A vaccine central in A(H5) influenza antigenic space confers broad immunity

Phylogenetic tree construction

All available A(H5) HA nucleotide sequences and accompanying metadata were downloaded from the GISAID Data Science Initiative⁶⁰ and the Bacterial and Viral Bioinformatics Resource Center (BV-BRC)⁶¹ databases on 5 May 2023. The HA sequences of three antigens present in-house (A/Vietnam/3218/2004, A/duck/Hong-Kong/1091/2011 and A/eurasian-wigeon/Netherlands/EMC-3/2014), which were not yet available through the above mentioned databases at the time, were added to the dataset manually and deposited to GISAID in hindsight. The dataset was then preprocessed using Pépinière (a Python (v.3.10.14) jupyter notebook available at GitHub (https://github.com/epiv-lab/pepiniere) and Zenodo⁶²). This included deduplication of sequences present in both datasets based on identical accession numbers, identification and extraction of the open reading frame (ORF) corresponding to the longest ORF, and removal of sequences (1) without metadata or (2) shorter than 90% of the mean ORF length. Identical sequences were then grouped, and only the earliest (by isolation date) representative was retained. After preprocessing, sequences were aligned using MAFFT (v.7.515)⁶³, and the alignment was trimmed to the start and stop codons of the majority of sequences. Trimmed sequences were again filtered to remove identical sequences (retaining the earliest). The resulting dataset contained 14,896 sequences (Supplementary Table 1) and was realigned using MAFFT v.7.515 and maximum-likelihood trees were generated using IQ-Tree2 (v.2.1.4_beta)⁶⁴ with the GTR + F + R10 model (chosen by ModelFinder⁶⁵, implemented in IQ-Tree2) and 10,000 UFboot bootstrap approximations⁶⁶. Trees were midpoint-rooted, annotated, and visualized using iTOL⁶⁷. The display item was generated using ggtree (v.1.4.11)⁶⁸ in R (v.4.4.3). Genetic clades were predicted using LABEL⁶⁹ with the H5v2023 pre-release 1 (2023-05-05) module, courtesy of S. Shepard, and this prediction was used to annotate the tree.

Cells

Cells were maintained as described previously²⁴. 293T cells (ATCC) were cultured in Dulbecco’s modified Eagle’s medium (DMEM) (Lonza) supplemented with 10% FCS (Sigma-Aldrich), 1× non-essential amino acids (Lonza), 1 mM sodium pyruvate (Gibco), 2 mM l-glutamine (Lonza), 100 U ml⁻¹ penicillin, 100 μg ml⁻¹ streptomycin (Lonza) and 0.5 mg ml⁻¹ geneticin (Invitrogen). Madin–Darby canine kidney (MDCK) cells (ATCC) were cultured in Eagle’s minimal essential medium (EMEM) (Lonza), supplemented with 10% FCS, 1× non-essential amino acids (Lonza), 1.5 mg ml⁻¹ sodium bicarbonate (Lonza), 10 mM HEPES (Lonza), 2 mM l-glutamine (Lonza), 100 U ml⁻¹ penicillin and 100 µg ml⁻¹ streptomycin (Lonza). Cells were cultured at 37 °C, 5% CO₂, and passaged twice weekly. Cells were not authenticated. All cell banks were tested negative for mycoplasma.

Generation of plasmids

To generate plasmids for recombinant virus production, viral RNA was isolated from in-house available virus isolate stocks using the High Pure RNA Isolation Kit (Roche) according to the manufacturer’s instructions. The RNA was then used to generate viral copy DNA (cDNA) using SuperScript IV reverse transcriptase (Thermo Fisher Scientific) according to the manufacturer’s instructions; from the cDNA, individual HA or NA gene segments were amplified using segment-specific PCR (polymerase chain reaction) primers⁷⁰ and the PfuUltra II Fusion HS DNA Polymerase (Agilent) according to the manufacturer’s instructions. Individual viral gene segments were cloned into a previously described modified pHW2000 plasmid⁷¹ by restriction-site-based cloning or seamless cloning using the GeneArt Seamless Cloning kit (Thermo Fisher Scientific). If the respective virus isolate was not present in-house, synthetic genes containing HA sequences with a monobasic cleavage site were synthesized by BaseClear or Integrated DNA Technologies. When applicable, specific mutations were introduced in the HA genes and /or the HA multibasic cleavage site (MBCS) was removed by site-directed mutagenesis using the PfuUltra II Fusion HS DNA Polymerase (Agilent) and specific primers. Throughout the Article, A(H5) numbering is used to refer to specific amino acid positions⁷². To produce reverse genetics plasmids for A(H5N6) A/Sichuan/26221/2014 (Sichuan, accession number: EPI_ISL_163493), all eight gene segments were amplified from cDNA by PCR using specific primers⁷⁰ and cloned into our bidirectional reverse genetics plasmid. To produce reverse genetics plasmids for A(H5N1) A/duck/Giza/15292S/2015 (Giza, accession number: EPI_ISL_257168), viral RNA was extracted, and all eight gene segments were amplified from cDNA by PCR using specific primers⁷⁰ and cloned into our bidirectional reverse genetics plasmid. Non-coding regions, which are not part of the abovementioned sequences, were sequenced after RNA circularization as described previously⁷³. The non-coding regions used are listed in Supplementary Table 11.

Benefit sharing of synthetic constructs and viruses

Before the start of this work, we discussed and publicly announced our plans to generate synthetic HA constructs and produce recombinant A(H5) viruses and A(H5)-specific ferret sera through the GISAID website (https://gisaid.org/collaborations/collaboration-on-h5-antigenic-cartography/). Specifically, we made the commitment that the synthetic HA constructs, reverse genetics viruses and ferret sera will be shared with the laboratories that contributed the genome sequence data to GISAID⁶⁰. We committed to publish the antigenic maps with open access to the public. We also indicated that reagents may be provided to other researchers, including National Influenza Centers and global reference laboratories, after assurance that the originating laboratory, where the clinical specimen or virus isolate was first obtained, and the submitting laboratory, where sequence data have been generated and submitted through the GISAID Data Science Initiative, are fully recognized, to ensure fair attribution of contributions to the results benefitting from the data. We are indebted to GISAID and all scientists contributing to the GISAID Data Science Initiative, without whom this work would not have been possible. We thank the governments and scientists of Austria, Bangladesh, Cambodia, China, Egypt, Germany, Ghana, Indonesia, India, Iraq, Japan, Nepal, Nigeria, Mongolia, Russia, Scotland, South Africa, Sweden, Turkey, United States and Vietnam for their contributions that made this research possible.

Biosafety

All experiments were reviewed by the Erasmus MC Institution Review Entity (IRE), in accordance with the US Government September 2014 Dual-Use of Research of Concern (DURC) policy. The Erasmus MC IRE concluded that the studies described here were not falling under any of the seven DURC categories. Recombinant viruses that contained the HA (without MBCS) and/or NA of interest in the background of the attenuated vaccine strain A/Puerto Rico/8/1934 (PR/8) or the high-yield version thereof (PR/8 HY, 76) (Supplementary Table 2) were handled under biosafety level 2 (BSL2) conditions in agreement with national regulations. HPAIV wild-type isolates used in HI assays (Supplementary Table 2) were handled under BSL3 conditions. For the ferret challenge experiments, H5N6_Sichuan and H5N1_Giza recombinant viruses, which contained all eight wild-type gene segments of a single virus, were produced. These experiments were performed in the enhanced animal biosafety level 3 (ABSL3+) facility of the Erasmus University Medical Center as described previously⁷⁴.

Recombinant virus production

Recombinant influenza viruses were generated by reverse genetics using eight bidirectional plasmids as described previously^24,71. One day before transfection, about 3 × 10⁶ 293T cells were seeded in gelatin-coated 10 cm culture dishes. Calcium-phosphate-mediated transfection was used to deliver a total of 40 μg of plasmid DNA per dish. About 16 h after transfection, the cells were washed once with PBS and fresh medium containing 2% FCS with 200–350 μg ml⁻¹ N-tosyl-l-phenylalanine chloromethyl ketone (TPCK)-treated trypsin (Sigma-Aldrich) was added. Virus stocks were generated by inoculating either MDCK cells or 11-day-old embryonated chicken eggs with dilutions of the supernatant collected from the 293T cells 3 days after transfection or virus isolates. Virus stock production in MDCK cells was performed using EMEM medium containing the same supplements as described above, but without FCS and with the addition of 20–35 μg ml⁻¹ TPCK-treated trypsin, referred to as infection medium. MDCK supernatants or embryonated egg allantoic fluids were collected 2–3 d.p.i. and centrifuged at 2,100g for 10 min to remove cellular debris. The presence of virus was confirmed by haemagglutination assays using 1% turkey red blood cells (TRBCs, from in-house turkeys) in PBS. Sequences from all plasmids and from the non-PR/8 and PR/8 HY genes, that is, HA and NA of interest, of all virus stocks were confirmed with Sanger sequencing using the BigDye Terminator v.3.1 Cycle Sequencing Kit (Applied Biosystems) and the 3500xL Genetic Analyzer (Applied Biosystems).

Virus titrations

Virus titrations were performed in MDCK cells as described previously⁷⁴. In brief, flat-bottom 96-well plates containing confluent MDCK cells were inoculated with tenfold serial dilutions of the samples and incubated for 1 h at 37 °C under 5% CO₂. Cells were washed once with PBS and 200 μl of infection medium was added to each well. After 3 days of incubation at 37 °C, 5% CO₂, the presence of virus in the supernatants was determined using HA assays to determine the TCID₅₀. Virus titrations of the virus stocks were performed in ten replicates, and those of respiratory swabs and tissue homogenates from the vaccination-challenge experiments in four replicates. Virus titres were read out blindly.

Vaccine production

Vaccines were produced with recombinant viruses containing a mutated or wild-type HA, without MBCS, in the PR/8 HY background. For the initial screen of CVAs, whole-inactivated vaccines were generated with the corresponding matched NA. For the vaccination-challenge experiment, split-inactivated vaccines were used. To isolate the effect of varying the HA antigen in these studies, the NA present in the split-inactivated vaccines was mismatched with that of the challenge virus. Specifically, the vaccines in the H5N1_Giza challenge contained the N6 NA of H5N6_Sichuan, and the vaccines in the H5N6_Sichuan challenge contained the N1 NA of H5N1_Giza.

Whole-inactivated and split-inactivated vaccines were generated as described previously⁵². Eleven-day-old embryonated chicken eggs were inoculated with the virus of interest. Allantoic fluid was collected 2 d.p.i. and centrifuged for 10 min at 2,100g to remove cellular debris. Subsequent centrifugation steps were performed at 124,000g (SW 32 Ti, Beckman Coulter) at 4 °C, unless indicated otherwise. The allantoic fluid was concentrated on a 60% sucrose cushion by centrifuging for 2 h. Subsequently, resuspended sucrose cushions from multiple tubes were pooled and loaded onto 60/50/40/30/20% sucrose gradients, and centrifuged overnight at the lowest deceleration setting. The virus band, located on top of the 30% sucrose layer, was collected, diluted in PBS and subsequently pelleted by centrifugation for 2 h to remove the sucrose. The pellet was dissolved in either PBS (whole-inactivated vaccines) or PBS with 2% N-decanoyl-N-methylglucamine (Mega10, Sigma-Aldrich) (split-inactivated vaccines) to solubilize the viral membrane proteins. Incubation with PBS with 2% Mega10 was performed for 1 h at 37 °C. For both whole- and split-inactivated vaccines, the dissolved pellets were transferred to dialysis chambers (Slide-A-Lyzer Dialysis Cassettes, 10 K MWCO, Thermo Fisher Scientific) and subsequently submerged in PBS containing 0.01% formalin for 3 days. Subsequently, the dialysis chambers were immerged in PBS for a day, during which the PBS was refreshed twice. The resulting vaccines were aliquoted and stored at −80 °C. Vaccine inactivation was confirmed by two serial blind passages on MDCK cells and/or in embryonated chicken eggs.

Total protein content was determined using the Pierce bicinchoninic acid total protein analysis kit (Thermo Fisher Scientific). For the whole-inactivated vaccines, the absolute and relative HA content was estimated from SDS–PAGE protein gels using a BSA standard and stained with Instant Blue (Expedeon). The absolute HA content of the split-inactivated vaccines was estimated from SDS–PAGE using a BSA standard and stained with Instant Blue (Expedeon). The relative HA content of split-inactivated vaccines was determined with mass spectrometry, using a protocol based on a previous study⁷⁵ with modifications as described previously⁵². Diluted vaccines (10 μl of 125 μg ml⁻¹ of total protein) were mixed 1:1 with 0.2% RapiGest (Waters) and denatured for 5 min at 100 °C. After cooling to room temperature, 5 μl of sequence-grade modified trypsin solution (0.4 μg μl⁻¹; Promega) was added, and samples were incubated at 37 °C for 2 h. Digests were allowed to cool, and 55 μl of 0.5% trifluoroacetic acid was added. The samples were subsequently analysed by a nano-liquid chromatography (nano-LC) Ultimate 3000 system (Thermo Fisher Scientific) coupled to the Orbitrap Fusion Lumos mass spectrometer (Thermo Fisher Scientific).

Data from initial screens were used to select three peptides for stable isotope (SI) labelling (LVLATGLR, VNSIIDK and TLDFHDSNVK), based on intensity, length, sequence and sequence conservation within the A(H5) HA subtype. Digested vaccines were spiked with SI-labelled peptides with heavy lysine or arginine (Pepscan) at a final concentration of 50 fmol μl⁻¹, and measured on the nano-LC system (Ultimate 3000; Thermo Fisher Scientific) combined with an Orbitrap Fusion Eclipse Tribrid mass spectrometer (Thermo Fisher Scientific). For each peptide, the ratio between the SI-labelled peptide and the endogenous peptide was calculated, and was subsequently used to determine the concentrations of the endogenous peptide in the vaccines. The HA concentration based on the three individual peptides was averaged for each vaccine, and subsequently used to determine the relative HA content of the vaccines.

Ferret experiments

Ferret (Mustela putorius furo) experiments were performed in strict compliance with the Dutch legislation on the protection of animals used for scientific purposes (2014, European Union directive 2010/63/EU implemented). Experiments were performed at the Erasmus Medical Center in Rotterdam, the Netherlands under a project license accredited by the Dutch competent authority (license number AVD101002015340). Study protocols were approved by the Erasmus Medical Center Animal Welfare Body (permit numbers 15-340-01, -04, -06, -22, -23 and -24). Ferrets were seronegative for Aleutian disease, seasonal influenza A(H1N1), A(H3N2) and B viruses. For all ferret experiments, animals were randomly allocated to the different groups. Ferret experiments were not performed blindly because regulations required knowledge of the animal treatment for biosafety reasons.

Serum production

Ferret sera were generated as described previously^24,25 in class III isolators under BLS3 conditions using recombinant viruses unless indicated otherwise (Supplementary Tables 2 and 3). Recombinant viruses were produced carrying the HA (without MBCS) and the closest matching NA present in-house, in the background of PR/8 or PR/8 HY. In brief, male ferrets (at least six months old) were inoculated intranasally by applying dropwise 250 μl of virus stock per nostril. Unless indicated otherwise (Supplementary Table 2), a boost was administered after 14 days, by subcutaneously injecting a total of 250 μl concentrated virus combined with 250 μl TiterMax Gold adjuvant (Sigma-Aldrich) at two different spots in the back.

The concentrated virus used for the subcutaneous boost was prepared by inoculating five 11-day-old embryonated chicken eggs per virus. The allantoic fluid was collected 2 d.p.i. and cleared from debris by centrifuging for 10 min at 2,100g. About 36 ml of the cleared allantoic fluid was concentrated by centrifuging for 2 h at 124,000g (SW 32 Ti, Beckman Coulter), and the resulting pellet was resuspended in 700 μl PBS. Ferrets were terminally bled through cardiac puncture 14 days after the subcutaneous boost, or 14 days after the intranasal inoculation if no boost injection was performed. Before virus inoculation, subcutaneous boost injection and the terminal bleed, ferrets were anaesthetized with ketamine and medetomidine (10 and 0.05 mg per kg body weight, respectively), the latter which was antagonized with atipamezole (0.25 mg per kg body weight).

The blood samples were collected in VACUETTE CAT Serum Separator Clot Activator tubes (Greiner Bio-One), incubated at least for 15 min to allow clotting, and centrifuged for 15 min at 2,000g to obtain the serum.

Vaccination–challenge studies

Vaccination experiments were performed similarly as described previously⁵². Six- to 12-month-old female ferrets were used for vaccination studies (n = 2 for vaccination-only experiments and n = 6 for vaccination–challenge experiments). Ferrets were vaccinated twice intramuscularly on day 0 (prime) and day 28 (boost) with 250 μl of whole or split-inactivated vaccine estimated to contain about 7.5 μg HA, adjuvanted with 250 μl AddaVax (InvivoGen), which was equally divided between the two hind legs. For the mock-vaccinated groups, animals were vaccinated with 250 μl of PBS adjuvanted with 250 μl AddaVax. Before each vaccination, a blood sample was obtained through the cranial vena cava and serum was obtained as described above (pre- and pre-boost sera). The pre-sera were tested in HI assays as described below against seasonal influenza A(H1N1), A(H3N2) and B viruses (using vaccine strains of the respective year), as well as PR/8 recombinant viruses carrying A(H5) HAs from the respective vaccines and challenge virus, if applicable. Pre-sera were negative in HI assays against the tested viruses. The pre-boost sera of the ferrets from the challenge experiments were titrated in HI assays against PR/8 recombinant viruses with three vaccine antigens employed in the respective study.

For vaccination-only experiments, ferrets were euthanized 28 days after the boost vaccination through a cardiac puncture, and post-boost sera were obtained from the whole blood as described above.

For vaccination–challenge experiments, DST micro-T temperature loggers (Star-Oddi) were surgically implanted into the abdominal cavity of the ferrets 14 days after the prime vaccination. Serum samples were collected from whole blood sampled through the cranial vena cava 1 week before inoculation (post-boost sera) (Extended Data Fig. 8 and Supplementary Tables 6 and 7). Vaccinated ferrets were then transferred to class III isolators for acclimatization a week before inoculation with the challenge virus. Ferrets were inoculated intranasally and intratracheally with wild-type recombinant viruses containing all eight segments of the respective challenge virus. The inoculation doses were 10^5.5 and 10^3.4 TCID₅₀ per animal for the H5N1_Giza and H5N6_Sichuan virus, respectively, divided over 3 ml intratracheally and 250 μl in each nostril. These doses were determined before the challenge in a pilot experiment using three ferrets per tested dose. The doses were selected to induce a reproducible and consistent infection of the upper and lower respiratory tracts. Subsequently, daily nose and throat swabs were collected under light ketamine anaesthesia, and body weight and activity level score were monitored daily as described previously^74,76. Body temperature was recorded every 10 min by the implanted temperature loggers. Then, at 4 d.p.i., ferrets were euthanized through cardiac puncture, after which tissues (selected based on virus detection in the pilot studies) were collected for virological and pathological analysis as described previously⁷⁴. For virological analysis, the right nasal turbinates, trachea, right bronchus, right lung lobes, tracheobronchial lymph node and liver (for both challenges), right cerebrum and right cerebellum (for the H5N1_Giza challenge only), and the spleen (for the H5N6_Sichuan challenge only) were collected. For pathological examination, the left nasal turbinates, trachea, left bronchus and left lung lobes were collected.

During blood collection, vaccination, virus inoculation and cardiac puncture, ferrets were anaesthetized with a mixture of ketamine and medetomidine, and antagonized with atipamezole, as described above.

Histopathology and IHC

After necropsy, tissues were stored in 10% neutral-buffered formalin (lungs after careful inflation with formalin) for at least 2 weeks, after which the tissues were embedded in paraffin. Slides (4 μm) were made, and subsequent slides were either stained with haematoxylin and eosin or used for IHC as described previously⁷⁷. In brief, after deparaffinization, antigen retrieval and blocking of endogenous proteases, slides were incubated for 1 h at room temperature with either a primary antibody against influenza A virus nucleoprotein (Hb65, American Type Culture Collection, H16-L10-4R5) or a mouse IgG2a isotype control (R&D, MAB003), diluted in PBS with 0.1% BSA (1:400 and 1:200, respectively). After three washes with PBS with 0.05% Tween-20, the slides were incubated for 1 h at room temperature with a goat anti-mouse IgG2a secondary antibody coupled to horseradish peroxidase (HRP) (Bio-Rad, Star133A), diluted 1:100 in PBS with 0.1% BSA. HRP was revealed using 3-amino-9-ethylcarbazole and a haematoxylin counterstain was performed. A lung section from an animal experimentally infected with 2009 pandemic A(H1N1) virus was used as positive control in each staining experiment.

The pathological changes and the presence of viral antigen in respiratory tissues were blindly scored in a semi-quantitative manner by a veterinary pathologist. Semi-quantitative assessment of influenza virus-associated inflammation in the lungs (four slides with longitudinal section and cross-section of cranial and caudal lobes per animal) was performed in a blinded manner on every slide as reported earlier⁷⁸. The extent of alveolitis and alveolar damage was scored as follows: 0, 0%; 1, 1–25%; 2, 25–50%; 3, >50%. The severity of alveolitis, bronchiolitis, bronchitis, bronchial adenitis, tracheitis and rhinitis were scored as follows: 0, no inflammatory cells; 1, few inflammatory cells; 2, moderate numbers of inflammatory cells; 3, many inflammatory cells. The presence of alveolar oedema, alveolar haemorrhage and type II pneumocyte hyperplasia were scored as follows: 0, no; 1, yes. Finally, the extent of peribronchial, peribronchiolar and perivascular infiltrates were scored as follows: 0, none; 1, one to two cells thick; 2, three to ten cells thick; 3, more than ten cells thick. Semi-quantitative assessment of influenza virus antigen expression in the lungs was performed as reported earlier⁷⁹. For the alveoli, 25 arbitrarily chosen fields of lung parenchyma of the 4 lung sections per animal were blindly examined by light microscopy, using a ×20 objective, for the presence of influenza virus nucleoprotein. The cumulative scores for each animal were presented as a percentage corresponding to the number of positive fields. The percentage of positive epithelium in the bronchi and bronchioles was estimated on all four lung slides and averaged per animal. The percentage of positively staining epithelium in the nose and trachea was estimated for one slide.

Serological assays

HI assays were performed with recombinant viruses in PR/8 or PR/8 HY background and virus isolates (Supplementary Table 2) as described previously²⁴ using in-house TRBCs. Sera were treated overnight at 37 °C with five volumes of a Vibrio cholerae filtrate (generated in-house) containing receptor-destroying enzyme, to prevent aspecific inhibition. After inactivation for 1 h at 56 °C, sera were adsorbed using an equal volume of 10% TRBCs for 1 h at 4 °C to prevent aspecific agglutination. Twofold serial dilutions of sera in PBS were prepared in round-bottom 96-wells plates starting at 1:20 in a volume of 50 μl. To each well, 25 μl of virus, diluted in PBS to 4 hemagglutinating units, was added. After incubation for 30 min at 37 °C, 25 μl of 1% TRBCs was added to each well. Plates were subsequently incubated for 1 h at 4 °C before reading the HI titre. The HI titre was determined as the reciprocal value of the highest serum dilution which completely inhibited TRBC agglutination. HI titres were read out blindly. For the calculation of GMTs, threshold titres of <10 were converted to 5 unless stated otherwise.

Virus neutralization assays were performed in MDCK cells as described previously^24,32. First, sera were incubated for 30 min at 56 °C to inactivate complement. Twofold serial dilutions of sera in PBS, starting at 1:10, were combined with 100 TCID₅₀ of virus, and incubated for 2 h at 37 °C. Subsequently, the virus–serum mixtures were added to flat-bottom 96-wells plates containing confluent MDCK cells previously washed once with PBS. After incubation for 2 h at 37 °C and 5% CO₂, cells were washed once with PBS, and 200 μl infection medium per well was added. The plates were incubated at 37 °C under 5% CO₂, and the presence or absence of virus in the supernatants was determined after 3  days using HA assays with TRBCs. The virus neutralization titre was determined as the reciprocal value of the highest serum dilution for which no virus in the supernatants was detected. Virus neutralization titres were read out blindly. Virus neutralization assays were performed in duplicate, and the arithmetic means of log₂ titres were calculated.

Antigenic cartography and antibody profiles

Antigenic maps were constructed from HI data using a multidimensional scaling algorithm as described previously¹⁶ using the R package Racmacs (v.1.2.3)⁸⁰. First, HI titres are converted to a distance matrix (HI table distances) by (1) dividing each HI titre by 10 and applying a log₂ transformation (hereafter, log₂-transformed HI titres); and (2) subtracting each log₂-transformed titre to the highest one for each serum. Secondly, multidimensional scaling algorithms are used to find the best set of map coordinates to represent the distances from the distance matrix most closely. For each optimization, antigens and sera, hereafter named points, are randomly placed in n-dimensional space, and coordinates are optimized from these starting conditions using the L-Broyden–Fletcher–Goldfarb–Shanno algorithm, minimizing the sum of the squared differences between HI table distances and the map distances (Euclidian distance between points in the n-dimensional space). In an antigenic map, every direction represents antigenic distance, and one antigenic unit corresponds to a twofold change in HI titre.

Unless stated otherwise, antigenic maps were computed using the ‘make.acmap’ function, with 1,000 optimization runs in three dimensions, and the minimum column basis set to zero. The antigenic map was validated using several tests, which are described in Supplementary Notes 2 and 3. Total map stress was extracted using the mapStress function and individual antigen stresses were extracted using the agStress function. HI table distances and pairwise antigen–serum Euclidian distances in the antigenic map were extracted using the tableDistances and mapDistances functions, respectively. When threshold HI titres (that is, <10) are converted to table distances in the process of making an antigenic map, the resulting values are not an exact distance but a ‘greater-than’ value, that is, thresholded distance. To include these points in the visualization in scatter plots of HI distances versus map distances (Extended Data Fig. 3a), and corresponding regression coefficient (R²) calculations (Extended Data Figs. 1c and 3a), these values were converted to the thresholded distance increased by 1 on the log₂ scale, for example, a threshold distance of >7 on the log₂ scale is converted to an 8. To compute distances between points in the map, antigen and serum coordinates were extracted using the agCoords and srCoords functions, respectively. The dist function (base R) was subsequently used to compute pairwise Euclidean distances between points in the map. The map centre was determined by computing the mean x, y and z of the antigen coordinates. Pairwise genetic hamming distances, that is, the number of amino acid differences between two antigens, were computed using the ‘stringDist’ function (method = ‘hamming’) from the Biostrings package (v.2.74.1)⁸¹.

To average and visualize the immune response of multiple ferrets belonging to the same experimental group (referred to as a mean serum), the GMTs of multiple vaccination sera against each individual antigen were computed to generate antibody profiles. Individual threshold titres were first converted to the closest possible numerical titre (for example, <10 to a 5) and, subsequently, GMTs were calculated. For the visualization of the reactivity of post-vaccination sera using the A(H5) antigenic map, antigenic maps were optimized with datasets containing the antigenic map HI data and HI data of a single post-vaccination serum or mean data of multiple post-vaccination sera as described above. The resulting maps including individual or mean post-vaccination sera data shared similar conformations, also corresponding to that of the antigenic map (the mean median Procrustes distance between each map with post-vaccination data and the antigenic map was 0.15 AU). Thus, to generate displays in which the positions of post-vaccination sera were visualized without changing the position of the antigens or sera in the antigenic map, maps, containing either individual or merged HI data, were superimposed onto the antigenic map using the mergeMaps function with the frozen-merge method. These maps were then used to visualize and analyse the reactivity of the post-vaccination sera to antigens in the antigenic map using custom R code (https://github.com/epiv-lab/H5-antigenic-evolution and Zenodo⁸²). For analysis of mutant antigens, antigenic maps were computed with datasets containing HI data of a single mutant antigen in addition to the antigenic map dataset. The resulting optimized maps were used to calculate the distances described in the text. For visualization in Supplementary Data 5b, superimposed maps were generated as described above for the visualization of post-vaccination sera. The displays in Fig. 3 and Supplementary Data 8–10 were generated by superimposing an optimized map containing an individual single or mean post-vaccination serum, as described above, on the map from Supplementary Data 5b. Supplementary Video 1 was generated using the antigenic map displayed in Fig. 1b and Supplementary Data 2 using custom R code (https://github.com/epiv-lab/H5-antigenic-evolution and Zenodo⁸²). In brief, individual frames were generated by taking screenshots of rotating r3js (v.0.0.2)⁸³ antigenic map displays using R package webshot2 (v.0.1.2)⁸⁴, which were subsequently assembled into a video using ffmpeg.

Resialylated TRBC assay

Resialylated TRBC assays were performed as described previously⁷⁴. The pellet of 1.25 ml of 1% TRBCs was resuspended in 62.5 μl PBS and incubated for 1 h at 37 °C with 50 μl of 1 mU μl⁻¹ V. cholerae neuraminidase (Roche) and 10 μl 0.1 M CaCl₂ to remove all sialic acids from the TRBCs. After two washes with PBS, the TRBCs were combined with 3.75 μl of 30 mM CMP-sialic acid (Merck), and either 5 μl of α2,3-sialyltransferase (recombinant human ST3GAL6, Fc Chimera, R&D systems) or 5 μl α2,6-sialyltransferase (Recombinant Human ST6GAL1 (amino acids 44–406) Protein, R&D Systems), and PBS up to 75 μl. Alternatively, 1 μg of in-house generated Pasteurella multocida sialyltransferase 1 (Pmst1) M144D (α2,3-sialyltransferase)⁸⁵ or Pmst1 M144L P34H (α2,6-sialyltransferase)⁸⁶ was used. TRBCs were incubated at 37 °C for 2 h with the commercial enzymes or for 4 h with the in-house generated enzymes. After resialylation, TRBCs were washed twice with PBS, and resuspended in PBS containing 1% BSA to a final concentration of 0.5% TRBCs. Besides α2,3- and α2,6-linked sialic-acid-specific TRBCs, untreated and VNCA-treated TRBCs were taken along as controls when assessing the binding preference of viruses in HA assays. Moreover, in each independent assay, a minimum of three control viruses was used: one with α2,3-linked sialic acid specificity, one with α2,6-linked sialic acid specificity, and one with dual binding specificity.

Data visualization and statistics

Data were visualized with Racmacs (v.1.2.3)⁸⁰, r3js (v.0.0.2)⁸³ and/or ggplot2 (v.3.5.1)⁸⁷ in R v.4.4.3 (used throughout). The interactive figure files in the Supplementary Information (Supplementary Data 2–10) were generated with flexdashboard (v.0.6.2)⁸⁸ in R. Statistical analyses were performed using the base R functions for the Kruskal–Wallis test and linear regressions. For the comparison of multiple experimental groups, the Kruskal–Wallis test was first performed. If positive, pairwise two-sided Dunn’s tests with Bonferroni correction were then performed (R packages FSA (v.0.10.0)⁸⁹ and dunn.test (v.1.3.6))⁹⁰ to assess the significance of differences between two experimental groups.

Reporting summary

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