Nasal RNA-scaffold-protein vaccine protects mice from human H5N1 clade 2.3.4.4b virus lethal infection and safeguards against vaccine-unmatched viruses

Summary

Background

Recent cases of H5N1 avian influenza spilling over into human populations have raised concerns once again about the unpredictable emergence of zoonotic viral diseases. This phenomenon highlights the persistent threat posed by viruses that jump from animals to humans, underscoring the need for preparedness for antigens that humans have not encountered. Nasal vaccines, which can effectively block transmission at the site of viral entry, represent a promising strategy and should be prioritised for stockpiling to halt zoonotic spillover.

Methods

Here, we immunised a preclinical animal model intranasally with the RNA-Scaffold-Protein Vaccine (RSPVac) and evaluated mucosal immunity and protection against avian-origin and pandemic human influenza viruses.

Findings

Nasal immunisation of strain-specific H5N1 and H1N1 RSPVac induced strong antibody responses in serum and mucosa, accompanied by mucosal antigen-specific T cells. We examined immune cells during early vaccination and found that RSPVac induced a strong but transient cytokine expression. Neutrophils comprising both regulatory and effector subsets were observed, indicating a controlled activation. Vaccinated animals were completely protected against vaccine-matched avian virus challenges without pathological signs. Protection was also observed against vaccine-unmatched H1N1 and H7N9 challenges, with a display of trained immune cell profiles. Finally, mutant RSPVac demonstrated an impaired vaccination outcome, suggesting a complex activation mechanism.

Interpretation

These findings demonstrate that RSPVac elicits potent mucosal immunity and safeguards against vaccine-unmatched infections, underscoring its potential as a non-infectious mucosal vaccine to prevent zoonotic virus transmission.

Funding

This study is supported by the Health@InnoHK initiative of the Innovation and Technology Commission of the Hong Kong Special Administrative Region Government.

Research in context.

Evidence before this study

We searched for the current H5 vaccine stockpile in the WHO inventory and found 1 out of 22 licenced human vaccines, which is a nasal vaccine candidate, which is the AstraZeneca FluMist. We searched PubMed for “2.3.4.4b vaccine” on 9 Sep 2025, narrowing the range to the year 2020–2025, and excluding reviews. Our search results showed that the majority of recent studies use LNP-mRNA and inactivated viruses as candidate vaccines, with a few studies using technologies such as replicating RNA and viral vectors. We also searched for “2.3.4.4b nasal vaccine” using the same condition. One study uses an NA-based subunit coupled with bacterial-derived adjuvants as a nasal vaccine in a preclinical model. Overall, almost all preclinical studies use the intramuscular immunisation. Regarding the avian influenza situation, “avian influenza” search on PubMed returned 4680 results. For the more recent publications, as of 9 Sep 2025, avian influenza can occur in domestic animals such as cats and dairy cattle. There are also reports of human infection cases by avian influenza viruses that are not on the WHO vaccine stockpile list, including H3N3, H9N2, and H10N5 human cases, indicating the threat of avian influenza is sporadic and unpredictable.

Added value of this study

This study presents a nasal vaccine candidate for H5N1 avian influenza. This technology is non-infectious (in contrast to live virus vaccines), and all components are derived from the virus (no artificial or chemical components, all virus-focused), suggesting a unique design. Importantly, this work addresses the current lack of nasal vaccine candidates against highly pathogenic avian influenza in humans and also within preclinical research. The immunogenicity data further support the evaluation of translation potential. In addition, this work highlights the flexibility and effectiveness of our vaccine platform, which enables rapid adaptation to newly emerging infectious diseases and helps meet the urgent need to prevent unpredictable zoonotic spillover events. Moreover, our data suggest an underlying mechanism for mucosal immunisation that warrants deeper research.

Implications of all the available evidence

The lack of mucosal vaccines against highly pathogenic avian influenza reflects a gap in the current preparedness for zoonotic virus spillover. The avian influenza pool in wildlife comprises viral antigens that humans may not have encountered, and preparedness is lacking for antigenically shifted viruses. Nasal vaccines represent an effective way to induce mucosal immunity that can halt virus transmission and animal-to-human jumping. This work presents a nasal vaccine that was evaluated in the BALB/c mouse model and found to protect against contemporary H5N1 clade 2.3.4.4b and also in the scenario with heterologous, vaccine-unmatched viruses. Robust mucosal immunity, a well-trained immune system, and regulated early activation demonstrated by our vaccine address the needs of candidates in the current preclinical stage and present potential for future clinical application.

Introduction

There was a rise in public health concerns regarding the spillover of H5N1 clade 2.3.4.4b avian influenza to humans in September 2024.¹ Despite the rarity of infection cases, the highly pathogenic H5N1 raises significant public health concerns due to its high mortality rate.2, 3, 4 In a more pressing matter, viral presence was found in dairy products, and high viral load was detected in cattle. Currently, there is no direct evidence of human-to-human transmission, and the WHO has regarded its threat as low. Yet, a total of 82 human cases have been confirmed in 2025.⁵ Some cases have no known animal exposure, reflecting that the transmission of recent avian influenza is highly unpredictable. It is nearly impossible to determine when the next spillover will be. The transmission of avian influenza in wild animals and livestock has never ceased. Moreover, as current seasonal vaccines offer predominantly strain-specific immunity of relatively short duration, the effectiveness of current seasonal vaccines to protect against zoonotic viruses is undermined. This greatly calls for preparedness for emerging and unpredictable influenza outbreaks and underscores the importance of advancing vaccine designs for superior prevention.

While many current vaccines against H5N1 (including WHO stockpiled vaccine candidates) are administered via intramuscular injection, they cannot induce effective mucosal immunity.⁶ Therefore, the beneficial role of nasal vaccines in providing protection and preventing transmission has been increasingly recognised.⁷ Mucosal neutralising antibodies, particularly IgA, can directly inhibit viral entry and neutralise the virus at the site of infection.^8,9 Stimulating mucosal immunity through nasal vaccination represents a promising strategy. Also, because of the non-invasive administration route, nasal vaccines may potentially reduce vaccine hesitancy. To date, there is a lack of nasal vaccines for avian flu infection; only 1 out of 22 human licenced vaccines listed in the WHO stockpile is a nasal vaccine.⁶

With the aim of developing a non-infectious and versatile nasal vaccine, we have developed the RNA-Scaffold-Protein Vaccine (RSPVac) during the SARS-CoV-2 pandemic and showed a strong activation of mucosal immunity in mice model.¹⁰ RSPVac is a protein-RNA vaccine that comprises purified recombinant proteins complexed with in vitro transcribed short, non-coding RNAs. The recombinant protein is a fusion construct consisting of the viral surface antigen and the viral nucleoprotein. The viral surface antigen refers to the receptor-binding protein (such as spike protein for SARS-CoV-2, and haemagglutinin for influenza) and serves as the primary neutralising target of the virus, while the nucleoprotein represents the most conserved antigen across viral subtypes. The RNA component is a non-coding short RNA with a sequence derived from viral genomes. This RNA not only functions as an immunostimulatory agent but, critically, acts as a scaffold facilitating the binding of the antigen-nucleoprotein fusion to form protein-RNA complexes. Importantly, all components of RSPVac are derived exclusively from viral elements, with defined sequences, and produced recombinantly, thereby excluding the use of artificial or chemical adjuvants. Therefore, RSPVac combines neutralising target, cross-protective epitopes, and adjuvant in one package. As all components can be derived from virtually any virus, RSPVac is a nasal vaccine platform for various viral diseases.

Here in this study, taking advantage of the versatile design, we applied RSPVac technology to the recent avian H5N1 clade 2.3.4.4b spillover to humans, generated nasal vaccines, and further characterised its immune response in the preclinical animal model. We generated influenza-specific RSPVac, targeting avian H5N1/Vietnam/1194 and also H5N1/Texas/37. We also generated pandemic H1N1-RSPVac to examine its application for seasonal flu. Prime-boost nasal immunisation of RSPVac in mice elicited robust systemic and mucosal antibody responses, T cell activation, and provided full protection against vaccine-matched virus challenge. Profiling of the early innate immune response following a single nasal administration revealed that immune activation peaked at 24 h post-vaccination and resolved by 48 h, indicating a controlled and transient inflammatory response, which indicates safety for nasal immunisation. Neutrophil influx during early activation also showed balanced regulatory and effector functions. Furthermore, we explored the cross-protection capabilities against vaccine-unmatched infections and observed that H5N1-vaccinated mice resisted lethal H1N1 and H7N9 infections with minimal symptoms. A well-trained immune landscape, comprising distinct IgA, enhanced helper T cells function, and increased MHC expression, was observed in mice bronchioalveolar lavage fluid (BALF) cells, reflecting the robustness of RSPVac immunisation. Additionally, mucosal T cell responses were negatively impacted by the use of mutant proteins in RSPVac, suggesting a potential mechanism for mucosal vaccination that merits further research. Collectively, the results highlighted RSPVac's potential to safeguard against vaccine-unmatched viruses. The robust immunogenicity elicited by influenza-specific RSPVac in mice supports its potential to advance protein-based mucosal vaccine development. Our results also indicate that RSPVac's mode of action remains an important direction in future research.

Methods

Protein expression, T7 RNA production, and vaccine formulation

Recombinant proteins were expressed and produced using the Expi293F expression system (Gibco, RRID: CVCL_D615) according to the manufacturer's instructions. Briefly, sequences encoding the following: “TPA secretion peptide-6xHis-HRV 3C protease-nucleoprotein-HA head” were cloned into a CMV-driven plasmid vector for fusion protein expression. Detailed design for Flu-RSPVac proteins is illustrated in Fig. 1. Mutant protein designs are described in Fig. 7. Plasmids were transfected into Expi293F cells with Expifectamine 293 reagents plus enhancers (Gibco). At 4-days post-transfection, culture supernatant containing the secreted protein was harvested by centrifugation, filtered, and passed through Ni Sepharose Excel (Cytiva), and eluted with imidazole-containing buffer. His tags were removed by enzyme treatment, and proteins were further purified by the Akta Pure Chromatography system with Superdex 200 Increase column (Cytiva). Final protein products were buffer exchanged to PBS/10% glycerol, concentrated, and stored at −80 °C until use.

Fig. 1.

Generating influenza-RSPVac. (A) Schematic diagram of influenza-RSPVac protein component. H5N1/1194 full-length nucleoprotein was fused with the HA head of respective influenza A viruses. A small Gly–Ser linker was used to connect both protein domains. The nuclear localization signal (NLS) on the nucleoprotein was mutated to facilitate protein secretion during production. (B) R326 design schematic. The R326 was derived from the sequence of the defective influenza A virus PB2 gene,¹⁴ and was further modified to mutate the start and stop codons. GGG was presented at the 5′ end because of T7 transcription. (C) SDS-PAGE and glycan analysis of RSPVac proteins. The slight decrease in size following PNGaseF digestion indicated that the recombinant protein was N-glycosylated. (D) Bioanalyzer validation of R326. (E) Outline of RSPVac generation. Recombinant proteins were expressed in mammalian cells, while the RNA was produced by in vitro transcription. Both components were purified and then mixed according to vaccine dosage. The resulting protein-RNA complexes were used for intranasal vaccination. (F) Conceptual illustration of RSPVac-H5-37, RSPVac-H5-1194, and RSPVac-H1. Each RSPVac consisted of three components: HA head antigen, the nucleoprotein, and R326 RNA. Figures E and F were created in BioRender.

Fig. 7.

Mutant proteins impacted RSPVac immunisation. (A) Schematic showing the disrupted formation of RSPVac using an RNA-binding-deficient mutant. Female BALF/c mice (n = 6) were nasally immunised with two doses of RSPVac-H5, 14-days apart. The RSPVac was produced with either the WT or mutant protein. Sera, BALF, and BALF-flushed cells were harvested 10 days post-2nd dose and analysed. (B) Fluorescence polarisation of WT and Mutant protein. After mutating major arginine residues, the RNA-binding affinity was reduced by 7-fold. (C–F) ELISA for anti-HA serum IgG, BALF IgA, Serum IgG subtypes, and BALF IgG subtypes. (G) IFNγ ELISpot analysis of BALF-flushed cells, stimulated with the immunogen (WT protein used to generate RSPVac-H5-1194). (A) Schematic of the phase separation mutant of the SARS2-RSPVac. To explore whether phase separation is related to RSPVac nasal immunisation, phase separation mutants (Psmut) were generated and used to generate SARS2-RSPVac. Female BALB/c mice were immunised using the same regimen with WT or Psmut RSPVac. Sera and BALF were harvested for downstream analysis. (B–F) ELISA for anti-spike RBD serum IgG, BALF IgG, BALF IgA, serum IgG1/IgG2a, BALF IgG1/IgG2a. (G) IFNγ ELISpot analysis of BALF-flushed cells, stimulated with the immunogen (protein used to generate WT SARS2-RSPVac). Statistical significance was determined by the Mann–Whitney test. p < 0.05 was considered significant. P-values were shown, and those that were considered statistically not significant were labelled not significant (ns). Figures A were created in BioRender.

For small RNA R326 production, the HiScribe T7 High Yield production kit (NEB) was used according to the manufacturer's instructions. Briefly, the linearised DNA template was mixed with kit components and incubated at 37 °C for 16 h, followed by DNase I (provided in kit) treatment for 15 min. The resulting RNA was purified by RNAiso (Takara), dissolved in 0.1M TE buffer, and stored at −80 °C until use. RNAs were characterised using Bioanalyzer with RNA 6000 Nano kit (Aglient).

To generate vaccines for immunisation, protein and RNA were thawed on ice. Once thawed, components were manually mixed at room temperature and incubated for 15 min. Each vaccine dose comprised 12 μg protein +25 μg ssRNA. The RSPVac (H5-37, H5-1194, H1) contains a recombinant fusion protein comprising the H5N1-1194 nucleoprotein and a virus-specific HA head, with the 326-bp ssRNA R326 (described in the text). Final buffer composition was normalised at PBS/2.5% (v/v) glycerol. Vaccines were then chilled on ice and equilibrated to room temperature right before animal immunisations were performed. The SARS2-RSPVac comprised the SARS-CoV-2 nucleoprotein, the spike receptor-binding domain, and the single-stranded RNA R268, which was described in previous studies.¹⁰ For mutant RSPVac, the identical vaccine formula was used except that mutant proteins with defective RNA-binding or phase–separation properties were used. Mutant RSPVac-H5-1194 contains the mutated H5N1-1194 nucleoprotein (described in text and Fig. 7) and virus-specific HA head fusion protein, and the ssRNA R326. The SARS2-RSPVac phase-separation mutant contains a fusion protein of mutated SARS2 nucleoprotein and spike receptor-binding domain, and the ssRNA R268.

Cultures and viruses

All viruses were cultured and propagated in MDCK cells (ATCC, RRID: CVCL_0422) and quantified by plaque assay. During propagation, 1 μg/mL TPCK-trypsin (Sigma) was added to facilitate virus replication. Virus strains used include: A/H1N1/Hong Kong/415742Md (H1N1/pdm09), A/H1N1/Puerto Rico/8, A/H5N1/Vietnam/1194, A/H7N9/Anhui/1. A/H5N1/TX/37 was recovered by reverse genetics. H1N1 virus cultures and related experiments were performed in a BSL-2 laboratory at the University of Hong Kong. H5N1 and H7N9 virus cultures and related experiments were performed in the BSL-3 containment laboratory at the University of Hong Kong, Queen Mary Hospital.

Zetaview nanoparticle tracking analysis

Nanoparticle characterisation was performed using the ZetaView® TWIN Nanoparticle Tracking Analyser (Particle Metrix). Briefly, the instrument was calibrated with PS100 nanoparticle beads. Samples were diluted 100-fold in PBS and injected into the instrument. Data was recorded in triplicate, and raw data generated by the software was extracted and plotted in GraphPad Prism 9. Particle size estimation was obtained from the software-generated report.

Mass photometry

Mass photometry measurements were performed using the TwoMP mass photometer (Refeyn) at the Centre for PanorOmic Sciences, Li Ka Shing Faculty of Medicine, University of Hong Kong. Briefly, sample glass slides obtained from Refeyn was coated with 0.1% poly-l-lysine (Sigma) for 1 min, air dried, and thoroughly washed with 0.22 μm filtered MilliQ H2O, and were immediately used for sample measurement. Instrument focus was obtained with buffer only, and samples were measured using the Regular profile. Samples were diluted 100-fold for measurement. The data obtained were analysed and plotted using Refeyn Discovery software.

Fluorescence polarisation

Purified proteins were 2-fold serially diluted and incubated with 50 nM of HEX-labelled 20-mer RNA¹⁰ at room temperature for 30 min in opaque 96-well microplates (SPL Life Sciences). Fluorescence polarisation was measured using BioTek Cytation 5 Multimode Reader (Agilent Technologies). Plots and the dissociation constant were analysed using GraphPad Prism 9.

Animal ethics approval and group size calculation

Female BALB/c mice (RRID: MGI:2161072) at 6–7 weeks were bred under an AAALAC International-accredited program at the Centre for Comparative Medicine Research, the University of Hong Kong, under Specific Pathogen Free (SPF) conditions. Research only proceeded following review and approval from the HKU Committee on the Use of Live Animals in Teaching and Research (CULATR, ref. 22–094) and under licence from the Hong Kong SAR Government's Department of Health. A total of 141 mice were used in this study. Protocols and experiments were designed according to ARRIVE guidelines. Animals were housed in individually ventilated cages under a 12:12 dark–light cycle within environmentally controlled rooms and were fed ad libitum with laboratory diets manufactured by LabDiet, USA. To set animal group sizes, power calculation following guidelines from the National Centre for Replacement, Refinement and Reduction of Animals in Research (NC3RS) was used. Briefly, based on prior vaccination experiments, a 2-fold difference was considered to be the minimal biologically relevant effect size. Variability was estimated by the square root of the pooled variances of previous immunological experiments. Statistical power was set as 0.8, and the significance level at 0.05. Power calculation was then performed for two-sided unpaired t tests. The calculation yielded a variance of 0.5491, the effect size as 1.845 (Cohen's d = log2(2)/0.5491), and a final N per group of 3.84. Based on the grouping of related studies in the field, we chose N = 6–9 per group to match the common standard for immunogenicity studies.

Animal immunisation and live virus challenge

On the day of immunisation, female 6–7 week-old mice were randomly assigned to control and vaccination groups. Mice were then anaesthetised by intraperitoneal injection of 100 mg/kg ketamine and 10 mg/kg xylazine. For intranasal immunisation, 20 μL of vaccine was instilled dropwise into the left nostril (when the animal's ventral side facing researcher). All vaccine dosages in this study consisted of 12 μg protein complexed with 25 μg ssRNA per dose, dissolved in PBS/2.5% glycerol as described above. The vaccines were freshly prepared and administered within 1–2 h.¹⁰ Animals were allowed to rest and recover following the vaccine administration. On the day of infection, animals were anaesthetised. 1000 plaque-forming unit (PFU) A/H1N1/Hong Kong/415742Md, 100 PFU A/H1N1/Puerto Rico/8, 30 PFU A/H5N1/Vietnam/1194, or 2 × 10⁵ PFU A/H7N9/Anhui/1 in a volume of 20 μL was intranasally inoculated into the mice's left nostril. Animals were weighed daily, and survival was monitored across 14-days. Disease severity scoring was performed for each animal based on observation of ruffled fur, hunched back, laboured breathing, and inactivity. Humane endpoint was reached when the animal experienced body weight loss greater than 30% or failure to stand upright. For H7N9/Anhui/1 infection, based on preliminary optimisation experiment and references,^11,12 we used the standard 20% weight loss cutoff to capture significant differences. At the experiment endpoint, surviving animals were euthanised by an overdose of anaesthesia.

Animal serum, BALF, and tissue harvest

For harvesting serum, blood was collected from the facial vein, followed by centrifugation. Sera were then collected and stored at −80 °C. On the day of BALF and tissue harvest, animals were euthanised by an overdose of anaesthesia. Bronchoalveolar lavage fluids (BALFs) were collected by instilling PBS twice into the mouse lungs by inserting a catheter into the trachea. The fluids were centrifuged, the cell pellets were pooled, and supernatants (not pooled, 1st instillation was used for analysis) were collected for analysis. For the analysis of cells in the whole lung, perfusion was performed. Perfused lungs were then digested and dissociated into single-cell suspensions using the Lung Dissociation kit (Miltenyi Biotec) and gentleMACS OctoDissociator (Miltenyi Biotec). Note that BALF was not collected for whole lung analysis. Dissociated cells were treated with BD Pharm lyse (BD Bioscience) to lyse red blood cells before downstream assays. For infection experiments, animal lungs and nasal turbinates were harvested and homogenised by TissueRuptor II (Qiagen). One lobe of the mice's lung was fixed in 4% formalin for 48 h and further processed for histology staining.

Viral titre and viral load detection

Viral titre in animal tissues was analysed by plaque assay. Briefly, tissue homogenates were serially diluted and inoculated into MDCK cells. After 1 h of inoculation, supernatants were removed, and 1% low melting agarose (Thermo) in 1× MEM medium was overlayed onto the cells. Plates were further incubated at 37 °C, 5% CO₂ for 60–72 h. Cells were fixed and stained with 0.5% crystal violet. Plaques were visualised and counted. Viral load in animal tissues was analysed by probe-based quantitative PCR (qPCR). Briefly, viral RNAs were extracted from tissue homogenates using the QIAamp viral RNA kit (Qiagen). qPCR was then performed using QuantiNova probe RT-PCR kit (Qiagen), with a Taqman probe specific to influenza A M gene (5′-HEX/TCAGGCCCCCTCAAAGCCGAG/BHQ1-3′), designed according to CDC. Reaction was performed on the LC480 instrument (Roche). For M gene copy number determination, a standard curve was constructed using a DNA plasmid carrying the H3N2 M gene. Viral titres, plaque-forming unit/mL/gram (PFU/mL/g), and viral loads M gene copy/gram (copy/g) were normalised to the average weight of the tissue (0.5 g for lung, 0.05 g for nasal turbinates).

Beads-based cytokine quantitation

Cytokine quantitation was performed using LEGENDplex Mouse Anti-Virus Response Panel (13-plex) (Biolegend) according to the manufacturer's instructions. Briefly, BALF was incubated with the antibody labelled beads, washed, stained, and analysed by BD Fortessa X-20 SORP cell analyser according to the instructions of the kit. Flow cytometry data were analysed by Flowjo v10 software.

ELISA

Antigen-specific antibodies were detected by Enzyme-Linked ImmunoSorbent Assay (ELISA). Briefly, recombinant proteins were coated on half-area flat-bottom high-binding plates (Corning, 3690), in 0.05 M bicarbonate buffer (pH 9.6) at 4 °C overnight. Coating conditions are as follows: 1 μg/mL recombinant H1N1/California/04/2009 haemagglutinin (SinoBiological, Cat# 11055-V08H) and 0.5 μg/mL recombinant H1N1/Hong Kong/415741Md nucleoprotein (in-house bacteria expression). Plates were blocked with 2.5 g/L casein buffer. Sera and BALFs were pre-diluted in PBS at 1:100 and 1:5, respectively, followed by 2-fold serial dilution, then added to the plate, and incubated at room temperature for 1 h. Detection antibodies were diluted in 0.5% PBST and incubated with the plate for 40 min at room temperature. Plates were washed with 0.05% PBST between sample and antibody incubation. Presence of antigen-specific antibodies was determined by horseradish peroxidase-based method using 1-Step Ultra TMB (Thermo). Colour development was quenched by 0.5 M sulphuric acid. Plates were read by Varioskan LUX (Thermo) at OD450. Antigen-specific antibody endpoint titres were calculated in GraphPad Prism 9 by 5-parameter logistical fit for an asymmetric sigmoidal model, and defined as the interpolated reciprocal of dilution with the same absorbance as the mean of blank wells plus 2 standard deviations. Detection antibodies used include: anti-mouse IgG Fc-HRP (1:10000, Cat# ab97265, Abcam, RRID: AB_10680426), anti-mouse IgG minimal x-reactivity (1:2000, Cat# 405306, Biolegend, RRID: AB_315009), anti-mouse IgA-HRP (1:1000, Cat# 626720, Invitrogen, RRID: AB_2533951), biotin anti-mouse IgG1 (1:2000, Cat# 406604, Biolegend, RRID: AB_315063), biotin anti-mouse IgG2a (1:2000, Cat# 407104, Biolgend, RRID: AB_345324), HRP-streptavidin (1:3000, Cat# 405210, Biolegend).

Haemagglutinin inhibition assay (HAI)

The ability of antibodies to inhibit virus haemagglutination was analysed by HAI assay.¹³ Briefly, all sera were treated with RDE II Seiken (Denka) overnight to remove interfering molecules and heat inactivated at 56 °C for 30 min 8 HAU per 50 μL of virus (H1N1/415742 (pdm09), H5N1/Vietnam/1194, and H5N1/TX/37) was mixed with 50 μL 2-fold serially diluted animal sera (starting from 1:10) in V-bottom plates, and incubated at room temperature for 1 h 0.5% turkey red blood cell (Rockland) was then added, and haemagglutination patterns were observed after 1-h incubation at room temperature. HAI titre was determined as the reciprocal of the highest dilution of sera showing complete haemagglutination.

Focus reduction neutralization assay (FRNT)

Antibody neutralisation was analysed by FRNT assay. Briefly, MDCK cells were plated in 96-well black plates (SPL). Sera samples were treated as stated above and serially diluted in plain medium. 1000 PFU/well of viruses were added and incubated with sera samples at 37 °C for 1 h. Cells were then washed with PBS and replenished with plain medium, and further incubated at 37 °C for 6 h, followed by fixation with 4% formalin. Plates were then washed, permeabilized with 0.1% NP40, blocked with 2% BSA/PBS, and stained with Pan Influenza A nucleoprotein mouse monoclonal antibody (1:2000, Cat# 40205-MM18, SinoBiological, RRID: AB_3677052) and detected by anti-mouse AlexaFluor 488 (1:2000, Cat# ab150105, Abcam, RRID: AB_2732856). Nucleoprotein-positive cells were detected and counted by Cytation 7 (Biotek) and Gen5 version 3.12. Neutralisation titres FRNT50 were calculated in GraphPad Prism 9 by 5-parameter logistical fit for an asymmetric sigmoidal model, and determined as the interpolated reciprocal of the dilution having 50% reduction of nucleoprotein-positive foci compared to control wells.

Intracellular cytokine staining

Peptide pools of HA head of 15-mer with 11 amino acids overlap and the specific NP 147-155aa peptide were obtained from custom peptide synthesis by Genscript. Isolated cells from mouse tissues were cultured in activation medium (RPMI-1640/10% FBS, 2 mM l-glutamine, 55 μM β-mercaptoethanol, 2 mM sodium pyruvate, 0.1 mM MEM non-essential amino acids) and stimulated with peptide pools (2 μg/mL/peptide) at 37 °C, 5% CO₂. After 2 h stimulation, brefeldin A (7.2 μg/mL) and monensin (2 μM) were added, followed by further culturing for 16 h. Cells were then blocked with Mouse BD Fc-blocker (1:200, Cat# 553141, BD Biosciences, RRID: AB_394656), washed with PBS/5%FBS, and stained with surface staining antibody cocktail: anti-CD3-PE-Cy7 (1:100, Cat# 25-0032-82, eBiosciences, RRID: AB_2815096), anti-CD4-FITC (1:400, Cat# 100510, Biolegend, RRID: AB_312713), anti-CD8a-APC-Fire750 (1:400, Cat# 100766, Biolegend, RRID: AB_2572113), anti-CD44-BV785 (1:100, Cat# 103059, Biolegend, RRID: AB_2571953). Cell viability was determined by Zombie Aqua (1:400, Cat# 423102, Biolegend). After washing, cells were then fixed and permeabilized with BD Cytofix/Cytoperm (BD Biosciences), washed, and intracellularly stained with anti-IFNγ-APC (1:100, Cat# 505810, Biolegend, RRID: AB_315404). After staining, cells were washed, passed through a 70 μm cell strainer, and analysed by BD Fortessa X-20 SORP cell analyser. Flow cytometry data were analysed by Flowjo v10 software.

IFNγ ELISPOT

BALF cells were stimulated with recombinant proteins and IFN-secreting cells were detected by ELISpot Flex Mouse IFN-γ (ALP) kit (Mabtech Cat# 3321-2A). 1 × 10⁴ BALF cells were cultured in activation medium (described above) on Multiscreen® 96 well mixed cellulose esters filter plate (Millipore, MAHAS4510) and stimulated with 10 μg of corresponding immunogen (the protein used to generate RSPVac) for 48 h. Plates were then washed, stained with kit antibodies, and spots were developed by Sigmafast BCIP/NBT (Sigma, B5655-25TAB) and counted by Cytation 7 (Biotek) or S6 Universal M2 (ImmunoSpot).

Single-cell RNA library construction and sequencing

Mice BALF cells were harvested, fixed with Chromium Single Cell Fixed RNA Sample Preparation Kit (10× Genomics) overnight at 4 °C, and submitted, without further storage, to 10× Flex RNA Expression Library preparation and sequencing at the University of Hong Kong, LKS Faculty of Medicine, Centre for PanorOmic Sciences (CPOS), Genomics Core. Each experiment group contains pooled cells from 3 mice. Briefly, fixed single-cell suspension goes through cell counting using Countess II FL Automated Cell Counter (Invitrogen). Probe hybridisation, single-cell encapsulation, and libraries were prepared by Chromium Fixed RNA Profiling Reagent Kit and Chromium Next GEM Chip Q Single Cell Kit (10× Genomics). Fixed cells in suspension were incubated with whole transcriptome probes containing probe barcodes for 18 h. After hybridisation, samples were pooled together and washed. Cells in suspension were loaded into individual wells of 10× Chromium Single Cell chip. Single cells were then encapsulated into Gel Beads-in-emulsion (GEM) by 10× Chromium iX. Fixed RNA Profiling Reagent Kit was used to perform downstream steps: probe pair hybridised to target RNA were ligated and extended in GEMs, followed by GEM recovery and amplification. Index PCR and SPRIselect size selection of the 10× barcoded, ligated probe products were performed according to the manufacturer's protocol. Library size, concentration, and quality control were determined by Qubit (Invitrogen) and Bioanalyzer (Thermo) assays. Illumina sequencing (Pair-End sequencing of 151bp) was performed on Illumina NovaSeq 6000.

Single-cell RNA sequencing data processing and automated cell type identification

Filtered barcodes and feature matrices were output using 10× CellRanger. Sequencing data were further analysed and quality checked using Seurat v5 on R 4.4.0. Briefly, samples were merged, and layers were integrated using IntegrateLayers function. Data was then filtered, normalised, scaled, and dimensionality reduced by PCA (RunPCA), and clustering (FindClusters) was performed. Dimension reduction was further performed using UMAP and TSNE methods (RunUMAP and RunTSNE). Barcodes and coordinates were exported from R into a Loupe Browser (10× Genomics) object using the LoupeR packages. Individual samples were distinguished based on the assigned original identity. Differential gene expression between clusters and cell types was determined by Loupe Browser 8 differential expression functions, based on the CellRanger pipelines. Cell type identities were validated by evaluating gene markers. Differential gene expression between cell types and sample groups was performed in Loupe Browser 8. Reduction plots, feature plots, violin plots, doughnut plots, and heatmaps were generated using Seurat, ggplot2, pheatmap, and paleteer functions in R as well as image captures in Loupe Browser.

Statistics calculation

Statistical tests for data were performed in GraphPad Prism 9, using the non-parametric Mann–Whitney test (for comparing two groups) and one-way ANOVA (for comparing multiple groups where normality applies). Statistical tests are indicated in the figure legend. P-values are indicated in figures, and p < 0.05 was considered significant.

Role of funders

This study is supported by the Health@InnoHK initiative of the Innovation and Technology Commission of the Hong Kong Special Administrative Region Government. The funders support the study financially and have no role in the study design, data collection, data analysis, interpretation, or writing of the report.

Results

Design of influenza RSPVac

For the influenza-specific RSPVac, we designed and expressed the antigen-nucleoprotein fusion protein according to the design formula.¹⁰ We adopted the haemagglutinin globular head (HA head) as the antigen for its neutralising potential, and paired it with influenza nucleoprotein (Fig. 1A). We expressed three versions of influenza-RSPVac, targeting A/H5N1/Vietnam/1194, A/H5N1/Texas/37, and A/H1N1/415742. The nuclear localisation signal was mutated to further facilitate protein production.^15,16 We found that the fusion protein with H5N1/1194 nucleoprotein expressed at the highest level, while those with H1N1 nucleoprotein did not express well (Supplementary Figure S1A). Hence, we adopted H5N1/1194 nucleoprotein for all designs. For the ssRNA, we extracted one of the defective-interfering virus sequences from our previous study.¹⁴ The start and stop codons were mutated to ensure that there are no coding sequences (Fig. 1B). The resulting RNA product is 326bp long, and thus was given the name R326. The expressed proteins were analysed in reducing and non-reducing conditions, and also with glycosidases (Fig. 1C). Band shifts were observed in all proteins after PNGaseF treatment, suggesting N-glycosylation. The produced ssRNA was analysed and validated using Bioanalyzer to ensure single-species RNAs were obtained (Fig. 1D). RSPVac was then generated using mammalian cell-expressed proteins and in vitro transcribed ssRNA (Fig. 1E). After purification and validation, proteins and ssRNA were mixed at a specific ratio (12 μg protein +25 μg R326), generating RSPVac-H5-37, RSPVac-H5-1194, and RSPVac-H1 (Fig. 1F).

RNA-binding ability was confirmed using fluorescence polarisation, with Kd estimated in the nanomolar range (Supplementary Figure S1B). Three versions of RSPVac showed highly comparable binding. Next, we characterised RSPVac using nanoparticle tracking (NTA) (Supplementary Figure S1C). Nucleoprotein is known to form dimers and oligomers in solution, visible under NTA, and the peak median was estimated to be 120.3 nm (±53.0). Interestingly, upon complexing with R326, the peak median increased to 147 nm (±74.2), and the number of particles drastically increased. This indicates protein forming larger size complexes by binding to the ssRNA scaffold. The effect can be visualised in the captured image by NTA. To estimate the stoichiometry of RSPVac, we employed the mass photometry technique. The proteins all showed a molecular weight corresponding to a dimer, and R326 also showed a dimer population (Supplementary Figure S1E and F). Upon mixing protein and ssRNA at the vaccine ratio (12 μg protein +25 μg ssRNA), a broad peak was observed, which corresponded to 427–478 kDa (Supplementary Figure S1G). Since the proteins were observed to be 100 kDa under denaturing conditions, and R326 monomer was ∼117 kDa, this ∼400 kDa specific peak indicated that the stoichiometry of RSPVac complexes can be estimated as 3:1 (Protein:RNA). The mass photometry data also revealed that RNA was in excess in the vaccine formulation, as seen with the peaks of similar ratiometric contrast compared with RNA alone. Based on the peak population estimation of the current formulation, excess RNA contributed to 70% of the mixture, while protein-RNA complexes contributed 30%.

Early activation by nasal RSPVac immunisation is transient and regulated

Understanding how the immune system reacts to nasal immunisation is essential for vaccine development, especially during early activation and reaction to adjuvants. Given that RNA contributes the majority in the formulation, we examined the cytokine responses at the early time point of RSPVac nasal immunisation. BALB/c mice were nasally immunised with RSPVac following the dosage we determined in our previous study,¹⁰ and BALFs were harvested 16-, 24-, and 48-h post-administration to examine early cytokine production (Fig. 2A). Cytokine profiling at the protein level was performed using a bead-based assay. As expected, RSPVac activated potent cytokine production upon administration to mice. Interferon α and γ (Fig. 2B), chemoattractants (Fig. 2C), and inflammatory cytokines (Fig. 2D) production increased at 16 h, and peaked at 24 h. This indicated an enriched environment for immune activation. Interferon β only showed mild induction. Notably, all cytokines returned to similar levels with mock-treated mice at 48 h, as indicated by statistical tests (the top p-value in Fig. 2, comparing mock and 48-h). Other cytokines, such as IL-10, IL-12, and GM-CSF, showed some statistically significant induction, but not as strong as the others (Fig. 2E). This indicated that RSPVac induced a strong but transient innate activation initially, which suggested a controlled response without leading to tissue damage.

Fig. 2.

Cytokine profiling of RSPVac at the early timepoint. (A) Female BALB/c mice (n = 3) were nasally immunised with RSPVac-H1. Mice BALF were harvested at 16, 24, and 48 h post-administration for cytokine analysis using a bead-based multiplex assay. (B) Interferon levels; (C) Chemoattractant levels; (D) Inflammatory-related cytokines; and (E) Other cytokines were shown. Data was shown as violin plots, showing all data points. Statistical significance was determined by using one-way ANOVA. p < 0.05 was considered significant, and the p-values of each comparison were shown, with brackets indicating the comparison. Statistically insignificant p-values were labelled with not significant (ns). Figure A was created in BioRender.

We then performed single-cell RNA sequencing to gain further insights into the initial innate activation. Mice were intranasally administered protein-only, RNA-only, and RSPVac for comparison. BALF-flushed cells harvested 16 h post-administration and subjected to analysis (Fig. 3A). Cytokine profiling of the resulting BALFs by bead-based assay revealed that protein-only samples were similar to mock and showed no induction of cytokine, as evidenced by the statistical testing (Fig. 3B). This indicates that protein by itself has low immunogenicity. RNA-only and RSPVac samples both elicited strong responses of CXCL10, IL6, IL1β, and IFNα, and a moderate response of CXCL1 (Fig. 3B, top row). Other cytokines displayed a small degree of induction but not statistically significant (Fig. 3B, down row). This reflects that RSPVac and the R326 component induce strong chemokines and interferon α that can modulate immune responses. The single-cell RNA sequencing analysis revealed a prominent neutrophil influx across protein-only, R326-only, and RSPVac groups (Fig. 3C–E). However, their expression profiles differed extensively. Protein-only samples mainly contained recruited neutrophils (Csf3r expressing) and metabolically active ECM-remodelling neutrophils (Thbs1, Plau, Fnip2), lacking specific activation and interferon (IFN) signatures, while R326-only and RSPVac samples exhibited increased infiltration of two distinct neutrophil populations: regulatory neutrophils (Ifi207, Gadd45b, Mreg) and effector neutrophils (Tagap, Plac8, Cxcl10). Inflammatory neutrophils (S100a8, S100a9, Retnlg) were present in similar amounts in all groups. For other cell types, there were limited differences between samples based on differential expression analysis, other than a moderate increase in interferon-stimulated genes (ISGs) and chemokine signals. Focussing on regulatory and effector neutrophils, we visualised the distribution of gene expression levels (Fig. 3F). These neutrophil subsets displayed elevated IFN signatures, including increased expression of ISGs, demonstrating their enhanced responsiveness and priming towards antiviral responses. Particularly, these two subsets had a significant increase in Cxcl10 and Cd274, which further indicated their strong chemotactic and immunomodulatory function. Differential gene expression comparing regulatory neutrophils and effector neutrophils deepened the understanding of these two subsets (Fig. 3G). Neutrophil subsets have been described to possess N1 and N2 phenotypes.¹⁷ Regulatory neutrophils were characterised by their elevated Fnip2, Plau, Mreg, and Thbs1, which reflected their specialisation in tissue remodelling and metabolic regulation, and resembled “N2-like” neutrophils that tend to resolve inflammation. Effector neutrophils were defined by their slightly increased ISGs, such as Ifitm6 and Ifih1, and also with increased proinflammatory markers, such as S100a8 and S100a9, and resembled “N1-like” neutrophils that tend to promote inflammation. The presence of both clusters reflected a well-balanced immune environment upon initial administration of either R326 alone or RSPVac. Notably, regulatory neutrophils in the RSPVac group showed distinct upregulation of Ifi207 and H2-Eb1 compared to R326-only (Fig. 3H). This indicated that RSPVac influenced specific gene signatures in this subset, enhancing the immune sensing, immunomodulation, and antigen presentation capabilities in neutrophils, which are also consistent with N2-like behaviour. Note that effector neutrophils showed no significant difference between RNA-only and RSPVac groups. Overall, a balanced neutrophil structure comprising regulatory and effector functions was induced by RSPVac at early nasal immunisation.

Fig. 3.

RSPVac immunisation recruits neutrophils with a distinct gene signature at an early time point. (A) RSPVac-H1, protein-only, R326-only, and PBS were intranasally administered to female BALB/c mice (n = 3), respectively. At 16 h post-administration, BALFs and BALF-flushed cells were harvested and proceeded with cytokine analysis and single-cell RNA sequencing (scRNA seq). (B) Cytokine profiling of BALFs using bead-based assays. Data was shown as violin plots, showing all data points. Similarly, statistical significance was determined by using one-way ANOVA. p < 0.05 was considered significant, and the p-values of each comparison were shown, with brackets indicating the comparison. Statistically insignificant p-values were labelled with not significant (ns). (C) TSNE reduction map of the scRNA seq performed on the BALF-flushed cells. The four samples, mock (PBS), protein-only, R326-only, and RSPVac-H1, were shown. Distinct clusters were coloured, and respective cell identities were under the maps. (D) Cell type proportions of each sample. Colour codes correspond to labels under the TSNE map. (E) Neutrophil population shown in doughnut plots. Data was extracted to focus on neutrophils. (F) Interferon-stimulating genes profile of respective neutrophil populations. Violin plots showing the expression distribution of log-normalised data processed in Seurat. Regulatory and Effector neutrophils were highlighted in red boxes. (G) Heatmap of differential expression between regulatory neutrophils and effector neutrophils. (H) Distinct Ifi207 and H2-Eb1 differential expression in regulatory neutrophils, between R326-only and RSPVac-H1 samples. Fold change in log2 scale, and p-values were indicated. Reduction maps, bar plots, doughnut plots, violin plots, and heatmaps were generated in R using Seurat and ggplots2. Differential expression was performed in Loupe Browser 8 based on the CellRanger pipeline. Figure A was created in BioRender.

Immunogenicity and protective immunity elicited by RSPVac against highly pathogenic H5N1

To examine influenza-RSPVac immunogenicity and protectiveness, BALB/c mice were vaccinated intranasally twice at a 14-day interval, using the previously established dosage¹⁰ (Fig. 4A). At 10 days post-2nd dose, sera and BALFs were harvested for antibody analysis. Cells enriched by BALF were collected for IFNγ ELISpot analysis. All influenza-RSPVac elicited strong antibody responses, including systemic anti-HA IgG (Fig. 4B), mucosal IgG (Fig. 4C) and IgA (Fig. 4D), as well as anti-NP antibodies (Fig. 4E). Protein-only groups showed weak antibody induction, and the mean values were statistically lower than those of the RSPVac groups. Serum Anti-HA IgG also showed a balanced IgG1/IgG2a ratio, implying induction of both Th1 and Th2 responses (Fig. 4F). IgG elicited against H1 were slightly skewed to IgG1 according to statistical testing, while H5 antibodies were highly balanced. Consistently, BALF cells from these vaccinated mice stimulated with HA proteins showed increased IFNγ production, indicating mucosal resident T cell activation (Fig. 4G). To further analyse the T cell responses, we harvested single cell suspensions from vaccinated mice lungs at day 7 post-2nd dose and stimulated with overlapping peptide pools of nucleoprotein originating from avian flu (H5N1 and H7N9) and pandemic (H1N1) influenza, and detected T cell activation (Fig. 4H). All three peptide pools of different influenza viruses were able to recall CD4+ and CD8+ memory cells and produce IFNγ (Fig. 4I and J). Furthermore, the T cell activation implied cross-reactive T cells that recognise common epitopes of the highly conserved nucleoprotein.

Fig. 4.

Mouse antibody and T cell responses following RSPVac nasal vaccination. (A) Vaccination regimen. Female BALB/c mice (n = 6–9) were intranasally vaccinated with 2 doses of RSPVac, 14 days apart. Sera, BALF, and BALF cells were harvested at day 10 post-2nd dose for analysis. Control mice received the protein component without RNA (protein-only). Data was shown as box plots with respective colours indicating vaccine given, showing all values. Error bars represented the highest and lowest values of each box. ELISA was used to measure antibody responses for (B) anti-HA serum IgG, (C) anti-HA BALF IgG, (D) anti-HA BALF IgA, (E) anti-NP, and (F) serum anti-HA IgG subtypes for each of the RSPVac, respectively. (G) Mucosal T cell responses measured by IFNγ+ ELISpot using the immunogen (the protein used to generate RSPVac) stimulation of BALF cells. (H) To analyse lung T cells, female BALB/c mice (n = 6) were intranasally vaccinated with 2 doses of RSPVac, 14 days apart. Lungs were harvested at day 7 post-2nd dose. Single cell suspensions were prepared, and the NP peptide pool of H5N1, H1N1, and H7N9 was used to stimulate lung cells. Resident T cell responses were analysed by intracellular staining and FACS analysis. NP-reactive (I) CD4+ and (J) CD8+ T cells were shown, comparing protein-only and RSPVac-vaccinated. Statistical significance was determined by the Mann–Whitney test. p < 0.05 was considered significant. P-values were shown, and those that were considered insignificant were labelled not significant (ns). Figures A and H were created in BioRender.

For neutralising antibodies, RSPVac-H1 elicited antibodies showed strong neutralising ability in FRNT, complemented by haemagglutination inhibition (HAI) function. For both H5N1-targeting RSPVac, only 2–3 mice showed neutralising ability (Fig. 5A and B). We found that this observation is similar to recent vaccine studies using mouse models, which have difficulties detecting neutralising antibodies against H5.18, 19, 20 This is possibly due to epitope limitation in mice. Nevertheless, we further examined RSPVac protectiveness against lethal virus infection. Vaccinated animals were challenged with lethal doses of vaccine-matched viruses. Body weight across 14 days and disease severity for the first 7 days post-infection were monitored (Fig. 5C). Consistent with the robust humoural responses, vaccinated mice were completely protected against the lethal virus challenges (Fig. 5C–I). We also observed that the human H5N1/Texas/37 virus displayed increased virulence compared to the H5N1/1194/Vietnam virus. During H5N1/Texas/37 infection, mock-vaccinated animals experienced significant pathology and required earlier euthanasia due to severe neurological symptoms (Fig. 5E). In contrast, RSPVac vaccinated mice showed little to no pathological signs against the clade 2.3.4.4b virus, indicating sufficient protection. We also observed undetectable viral loads in the lung and nasal turbinates of vaccinated animals during H5N1 (Fig. 5J–K) and H1N1 (Fig. L–M) infection, which indicated full protection despite neutralising antibodies not being detectable. For mice vaccinated with protein-only, although some levels of antibody responses were observed, it did not confer protection against H5 and H1 infection (Supplementary Figure S2).

Fig. 5.

Protection against lethal virus infection elicited by RSPVac. (A) HAI and (B) FRNT50 measurement for neutralising antibodies in the sera samples of Fig. 4. HAI was shown as box plots, showing all points. N.D., not detected. FRNT50 was shown as individual data points, with error bars showing SD. (C) To examine RSPVac protection against virus infections, female BALB/c mice (n = 6) were intranasally vaccinated with two doses of RSPVac, 14 days apart. Control mice received PBS only. It was then followed by nasal challenge with a lethal dose of virus (H5N1/TX/37 and H5N1/VN/1194: 30 PFU/mouse; H1N1/pdm09: 1000 PFU/mouse) at 14 days post-2nd dose. Body weight was monitored for 14 days. Disease severity scoring was conducted from day 0–7. Lungs and nasal turbinates (n = 6) were harvested to test for viral loads. Body weight changes, survival, and disease severity for (D–E) H5N1/TX/37, (F–G) H5N1/Vietnam (VN)/1194, and (H–I) H1N1/pdm09 infection were shown. Disease severity was determined by assessing symptoms, including ruffled fur, hunched back, laboured breathing, and inactivity. Diseases in individual mice were shown. Colour intensity indicates disease severity. A cross indicates animal death. Mice were euthanised when body weight loss exceeded 30%. Lungs and nasal turbinates viral loads and virus titres (J–K) H5N1 and (L–M) H1N1 infection were shown as bar plots, with error bars indicating SD. Viral titres PFU/mL/g and viral loads M gene copy/g were normalised to the average weight of the tissue. Figure C was created in BioRender.

A well-trained immune landscape prepares vaccinated mice for unmatched challenges

Mice vaccinated with RSPVac-H5-1194 were also lethally challenged at day 14 post-2nd dose by avian H7N9/AH1 and distant H1N1/PR8, to simulate a vaccine-unmatched situation in humans when an unpredictable zoonotic jump occurs (Fig. 6A). To further understand how RSPVac protected against vaccine-unmatched influenza infection, we collected BALF flushed cells at day 6 post-infection from mock-vaccinated and RSPVac-H5-1194 vaccinated mice that were infected with vaccine-matched (H5N1/1194) and vaccine-unmatched viruses (H7N9/AH1 and H1N1/PR8). Cells were then processed and analysed by single-cell RNA sequencing (Fig. 6B).

Fig. 6.

Well-trained immune system and protection against vaccine-unmatched viruses. Two experiments were set up to examine vaccine-unmatched infection. (A) Female BALB/c mice (n = 6) were nasally immunised with 2 doses of RSPVac-H5-1194, 14 days apart. And followed by nasal challenge with lethal doses of H1N1/PR8 (100 PFU/mouse) and H7N9/AH1 (2 × 10⁵ PFU/mouse) at 14 days post-2nd dose. Mock-vaccinated mice received PBS only. Body weight and survival were monitored for 14 days. (B) In addition, another batch of mice (n = 3) received the same vaccination and were infected with vaccine-matched H5N1/1194 (30 PFU/mouse) and vaccine-unmatched H1N1/PR8 (100 PFU/mouse), and H7N9/AH1 (2 × 10⁵ PFU/mouse). BALF-flushed cells were harvested on day 6 post-infection, followed by processing and analysis by single-cell RNA sequencing. (C and D) Body weight changes and survival for vaccine-unmatched H1H1 and vaccine-unmatched H7N9 infection. Mice were euthanised when body weight loss exceeded 30% for H1N1/PR8, and 20% for H7N9/AH1 due to differences in pathology. (E) TSNE reduction map of mock, RSPVac-H5-1194 vaccinated, mock infected H5N1/1194, and RSPVac-H5-1194 vaccinated + H5N1/1194 infected samples. Cell types were labelled with respective colours and shown under the plots. (F and G) TSNE maps for mock and vaccinated mice infected by H1N1/PR8 and H7N9/AH1 were shown. Gene expression of different cell types was analysed using Loupe Browser 8 differential expression (based on CellRanger). Heatmaps for differential gene expression profiles for (H–J) macrophages, (K) B cells, (L) CD4+ T cells, and (M) NK-like CD8+ T cells were shown and labelled for the corresponding infection samples. Key genes described in the text were highlighted in black boxes. Figure A and B was created in BioRender.

For lethal virus challenges, vaccinated mice showed body weight loss for both challenges, and subsequently recovered. Protection for H7N9/AH1 is relatively better than for H1N1/PR8, as shown by the reduced weight loss and faster recovery (Fig. 6C and D). These observations indicated that vaccine-unmatched viruses cannot be neutralised initially, but the immunity was sufficient to rescue the vaccinated animals. This protective effect is consistent with, and most likely accounts for, the cross-reactive NP-specific T cells observed earlier.

During H5N1/1194 infection, the immune cell landscape in mock-vaccinated mice was significantly altered (Fig. 6E). Compared to non-infected mice, a significant influx of macrophages, neutrophils, NK cells, and B/T cells was observed. Inflammatory cytokines were expressed by innate cells. This indicated that the immune system was actively responding to a new threat and developing adaptive responses to counter virus invasion. In contrast, vaccinated mice were protected from a vaccine-matched virus when compared to uninfected mice. Resident B and T cells were observed in the RSPVac-vaccinated mice's BALF. Differential gene expressions of infected and non-infected RSPVac-vaccinated mice did not show significant differences, suggesting that the virus infection had minimal impact. These data reflected that RSPVac elicited a strong protection against the matched virus and neutralised the viral challenge.

Infection with vaccine-unmatched viruses provided a different perspective when neutralisation is absent. Under H7N9 and H1N1 infection on H5N1-vaccinated mice, the immune landscape and cell types between mock and vaccinated mice were highly similar (Fig. 6F and G). Vaccine-unmatched infection caused a significant influx of neutrophils, monocytes, NK cells, inflammatory/regulatory macrophages, and also B/T cells. This suggested that the immune system cannot neutralise the virus immediately, and infection was established. This is expected and consistent with the delayed body weight recovery during infection as observed earlier. To further understand how the immune system reacted and led to recovery, we examined the differential gene expression in different cell types and compared between mock and vaccinated (Fig. 6H-M). Neutrophils, monocytes, and NK cells did not show significant differential gene expression. Three cell types showed distinct differences: macrophages, B cells, and T cells. For both H1N1 and H7N9 infection, firstly, macrophages in vaccinated mice showed increased MHC-II expression (Cd74, H2-Aa, H2-Eb1, H2-Ab1), indicating their increased antigen-presentation abilities to the adaptive immune system (Figure H–J). Secondly, for B cells, only vaccinated mice showed Igha and Jchain expression (Fig. 6K). This reflected that the vaccinated mouse effectively produced IgA against influenza but not the mock mice, reflecting IgA B cell memory and indicating RSPVac activation breadth. Lastly, CD4+ T cells in vaccinated mice showed a specific pattern, with elevated Lag3, Il10, Ccl4, Cxcr6 during H1N1 infection, and Rbpj, Ccl3, Ccl4, Itgb1, Cxcr6 during H7N9 infection (Fig. 6L). These profiles reflected an enhanced CD4+ helper T cell function by combining regulatory functions and preventing excessive activation (Lag3, Rbpj, Il10), as well as tissue homing and residency (Ccl3, Ccl4, Cxcr6). For the NK-like CD8+ T cells, Trbv4, Ccl5, S100a4 are common signatures between H1N1 and H7N9 infection (Fig. 6M). This indicated activation of a specific, but common T cell repertoire and enhanced tissue homing for these cytotoxic T cells. These gene signatures reflected enhanced T cell functionality. Collectively, this showed that RSPVac nasal immunisation induced a well-trained immune landscape that involved enhanced innate and adaptive immune cells, which helped to defend the vaccinated animals against vaccine-unmatched infection when neutralisation is not available.

The biomolecular properties of RSPVac influence vaccine effectiveness

The well-trained immune system induced by RSPVac observed during vaccine-unmatched infection warrants further investigation into the mechanism of protein-RNA complexes immunisation. In the current formulation, R326 is in excess compared to protein (Fig. 1). Our data on the early immune response profiling suggested that R326 itself can induce potent cytokines as well (Fig. 3). These data implied that the R326 may be accountable for the robust mucosal activation, while the inflammation reaction is controlled. However, there were still subtle differences when protein-RNA complexes were presented (Fig. 3H). The functional role of protein-RNA complexes, therefore, needs to be studied more extensively. To this end, we further tested mutant nucleoproteins with defective RNA-binding abilities in RSPVac and examined the immunisation outcome (Fig. 7A). Arginine residues in the H5N1-nucleoprotein were substituted to reduce RNA affinity by 7-fold^15,16 (Fig. 7B). Mutant RSPVac was generated using a fusion protein consisting of mutated H5N1-nucleoprotein and H5-1194 HA head, and R326. Mice were nasally immunised with wildtype (WT) and mutant RSPVac-H5-1194 using the same “two-dose, 14 days apart” regimen as above. The same amount of RNA was present in each dose of vaccine, but protein-RNA complex formation was disrupted in RSPVac with RNA-binding mutant. Critically, we observed that the vaccination outcome was influenced by the interaction between proteins and ssRNAs. Serum anti-HA IgG levels were similar between groups, while BALF anti-HA IgA was slightly, yet significantly, reduced in the mutant group (Fig. 7C and D). Mutant immunisation also caused a mild imbalance in serum anti-HA IgG1/IgG2a ratios compared to the balanced response in WT, with further reduction of IgG2a in BALF (Fig. 7E and F), indicating disrupted mucosal antibody subtype balance. Moreover, BALF T cell responses were significantly impaired in mutant immunised mice (Fig. 7G). This reflected that the defective-binding property weakened the mucosal T cell response in mice vaccinated with mutant RSPVac. This reflected that RNA binding by the protein, and subsequently complex formation, is critical for inducing balanced antibody and mucosal T cell responses. Furthermore, we conducted an additional experiment to explore the underlying mechanism. Viral nucleoproteins and RNAs can undergo liquid–liquid phase separation, as shown for the SARS2 nucleoprotein.²¹ Influenza nucleoprotein also exhibits oligomerization and contains intrinsically disordered regions. We investigated the role of phase separation in RSPVac's immunisation by generating a phase separation-deficient SARS2-RSPVac mutant (Psmut, with nucleoprotein deleting residues 369–390), which retains RNA-binding but lacks phase separation ability (Fig. 7H). The WT SARS2-RSPVac was the same as documented in our previous study, consisting of SARS2 nucleoprotein and spike receptor binding domain, and the SARS2 ssRNA.¹⁰ The Psmut used the same configuration but with a mutant protein. While antibody responses in both serum and BALF were unaffected (Fig. 7I–M), mucosal T cell responses were significantly impaired (Fig. 7G). This suggested that proper antibody induction depends on proper RNA binding, whereas T cell activation was potentially influenced by both RNA binding and phase separation. This indicates that protein-RNA complexes are crucial for balanced immunisation and also implies an activation mechanism for RSPVac nasal immunisation that warrants deeper characterisation.

Discussion

Influenza has been a major public health burden for many decades and has not slowed down, even after the COVID-19 pandemic, despite increased public hygiene awareness. Moreover, the recent spillover of avian H5N1 clade 2.3.4.4b to humans has reignited concerns about influenza potentially causing the next global pandemic.^1,4 Vaccines continue to play an irreplaceable role in influenza control, as demonstrated by the rapid responses of pharmaceutical companies and the WHO in providing candidate vaccines for avian flu. Although this recent outbreak did not escalate into an uncontrollable pandemic, human cases continued to emerge occasionally. During 2024, a chain of human cases emerged in the US and other countries, but the risk diminished over time. However, 12 cases have emerged since January 2025.²² Other avian spillovers to humans involving non-stockpiled subtypes have occurred over the years, including H3, H9, and H10 viruses.23, 24, 25, 26, 27 It remains crucial to invest in the development of improved vaccines to protect against zoonotic influenza when urgently needed. Intranasal vaccines have long been advocated for respiratory diseases due to their superior ability to induce mucosal IgA and neutralising antibodies, which can effectively prevent viral transmission.⁸ In contrast, the stockpiled H5N1 vaccines are mostly intramuscular inactivated vaccines, which do not elicit mucosal responses. Live attenuated virus vaccine is the only nasal vaccine approved for human use to date. While careful design can minimise the risk of reassortment with natural viruses, these vaccines can still cause disease-like symptoms and often have reduced efficacy in humans.^28,29 Novel vaccine technology is needed to advance nasal vaccine development, enabling safer vaccination and stronger mucosal immunity to halt virus transmission.

Recombinant protein vaccines have the potential to avoid the risks associated with attenuated viruses and can also potentially boost existing immunity, but are often limited by low immunogenicity. Numerous strategies have been explored to enhance their immunogenicity, including increasing molecular size, fusing with immunogenic partners, and incorporating adjuvants.³⁰ To address the challenges and design a better nasal vaccine, we developed the RSPVac. This approach combines increased molecular bulk, inclusion of both neutralising and cross-protective epitopes, and adjuvants in a single formulation. Importantly, all components are derived exclusively from viral sources, rather than artificial, plant, or chemically synthesised materials, to enhance specificity and reduce non-specific immune responses. In our previous work, we successfully generated a SARS2-RSPVac and demonstrated strong protective immune responses in a mouse model.¹⁰

In this study, to respond to the threat of recent H5N1 spillover and the need for vaccine stockpile preparedness, we applied the RSPVac technology to generate influenza vaccines and examined their versatility to prepare for emerging infectious diseases. RSPVac induced robust antibody responses in both systemic and mucosal systems (Fig. 4). Resident T cell responses in the lung were also induced, including both HA-specific and NP-specific. We collected samples at day 7–10 post-2nd dose to capture the peak adaptive responses. Vaccinated mice were completely protected against vaccine-matched H5N1 and H1N1 lethal challenges; whereas protection against disease weight loss was incomplete following challenges by vaccine-unmatched pandemic H1N1 and avian H7N9, but mice were protected from mortality (Figs. 5 and 6). We selected day 6 post-infection for single-cell RNA sequencing analysis to examine the protective effect of RSPVac when vaccine-unmatched infection was not controlled by neutralisation and manifested into tissue damage. The activation of cross-reactive NP T cells and the expression of the common Trbv4 gene in CD8+ T cells indicated that NP immunity played a major role during cross-protection. This also indicates the functional design of including nucleoprotein in RSPVac, not only for RNA-binding to generate immunogenic protein-RNA complexes, but also for protection when neutralisation for vaccine-unmatched viruses is unavailable. This is important to control avian spillover infections, when the HA may not be sufficiently neutralised by the current human immunity. Infected cells can be effectively eliminated before the highly pathogenic avian flu reaches a critical stage of infection, as nucleoprotein similarity is more than 90%.^31,32 To further enhance the efficacy of RSPVac in future development, more neutralising epitopes can be incorporated in the design. For instance, including the HA stalk in the protein design. Nevertheless, in the absence of neutralising antibodies, NP immunity elicited by RSPVac may provide important protection against severe diseases in the mouse model. The HA head domain could possibly provide epitopes for eliciting cross-reactive antibodies, but may not be observed in mice due to the inherent limit in epitope breadth. Moreover, the innate immune activation we observed could also play a role in controlling vaccine-unmatched infections. These warrants further investigations to deepen the understanding of RSPVac mechanisms.

Using mutant proteins, we demonstrated that altering the biomolecular properties of the protein-RNA complex impacted vaccination outcomes. Specifically, reduced RNA affinity and the loss of phase separation in RSPVac impaired mucosal T cell responses (Fig. 7G and N). These findings align with existing research and hypotheses in the field. The RNA-binding capacity of the protein may be crucial for co-delivering antigen and adjuvants, potentially enhancing cross-presentation and T cell activation.33, 34, 35 Regarding phase separation, it is plausible that liquid–liquid phase separation (LLPS) of RSPVac at immune cell membranes facilitates receptor clustering, thereby amplifying activation signalling.36, 37, 38, 39 While these remain hypotheses based on current understanding, they warrant comprehensive mechanistic investigation. Our results highlight a potentially complex mechanism that could provide new insights for mucosal vaccine design, though further detailed studies are much needed. Exploring the link between biophysical properties and immune activation remains an important area for future research.

There are several limitations in this current study. The current configuration of RSPVac contains excess RNA, and our data indicate that R326 contributes substantially to immune activation. We observed a transient and regulated profile during early activation in preclinical models; however, the implications for human safety remain uncertain and require further evaluation. Moreover, standardising vaccine production and vaccine stability still requires deeper research and development. Despite the utilisation of various physical assays in this study, standardising RSPVac in this configuration can prove difficult in manufacturing. Therefore, optimisation of methods should be investigated in future studies.

We also acknowledge several limitations in the current methods. While this study focused on the primary immunisation and immune response of RSPVac, the relatively short interval between the prime-boost regimen and virus challenge limits the assessment of the longevity of RSPVac-elicited protection. The likelihood and duration of an individual being infected after vaccination may be impossible to determine, and thus, the durability of RSPVac intranasal vaccination must be evaluated in future studies. In particular, characterising memory B and T cell populations and their recall capacity upon H5N1 exposure will be essential for understanding the potential of RSPVac. This also raises the issue regarding pre-existing immunity in humans. Resident memory immunity induced by prior influenza infection can help with virus clearance but would also affect vaccine efficacy, as immune memory tends to be biased toward previously encountered strains, resulting in reduced responses to contemporary variants. Naïve mice were used in this experimental study, which limits the relevance of the results to the majority of human cases having a certain degree of prior immunity. While our power calculation indicated that approximately 4 mice per group (rounded up from 3.84) would be sufficient to detect the primary effect of interest, we chose to use N = 6–9 mice per group to align with common practice in animal vaccine-immunogenicity studies and to enhance statistical reliability and reproducibility, and also reduce the risk of having to repeat experiments that would require additional animals. In this study, we focused on the primary immunisation efficacy; however, an additional full set of prime-boost study of RSPVac in comparing the naive and pre-immunised animals will provide additional insights into the further development of this vaccine in humans. Although this study examines immunogenicity in mice and includes control groups, incorporating blinding during data acquisition would be a valuable improvement to reduce potential bias and further enhance the rigour and reproducibility of the findings. Additional animal models, such as ferrets and macaques, should be tested. Multivariate analysis, including other cofounding factors such as sex and age, should also be investigated. Dosing, safety, and durability of immune responses should be determined for progressing to Phase I trials. Overall, we showed the effectiveness of intranasal RSPVac targeting highly pathogenic avian H5N1 influenza in a preclinical mouse model and demonstrated that RSPVac design is promising for responding to newly emerging infectious diseases.

Contributors

J.Y.L and K.H.K conceptualised the vaccine design, protein and ssRNA design, and animal experiment design. J.Y.L and S.K.C expressed, produced, and purified the recombinant proteins and RNA. J.Y.L carried out the animal immunisation and animal sample harvesting. J.Y.L and C.K.Y performed experiments in the physical containment level 3 laboratory. J.Y.L, S.K.C, and P.H.C performed the downstream immunological assays (ELISAs, ELISpots, intracellular staining, and flow cytometry) and acquired data. Data analysis was performed by J.Y.L and K.H.K. J.Y.L prepared the manuscript original draft and figures, while K.H.K reviewed and edited. Manuscript revision was reviewed and done by J.Y.L, K.H.K, and K.Y.Y. All authors have read and approved the final version of the manuscript. JYL and KHK both have access to the data of this manuscript and verified the data.

Data sharing statement

Single-cell RNA sequencing data are deposited at NCBI BioProject (PRJNA1169636, PRJNA1170417). All raw data and materials can be provided on request. All data needed to evaluate the conclusion of this study are present in this paper and the Supplementary Materials.

Declaration of interests

Authors Joy-Yan Lam, Kwok-Yung Yuen, and Kin-Hang Kok are listed as inventors on patent applications (US 18/643883, EU 24172271.9, CN202410502484.3A, JP2024070290A) related to the vaccine technology described in this work are pending at the time of manuscript preparation. The other authors declare that they have no competing interest.