Characterization and immunogenicity of nanoparticle vaccines based on Clade 2 and Clade 3 sarbecovirus S2 proteins

Abstract

The emergence of SARS-CoV-2 and its subsequent variants in addition to the previous SARS-CoV-1 outbreak indicates the importance of developing broadly protective sarbecovirus vaccines. To date, broadly protective vaccines have primarily focused on clade 1 sarbecoviruses including SARS-CoV-2 variants and SARS-like animal viruses. The discovery of clade 2 and clade 3 sarbecoviruses capable of infecting human cells highlights a need to preemptively develop vaccines that can protect against these viruses. Here, we develop stabilized multivalent subunit vaccines from clade 2 and clade 3 sarbecovirus S2 proteins and evaluate their immunogenicity. Clade 2 and clade 3 S2-subunit vaccines elicit cross-reactive antibodies in mice capable of binding to clade 1, 2, and 3 sarbecovirus antigens. Female mice immunized with these S2-based vaccines also provide protection against sarbecovirus challenges from clades 1a and 1b, including a mouse-adapted SARS-CoV-2 strain, XBB, and WIV1.

Introduction

The recent pandemic caused by severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) has led to approximately 7.1 million deaths as of January 2025¹. The spike (S) glycoprotein, split into S1 and S2 subunits, covers the surface of the coronavirus virion and is responsible for mediating viral entry into cells². The receptor binding domain (RBD) is found within the S1 region and binds to angiotensin-converting enzyme 2 (ACE2) on cells to initiate entry³. This interaction prompts the S2 subunit to undergo conformational changes that lead to the fusion of the viral membrane with the membrane of the host cell. As a response to the worldwide spread of SARS-CoV-2, mRNA vaccines which encode the full S protein were developed and globally distributed. Due to its immunodominance, the S1 subunit has been the primary target of neutralizing antibodies elicited by these mRNA vaccines⁴. While this neutralizing antibody response against the S1 subunit was initially effective in curbing the spread and lethality of the virus, and an overall increase in population immunity due to widespread infection and vaccination has helped keep SARS-CoV-2 lethality low⁵, new strains with mutations primarily in this subunit reduced the efficacy of these vaccines^6,7 and continue to cause recurrent waves of infection.

There have been several approaches for developing broadly protective vaccines against sarbecoviruses. Many of these strategies use mixtures of either full S proteins^8,9 or RBDs from various sarbecoviruses^10–12. However, these vaccines tend to elicit an immune response that targets the S1 subunit which is more variable than the S2 subunit. A promising alternative approach is to develop vaccines that target the S2 subunit of the S protein, which is much more conserved relative to the S1 subunit (Fig. 1b, c). To that end, various vaccines have been successfully developed that aim to direct the immune response to the S2 subunit^13–20.

Fig. 1. Analysis of Sarbecovirus Sequences.

a Amino acid phylogenetic tree of selected Sarbecovirus S proteins. The number of mutation events per residue is represented by the scale bar. Branch and node color indicate the clade assignments (Clade 1a: blue; Clade 1b: dark red; Clade 2: violet; Clade 3: teal; Clade 4: orange). b Amino acid sequence identity of selected Sarbecovirus S1 subunits. c Amino acid sequence identity of selected Sarbecovirus S2 subunits.

However, the S2-based sarbecovirus vaccine efforts to date have primarily targeted clade 1 sarbecoviruses^13–20. Sarbecoviruses from the other clades (Fig. 1a) also have the potential to infect humans. Several clade 2 and clade 3 sarbecoviruses have been shown to infect human cells naturally or with the addition of exogenous proteases. RBDs from clade 2 viruses As6526 and Rs4081^21,22 have been shown to exhibit protease-mediated, ACE2-independent entry. The clade 3 sarbecovirus Khosta-1 has similarly been shown to infect human ACE2-expressing baby hamster kidney (BHK) cells with the addition of exogenous proteases²³. A chimeric spike containing the RBD from Khosta-2, another clade 3 sarbecovirus, has been shown in vitro to infect cells expressing human ACE2 even without the presence of proteases^23,24. Other clade 3 viruses such as PRD-0038, PDF-2370, and BtKY72 similarly have been shown in vitro to utilize the human ACE2 as an entry receptor^23,25. The binding of PRD-0038 and BtKY72 to human ACE2 has also shown to be enhanced with the incorporation of two amino acid mutations (K493Y and T498W)²⁶. Khosta-1 has also shown a similar ability, where a single point mutation (T498W) in its RBD broadened its tropism from bat to human ACE2²⁶. Very little is known currently about clade 4 viruses, and none currently identified have been shown to infect human cells.

Perhaps unsurprisingly, sera from individuals who were immunized with two doses of either the Moderna or Pfizer-BioNTech vaccines failed to neutralize a pseudotyped chimeric SARS-CoV-2 spike virus with the Khosta-2 RBD²³. Efforts to develop clade 2 and clade 3 vaccines have been limited, especially those targeting the conserved S2 subunit. Lee et al. developed a PRD-0038 (clade 3) S trimer vaccine and found that sera from immunized mice had potent neutralizing activity against PRD-0038 and Khosta-1 pseudoviruses but little neutralizing activity against clade 1 pseudoviruses²⁷. Lee et al. developed a stabilized version of the S2-subunit of the PRD-0038 S protein; however, no immunogenicity characterization was reported for this construct¹⁹.

Here, we generated stabilized clade 2 and clade 3 sarbecovirus S2 antigens and multivalently presented them on virus-like particles (VLPs) as vaccine constructs. We found that these clade 2 and clade 3 S2 antigens as well as the VLP-S2 vaccines were recognized by the broadly cross-reactive S2-binding S2P6 antibody. Immunization with the VLP-S2 vaccines elicited cross-reactive antibodies that bound to clade 1, clade 2, and clade 3 sarbecovirus S and S2-subunit antigens. Mice immunized with two doses of clade 2 or clade 3 VLP-S2 vaccines were also protected against challenges with SARS-CoV-2, XBB, and a SARS-like bat-CoV WIV1.

Results

Selection of S2 antigens for immunizations

Given the high homology between clade 2 S2 subunits, we opted to select a single S2 antigen from this clade. We opted for the S2 subunit from YN2013, because of its lower homology to the SARS-CoV-2 S2 subunit relative to that of HKU3-8 (Fig. 1c). For clade 3, we selected the S2 subunit from Khosta-2, because Khosta-2 was one of the clade 3 sarbecoviruses most capable of infecting cells using human ACE2²³. We also chose the S2 subunit from BtKY72 as BtKY72 was one of the clade 3 sarbecoviruses with the most dissimilar S2 subunit to Khosta-2 (Fig. 1c).

Generation and characterization of S2 antigens and VLP-S2 vaccines

We have previously developed streptavidin-coated MS2 protein-based VLPs capable of multivalently presenting biotinylated antigens (including the Zika virus envelope protein domain III²⁸, Influenza virus hemagglutinin²⁹, and various sarbecovirus S and S2-subunit proteins^9,20,30) as vaccine constructs. The MS2 coat protein self-assembles from 90 MS2 homodimers, and the addition of an AviTag into the surface loop of each dimer allows for site-specific biotinylation^{9,13,20,28–30}. These biotinylated MS2 VLPs are then mixed with an excess of streptavidin via drop-wise addition and purified by size exclusion chromatography (SEC) to obtain streptavidin-coated MS2 VLPs.

Biotinylated S2 antigens were also developed as described previously^13,20. S2 antigens were modified to include a C-terminal trimerization domain, a C-terminal AviTag for biotinylation, and a C-terminal his-tag for purification. The trimerization domain is approximately 30 amino acids and originates from the fibritin protein of the T4 bacteriophage. The domain forms a stable trimeric β-hairpin propeller structure that induces the trimerization of the other proteins³¹, in our case, the S2 subunits of various sarbecovirus spikes. The trimerization domain or tags were not removed prior to characterization and immunization.

In addition, S2 antigens were modified to include mutations in the S2’ cleavage site^13,20 and the HexaPro proline mutations³². For the clade 3 antigens BtKY72 and Khosta-2, an additional 3 mutations (F744W, F883L, and N940D for BtKY72; F739W, F878L, N935D for Khosta-2) were added to stabilize the S2 (determined using the Rosetta molecular modeling program following the point mutation method reported by Thieker et al.³³). Additionally, as Lee et al. previously found that truncating the N-terminus enhanced the expression of their SARS-CoV-2 S2 antigen¹⁹, we generated a truncated version of the BtKY72 S2 antigen (dubbed BtKY72 Tr. S2). We also similarly truncated our Khosta-2 S2 antigen. The truncated BtKY72 S2 antigen included amino acids 682–1193 while the truncated Khosta S2 antigen included amino acids 677–1188.

Plasmids encoding YN2013 S2, BtKY72 S2, BtKY72 Tr. S2, and Khosta-2 Tr. S2 were transfected into Expi293F mammalian cells, and the secreted proteins were purified by IMAC and SEC and then biotinylated in vitro. The VLP-S2 vaccines (Fig. 2a) were generated by mixing these S2-antigens with the MS2-SA scaffold at a determined optimal stoichiometric ratio.

Fig. 2. S2 and VLP-S2 characterization.

a Schematic of S protein and VLP-S2. b SDS-PAGE analysis of individual S2 antigens and VLP-S2s. Characterization by SDS-PAGE was performed twice using different preparations of each sample with similar results. c Representative dynamic light scattering characterization of S2 antigen, MS2, and VLP-S2 constructs. d S2P6 binding characterization of S2 antigens and VLP-S2 constructs (mean ± SD, n = 3), one assay with three technical replicates. e Represenative size exclusion chromography chromatogram of BtKY72 S2 and BtKY72 Tr. S2 antigens, both purified from transfection volumes of 90 mL. f Total yield of protein expected from 1 L of media, extrapolated from separate 50 mL volume expressions (mean ± SD, n = 3). g Melting temperature analysis of BtKY72 S2 and BtKY72 Tr. S2 antigens.

The S2 antigens were characterized by SDS polyacrylamide gel electrophoresis (SDS-PAGE), dynamic light scattering (DLS), and enzyme-linked immunosorbent assay (ELISA) both individually and when presented on the MS2-SA scaffold. The S2 proteins and VLP-S2 constructs were shown to be pure on the SDS-PAGE gel (Fig. 2b). The DLS measurements showed theVLP-S2 constructs had a diameter between 75 and 85 nm (Fig. 2c). This diameter measurement is also consistent with the measured sizes of the S2 antigens (18-20 nm) and the MS2-SA (40–45 nm) (Fig. 2c). These DLS measurements were also similar to previous characterizations of clade 1 S2 antigens and VLP-S2 constructs²⁰. The clade 2 and clade 3 S2 antigens were also recognized by the broadly cross-reactive S2-binding S2P6³⁴ and 3A3³⁵ antibodies, and this binding was retained when the S2 subunits were multivalently presented on the MS2-SA scaffold (Fig. 2d and Supplementary Fig. 2)²⁰.

We found that the truncation of the clade 3 BtKY72 S2 antigen increased its overall yield, similar to what Lee et al. reported¹⁹ with their SARS-CoV-2 S2 construct (Fig. 2e, f). Additionally, we found the truncation did not impact the melting temperature of the BtKY72 S2 antigen (Fig. 2g). As indicated above, the truncation also did not impact the antigen purity or S2P6 binding (Fig. 2b, d).

Clade 2 and clade 3 VLP-S2 vaccines elicit cross-reactive antibodies against sarbecovirus clade 1, clade 2, and clade 3 antigens

We next characterized the IgG antibody response elicited by our VLP-S2 vaccines. Groups of C57BL/6 mice (10–12 week-old, females) were immunized with either VLP-YN2013 S2, VLP-BtKY72 S2, VLP-BtKY72 Tr. S2, VLP-Khosta-2 Tr. S2, or VLP-Control (7.5 µg of S2 antigen per mouse) adjuvanted with AddaS03 and poly I:C (Invivogen, equal volume of AddaS03 plus 100 µg of poly I:C). After 4 weeks, sera were collected and tested for binding against various sarbecovirus spike protein antigens. All VLP-S2 immunized mice elicited high titers against the selected clade 1, clade 2 and clade 3 sarbecovirus antigens (Fig. 3). VLP-S2 immunized mice had antibody endpoint titers significantly higher than those for VLP-Control-immunized mice. There was no significant difference in endpoint titers between the VLP-S2 immunized mice groups themselves. Sera were also tested for reactivity against an H1 hemagglutinin containing the same trimerization domain, AviTag, and hexahistidine tag as the S2 proteins, with little to no binding detected (Supplementary Fig. 3).

Fig. 3. Endpoint titers of VLP-S2-immunized mice.

Antibody endpoint titers of sera from C57BL/6 mice immunized with VLP-S2 vaccines against S and S2 proteins (geometric mean ± geometric SD, n = 4 biological replicates for VLP-YN2013 S2 and VLP-BtKY72 S2, and n = 5 biological replicates for VLP-BtKY72 Tr. S2 and VLP-Khosta-2 S2). ****P < 0.0001, ns: not significant. Significance was determined by one-way ordinary ANOVA and Tukey post hoc multiple comparisons between groups. Detection limit (dashed line) = 20. • indicate data from individual mice.

Clade 2 and clade 3 VLP-S2 vaccines protect against challenges with clade 1a and clade 1b sarbecoviruses

While we cannot test the ability to protect against challenges with clade 2 and clade 3 sarbecoviruses in vivo, we tested the ability of the clade 2 and clade 3 VLP-S2 vaccines to protect mice against challenges with clade 1 sarbecoviruses given the conserved nature of the S2 subunits. Specifically, mice were challenged with either a mouse-adapted strain of SARS-CoV-2 (MA10) (clade 1b), the SARS-CoV-2 Omicron subvariant XBB (clade 1b), or the bat coronavirus WIV1 (clade 1a). WIV1 was selected as it has been identified as poised for human emergence with pandemic potential³⁶. Groups of mice (C57BL/6 for MA10 or K18-hACE2 for XBB/WIV1, 10–12 week old, females) were immunized with either VLP-YN2013 S2, VLP-BtKY72 S2, VLP-BtKY72 Tr. S2, VLP-Khosta-2 Tr. S2, or VLP-Control (7.5 µg of S2 antigen per mouse) adjuvanted with AddaS03 and poly I:C (Invivogen, equal volume of AddaS03 plus 100 µg of poly I:C). After 4 weeks, they were given a second dose before being challenged intranasally 4 weeks later with 10⁵ plaque-forming units (PFU) (Fig. 4a). Virus titers from lung and nasal turbinate tissue were quantified by plaque assay 3 days after challenge.

Fig. 4. Protective efficacy of VLP-S2 vaccines.

a Schematic of vaccine immunization regimen. b Lung virus titers of two-dose VLP-S2-immunized C57BL/6 mice three days after infection with 10⁵ PFU of SARS-CoV-2 MA10 (mean ± SD). n = 8 biological replicates for VLP-Control, n = 3 biological replicates for VLP-YN2013 S2, n = 4 biological replicates for VLP-BtKY72 S2, n = 5 biological replicates for VLP-BtKY72 Tr. S2 and VLP-Khosta Tr. S2. ****P < 0.0001, ns: not significant., determined by one-way ordinary ANOVA and Tukey post hoc multiple comparisons between groups. Detection limit (dashed line) = 1.3 log₁₀ pfu/g. ● indicate data from individual mice. c Lung virus titers of two-dose VLP-S2-immunized K18-hACE2 mice three days after infection with 10⁵ PFU of XBB (mean ± SD). n = 9 biological replicates for VLP-Control, n = 4 biological replicates for VLP-BtKY72 S2, and n = 5 biological replicates for the rest. ****P < 0.0001, ns: not significant., determined by one-way ordinary ANOVA and Tukey post hoc multiple comparisons between groups. Detection limit (dashed line) = 1.3 log₁₀ pfu/g. ● indicate data from individual mice. d Lung virus titers of two-dose VLP-S2-immunized K18-hACE2 mice three days after infection with 10⁵ PFU of WIV1 (mean ± SD). n = 8 biological replicates for VLP-Control, and n = 4 biological replicates for rest. *P < 0.05, ***P < 0.001, ****P < 0.0001, determined by one-way ordinary ANOVA and Tukey post hoc multiple comparisons between groups. Detection limit (dashed line) = 1.3 log₁₀ pfu/g. ● indicate data from individual mice. VLP-YN2013 S2 vs. VLP-BtKY72 S2: P = 0.0456; VLP-YN2013 S2 vs. VLP-BtKY72 Tr S2: P = 0.0204; VLP-Control vs. VLP-BtKY72 S2: P = 0.0004; and VLP-BtKY72 Tr S2 vs. VLP-Khosta Tr S2: P = 0.0002.

MA10-challenged C57BL/6 mice immunized with VLP-S2 vaccines all had lung viral titers that were significantly lower than those for mice immunized with the VLP-Control (Fig. 4b). Mice immunized with the clade 2 S2 subunit antigen (VLP-YN2013 S2) had a mean lung viral titer more than 430,000-fold lower than that for VLP-Control immunized mice. Mice immunized with clade 3 VLP-S2 constructs had mean lung viral titers at least 165,000-fold lower than the mean lung viral titer of VLP-Control immunized mice. VLP-S2 immunized mice did show a significant albeit minor decrease in viral titers in the nasal turbinates relative to control (Supplementary Fig. 4b).

Similarly, VLP-S2-immunized K18-hACE2 mice challenged with XBB all had lung viral titers significantly lower than those for mice immunized with VLP-Control (Fig. 4c). The mean viral titers of VLP-YN2013 S2, VLP-BtKY72 S2, VLP-BtKY72 Tr. S2 and VLP-Khosta-2 Tr. S2 were more than 40, 45, 145, and 30-fold (respectively) lower than the mean viral titer of VLP-Control immunized mice. There was no statistically significant difference between the lung viral titers of the VLP-S2-immunized mice. No significant decrease in virus in the nasal turbinates was detected from VLP-S2 immunized mice (Supplementary Fig. 4c). Overall, these results indicate that VLP-S2 vaccines from clade 2 and clade 3 sarbecoviruses can protect against clade 1b sarbecoviruses.

For clade 1a WIV1 challenges, K18-hACE2 mice immunized with VLP-S2 vaccines again had lung viral titers significantly lower than those for mice immunized with the VLP-Control (Fig. 4d). The mean lung viral titers of VLP-BtKY72 S2 and VLP-Khosta-2 Tr. S2 immunized mice were approximately 30 and 65-fold lower, respectively, than those for mice immunized with VLP-Control. There was no statistically significant difference in lung viral titers between VLP-BtKY72 S2 and VLP-Khosta-2 Tr. S2-immunized mice. Mice immunized with VLP-YN2013 S2 had a mean viral titer over 330-fold lower than those for mice immunized with VLP-control. The greatest decrease in mean lung viral titers relative to controls was seen in VLP-BtKY72 Tr. S2-immunized mice, with mean lung viral titers over 4,500-fold lower than those in mice immunized with VLP-Control. VLP-BtKY72 Tr. S2-immunized mice also had lung viral titers significantly lower than those for mice immunized with the other 3 VLP-S2 vaccines. All VLP-S2-immunized groups showed significantly lower viral loads in the nasal turbinates compared to VLP-Control (Supplementary Fig. 4d). Collectively, these results demonstrate that VLP-S2 vaccines from clade 2 and clade 3 sarbecoviruses can also provide protection against clade 1a sarbecoviruses.

To probe the mechanisms underlying this protection, we conducted neutralization tests and in vitro Fc-activity assays using sera from single-dose immunized C57BL/6 mice. Sera from VLP-S2-immunized mice failed to neutralize an early isolate of SARS-CoV-2 (Supplementary Fig. 5), indicating that neutralization is unlikely to be a significant cause of the observed protection. We also tested the ability of sera from VLP-S2-immunized mice to activate FcγR-mediated signaling using antibody-dependent cellular cytotoxicity (ADCC) reporter bioassays. Sera from mice immunized with VLP-YN2013 S2, VLP-BtKY72 S2, or VLP-BtKY72 Tr. S2 were all capable of significantly activating ADCC signaling in effector cells co-cultured with CHO cells expressing the SARS-CoV-2 spike protein (Supplementary Fig. 6). While some activity was seen for sera from mice immunized with VLP-Khosta-2 Tr. S2, the difference did not achieve statistical significance relative to control. As such, while ADCC may be a contributing factor to protection, other mechanisms such as the classical complement pathway, antibody dependent cellular phagocytosis (ADCP), or adaptive cellular immunity may also contribute.

Discussion

There has been minimal development of vaccines against clade 2 and clade 3 sarbecoviruses despite several such viruses being capable of infecting human cells. Given the global impact of previous sarbecoviruses, most notably SARS-CoV-1 in 2003 and SARS-CoV-2 in 2019, there is an appreciable need to preemptively develop and characterize vaccines targeting clade 2 and clade 3 sarbecoviruses. Here, we developed clade 2 and clade 3 S2 vaccines and showed that they can elicit a robust and highly cross-reactive antibody response in mice (Fig. 3).

The protective efficacy of clade 2 and clade 3 S2 antigens against sarbecovirus challenges in vivo had not been evaluated prior to this work. While we were not able to evaluate how well these vaccines would protect against challenges from viruses in their respective clades, due to limitations in viral challenge models, we showed clade 2 and clade 3 S2-based vaccines can protect against clade 1 sarbecovirus challenges. Hsieh et al. previously reported that mice immunized with a SARS-CoV-2 (clade 1b) S2-based vaccine were protected to a greater extent against a clade 1b (SARS-CoV-2) challenge than against a clade 1a (SARS-CoV-1) challenge¹⁸. Similarly, we have previously shown that our clade 1a VLP-S2 vaccines protect mice against clade 1a challenges to a greater extent than clade 1b VLP-S2 vaccines²⁰. Given the protection seen for the clade 2 and clade 3 S2 vaccines against clade 1 sarbecoviruses in this work and in previous work, we hypothesize that these vaccines would provide even greater protection against viruses in their respective clades; however, direct evidence from challenge experiments with clade 2 and clade 3 sarbecoviruses will be needed to prove this hypothesis.

Lee et al. previously reported truncating the SARS-CoV-2 subunit enhanced its expression without compromising its binding to SARS-CoV-2 antibodies¹⁹. Lee et al. similarly included this truncation into their clade 3 S2 antigen (PRD-0038) but did not compare its expression to an untruncated PRD-0038 S2 antigen or report any further characterization for it¹⁹. We similarly found that truncating another clade 3 antigen BtKY72 S2 moderately enhanced its protein yield without compromising its immunogenicity or protective efficacy. Collectively, these results suggest that a similar truncation should be considered in future S2-based vaccine designs.

While our vaccines reduced viral titers in sarbecovirus-challenged mice, titers were not reduced to non-detectable levels. As such, further optimizing on the vaccine regimen needs to be done on these VLP-S2 vaccines, either by increasing the dosage amount, dosage frequency, or adjuvant used. Our future efforts will include expanding upon S2-based vaccines to include antigens from clade 4 sarbecoviruses. The data obtained from this study serve to contribute to the development of vaccines that can provide protection against viruses from all clades of the Sarbecovirus subgenus.

A limitation of this study is the usage of mouse models to evaluate the immune response of our VLP-S2 vaccines. Conclusions made in this paper may not directly translate directly into humans. Furthermore, while the cross-reactivity of elicited antibodies was tested against various sarbecovirus antigens, antibodies were not tested against all potential antigens. Another limitation of the study was the low number of mice used per group, despite the immune responses of the VLP-S2 immunized mice being consistent across the same experimental group. Additionally, we were unable to evaluate these clades 2 and clade 3 S2 vaccines against clade 2 and clade 3 sarbecovirus challenges. While we anticipate these vaccines would protect against sarbecoviruses from their respective clades to a greater extent than clade 1 sarbecoviruses tested here, this is not a guarantee. This study primarily serves to preemptively evaluate the immunogenicity and potential protective efficacy of S2-based vaccines against clade 2 and clade 3 sarbecoviruses should they ever lead to outbreaks in humans.

Methods

Approval of animal studies

Immunization and challenge studies at the University of Wisconsin-Madison were performed under an approved protocol (Protocol Number: V006426) reviewed by the Institutional Animal Care and Use Committee. To minimize pain, virus infections were performed under isoflurane anesthesia. Group sizes were determined based on previous S2-antigen vaccine studies, and no sample-size calculations were performed prior to the study in order to determine the power of these studies. Animal groups were not blinded to the researchers.

Expression of S2 proteins

DNA encoding the S2 subunit regions of the YN2013 (AIA62330.1, residues 646-1233, BtKY72 (APO40579.1, residues 673–1257) and Khosta-2 (QVN46569.1, residues 667–1253) spike proteins were modified to include six proline mutations based on the SARS-CoV-2 HexaPro spike protein³². The S2’ protease cut site of the S2 antigens was also eliminated with additional mutations similar to our previous S2-constructs³². Clade 3 antigens from BtKY72 and Khosta-2 were modified to include 3 additional mutations (F744W, F883L, N940D) determined by the Rosetta molecular modeling program following the point mutation method reported by Thieker et al³³. Another version of the BtKY72 S2 antigen was generated with a truncation in the N-terminus. This truncation was also added to the Khosta-2 S2. These four DNA segments (YN2013 S2, BtKY72 S2, BtKY72 Tr. S2, and Khosta-2 Tr. S2) were cloned by Gene Universal, Inc. (Newark, DE) into pcDNA3.1(-) between the Ncol and Xhol restriction sites with the addition of an N-terminal mouse Ig Kappa signal peptide and C-terminal T4 fibritin trimerization domain, AviTag, and hexahistidine tag similar to our previous S2 constructs^13,20.

These plasmids were transfected into Expi293F cells (RRID: CVCL_D615) using the ExpiFectamine Transfection Kit (Thermo Fisher Scientific) per manufacturer protocol. Cell cultures were centrifuged 5 days after transfection at 6000 × g for 10 minutes. The supernatant was dialyzed into PBS for 2 h, then dialyzed for a further 2 hours with fresh PBS. The dialyzed supernatant and 1 mL of HisPure Ni-NTA resin (Thermo Fisher Scientific) were mixed together overnight on a stir plate. The mixture was loaded onto a gravity flow column (G-Biosciences) and then washed with 90 mL of binding buffer (150 mM Tris, 150 mM NaCl, 20 mM Imidazole, pH 8). For collecting the hexahistidine-tagged S2 proteins, 3 mL of elution buffer (150 mM Tris, 150 mM NaCl, 400 mM Imidazole, pH 8) was loaded onto the column and allowed to incubate with the resin before being collected. This elution process was repeated two more times to collect a total eluate of 9 mL. This eluate was concentrated down using a 10 kDa MWCO spin filter (Millipore Sigma) to 1 mL and then further purified by size exclusion chromatography using a Superdex 200 Increase 10/30 column in PBS. The bicinchoninic acid assay (BCA) assay (Thermo Scientific) was used to determine the final protein concentration.

Computational stabilization of BtKY72 and Khosta-2 antigens

Stabilized BtKY72 and Khosta-2 S2 proteins were designed using the point mutation protocol developed by Thieker et al.³³. First, a model of the BtKY72 S2 ectodomain with a mutated S2’ site and six proline mutations based on the SARS-CoV-2 HexaPro spike protein was created using AlphaFold-Multimer^37,38 v2.3.1(Google DeepMind) using the sequence databases BFD, MGnify 2022_05³⁹, UniRef30 2021_03, UniRef90⁴⁰ (downloaded February 2023), and UniProt (downloaded March 2023). Templates from the Protein Data Bank were used with a cutoff date of January 1, 2023. The BtKY72 S2 model was then used as input for the point mutation protocol from Thieker et al. included in Rosetta 2023.6⁴¹. All residues with less than 10% relative solvent accessible surface area as calculated using PyMOL 2.5 (Schrodinger) and which were not included in the HexaPro mutations, disulfide bonds, glycosylation sites, or broadly neutralizing epitopes were analyzed using this approach. From this search, three mutations—F744W, F883L, and N940D—were selected for use in combination with the HexaPro proline mutations and the S2’ mutation already included.

The truncated BtKY72 S2 antigen was then designed by truncating the first 10 residues of the stabilized BtKY72 S2 antigen. In silico stability was then compared by generating new predicted structures using AlphaFold and comparing their Rosetta energy scores after relaxation into the ref2015 score function⁴² using the FastRelax application.

The Khosta-2 S2 antigen was designed by aligning the Khosta-2 spike protein sequence to the truncated BtKY72 S2 antigen and transferring the HexaPro, S2’, and stabilizing mutations (F739W, F878L, N935D) to the Khosta-2 spike.

Expression and purification of MS2

The expression and purification of MS2 have been described previously^9,13,29,30. DNA encoding a single chain dimer of the MS2 coat protein was cloned between the XhoI and Ndel restriction sites into the pET-28b vector. An AviTag was inserted between the 14th and 15th residues of the second MS2 protein monomer. A plasmid encoding BirA biotin-protein ligase was co-transformed alongside the MS2 plasmid into BL21(DE3) competent E. coli (New England Biolabs) per the manufacturer’s protocol. A 5 mL starter culture using 2xYT media was made using a single colony and left overnight at 37 °C in a rotating incubator. The starter culture was added to 1 L of 2xYT media and grown at 37 °C at 225 rpm until the culture reached an optical density between 0.6–0.8. The culture was then induced by adding 1 mL of 1 M isopropyl β-D-1-thiogalactopyranoside (IPTG) (Fisher BioReagents). Immediately after, 50 mM D-biotin was added to the culture, and the incubation temperature was reduced to 30 °C and left overnight^9,13,29,30. The culture was centrifuged at 7000 × g for 7 min, and the cell pellet was resuspended in 25 mL of lysis buffer (20 mM Tris base, 0.5 mg/mL lysozyme, 125 units of EMD Millipore benzonase, a quarter of a SigmaFast EDTA-free protease inhibitor cocktail tablet, pH 8). This lysate was incubated at 4 °C on a rocker for 20 minutes, after which sodium deoxycholate (Alfa Aesar) was then added so the final concentration was 0.1% (w/v). The lysate was sonicated on ice at 35% amplitude for 3 min with 3-s pulses (Sonifier S-450, Branson Ultrasonics)^9,13,29,30. This sonication process was repeated one additional time. The lysate was then centrifuged at 19,000 × g for 30 min, and the supernatant was collected and centrifuged at the same speed and duration a second time. The supernatant was then collected and diluted with 20 mM Tris Base, pH 8 to reach a total volume of 100 mL, and loaded onto four HiScreen Capto Core columns (Cytvia) in series using an AKTA Start system. To elute the MS2 protein, the columns were washed with 5 column volumes (CVs) of 20 mM Tris Base, pH 8. Fractions corresponding to the MS2 protein were collected and concentrated down using a 10 kDa MWCO spin filter (Millipore Sigma) and purified further using a Superdex 200 Increase 10/30 column in 20 mM Tris, 20 mM NaCl, pH 8^9,13,29,30. The bicinchoninic acid assay (BCA) assay (Thermo Scientific) was used to determine final protein concentration.

In vitro biotinylation of AviTagged MS2 and S2 proteins

The BirA biotin-protein ligase standard reaction kit (Avidity LLC) was used to biotinylate the MS2 and S2 proteins as described previously^9,13,20. Biomix B (ATP, biotin, and magnesium acetate) and BirA were added to the proteins, which were then mixed together overnight at 4 °C and supplemented again with Biomix B the next day. After 2-h incubation at 37 °C the next day, biotinylated proteins were then purified via size exclusion chromatography using a Superdex 200 Increase 10/30 column in PBS to remove BirA and excess biotin. The bicinchoninic acid (BCA) assay (Thermo Scientific) was used to determine final protein concentration.

Assembly and purification of MS2-SA VLPs

The expression, refolding, and purification of streptavidin (SA) have been extensively described previously^{9,13,29,30,43,44}. The SA expression plasmid, pET21-Streptavidin-Glutamate_Tag, was a gift from Mark Howarth (Addgene plasmid # 46367; http://n2t.net/addgene:46367; RRID:Addgene_46367)⁴³. A solution of SA (at a minimum concentration of 30 mg/mL) was rapidly stirred in a glass vial to which biotinylated MS2 (at a maximum concentration of 700 µg/mL) was added in 2.5 µL increments. The MS2 and SA were mixed together so that the SA was at a 20x molar excess to that of the MS2. The MS2-SA mixture was loaded onto a Superdex 200 Increase 10/30 size exclusion chromatography column in PBS to separate the excess SA from the MS2-SA VLPs. The concentration of SA on MS2-SA VLPs was quantified using SDS polyacrylamide gel electrophoresis (SDS-PAGE). The VLPs were heated for 30 minutes at 90 °C after being mixed with Nu-PAGE lithium dodecyl sulfate (LDS) sample buffer (Invitrogen). SA standard samples of known concentrations were loaded onto a polyacrylamide gel alongside the VLP samples. A standard curve relating the band intensities of the SA standard samples to their known concentrations was compared to the SA band intensities of the VLPs to determine their concentration^9,13,29,30.

Preparation of VLP-S2

VLPs were generated as described previously^9,13,29,30. Analytical SEC was used to determine the optimal ratio of biotinylated VLP and S2 proteins at which no excess unbound S2 was detected. Varying amounts of the MS2-SA VLP were mixed with 8 µg of the S2 antigen, diluted in PBS for a total volume of 950 µL, and loaded onto a Superdex 200 increase 10/300 SEC column (Cytiva). The proteins were eluted at 0.65 mL/min with a full column volume of PBS while the absorbance was monitored at 210 nm. The ratio that included the least amount of MS2-SA without a peak corresponding to excess S2 antigen was considered the optimal ratio and used to generate VLP-S2.

SDS-PAGE

PNGase F (New England BioLabs) was used to deglycosylate protein samples, which were then mixed with 5 µL of Nu-PAGE lithium dodecyl sulfate (LDS) sample buffer (Invitrogen) and 2 µL of a 1:1000 dilution of beta-mercaptoethanol as described previously^13,20. Protein samples were heated at 98 °C for 30 min before being loaded into a 4–12% Bis-Tris gel (Invitrogen) alongside a PageRuler Plus Prestained Protein Ladder (Thermo Scientific). The gel was run for 50 min at 120 V in MES-SDS buffer, stained with Imperial Protein Stain (Thermo Scientific) for 30 min, and then destained in DI water overnight. The ChemiDoc MP imaging system and Image Lab 5.2.1 software (Bio-Rad) were used to image the gels.

Expression of S2P6 and 3A3 antibodies

The S2P6 antibody³⁴ and 3A3 antibody³⁵ were expressed as described previously⁹. The antibody expressed using the ExpiFectamine Transfection Kit (Thermo Fisher Scientific) in Expi293F cells (according to manufacturer instructions) after its light and heavy chains were cloned into the TGEX-HC and TGEX-LC vectors (Antibody Design Labs), respectively. The transfected cultures were incubated with shaking at 37 °C and 8% CO₂ in a humidified incubator. Six days post-transfection, cells were centrifuged at 6000 × g for 15 min, and the supernatant was diluted in MabSelect Binding Buffer (20 mM sodium phosphate, 150 mM NaCl, pH 7.2). The supernatant was loaded onto a 1-mL MabSelect SuRe column (Cytiva) connected to an ÄKTA Start to purify the protein according to the manufacturer’s instructions, and the antibody was subsequently purified again by SEC using a Superdex 200 Increase 10/30 column. The eluted protein was then dialyzed in PBS, and the BCA assay (Thermo Scientific) was used to determine final protein concentration.

S2 and VLP-S2 ELISA

ELISA characterization was done as described previously^9,13,29,30. Into each well of a Nunc Maxisorp 96-well plate, a solution containing 0.1 µg of S2 protein in 100 µL of PBS was added and incubated for 1 h. The protein solution was removed, and then the plate was blocked with 200 µL per well of a 5% BSA (EMD Millipore) solution in PBST (PBS with 0.05% v/v Tween-20). The plate was incubated for 1 hour. The BSA solution was then discarded, and the wells were washed three times with 100 µL each of PBST. For Fig. 2, the primary antibody S2P6 (at 1.9 mg/mL in a stock solution) was diluted 1:30000 in a 1% BSA solution in PBST. For concentration-dependent ELISAs, S2P6 or 3A3 antibodies were diluted to their final concentration in a 1% BSA solution in PBST.

A total of 100 µL of this diluted antibody solution was added to each well, incubated for 1 hour, and then removed. The plates were washed again three times with PBST. HRP-conjugated goat anti-human secondary antibody (MP Biomedicals) were then diluted 1:5000 in 1% BSA in PBST, and 100 µL of this diluted secondary antibody was added to each well. After one hour of incubation, the secondary antibody was removed, and the plates were washed three times with 100 µL per well of PBST. After washing, 100 µL of TMB substrate (Thermo Scientific) was added to each well, and 160 mM sulfuric acid was added 3 minutes later to stop development. The plates were read using a Synergy H4 plate reader (BioTek) with Gen5 2.07 software (BioTek) at an absorbance of 450 nm^9,13,29,30.

Dynamic light scattering

A Zetasizer Nano (Malvern) was used to measure dynamic light scattering. The proteins (either S2 only, VLP only, or VLP-S2) were diluted to a final concentration of 5 µg of S2 in 100 µL of PBS. The MS2-SA was diluted so there was approximately 2 µg of SA in 100 µL. The diluted protein, in a volume of 100 µL was added to a UVette (Eppendorf). A total of 13 acquisitions were collected at 25 °C per sample. Results were displayed as percent volume.

Nano differential scanning fluorimetry

Nano differential scanning fluorimetry (nanoDSF) was performed using a Prometheus NT.48 instrument (NanoTemper) using PR.ThermControl (NanoTemper) as control software. Data analysis was conducted using PR.StabilityAnalysis (NanoTemper). S2 proteins were concentrated to 125 µg/mL and loaded into Prometheus Standard Capillaries (NanoTemper). Fluorescence emission at 330 nm was measured over a temperature range of 20 °C–70 °C at a temperature ramp rate of 2 °C per min and an excitation power of 100%. Three technical replicates per protein were conducted from a single preparation.

Amino acid identity and phylogenetic trees

Phylogenetic trees and identity tables were generated as described previously⁹. Spike amino acid sequences were retrieved from GenBank and aligned using Clustal Omega 1.2.3 (Conway Institute, UCD Dublin). The percentage of identical amino acids for each paired comparison was calculated using these alignments, and maximum likelihood phylogenetic trees were generated using PhyML 3.3.20220408 (Stephane Guindon, University of Montpellier) using a maximum parsimony starting tree and the LG amino acid substitution model. TreeViewer 2.0.1 (Giorgio Bianchini, University of Bristol) was used to visualize phylogenetic trees.

Biosafety and containment for coronaviruses

Research with sarbecoviruses was performed under biosafety level 3 agriculture (BSL-3 AG) containment at the Influenza Research Institute with an approved protocol reviewed by the University of Wisconsin-Madison’s Institutional Biosafety Committee. The laboratory is designed to meet and exceed the standards outlined in Biosafety in Microbiological and Biomedical Laboratories (6th edition).

Cell lines and viruses

Vero E6 TMPRSS2 cells (National Institute of Infectious Diseases, Japan) were used to propagate all virus stocks and were maintained in high glucose Dulbecco’s modified Eagle’s medium (DMEM) containing 10% fetal bovine serum (FBS) and antibiotic/antimycotic solution along with G418 (1 mg/ml). Virus titrations of tissue samples were performed on Vero E6 TMPRSS2-T2A-ACE2 cells (NIAID Vaccine Research Center; Dr. Barney Graham) overlaid with 1% methylcellulose media. The cells were maintained in DMEM supplemented with 10% FBS, 10 mM HEPES (pH 7.3) and antibiotic/antimycotic solution along with puromycin (1 µg/ml). Both cell lines were tested monthly for mycoplasma contamination by PCR and were confirmed to be mycoplasma-free.

In the mouse challenge studies, the following viruses were used, SARS-CoV-2 mouse-adapted strain (MA10 variant), XBB (hCoV-19/Japan/TY41-795/2022 [Accession ID: EPI_ISL_16355653), and WIV1 (GenBank: KF367457).

Mouse immunizations and challenge studies

C57BL/6 and K18-hACE2 mice were purchased from Jackson Laboratories. All mice were female and at the age of 10–12 weeks at the start of the studies. Mice were immunized by subcutaneous injection with a total volume of 250 µl of an equal volume mixture of VLP vaccine preparations (7.5 µg of each S2 antigen) and adjuvant mixture (AddaS03 with 100 µg/mouse of poly I:C [InvivoGen]). For serum collection, blood was obtained via terminal cardiac puncture while mice were under deep isoflurane anesthesia.

For challenge studies, mice were infected intranasally with the indicated virus at 10⁵ plaque-forming units (pfu) of virus (25 µl of total volume). Three days after infection, animals were humanely sacrificed by an overdose of isoflurane and tissues were collected to determine virus titers.

Enzyme-linked immunosorbent assay (ELISA) for mouse sera

To perform ELISA, Nunc Maxisorp 96-well plates were coated overnight at 4 °C with 50 μL per well with the indicated antigen at a concentration of 2 μg/mL in PBS (0.2 μg S2 per well). After blocking with PBS containing 0.1% Tween 20 (PBS-T) and 3% milk powder, the plates were incubated with heat-inactivated sera diluted in PBS-T with 1% milk powder (2-fold dilution series starting at 1:320). A mouse IgG secondary antibody conjugated with horseradish peroxidase (1:3000 dilution, ThermoFisher) was used for detection. Plates were developed with SigmaFast o-phenylenediamine dihydrochloride solution (Sigma), and the reaction was stopped with the addition of 3 M hydrochloric acid. The absorbance was measured at a wavelength of 490 nm (OD₄₉₀). Background absorbance measurements from pooled sera collected from naïve mice were subtracted from sera collected after immunization. IgG antibody endpoint titers were defined as the highest serum dilution with an OD₄₉₀ cut-off value of 0.15.

Tissue viral titration

Lung and nasal turbinate tissues from challenged mice were homogenized and clarified before being added to confluent Vero E6/TMPRSS2-T2A-ACE2 cells as undiluted or 10-fold serial dilutions. The mixture was incubated for 30 minutes before the tissue homogenate was removed. Cells were washed and then overlaid with 1% methylcellulose solution in DMEM with 5% FBS. Plates were incubated for three days before fixation with 20% methanol and plaque staining with crystal violet.

Focus reduction neutralization test

Neutralization assays were carried out as previously described. Sera from single-dose VLP-S2 or VLP-Control immunized C57BL/6 mice were serially diluted starting at a 1:20 dilution and mixed with 1000 focus-forming units of an early isolate of SARS-CoV-2, incubated for 1 hour at 37 °C, and then loaded onto Vero E6/TMPRSS2 cells and incubated again for 1 h at 37 °C. An equal volume of methylcellulose solution was then added before incubation for 16 hours at 37 °C. Cells were then fixed with formalin and stained with an anti SARS-CoV-1/2 nucleoprotein monoclonal antibody (Sigma-Aldrich, MA5-29982, 1:10,000 dilution) followed by an HRP-conjugated goat anti-mouse secondary antibody. (ThermoFisher, catalog #31430, 1:2000 dilution). TrueBlue Substrate (SeraCare Life Sciences) was then added to stain infected cells. Cells were washed with distilled water and dried. Foci counts were determined using an ImmunoSpot S6 Analyzer with ImmunoCapture and BioSpot software (Cellular Technology).

ADCC reporter assay

ADCC reporter assays were performed using mFcγRIV ADCC Reporter Bioassay Kits (Promega) according to the manufacturer’s instructions. The activation of ADCC signaling in effector cells was measured using CHO cells expressing the SARS-CoV-2 spike (Promega). Briefly, CHO cells were plated at a density of 10,000 cells/well in a flat-bottom white 96-well plate (Corning). The cells were used as target cells 24 h later. The medium of the target cells was replaced with mouse serum five-fold diluted (1/20–1/12500) in the assay buffer (Promega). Effector cells were added to the antibody-treated target cells and incubated for 6 h at 37 °C. The firefly luciferase activity was then measured using luciferase assay reagents (Promega).

Statistics and reproducibility

Characterization of S2P6 binding by ELISA (Fig. 2d) and concentration-dependent characterization of S2P6 and 3A3 binding by ELISA (Supplementary Fig. 2) were each conducted once in triplicate for each condition and presented as mean ± SD. Melting temperature analysis (Fig. 2g) was done in triplicate with similar results.

No sample-size calculations for power analysis were done for these studies as group sizes were determined based on previous S2-antigen vaccine studies. Animals were not blinded to researchers.

Endpoint titers (Fig. 3) were determined using sera from each mouse to conduct a single assay. One animal from each of the VLP-Control, VLP-YN2013 S2, and VLP-BtKY72 S2 cohorts was removed during the vaccination phase due to veterinary recommendations, resulting in n = 4 for these groups. The remaining cohorts retained all five animals (n = 5). The data is presented as geometric mean ± geometric SD. Significance was determined by an ordinary one-way analysis of variance (ANOVA) and Tukey’s multiple comparisons between groups (α = 0.05).

Virus titers in the lungs of VLP-S2 immunized mice (Fig. 4b-d) were presented as mean ± SD. This study was conducted in two staggered experimental blocks; VLP-Control-immunized mice were included in each block to ensure reproducibility, resulting in a larger shared VLP-Control cohort. Vaccine group sizes differ due to 1–2 animals being removed from select groups during the vaccination phase due to veterinary recommendations. For MA10 challenge mice groups: n = 8 for VLP-Control, n = 3 for VLP-YN2013 S2, n = 4 for VLP-BtKY72 S2, and n = 5 for VLP-BtKY72 Tr. S2 and VLP-Khosta Tr. S2. For XBB challenge groups, n = 9 for VLP-Control, n = 4 for VLP-BtKY72 S2, and n = 5 for the remaining groups. For WIV1 challenge groups: n = 8 for VLP-Control, and n = 4 for the rest. Significance was determined by one-way ordinary ANOVA and Tukey post hoc multiple comparisons between groups (α = 0.05).

Author contributions

All authors P.J.H., R.S.P., A.D., T.W., A.Y., K.C., A.S., C.R., P.C.W., Y.K., and R.S.K. were involved in the study design. P.J.H., R.S.P., A.D., T.W., A.Y., K.C., A.S., and C.R. performed data collection and data analysis. P.J.H., R.S.P., A.D., A.Y., and R.S.K. wrote the manuscript. All authors read and approved of the final manuscript.

Data availability

Data supporting the conclusions of this paper can be found within the paper, Supplementary Information, and Supplementary Data 1 file.

Competing interests

The authors declare the following competing interests: Y.K. has received unrelated funding support from Daiichi Sankyo Pharmaceutical, Toyama Chemical, Tauns Laboratories, Inc., Shionogi & Co. LTD, Otsuka Pharmaceutical, KM Biologics, Kyoritsu Seiyaku, Shinya Corporation, and Fuji Rebio. The other authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41541-025-01333-4.