Single‐dose immunisation with a multimerised SARS‐CoV‐2 receptor binding domain (RBD) induces an enhanced and protective response in mice

The COVID‐19 pandemic, caused by the SARS‐CoV‐2 coronavirus, has triggered a worldwide health emergency. Here, we show that ferritin‐like Dps from hyperthermophilic Sulfolobus islandicus, covalently coupled with SARS‐CoV‐2 antigens via the SpyCatcher system, forms stable multivalent dodecameric vaccine nanoparticles that remain intact even after lyophilisation. Immunisation experiments in mice demonstrated that the SARS‐CoV‐2 receptor binding domain (RBD) coupled to Dps (RBD‐S‐Dps) elicited a higher antibody titre and an enhanced neutralising antibody response compared to monomeric RBD. A single immunisation with RBD‐S‐Dps completely protected hACE2‐expressing mice from serious illness and led to viral clearance from the lungs upon SARS‐CoV‐2 infection. Our data highlight that multimerised SARS‐CoV‐2 subunit vaccines are a highly efficacious modality, particularly when combined with an ultra‐stable scaffold.

The Spike protein, which is embedded in the virus' membranous envelope, is essential for the interaction with human angiotensin-converting enzyme 2 (hACE2) [6]. It is the interaction with hACE2 that is thought to initiate the process that leads to cell entry of viral RNA and infection [7]. The Spike protein is translated as a single polypeptide that is proteolytically processed into its two subunits, S1 and S2. The Spike of SARS-CoV-2 is a trimer consisting of three S1-S2 heterodimers [8]. For membrane fusion between the cell and the virus to occur, two cleavage events within the Spike complex are required [6]. A protease cleavage site located between S1 and S2 is cleaved by the producer cell's proprotein convertase furin during virus assembly [9] (Fig. 1A). The second cleavage site is located in the S2 domain at position R797, and its hydrolysis by the target cell's surface protease TMPRSS2 triggers membrane fusion and cell entry [9].
The SARS-CoV-2 receptor binding domain (RBD) is located within the S1 subunit of the Spike. It is the RBD that interacts directly with the host cell via the hACE2 receptor [6]. It is therefore not surprising that antibodies directed against the RBD and that overlap with the ACE2 binding region are strongly neutralising, making the RBD a promising subunit vaccine candidate [6,10]. The RBD is glycosylated and contains four disulphide bridges that contribute to its stability, necessitating its expression in mammalian cells, as is also the case for the Spike.
To end the pandemic, vaccines are by far the most promising approach and vaccine developments, clinical trials, approvals, and mass roll-outs are in progress. So far, until May 2021, 89 COVID-19 vaccines have been tested in clinical trials. Of those, 36 are undergoing safety trials, 27 are in the phase of largescale testing, 6 vaccines are authorised for limited use, and 8 vaccines are fully approved [2]. All approved vaccines show good-to-excellent protection against severe illness, and preliminary data have shown that virus transmission is significantly reduced in vaccinated individuals [11,12]. Most of the approved vaccines and those in late-stage trials are mRNA-based, vector-based, inactivated viruses or DNA vaccines [11]. Vector-and RNA-based vaccines can often be rapidly developed because they deliver the immunogen coding sequence rather than the immunogen itself. Currently, only one vaccine candidate in late phase trials is a protein-based subunit vaccine, Novavax [13]. Some subunit vaccines are amenable to processes such as lyophilisation that remove the need for a complex storage and distribution cold-chain. As such, they provide substantial advantages over nucleic acid-based vaccines in the quest for complete and global vaccination. A second challenge facing global vaccination is the emergence of viral variants, some of which are more infectious and/or cause more severe illnesses, and reduce the efficacy of existing vaccines [14][15][16][17]. Repeat vaccinations directed against these variants, but that use the same type of vaccine, could be problematic. This is because immunity is generated against the vaccine vector itself, neutralising it before it can deliver its immunogen cargo [18]. It is anticipated that in future, several different types of vaccines will be required to cope with emerging variants of SARS-CoV-2.
Previous work has shown that protein-based subunit vaccines directed against SARS-CoV-2 deliver high antibody responses in animal models [19,20]. Furthermore, subunit antigens have the potential to deliver a cheaper, boostable, and more robust alternative to nucleic acid-based vaccines [21][22][23][24][25][26][27][28][29][30]. To explore the development of stable and efficient subunit vaccine candidates, here we covalently linked SARS-CoV-2 proteins expressed in mammalian and bacterial cells with bacterially expressed Dps from the hyperthermophilic archaeon Sulfolubus islandicus [31]. Immunisation using SARS-CoV-2 RBD linked to Dps (RBD-S-Dps) proved to be highly effective in eliciting an immune response, including after lyophilisation, and to produce neutralising antibodies that inhibit cell entry in vitro. Furthermore, transgenic K18-hACE2 mice infected with SARS-CoV-2 were completely protected from serious illness following a single immunisation with RBD-S-Dps.

Materials and methods
Cloning, expression and purification of the protein components

SpyC-Dps
A hexa-histidine tag was fused to DN1-SpyCatcher, which was subsequently linked to the Dps from S. islandicus (ORF SIL_0492, GenBank AGJ61963.1), separated by a GSEGSSGG-linker (Table S1, SpyC-Dps). The sequence was codon optimised for the expression in Escherichia coli, and the gene was cloned into the pOPINS vector by Gibson assembly. The plasmid encoding for SpyC-Dps was transformed into C43(DE3) E. coli. Cells were grown at 37°C in 2xYT medium to an OD 600 of 0.8. Protein production was induced with 1 mM IPTG for 6 h. Cells were harvested at 4500 g for 25 min at room temperature (RT). Cells were shock-frozen in liquid nitrogen (LN2) and stored at À80°C. Cells producing SpyC-Dps were resuspended in T-buffer1 (30 mM Tris, 250 mM NaCl, pH 8.0) with one tablet of Complete Protease Inhibitors (Roche, Basel, Switzerland) per 10 g cells wet weight. Cell disruption was carried out using sonication for 7.5 min 'on' time, using a 50% duty cycle. Cell debris were removed by centrifugation at 20 000 g for 30 min at RT. The supernatant was loaded onto a HisTrap FF affinity chromatography column (Cytiva, Freiburg im Breisgau, Germany). Washing was carried out for 17 column volumes (CV) with T-buffer1 plus 110 mM imidazole. The protein was eluted with T-buffer1 containing 400 mM imidazole. Purity of fractions was examined by SDS/PAGE, and the purest fraction were pooled and concentrated using a Vivaspin Turbo centrifugal concentrator (100 000 MWCO, Sartorius, G€ ottingen, Germany). Concentrated sample was loaded onto a size exclusion column (SEC, Sephacryl S-400, Cytiva), with PBS as the running buffer. Purity was examined by SDS/ PAGE, and the sample was frozen in LN2 and stored at À80°C.

SpyT2-NP
The NP (amino acids 48-364; GenBank: MN908947; NP) was cloned into the vector pOP-TH and N terminally equipped with a hexa-histidine tag [32]. A SpyTag2 sequence separated by GS-rich linkers was inserted between the hexahistidine tag and NP ( Fig. 1 and Table S1, SpyT2-NP). The vector encoding for SpyT2-NP was transformed into E. coli C41(DE3) cells. For protein expression, cells were grown at 37°C in 2xYT medium to an OD 600 of 0.7. Protein production was induced with 1 mM IPTG for 6 h. Cells were harvested at 4500 g for 25 min at 4°C. Cells were frozen in LN2 and stored at À80°C. SpyT2-NP-producing cells were resuspended in T-buffer2 (50 mM Tris, 1 M NaCl, 10 mM imidazole, 2 mM DTT, pH 8.0) with Complete Protease Inhibitor added (1 tablet per 10 g cells wet weight). Cells were lysed by sonication (3 min total 'on' time, duty cycle 50%).

Spike-SpyT2 and Spike
To express the ectodomain of the stabilised prefusion Spike protein trimer [33] with only one subunit carrying the Spy-Tag2 tag, two constructsone with and one without a SpyTag2were made. First, a gene encoding residues 16-1208 of SARS-CoV-2 Spike protein (GenBank: MN908947) with proline substitutions at residues 986 and 987, a GSAS substitution at the furin cleavage site (residue 682-685), a C-terminal T4 fibritin trimerisation motif, a GGSGGS linker, an HRV3C protease cleavage site, a GGS linker and an AviTag, was synthesised and cloned into the lentiviral expression vector pHR-SFFV [34][35][36] downstream of the sequence encoding the chicken RPTPr secretion signal peptide (cRPTPrSP) [37]. Then, either a GGS linker and a hexa-histidine tag, or a GGS linker, an octa-histidine tag, a GGSGGSGGS linker and a SpyTag2 were inserted after the AviTag sequence to form two Spike constructs, with and without a SpyTag2 (Table S1, Spike-SpyT2 and Spike, respectively). For protein expression and purification, see the next paragraph.

RBD-SpyT2
A gene encoding residue 332-529 of SARS-CoV-2 Spike protein (constituting the receptor binding domain, RBD) was synthesised and cloned downstream of cRPTPr of the pHR-SFFV vector and a GGSGGS linker, an AviTag, a GGS linker, an octa-histidine tag, a GGSGGSGGS linker and a Spy-Tag2 were inserted at the 3 0 end of the gene (Table S1, RBD-SpyT2). The vectors for Spike-SpyT2, Spike and RBD-SpyT2 were used for protein production in the mammalian lentiviral expression system [34][35][36]. The DNA of the constructs was mixed with the lentiviral envelope and packaging vectors pMD2-G and psPAX2c (Addgene, Watertown, MA, USA) and polyethylenimine (PEI, Sigma Aldrich, St. Louis, MO, USA) to transiently transfect HEK 293T Lenti-X cells (Takara Bio, Kusatsu, Japan) to make lentiviral particles. To make Spike trimer protein with only one subunit carrying a SpyTag2, the DNAs of constructs Spike and Spike-SpyT2 were used at a molar ratio 3 : 1. The virus particles produced were used to infect HEK 293S GnT1 À/À cells (for Spike-SpyT2) or Expi 293 cells (for RBD-SpyT2). The infected cells were then expanded to obtain 3 L cultures, and conditioned media were harvested and sterile filtered (0.22 lm). The supernatant was concentrated and the buffer exchanged to 25 mM Tris pH 8.0, 300 mM NaCl using an € Akta flux tangential flow system (Cytiva). The conditioned supernatant was then loaded onto a HisTrap column (Cytiva) and washed and eluted with 50 and 250 mM imidazole in the same buffer, respectively. Eluted fractions were checked by SDS/PAGE, pooled and further purified in PBS buffer by SEC on Superdex 200 for RBD and Superose 6 for trimeric Spike protein (both Cytiva). Peak fractions were checked by SDS/PAGE again and frozen in LN2 and stored at À80°C.

Coupling and purification of multimerised complexes
For the preparations of Ag-S-Dps complexes, comprising RBD-S-Dps, NP-S-Dps and Spike-S-Dps the antigens: RBD-SpyT2, SpyT2-NP and Spike/Spike-SpyT2, and the scaffold protein SpyC-Dps were diluted in PBS buffer + 250 mM NaCl to 0.2-1 mgÁmL À1 and mixed. To achieve full occupancy of SpyC-Dps with the antigens, the molar ratio for SpyC-Dps to RBD-SpyT2 was 1 : 1.3, for SpyT2-NP 1 : 2 and for trimeric Spike/Spike-SpyT2 1 : 2.5. Reactions were left for~5 min at RT, and covalent coupling between SpyCather2 and SpyTag2 was checked by SDS/PAGE. Subsequently, samples were concentrated using Vivaspin Turbo concentrators (100 000 MWCO). Antigen-decorated SpyC-Dps complexes were separated from the excess antigens by SEC in PBS + 250 mM NaCl on a Superose 6 Increase column (Cytiva). Fractions were checked again for purity by SDS/PAGE, frozen in LN2 and stored at À80°C.

Negative-stain electron microscopy
Proteins were diluted in PBS to concentrations of 0.012 mgÁmL À1 . Three microlitre of the solution was applied to a glow-discharged carbon-coated grid and immediately blotted. For the staining, 10 lL of 2% (w/v) uranyl formate were applied and removed immediately by blotting the grid with filter paper. Images were collected on a FEI Tecnai Spirit 120 kV electron microscope, equipped with a CCD detector.

In vitro human plasma stability assay
The in vitro stability of RBD-S-Dps was studied in clotted human plasma (MD Biomedicals, Taipei City, Taiwan; cat.

Lyophilisation of samples
An aliquot of RBD-S-Dps of 120 µL (at a protein concentration of 1.4 mgÁmL À1 , in PBS buffer plus additional 250 mM NaCl) was divided into a 40 µL control and a second aliquot of 80 µL.

Spike-pseudotyped neutralisation assays with mouse sera
HEK 293T-hACE2-TMPRSS2 cells were described previously [9]. Cells were plated into 96-well plates at a density of 0.75 9 10 3 cells per well and allowed to attach overnight. Twenty microlitre pseudovirus-containing supernatant was mixed with 2 µL dilutions of heat-inactivated mouse sera and incubated for 40 min at RT. Ten microlitre of this mixture was added to cells. Seventy-two hour later, cell entry was detected through the expression of GFP by visualisation on an Incucyte S3 live cell imaging system (Sartorius). The per cent of cell entry was quantified as GFP positive areas of cells over the total area covered by cells. Entry inhibition by the sera was calculated as per cent virus infection relative to virus only control.

ELISA assays
96-well plates (Nunc) were coated overnight with 5 µgÁmL À1 of the indicated antigens. Plates were blocked with MPBST: 2% (v/v) milk in PBS, 0.05% Tween-20. Polyclonal sera from individual mice (challenge experiment) or mouse sera pooled within the same group (mouse immunisation) were diluted as indicated with MPBST and incubated for 45 min on antigen-coated plates. Plates were washed with MPBST, and bound antibodies were detected with goat anti-mouse IgG-HRP (Jackson Immunoresearch, West Grove, PA, USA, #115-035-071).

Cell culture and virus
UK strain of SARS-CoV-2 (hCoV-2/human/Liverpool/ REMRQ0001/2020; PANGO lineage B) was used and grown to P4 in Vero E6 cells [39]. The intracellular viral genome sequence and the titre of virus in the stock were determined by direct RNA sequencing (GenBank: MW041156). The virus stock did not contain a deletion of the furin cleavage that has been described previously during passage [40].

RNA extraction and DNase treatment
The upper lobes of the right lung were dissected and homogenised in 1 mL of TRIzol reagent (TFS) using a Bead Ruptor 24 (Omni International, Kennesaw, GA, USA) at 2 mÁs À1 for 30 s. The homogenates were clarified by centrifugation at 12 000 g for 5 min before full RNA extraction was carried out according to manufacturer's instructions. RNA was quantified and quality assessed using a Nanodrop (TFS) before a total of 1 lg was DNase treated using the TURBO DNA-free kit (TFS) as per manufacturer's instructions.

qRT-PCR for viral load
Viral loads were quantified using the GoTaq Ò Probe 1-Step RT-qPCR System (Promega, Madison, WI, USA). For quantification of SARS-COV-2, the nCOV_N1 primer/ probe mix from the SARS-CoV-2 (2019-nCoV) CDC qPCR Probe Assay (IDT) was utilised whilst the standard curve was generated via 10-fold serial dilution of the 2019-nCoV_N_Positive Control (IDT) from 10 6 to 0.1 copies/reaction. The E sgRNA primers and probe have been previously described [41] and were utilised at 400 and 200 nM, respectively. Murine 18S primers and probe sequences were utilised at 400 and 200 nM, respectively. The IAV primers and probe sequences were published as part of the CDC IAV detection kit (20403211; Centre for Disease Control and Prevention, Atlanta, GA, USA). The IAV reverse genetics plasmid encoding the NS segment was diluted 10-fold from 10 6 to 0.1 copies/reaction to serve as a standard curve. The thermal cycling conditions for all qRT-PCRs were as follows: 1 cycle of 45°C for 15 min and 1 cycle of 95°C followed by 40 cycles of 95°C for 15 s and 60°C for 1 min The 18S standard was generated by the amplification of a fragment of the murine 18S cDNA using the primers F: ACCTGGTTGATCCTGCCAGGTAGC and R: GCATGCCAGAGTCTCGTTCG. Similarly, the E sgRNA standard was generated by PCR using the qPCR primers. cDNA was generated using the SuperScript IV reverse transcriptase kit (TFS) and PCR carried out using Q5 High-Fidelity 2X Master Mix (New England Biolabs, Ipswich, MA, USA) as per manufacturer's instructions. Both PCR products were purified using the QIAquick PCR Purification Kit (Qiagen, Hilden, Germany) and serially diluted 10-fold from 10 10 to 10 4 copies/reaction to form the standard curve.

Histology and immunohistology
The left lung lobes were fixed in formal saline for 24 h and routinely paraffin wax embedded. Consecutive sections (3-5 µm) were either stained with haematoxylin and eosin (HE) or used for immunohistology (IH). IH was performed to detect SARS-CoV-2 antigen and leukocyte subtypes,

Three multimerised SARS-CoV-2 antigen complexes
We aimed to find a stable, convenient, and nonbacterial display scaffold that would allow the display and multimerisation of a range of SARS-CoV-2 antigens (Fig. 1A). Multimerisation has been used for many years to increase the immunogenicity of different antigens through multivalency, and this approach has also been shown recently to work well with SARS-CoV-2 antigens [20,21,25,28]. For the purpose of stable multimerisation, we identified Dps (ORF SIL_0492) from S. islandicus. The source organism is an archaeon, which prefers pH~3 and, as a hyperthermophile, has adapted to grow optimally at temperatures of around 80°C. The intrinsic thermostability and environmental robustness of S. islandicus Dps make it an outstanding candidate for the development of a multimerisation scaffold. Dps, a member of the ferritin-like protein family, selfassembles into hollow, dodecameric spheres with 12 subunits, which are roughly 10 nm across [31]. Most ferritins assemble larger spheres with 24 subunits. Also, in contrast to bona fide ferritin scaffolds, both the N and the C termini of Dps are accessible on the outside of the sphere.
We aimed to test whether Dps could efficiently display Spike, RBD and also NP antigens of SARS-CoV-2 (Fig. 1A). Spike and RBD could not be expressed in folded form in E. coli, whereas NP as well as Dps expressed and folded well in E. coli. Expression of soluble and multimeric antigens genetically fused to Dps in mammalian cells (or E. coli) was unsuccessful, and therefore, we decided to employ the SpyCatcher/SpyTag system to attach Dps to different antigens. The SpyCatcher/SpyTag system forms isopeptide bonds between amino acid side chains of the catcher domain and the peptidic tag [42,43]. DN1-SpyCatcher [44] was fused genetically to the N terminus of Dps, separated by an eight amino acid long GS linker and a hexa-histidine tag added for purification purposes (SpyC-Dps, Fig. 1B,C). We chose Nterminal linkage to Dps, SpyC-Dps, rather than Dps-SpyC since the coupling reactions were more efficient, but we did not explore this in any detail. Both the N and C terminus of Dps are on the outside of the sphere and are accessible for covalent coupling. For the antigens, SpyTag2 sequences were fused either at the N or C termini, based on steric considerations (RBD-SpyT2, SpyT2-NP, Spike-SpyT2). Conjugation of stabilised and trimeric Spike-SpyT2 to the dodecameric SpyC-Dps could potentially lead to unwanted polymerisation due to the multivalency of both partners. To overcome this problem, and to obtain a biochemically defined sample, we co-transfected HEK 293T Lenti-X cells with two different plasmids in a 3 to 1 ratio, one expressing a SpyT2 version and one without SpyT2. This favoured the expression of Spike trimers in which only one of the monomers contains the SpyTag. Stabilised, trimeric, and on average monovalent Spike-SpyT2 and also RBD-SpyT2 were purified from conditioned media of HEK 293S GnT1 À/À (for Spike-SpyT2) or Expi 293 (for RBD-SpyT2) cell cultures. SpyC-Dps and SpyT2-NP were purified from the cytosol of E. coli cells transformed with the appropriate plasmids. All constructs possess histidine tags and were purified by immobilised metal affinity chromatography and at least one additional size exclusion step (SEC). Sequences of all proteins used can be found in Table S1. Expression yields were excellent in all cases: SpyC-Dps yielded~120 mgÁL À1 culture, RBD-SpyT2~40 mgÁL À1 culture, stabilised trimeric and monovalent Spike-SpyT2~13 mgÁL À1 culture and SpyT2-NP~60 mgÁL À1 culture of pure proteins ( Fig. 2A).
To achieve efficient coupling of scaffold and antigens, a molar excess of each of the three purified antigens (RBD-SpyT2, SpyT2-NP, Spike-SpyT2) was mixed with SpyC-Dps to facilitate covalent coupling. Subsequent removal of excess antigens was accomplished by SEC using a Superose 6 column (Fig. 2B). Coupling efficiency was analysed by SDS/PAGE, followed by Coomassie staining (Fig. 2C). When the coupled samples were mixed with denaturating SDS sample buffer without additional heating, we detected high molecular weight complexes that we suggest represent dodecameric assemblies caused by Dps that survive SDS treatment ('RT' lanes). Heating the samples to 99°C led to the disappearance of the higher bands ('99' lanes), confirming both the (SDS-) stability and the purity of the coupled and multimerised protein samples. Note that there were no bands showing uncoupled SpyC-Dps in any of the three Ag-S-Dps samples, meaning that coupling used all 12 available Dps subunits.
Next, we analysed the integrity of the scaffold after the coupling reactions, as well as homogeneity by electron microscopy (Fig. 2D). For the scaffold alone, SpyC-Dps, we observed the expected small and well-dispersed~10 nm Dps spheres. Similar homogeneity and monodispersity were observed for all three coupled Ag-S-Dps versions, RBD-S-Dps, NP-S-Dps and Spike-S-Dps. The Ag-S-Dps complexes were larger than the scaffold alone as the Dps spheres were densely surrounded by extra densities, indicating the success of the coupling and the structural integrity of Ag-S-Dps complexes after the coupling reactions. We note that no aggregation was observed for Spike-S-Dps, indicating that the co-transfection approach produced mostly trimeric Spike proteins with only one SpyTag2 present. Taken together, we showed that the scaffold and the three antigens could be produced easily and at high yields and resulted in biochemically pure and defined Ag-S-Dps proteins that display 12 antigens on each Dps scaffold.
To determine whether the coupled Ag-S-Dps complexes were stable in blood plasma for immunisations, we mixed the RBD-S-Dps complex with human serum (clotted, not heat inactivated, at a 1 : 3 volume ratio). The RBD-S-Dps complex was remarkably stable, with 50% remaining intact after~40 h at 37°C (Fig. S1A,  B). Given the stability of the Dps scaffold both in serum and when exposed to denaturing conditions (SDS/PAGE, 'RT' lane) (Fig. 2C), we next investigated whether the coupled RBD-S-Dps sample would survive lyophilisation and subsequent re-solubilisation. A lyophilised, dry sample would facilitate prolonged storage even in the absence of refrigeration. We therefore freeze-dried RBD-S-Dps and after rehydration found no evidence of precipitation or significantly reduced protein concentration by SDS/PAGE (Fig. S1C). There was also no disappearance of the SDS-stable high molecular weight band, indicating Dps sphere integrity was maintained after rehydration. Finally, electron microscopy analysis showed the rehydrated sample to be indistinguishable from the starting material with no evidence of disintegration or aggregation (Fig. S1D).

Multimerisation by Dps greatly enhances immunogenicity, especially for RBD
Having obtained the three multimerised antigen-Dps complexes (Ag-S-Dps), we tested whether they induce a stronger immune response than their monomeric equivalents. We immunised mice with the following protocol: five male C57BL/6J mice per group were given 50 µg protein subcutaneously on days zero and 23, and 25 µg on day 64 (using CpG 1668 as an adjuvant) (Fig. 3A). Blood samples were taken on days 13 (1st bleed), 34 (2nd bleed), and 74 (3rd bleed). After the 1st boost on day 34, antigen-specific antibodies were detected in the sera from the mice by ELISA (Fig. 3B). Substantially higher antibody titres were detected with multimerised Dps-fused RBD and NP. Multimerisation improved Spike titres only modestly, which may be expected given that Spike is already a trimer without Dps. After 74 days, and the second boost, the specific antibody titres were further increased. Spike induced the weakest response and multimerisation had the smallest effect. In contrast, RBD-S-Dps and NP-S-Dps induced substantial increases in antibody titres compared to the nonmultimerised versions. We also analysed sera for antibodies against the scaffold protein itself (SpyC-Dps). Sera from mice immunised with coupled Ag-S-Dps complexes showed measurable but low antibody titres against SpyC-Dps. Anti-SpyC-Dps responses remained low even after the second boost, suggesting that the scaffold itself is poorly immunogenic and that in the context of the fusions the antibody response is largely directed against the viral antigens displayed on the surface. Furthermore, we tested if lyophilisation affects the ability of RBD-S-Dps to induce an anti-Spike response. Figure S1E shows that single or double lyophilisation do not systematically affect the immunogenicity of RBD-S-Dps. Taken together, the data show that multimerised Ag-S-Dps complexes produce substantial improvements in antibody titres over the uncoupled antigens. Overall, the strongest response was observed for RBD-S-Dps and the strength of the response was not affected by lyophilisation.
Next, we tested the neutralisation activity of antibodies produced by the mice immunised with RBD-S-Dps, RBD-SpyT2, Spike-S-Dps and Spike-SpyT2. The mouse sera within each group were pooled at day 34 (2nd bleed) or 74 (3rd bleed) and analysed using a pseudovirus infection assay (note that NP-directed sera will not have an effect in this assay because pseudotyped viruses do not contain NP). In this assay, a lentiviral vector expressing GFP is pseudotyped with Spike protein from SARS-CoV-2 to obtain virions that display Spike in their envelope and infect cells in an ACE2-dependent manner. As seen in Fig. 3C, the day 34 sera pool of the multimerised RBD-S-Dps group protected against pseudovirus infection up to a dilution of 1 : 400, whereas the monomeric RBD-SpyT2 only showed a protective effect at a 1 : 100 dilution, and even then, it only reduced infection by~50%. Sera from mice immunised with multimeric Spike-S-Dps also protected against infection, whilst Spike-SpyT2 sera were unable to neutralise at any of the dilutions tested. The sera taken after 74 days had  neutralisation activity (Fig. 3D). The sera from RBD-S-Dps-immunised mice gave the strongest protection: even at a 1 : 6400 dilution only~10% infection could be detected. At this 1 : 6400 dilution, the monomeric RBD-SpyT2 and Spike-S-Dps sera gave very little neutralisation. Whilst pseudoviruses are widely used to test the neutralisation activity of SARS-CoV-2 antisera, they are based on a lentiviral rather than coronavirus particle and do not recapitulate live virus replication. We therefore tested whether antibodies raised against multimeric RBD-S-Dps were capable of blocking a spreading infection of a primary clinical isolate of SARS-CoV-2. Viral replication was measured by RT-qPCR using probes against NP (gRNA) or E (sgRNA). RBD-S-Dps antisera from five different animals all potently inhibited SARS-CoV-2 (Fig. S2A,B). In contrast, the potency of antisera raised against RBD-SpyT2 varied considerably between mice. We conclude that immunisation with RBD-S-Dps not only produces the highest titre antibodies (Fig. 3B), but also the most neutralising (Fig. 3C,D) and with reliable potency against live virus (Fig. S2A,B). We also used ELISpot to test splenocytes from immunised mice culled on day 74 for their T-cell responses. We observed a statistically significant increase in the proportion of Spike-specific splenocytes in RBD-S-Dps immunised mice versus SpyC-Dps alone, but not between RBD-SpyT2 and SpyC-Dps alone, or between RBD-S-Dps and RBD-SpyT2 (Fig. S2C). However, given the relatively weak response, we reasoned it would not be possible to properly assess whether multimerisation improves Tcell immunity with this version of the Dps scaffold and so focussed on the antibody response.

Single-shot immunisation with multimerised RBD-S-Dps protects mice against SARS-CoV-2 infection
Encouraged by these results, we wanted to know whether antigen display on our Dps scaffold would induce a sufficiently strong antibody response to protect animals from SARS-CoV-2 infection. We selected our most potent immunogen, RBD-S-Dps, and used it to immunise mice transgenic for human ACE2 (K18-hACE2) [45]. As a single-dose vaccination regime offers many downstream logistical and practical benefits, we opted to immunise only once and then challenge with SARS-CoV-2 on day 28 (Fig. 4A). We immunised subcutaneously six K18-hACE2 mice with RBD-S-Dps, six with RBD-SpyT2 and six with PBS (always three female and three male mice), each with 25 µg of the immunogens (except PBS control), plus CpG adjuvant. The anti-Spike antibody response following immunisation was measured by ELISA on days 13 and 24 (before challenge) and on day 35 (7 days postchallenge). A strong anti-Spike antibody titre was detected in RBD-S-Dps-immunised mice, but almost none for either RBD-SpyT2 or PBS (Fig. 4D). Antibody titres remained high for RBD-S-Dps at days 24 and 35. On day 28, animals were challenged with 10 4 PFU SARS-CoV-2. Mice in the PBS control and RBD-SpyT2-immunised groups began to show clinical signs of illness and a decline in body weight from day four postinfection (Fig. 4B), consistent with previous reports of infection in na€ ıve animals [45]. In contrast, mice immunised with multimerised RBD-S-Dps maintained body weight until the day seven end point. There was a statistically significant difference in weights between the RBD-S-Dps-immunised and PBS control groups from day four, and between RBD-S-Dps-and RBD-SpyT2-immunised mice from day five (Fig. 4C). There was no significant difference in weight loss between the RBD-SpyT2-immunised mice and PBS controls at any time point, suggesting that, unlike RBD-S-Dps, nonmultimerised RBD does not provide protection after only a single vaccination. All animals were culled on day seven postinfection and tissues collected for analysis. As mentioned, there were no significant changes in anti-Spike antibody levels pre-versus postchallenge, indicating that mostly antibodies raised during the immunisation contributed to the immune response during the infection (Fig. 4D). SARS-CoV-2 infection of the lung was quantified by plaque assay and genomic and subgenomic qPCR, using probes against the viral genes NP and E, respectively. There were significantly lower levels of infectious virus in the lungs of mice immunised with RBD-S-Dps, compared to either RBD-SpyT2 immunised or PBS control groups (Fig. 4E). A broadly similar pattern was observed when quantifying virus using probes were tested for binding activity by ELISA against Spike-SpyT2, NP-SpyT2 or polymeric scaffold, SpyC-Dps. In all cases did the multimerised Ag-S-Dps complexes produce more antibodies than the nonmultimerised versions. RBD-S-Dps and NP-S-Dps produced very strong responses. (C) Pseudoviral cell entry neutralisation assay with sera from the 2nd bleed. Sera from immunised mice were tested for neutralisation activity against a Spike-pseudotyped lentiviral GFP vector (hence NP-S-Dps sera will not neutralise). This is a lentiviral vector that incorporates a mutant Spike protein lacking 19 residues from the C terminus (D19) into the envelope of the budding particle (see Materials and methods). Infection was measured 72 h after vector addition by quantifying GFP expression in HEK 293T ACE2/TMPRSS2 target cells. The data were normalised to the mean infection level in the absence of virus, which was set to 100%. This is a relative measure and 100% does not indicate that all cells are infected. The multimerised RBD-S-Dps and Spike-S-Dps showed strong neutralisation, in contrast to their nonmultimerised precursors. (D) Same as (C) but using sera from the 3rd bleed. Neutralisation activity is enhanced in all sera, and the differential between multimerised and nonmultimerised antigens remains. Overall, RBD-S-Dps showed the strongest neutralisation activity. Data were measured in triplicates and are given as the mean with error bars representing the standard error of the mean. against NP or E (Fig. 4F,G). However, we noted a marked difference in the amounts detected between male vs female mice. Female RBD-S-Dps-immunised mice had significantly fewer genomic and subgenomic transcripts, compared to mice from other groups and their male equivalents (Fig. 4F,G and Fig. S3A,B). We attempted to correlate this with differences in antibody titres, but whilst there was a trend towards lower titres in male mice, particularly just before and just after the challenge, this was not significant (Fig. S3C).
A larger group size would be needed to confirm this result. Despite these sex-dependent differences in the qPCR data, the near-absence of infectious virus in both male and female RBD-S-Dps immunised mice, as measured by plaque assay (Fig. S3D), suggests they were both highly protected. Finally, we examined the lungs of mice from the different groups for histopathological changes and for viral antigen expression using an anti-NP antibody to reveal sites of replication (Fig. 4H, Fig. S4) and immune cell infiltration (Fig. S5). A detailed description is provided in the Appendix S1. In summary, lungs from RBD-SpyT2immunised mice or PBS control mice showed substantial and widespread NP expression mainly in pneumocytes ( Fig. 4H and Fig. S4), indicative of viral replication throughout the lobe and consistent with the high virus levels measured in these animals ( Fig. 4E-G). There was also evidence of pneumocyte degeneration and syncytial cell formation, as has been reported in COVID-19 cases postmortem [46]. Multifocal leukocyte infiltration was observed, particularly in PBS control animals, dominated by macrophages, followed by T cells (mainly CD4+ and lesser CD8+ cells), B cells and neutrophils (Figs S4 and S5). This is reminiscent of the hyperinflammation in postmortem reports of lethal COVID-19 associated with immunopathology [47]. In contrast, the lungs of mice protected by multimerised RBD-S-Dps were either almost or entirely clear of NP expression (Fig. S4) and pathological changes (female mice), or showed only mild changes consistent with those observed in the PBS control animals (Fig. S5), and markedly reduced NP expression. Taken together, these data indicate that immunisation with RBD-S-Dps is highly protective against SARS-CoV-2 in hACE2-expressing mice, even after a single dose, whilst monomeric RBD-SpyT2 is not.

Discussion
Here, we have shown that the ferritin-like protein Dps from the hyperthermophilic archaeon S. islandicus possesses exceptional qualities as a SARS-CoV-2 subunit vaccine scaffold. We combined Dps with the Spy-Catcher/SpyTag system in order to create a 'plugand-play' system that allows the rapid and facile synthesis of highly stable multimeric subunit vaccines.
Mixing the SpyCatcher-Dps protein with any compatible SpyTag antigen leads to the assembly of highly monodisperse nanoparticles displaying exactly 12 antigens. Using this approach, we have produced subunit vaccines based on Spike, RBD or NP from SARS-CoV-2 and tested them in immunisation and viral challenge experiments. In each case, the Dpsdisplayed antigens out-performed their nonmultimerised equivalents and induced a more rapid and potent antibody response. Subunit vaccines offer distinct advantages in cost, simplicity, production capacity, storage, transport and administration over nucleic-acid based vaccines [48]. Principle amongst these considerations is stability, with currently used vaccines such as those from Pfizer-BioNtech (Pfizer, New York City, New York, USA; BioNTech, Mainz, Germany), Moderna (Moderna, Cambridge, MA, USA), And Oxford-AstraZeneca (Oxford University, Oxford, UK; AstraZeneca, Cambridge, UK) requiring a À80°C or À20°C cold-chain. In countries without a highly developed logistical and medical infrastructure, this represents a significant impediment to vaccination. Whilst subunit vaccine development currently lags behind nucleic-acid based equivalents, there is good evidence that such vaccines are nevertheless effective at inducing a protective response. SARS-CoV-2 RBD by itself [49] or in simple fusions such as to IgG Fc [50] has been shown to elicit SARS-CoV-2-neutralising antibodies. Antigen multimerisation increases neutralising titres, for instance when using ferritin as a scaffold [23,28,29]. Our finding of a 3-log increase in the neutralisation of live SARS-CoV-2 in vitro upon RBD multimerisation compares favourably with the 2-log increase in neutralisation of Spike-pseudotyped lentivirus observed by these earlier studies. However, neither study examined whether their multimerised scaffolds provided protection against SARS-CoV-2 in a live virus in vivo infection model. A live virus challenge experiment in mice was performed by Ma et al. [27], who tested their ferritin-multimerised RBD vaccine using a prime/boost strategy and found that whilst monomeric RBD dramatically reduced viraemia upon SARS-CoV-2 challenge, multimerisation decreased it 1-log further. A direct comparison with our data is difficult but the Dps scaffold presented here performs at least as good, if not better, as the difference in viraemia after vaccination with monomeric vs multimeric antigen we observed was in the order of several logs. Moreover, our data suggest that multimerisation is particularly beneficial when a single vaccine dose is used.
Nonferritin scaffolds have also been used to multimerise antigens. For instance, virus-like icosahedral particles display 60 antigen copies (e.g. I3-01) [51]. When fused directly to viral antigens [52], or using the SpyCatcher/SpyTag system [19,53], the I3-01 scaffold has been shown to induce a neutralising antibody response. Our scaffold differs from both ferritin and nonferritin scaffolds used previously to deliver SARS-CoV-2 immunogens in several important aspects. First, because we have used a thermostable protein it is intrinsically more stable, providing potential benefits to vaccine transport and storage and also to immunogen stability in vivo. Second, it is smaller than other scaffolds (< 10 nm vs > 10 nm for ferritin or 25 nm for the I3-01 nanoparticle), making it an easier cargo for cellular uptake. Third, it displays fewer copies than ferritin or I3-01 (12 vs 24 or 60, respectively), allowing the selection of higher-affinity B cells and avoiding the activation of off-target (and possibly cross-reactive) Bcell competitors [54]. Fourth, in contrast to bona fide ferritin scaffolds, the N and the C termini of Dps are both accessible on the outside of the sphere. This allows, at east in principle, for the conjugation of two discrete antigens onto a single scaffold, for example both SARS-CoV-2 Spike/RBD and NP to be displayed simultaneously.
Importantly, we have provided here data demonstrating the benefit of antigen multimerisation in inducing not just neutralising antibodies but an immune response capable of providing in vivo protection. In our SARS-CoV-2 challenge experiments, we found that RBD alone failed to protect mice, which displayed continued weight loss and high viral loads in the lungs. In contrast, our Dps-based vaccine displaying RBD completely protected mice from SARS-CoV-2-associated pneumonia and disease after only a single immunisation. We noted however a difference in Dps-RBD induced protection between male and female mice, with the latter having lower viral loads and hardly any pulmonary changes. Trial data for both mRNA and vector-based vaccines have not been disaggregated by sex but data on SARS-CoV-2 infection show that men are more at risk of severe adverse conditions, hospitalisation, and death [55,56]. Our results support the consideration of sex as a variable in vaccine trials [57].
Further research is needed to develop the Dps scaffold into a bona fide vaccine for SARS-CoV-2 and other viruses. Replicating the robust neutralising antibody response and high level of protection achieved in mice from a single dose in primates and humans will be crucial, especially given that the hACE2 transgenic mouse model is not perfect. It does better recapitulate severe infection in humans than the Syrian hamster model but does not allow respiratory transmission [58]. The mouse model also has the advantage that it shows signs of neurological spread of the virus, something that is underappreciated in the human disease.
Whilst most studies of subunit vaccines have focused on antibodies, long-lasting protection is likely to be dependent upon stimulating CD4+ and CD8+ T cell immunity [59,60]. Data from current vaccine trials and roll-outs have yet to be fully analysed but a correctly balanced T-cell response appears associated with recovery from acute infection and the avoidance of hospitalisation and severe virus-induced immunopathology [61]. Fortunately, the analysis of Tcell epitopes from SARS-CoV-2 convalescents [62] provides a basis for engineering subunit vaccines specifically to engage both B and T cells. In this context, the ability of our Dps scaffold to display antigens at both termini may prove particularly beneficial. In addition to ensuring a well-balanced immune response in humans, a more thorough investigation into the longterm stability, storage, and reconstitution of lyophilised material is required to demonstrate that a Dpsbased vaccine is suitable for use in regions with limited infrastructure.
Future work notwithstanding, our data add to a body of evidence that subunit-based vaccines represent a viable choice as a vaccine modality for SARS-CoV-2. Whilst other vaccine formats are significantly more advanced, subunit approaches such as Dps offer distinct advantages in simplicity of production, requiring no proprietary technology, robustness of material and potency of protection.

Author contributions
Study design and conception: RS, ARA, JPS, LCJ, JL, with further study design by JAH, AO. Bacterial expression, purification, coupling reactions, EM analysis and stability assays: RS, with AG-L (EM, lyophilisation) and LK (NP protein). Design of Spike and RBD constructs, protein production in HEK cells, purification: VTC. In vivo challenge experiment and downstream analysis: JJC and PS. Antibody analysis and neutralisation assays: MV and AA. Histopathology: AK. Manuscript preparation and editing: RS, JJC, PS, AK, JPS, LCJ, JL, with further editing by JAH, AO, and ARA.

Data accessibility
The data that support the findings of this study are available from the corresponding authors (lcj@mrclmb.cam.ac.uk; jyl@mrc-lmb.cam.ac.uk) upon reasonable request.
lyophilised RBD-S-Dps. After collecting 1st and 2nd bleeds, ELISA assays on pooled samples per experimental group detected and quantified antibodies against Spike-SpyT2 (assayed in duplicate). Error bars represent the standard error of the mean (n = 2). Fig. S2. A, B): Vero cells expressing ACE2 and TMPRSS2 were infected with SARS-CoV-2 in the presence of serial dilutions of antisera. Viral replication was then determined after 24 h by RT-qPCR using probes for gRNA (A) or sgRNA (B). Each point represents sera from an individual mouse. C) ELISpot measuring IFNg production in splenocytes from mice 74 days post-immunisation upon stimulation with a peptide library covering Spike protein. Error bars depict the mean +/standard error of the mean. Fig. S3. Mice were immunised with RBD-S-Dps, RBD-SpyT2 or given PBS control on day 1 and then challenged with SARS-CoV-2 on day 28. Fig. S4. Lung, left lobe, K-18 hACE2 mice at day 7 post infection. Histological changes and SARS-CoV-2 antigen expression. Fig. S5. Lung, K18-hACE2 mice. Composition of the inflammatory infiltrates. Table S1. Amino acid sequences of the proteins used in this work. Appendix S1. Supplemental results.