection within 5 years prior to the current study showed significantly higher levels of pre-existing neutralizing, NA specific and ADCC inducing antibodies. H5N1 vaccination boosted antibodies to comparably high levels in all subjects ( Figure 3C ). Collectively, the H5N1 vaccines induced robust and multifaceted antibody responses regardless of age, vaccination/infection history or pre-existing antibody levels.

To study whether the H5N1 vaccine induced antibodies are protective against infection, we performed passive serum transfer and virus challenge in a murine model. Briefly, we pooled pre-vaccination sera from groups 1.5μg + and 30μg + as D0 sera, and post-vaccination sera were pooled within each group from all vaccinees in the 1.5μg + and 30μg + groups for days 7, 21 and 42. Pooled immune sera or saline were then transferred into mice one day prior to a viral challenge (5×MLD₅₀) with the NIBRG-14 virus. Body weight was monitored for 14 days after challenge ( Figure 4A ). Mice that received saline or D0 sera became ill and rapidly lost weight. All 5 mice receiving saline and 2 out of 6 mice receiving D0 sera died between days 7 and 8 after challenge. By contrast, all mice that received post-vaccination sera survived. Mice that received sera 7 and 21 days post-vaccination from group 1.5μg + showed mild symptoms and lost a maximum of 16% and 9.3% of body weight, respectively, although they quickly recovered. Mice receiving day 42 sera from the group 1.5μg + and sera at all 3 time points from the group 30μg + displayed no symptoms of disease and had < 3% body weight loss at maximum ( Figures 4B–D ).

We measured the lung weight and viral load at 3 and 5 days after challenge. Mice that received days 21 and 42 sera from group 30 μg + showed lower lung weight at 5 days post challenge, and lower viral loads at days 3 and 5, compared to the mice receiving saline ( Figures 4E, F ). In addition, we observed good correlations between the lung viral load and body weight loss, as well as between the lung viral load and the lung weight ( Figures 4G, H ). To summarise, immune sera elicited by H5N1 vaccination, especially the adjuvanted high dose (30μg +), conferred full protection against a lethal in vivo challenge.

Our final aim was to study which of the H5N1 vaccine elicited antibody responses correlated with in vivo protection. We included the widely used hemagglutination inhibition (HI) titer, neutralizing antibody titer measured in pseudotype-based assay (PN titer), hemagglutinin stalk specific antibody level (HA stalk IgG) and neuraminidase inhibition (NI) titer as candidates for correlates of protection. As the Fc receptors involved in ADCC are different between humans and mice, the ADCC inducing antibody titer was not included in analyses.

Importantly, we observed inverse correlations between the antibody levels in the pooled human sera that mice received by passive transfer and murine body weight loss, as well as lung viral load after challenge ( Figures 5A, B ). Of note, the NI titer had the closest relationship with both murine body weight loss and lung viral load, followed by PN titer. Whilst the HI titer and HA stalk IgG level showed lower degrees of correlation with in vivo protection, determined by the Peason’s r values in correlation analyses.

We further dissected the contribution of each antibody parameter to the predictions of in vivo protection. The adjusted R square and the standard error of the estimate in the multiple linear regression analyses were used to determine the level of fit in different prediction models. When predicting the maximum body weight loss, PN titer and NI titer alone or combined were among the best-fit models ( Figure 5C and Supplementary Table 2 ). In the lung viral load predictions, NI titer alone, NI and PN titer combined, and all 4 antibody parameters together were the top 3 models ( Figure 5D and Supplementary Table 2 ). Notably, NI titer contributed significantly to the top models predicting both body weight loss and lung viral load ( Figures 5C, D and Supplementary Table 3 ). Together, our correlation and regression analyses on the in vivo protection provided by pooled human sera in murine challenge model cohesively demonstrated that PN and NI titers act as correlates of protection. Further study is needed to confirm the roles of PN and NI titers as correlates of protection in human infection scenario.