Strategy to develop broadly effective multivalent COVID-19 vaccines against emerging variants based on Ad5/35 platform

Significance

The emergence of highly diverse sub-lineages within the SARS-CoV-2 Omicron variant poses a significant threat to the efficacy of existing COVID-19 vaccines, leading to breakthrough infections and re-infections. Therefore, developing new multivalent vaccines that can provide broad neutralization of circulating and emerging Omicron variants is crucial. Herein, we developed bivalent and trivalent Omicron variant-specific vaccines using phylogenetic trees and antigenic cartography and demonstrated their superior ability to neutralize a wide range of variants. The AdCLD-CoV19-1 XBB/BN.1/BQ.1.1 or XBB.1.5/BN.1/BQ.1.1 trivalent vaccine could elicit broader neutralizing antibodies against newly emerged and prevailing circulating Omicron variants, including XBB.1.5, XBB.1.16, EG.5.1, FL.1.5.1, and BA.2.86. Determining future booster combinations before new variants become widespread will be a crucial challenge in combating COVID-19.

Abstract

The severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) Omicron strain has evolved into highly divergent variants with several sub-lineages. These newly emerging variants threaten the efficacy of available COVID-19 vaccines. To mitigate the occurrence of breakthrough infections and re-infections, and more importantly, to reduce the disease burden, it is essential to develop a strategy for producing updated multivalent vaccines that can provide broad neutralization against both currently circulating and emerging variants. We developed bivalent vaccine AdCLD-CoV19-1 BA.5/BA.2.75 and trivalent vaccines AdCLD-CoV19-1 XBB/BN.1/BQ.1.1 and AdCLD-CoV19-1 XBB.1.5/BN.1/BQ.1.1 using an Ad5/35 platform-based non-replicating recombinant adenoviral vector. We compared immune responses elicited by the monovalent and multivalent vaccines in mice and macaques. We found that the BA.5/BA.2.75 bivalent and the XBB/BN.1/BQ.1.1 and XBB.1.5/BN.1/BQ.1.1 trivalent vaccines exhibited improved cross-neutralization ability compared to their respective monovalent vaccines. These data suggest that the developed multivalent vaccines enhance immunity against circulating Omicron subvariants and effectively elicit neutralizing antibodies across a broad spectrum of SARS-CoV-2 variants.

Since its first report in December 2019, the COVID-19, caused by severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), has spread worldwide. Although the pandemic has significantly abated, cases of COVID-19 and related deaths are still reported worldwide (1, 2). To combat the COVID-19 pandemic, numerous companies rapidly developed vaccines to reduce infection rates and COVID-19-related deaths. However, SARS-CoV-2 variants, such as Alpha, Beta, Gamma, Delta, and Omicron, continued to emerge, and the protective effect of COVID-19 monovalent vaccines based on the ancestral strain was reduced. In particular, COVID-19 monovalent vaccines showed a substantial loss in their neutralizing activity against the Omicron variants, which became the dominant species after being first reported in South Africa in November 2021 (3).

To improve responses against the Omicron variants, Moderna and Pfizer-BioNTech developed bivalent vaccines that contain two mRNAs encoding an ancestral strain and either BA.1 or BA.4/5 spike protein. These bivalent vaccine boosters showed greater induction of the immune response than the ancestral strain monovalent vaccine against BA.1 and BA.4/5. Therefore, BA.1- or BA.4/5-matched bivalent mRNA vaccines increased the breadth of neutralization (4–6).

However, BA.2.75, which has selective advantages relative to BA.5, like ACE2-binding and immune evasion, became highly prevalent especially in Asia after September 2022. At the end of 2022, several variants derived from BA.5, such as BQ.1, BQ.1.1, and those derived from BA.2.75, such as BN.1, BR.2.1, and XBB, co-circulated (7, 8). These newly emerging Omicron subvariants such as BA.2-, BA.5-, and BA.2.75-derived and recombinant XBB subvariants have evolved with additional novel spike mutations, which affect vaccine effectiveness (9). According to the results of a recent study, the currently used mRNA bivalent vaccine based on an ancestral strain and BA.1 or BA.4/5 spike proteins shows low neutralization rates against the new variants (9, 10). The neutralization activity against BA.2.75.2, BQ.1, BQ.1.1, XBB, and XBB.1 was markedly reduced in participants who had received three doses of the wild-type mRNA vaccines (Pfizer BNT162b2 or Moderna mRNA-1273) and received a fourth dose of the bivalent BA.5 booster. Therefore, a new vaccine development strategy is required to tackle the newly emerging variants and provide broad neutralization.

We previously developed AdCLD-CoV19-1, a replication-defective chimeric adenoviral vector (Ad5/35)-based COVID-19 vaccine that expresses the spike protein of the SARS-CoV-2 ancestral strain (11, 12). For better antigen delivery, we developed an Ad5/35-vector, which is replaced with a serotype 35 fiber based on the backbone of serotype 5 adenovector. Additionally, we inserted a linker peptide into the cleavage site, which leads to a more stable expression of the S protein in transduced cells and enhances S protein–specific immune responses. Therefore, our vaccine can effectively deliver spike genes to antigen-presenting cells through CD46 binding, which leads to effective stimulation of CD4⁺ T cells, CD8⁺ T cells, and B cells in either direct or indirect ways (13).

In this study, we developed the AdCLD-CoV19-1 BA.5/BA.2.75 bivalent vaccine and the AdCLD-CoV19-1 XBB/BN.1/BQ.1.1 and AdCLD-CoV19-1 XBB.1.5/BN.1/BQ.1.1 trivalent vaccines, which are replication-defective chimeric adenoviral vector (Ad5/35)-based COVID-19 vaccines. According to the wave of SARS-CoV-2 variants, we selected BA.5 and BA.2.75 variants for the bivalent vaccine and XBB (or XBB.1.5), BN.1, and BQ.1.1 variants for the trivalent vaccine based on a phylogenetic tree constructed using spike sequences and antigenic maps constructed using neutralization data. Furthermore, we evaluated the neutralizing ability of the bivalent and trivalent vaccines against newly emerged and prevailing circulating Omicron variants and compared the results to those of monovalent vaccines. This study will provide evidence to support the further clinical development of multivalent vaccines.

Results

Immunogenicity of the Bivalent Vaccine AdCLD-CoV19-1 BA.5/BA.2.75 in Mice.

We constructed modified AdCLD-CoV19-1 vaccines carrying spike mutations for the Omicron subvariants BA.5 and BA.2.75. We administered BALB/c mice a single intramuscular injection, and the total particle (VP) amount of the bivalent vaccine was equal to that of the monovalent vaccine (Fig. 1A). We performed a SARS-CoV-2 spike protein–expressing pseudotyped lentivirus-based neutralizing assay (Fig. 1 B–G and SI Appendix, Fig. S1 and Table S1). Mice vaccinated with the bivalent vaccine showed greater neutralizing activity against BA.1, BA.2, BA.2.75, and BA.5 than monovalent vaccinated mice (Fig. 1 B–D). The AdCLD-CoV19-1 BA.5 monovalent vaccine substantially induced neutralizing antibody titers against the spike-matched BA.5 pseudo-virus; however, antibody titers against BA.2.75, which has spike protein sequences that are distant from BA.5, were less (Fig. 1C). The AdCLD-CoV19-1 BA.2.75 monovalent vaccine also significantly induced neutralizing antibody titers against the spike-matched BA.2.75 pseudo-virus but those against BA.5 were less (Fig. 1D). We also compared the neutralizing activity against new emerging variants, such as BF.7, BQ.1, BN.1, and XBB; the bivalent vaccine AdCLD-CoV19-1 BA.5/BA.2.75 induced greater neutralizing responses than the monovalent vaccines (Fig. 1 E–G). Collectively, the bivalent vaccine AdCLD-CoV19-1 BA.5/BA.2.75 could elicit broader neutralizing antibody responses than the monovalent vaccines. To achieve a high magnitude and breadth of neutralization (results better than those of the bivalent vaccine AdCLD-CoV19-1 BA.5/BA.2.75) against recent variants, such as BN.1 and XBB, we developed an additional multivalent vaccine.

Fig. 1.

Neutralizing antibody responses of AdCLD-CoV19-1 platform-based BA.5-, BA.2.75-specific monovalent, and BA.5/BA.2.75 bivalent vaccines. (A) Study schedule. Naive BALB/c mice were vaccinated with 1 × 10⁹ VP of the BA.5/BA.2.75 bivalent vaccine or monovalent vaccine with a single intramuscular injection, and sera were collected at the indicated time points. (B–D) Neutralization of the Omicron BA.1, BA.2, BA.2.75, and BA.5 pseudoviruses by the sera of the mice 4 wk after vaccination with the BA.5/BA.2.75 bivalent or BA.5-, BA.2.75-specific monovalent vaccine. (E–G) Neutralization of the Omicron BF.7, BQ.1, BN.1, and XBB pseudoviruses by the sera of the mice 8 wk after vaccination with the BA.5/BA.2.75 bivalent or BA.5-, BA.2.75-specific monovalent vaccine. Geometric mean neutralization titers (GMT) are listed for each variant. “N” indicates the negative control group, which consists of naïve mice. See also SI Appendix, Table S1 and Figs. S1 and S7.

Antigenic Cartography of SARS-CoV-2 for Multivalent Vaccine Antigen Selection.

To refine the selection of variants for the multivalent vaccines, we constructed a phylogenetic tree using spike sequences of ancestral and Omicron strains, including the latest variants, such as BQ.1.1, XBB.1.5, BN.1, and CH.1.1.1. Excluding previously widespread but displaced variants, such as the ancestral strain, BA.1, and BA.2, the unrooted phylogenetic tree by maximum likelihood analysis revealed three distinct clusters: BA.5, BA.2.75, and XBB (Fig. 2A). Next, we generated antigenic maps constructed using neutralization data to visualize and quantify how different variants are antigenically related. We administered a single intramuscular injection of the modified monovalent AdCLD-CoV19-1 vaccines carrying spike mutations in each of the Omicron subvariants, including BA.1, BA.2, BA.2.12.1, BA.2.75, BA.4.1, and BA.5, to BALB/c mice. We obtained their sera at 4 and 8 wk post-vaccination and performed a SARS-CoV-2 S protein–expressing pseudotyped lentivirus-based neutralizing assay (SI Appendix, Table S2). Using neutralizing antibody titers, we generated an antigenic map (Fig. 2B). The degree of antigenic similarity is represented by the relative distance between points, with SARS-CoV-2 variants indicated with a circle and vaccinated sera marked with a square on the map. The map distances between points more closely together indicate higher cross-neutralization and antigenic similarity. Through this antigenic map, we obtained map distances and found that XBB, BN.1, and BQ1.1 were the most distant from each other (SI Appendix, Table S3). These variants contain several different mutations in the spike protein (Fig. 2C). XBB is a recombinant of BJ.1 (BA.2.10.1 sub-lineage) and BM.1.1.1 (BA.2.75 sub-lineage), BN.1 is a sub-lineage of BA.2.75, and BQ.1.1 is a sub-lineage of BA.5. Combining the phylogenetic tree and antigenic cartography results, we produced a trivalent vaccine using XBB-, BN.1-, and BQ.1.1-specific vaccines.

Fig. 2.

Clustering based on the sequence and immunogenicity of the vaccine. (A) Circular phylogenetic tree of the SARS-CoV-2 ancestral strain and Omicron subvariants spike protein. The phylogenetic tree was created using the maximum likelihood method based on the JTT matrix-based model. The four main clusters are shown in different colors: black = WT, BA.1, and BA2; blue = BA.2.75 subvariants; violet = BA.5 subvariants; green = XBB subvariants. XBB, BQ.1.1, and BN.1 which were selected for the trivalent vaccine, are shown in red. (B) SARS-CoV-2 variant spike antigenic map constructed using single dose neutralization data. The antigenic map was computed with the Racmacs package in R using 1,000 optimizations. Viruses shown as circles and antisera as squares are positioned on the map and distances between them are inversely related to the antibody titers. The grid in the background scales to a twofold dilution of antisera in the titrations. See also SI Appendix, Tables S2 and S3. (C) Differences in the amino acid substitution in the genomes of the SARS-CoV-2 variants used in the trivalent vaccine. Positions of the mutations and invariant (yellow) and variant (blue) non-Wuhan-Hu-1 are indicated. Deletions are indicated by dashes. NTD, N-terminal domain; RBD, receptor-binding domain; FP, fusion peptide; HR1, heptad repeat 1.

Immunogenicity of the Trivalent Vaccine AdCLD-CoV19-1 XBB/BN.1/BQ.1.1 in Mice.

The trivalent vaccine, AdCLD-CoV19-1 XBB/BN.1/BQ.1.1, comprised one-third of the targets (XBB, BN.1, and BQ.1.1), with a total VP amount equivalent to that of the monovalent vaccine. We administered a single intramuscular injection to BALB/c mice and performed a SARS-CoV-2 spike protein–expressing pseudotyped lentivirus-based neutralizing assay (Fig. 3A). Mice vaccinated with the trivalent vaccine showed greater neutralizing activity than the monovalent vaccinated mice against all the Omicron subvariants that we tested (Fig. 3B and SI Appendix, Figs. S2–S5 and Table S4). The monovalent vaccines induced neutralizing antibody responses against antigens closely related to the antigen used in the vaccine but did not elicit neutralizing antibody responses against distant antigens. For example, serum from the mice immunized with the AdCLD-CoV19-1 XBB monovalent vaccine efficiently neutralized XBB and XBB.1.5 but was less effective against BA.5, BA.2.75, and their sub-lineages (Fig. 3C). The AdCLD-CoV19-1 BN.1 monovalent vaccine induced neutralizing antibody responses against BN.1 and BA.2.75 but had less activity against BQ.1.1, XBB, and XBB.1.5 (Fig. 3D). Mice immunized with the AdCLD-CoV19-1 BQ.1.1 monovalent vaccine adequately neutralized BQ.1.1 and BA.5 but the responses against BA.2.75, XBB, and their sub-lineages were decreased (Fig. 3E). We also analyzed for neutralizing activity using a vesicular stomatitis virus (VSV)-based neutralization assay with pseudoviruses displaying spike proteins of BA.4/5, BN.1, BQ.1.1, and XBB.1.5 (SI Appendix, Fig. S6 and Table S5). The neutralizing activity seen with the trivalent vaccine AdCLD-CoV19-1 XBB/BN.1/BQ.1.1 was better than that seen with monovalent vaccines. Consequently, based on data from the pseudotyped lentivirus- and VSV-based neutralizing assays, the trivalent vaccine AdCLD-CoV19-1 XBB/BN.1/BQ.1.1 could elicit a broad neutralizing antibody response against currently circulating new variants.

Fig. 3.

Neutralizing antibody responses of the AdCLD-CoV19-1 platform-based monovalent and trivalent vaccines. (A) Study schedule. Naive BALB/c mice were vaccinated with 1 × 10⁹ VP of the XBB/BN.1/BQ.1.1 trivalent vaccine or monovalent vaccine with a single intramuscular injection and sera were collected at the indicated time points. (B–E) Neutralizing responses of the XBB/BN.1/BQ.1.1 trivalent vaccine or monovalent vaccine against the Omicron subvariants. A neutralization test was performed with samples obtained at 2, 4, 6, and 8 wk; data from these time points are shown. Geometric mean neutralization titers (GMT) are listed for each variant. N indicates the negative control group, which consists of naïve mice. See also SI Appendix, Figs. S2 and S5 and Table S4.

Antigenic Cartography for Evaluating the Effectiveness of the Trivalent Vaccine.

To visualize differences in antigenic relationships among SARS-CoV-2 variants by vaccination, we generated individual antigenic maps using neutralizing antibody titers from each specific vaccination group, as well as a consolidated antigenic map obtained by combining all neutralizing antibody titers from each vaccination group (Fig. 4 A–E). Titers were accurately represented as antigenic distances on the antigenic maps and confirmed that the data were well represented in two dimensions (SI Appendix, Fig. S7). Through the antigenic map generated using neutralizing antibody titers at 4, 6, and 8 wk after vaccination with the XBB, BN.1, and BQ.1.1 monovalent vaccines and the trivalent vaccine, the dataset presented in SI Appendix, Table S4, we observed that the positions of the data points representing sera from mice vaccinated with monovalent vaccines, indicated by squares, are noticeably concentrated around the respective antigens used for the vaccines (Fig. 4A). However, sera elicited by the trivalent vaccine were not biased compared with the monovalent vaccine, and the sera–variant distances were lower than those seen with the monovalent vaccines. We generated antigenic maps for each vaccinated group and calculated the antigenic distance represented on these maps to better visualize the differences in neutralization breadth between the vaccines. Each grid square represents one antigen unit (AU) and represents a two-fold change in the neutralization titer. Results of the antigenic map of the trivalent vaccine showed that the distances between variants and between serum and variants decreased compared with the results of the monovalent vaccines (Fig. 4 B–E). When we compared the geometric mean of all the distances between sera and variants, that of the trivalent vaccine was 1.2 AU, whereas that of the monovalent vaccines was 2.2 AU (XBB), 1.6 AU (BN.1), and 2.2 AU (BQ.1.1) (Fig. 4F and SI Appendix, Table S6). We also calculated antigenic distances between the variants based on the antigens used for each monovalent vaccine and converted them into fold changes for comparison (Fig. 4G and SI Appendix, Table S7). When we compared the geometric mean of the antigenic distance relative to the XBB antigen, the fold change of the trivalent vaccine was 2.7, but that of the monovalent vaccines was 6.62 (XBB), 5.4 (BN.1), and 15.63 (BQ.1.1). For the geometric mean of the antigenic distance relative to the BN.1 antigen, the fold change of the trivalent vaccine was 3.37, but that of the monovalent vaccines was 4.72 (XBB), 3.6 (BN.1), and 11.11 (BQ.1.1). For the geometric mean of the antigenic distance relative to the BQ.1.1 antigen, the fold change of the trivalent vaccine was 2.75, but that of the monovalent vaccines was 7.52 (XBB), 9.7 (BN.1), and 6.2 (BQ.1.1). Therefore, the trivalent vaccine had the shortest antigenic distance, calculated based on any of the XBB, BN.1, and BQ.1.1 antigens used in the monovalent vaccines. In addition, when we compared the geometric mean of the antigenic distances of 28 combinations for 8 antigens represented on maps, the fold change of the trivalent vaccine was the shortest at 3.1, and that of the monovalent vaccines was 7.8 (XBB), 4.8 (BN.1), and 9.4 (BQ.1.1) (SI Appendix, Table S7). We also verified that the trivalent vaccine AdCLD-CoV19-1 XBB/BN.1/BQ.1.1 is effective not only against the XBB.1.5 and XBB.2.3 variants that were prevalent in early 2023 but also against newly emerging variants such as EG.5.1, FL.1.5.1, and the BA.2.86 variant, which became prevalent in late 2023. Moreover, the neutralizing activity of the trivalent vaccine was better than that of the monovalent vaccines for these variants (Fig. 5 and SI Appendix, Figs. S8–S10 and Tables S8 and S9). These data indicate that the AdCLD-CoV19-1 XBB/BN.1/BQ.1.1 trivalent vaccine decreased antigenic distances between variants and enhanced cross-neutralizing activities to the SARS-CoV-2 variants.

Fig. 4.

Antigenic maps with the trivalent XBB/BN.1/BQ.1.1 vaccine (A) SARS-CoV-2 variant spike antigenic map constructed using neutralization data from single-dose groups of the XBB, BN.1, BQ.1.1 monovalent and trivalent vaccines. (B–E) SARS-CoV-2 variant spike antigenic map constructed using neutralization data from single-dose groups of the XBB/BN.1/BQ.1.1 trivalent vaccine or XBB, BN.1, and BQ.1.1 monovalent vaccines. Dashed circles represent the overall antigenic distances. (F) The antigenic map-based calculation geometric mean result of the relative distances between variant sera. (G) The antigenic map-based calculation result of relative distances between the variants. See also SI Appendix, Fig. S7 and Tables S6 and S7.

Fig. 5.

Neutralizing antibody responses to the monovalent and trivalent vaccines measured using pseudotyped lentiviruses. (A) Study schedule. (B–E) Naive BALB/c mice were vaccinated with 1 × 10⁹ VP of the XBB/BN.1/BQ.1.1 trivalent vaccine or monovalent vaccine with a single intramuscular injection, and serum samples were collected at indicated time points. The neutralizing activities in the collected sera were measured using pseudotyped lentivirus. Geometric mean neutralization titers (GMT) are listed for each variant. N indicates the negative control group, which consists of naive mice. See also SI Appendix, Table S8.

Immunogenicity of Booster Injection of the Trivalent Vaccine AdCLD-CoV19-1 XBB.1.5/BN.1/BQ.1.1 in Non-human Primates.

To determine whether the booster injection of trivalent vaccine AdCLD-CoV19-1 XBB.1.5/BN.1/BQ.1.1 could elicit broader neutralizing activity than the monovalent vaccine, we delivered 5 × 10¹⁰ VP of the bivalent vaccine AdCLD-CoV19-1 Wuhan-1/BA.5 to non-human primates (NHPs). After a 16-wk interval, we reimmunized the NHPs with a booster injection of the trivalent vaccine AdCLD-CoV19-1 XBB.1.5/BN.1/BQ.1.1 or an equivalent dose of monovalent vaccine AdCLD-CoV19-1 XBB.1.5 or AdCLD-CoV19-1 BQ.1.1. NHPs reimmunized by booster injection of the trivalent vaccine broadly induced neutralizing antibody responses compared with those vaccinated with the monovalent vaccines (Fig. 6 and SI Appendix, Table S10). The fold-change in neutralizing antibody titers was measured before and 2 and 4 wk after the booster injection. The response against the BA.2.86 variant was comparable between the NHPs that received the trivalent vaccine and those that received the monovalent vaccines. However, the responses against the Wuhan-1, BA.5, BQ.1.1, BN.1, XBB.1.5, and EG.5.1 variants were higher in NHPs vaccinated with the trivalent vaccine. These results suggest that trivalent vaccine could elicit a broad neutralizing antibody response against currently prevalent new variants even when it is delivered as a booster injection.

Fig. 6.

Booster shot evaluation of the monovalent and trivalent vaccines in macaques. (A) Study schedule. (B–D) Cynomolgus macaques were vaccinated with 5 × 10¹⁰ VP of the bivalent vaccine AdCLD-CoV19-1 Wuhan-1/BA.5, and after 16 wk, reimmunized with a booster injection of trivalent vaccine AdCLD-CoV19-1 XBB.1.5/BN.1/BQ.1.1 or an equivalent dose of monovalent vaccine AdCLD-CoV19-1 XBB.1.5 or AdCLD-CoV19-1 BQ.1.1. Serum samples were collected at indicated time points and the neutralizing activities in the collected sera were measured using pseudotyped lentivirus. Geometric mean neutralization titers (GMT) are listed for each variant. See also SI Appendix, Table S10.

Discussion

Adenoviral vector-based platforms have been used as vaccine strategies against infectious diseases, such as influenza, Zika, AIDS, Ebola, malaria, and COVID-19 (14–19). Adenoviral vector-based vaccines have been studied mainly on Ad5, but much of the human population has already been exposed to Ad5 and has anti-Ad5 antibodies (20). Barouch et al. reported frequent observation of high Ad5-neutralizing antibody titers in sub-Saharan Africa and Southeast Asia. In contrast, Ad35-neutralizing antibody titers were infrequent and low in all the regions they studied. Furthermore, Ad5-neutralizing antibodies are known to impair the immune response to Ad5 vector-based vaccines (21). To circumvent this problem, we developed a vaccine using a chimeric adenovirus 5/35 platform in which the fiber knobs, the receptor-interacting domain of the fiber protein, were substituted with Ad35 (11).

To determine the optimal combination of multivalent vaccines, we used a phylogenetic tree and antigenic cartography, which has been used extensively for clustering influenza viruses (22). Because hemagglutinin, which constitutes the glycoprotein expressed on the surface of influenza viruses, undergoes persistent and rapid evolution, the World Health Organization (WHO) yearly identifies the most suitable combination of seasonal influenza vaccines against new emerging variants. The WHO Collaborating Centres for Influenza routinely generate an antigenic map for the selection of influenza vaccine strains using hemagglutination inhibition (HI) assays (23, 24). In this study, by using antigenic maps based on currently circulating Omicron subvariants, we developed the bivalent vaccine AdCLD-CoV19-1 BA.5/BA.2.75 and the trivalent vaccines AdCLD-CoV19-1 XBB/BN.1/BQ.1.1 and AdCLD-CoV19-1 XBB.1.5/BN.1/BQ.1.1 using an Ad5/35 platform-based non-replicating recombinant adenoviral vector. It should be noted that the antigenic mapping of our study has several limitations. First, as the results were based on a limited number of samples, increasing the dataset size is necessary to improve the accuracy of the antigenic cartography. Second, antigenic cartography only focuses on neutralizing antibody responses and does not consider other components of the immune system, such as T cell responses, which play a crucial role in providing protection against viral infections. In the primary single-dose immunization model using naive BALB/c mice, we observed that the monovalent vaccine may have the potential to elicit a more focused neutralizing antibody response against homologous viruses than the trivalent vaccine in some cases. However, in the booster model, we showed that the multivalent vaccine not only extends the neutralization breadth but also increases each variant-specific response compared to the monovalent vaccine against most of the variants we tested. This is presumed to result from the combined effect of the primary and memory immune responses. The administration of multivalent vaccines could carry multiple antigens to antigen-presenting cells, thereby might lead to the presentation of different kinds of T cell epitopes to T cells, activating T cells with diverse repertoires. Considering the pivotal role of CD4⁺ T cells in B cell-mediated humoral immune responses, the activation of CD4⁺ T cells with different repertoires through the administration of a multivalent vaccine might contribute to diverse and robust antibody responses in monkeys upon booster doses. However, additional research is required to prove this assumption. Third, although this study showed that the multivalent vaccine, a combination using an antigenic map generated from the results of a single vaccine administration model, elicited a broader neutralization activity than the monovalent vaccine even when used as a booster injection, the cross-neutralization potential may vary in humans based on individual infection and vaccination histories. Despite these limitations, antigenic cartography is a powerful analytical method for selecting variants for use in vaccines by visually representing how viruses are antigenically related and for quantifying antigenic distances (25).

We found that even a single dose of our multivalent vaccines could efficiently produce broadly neutralizing antibodies against most of the variants we tested and reduced antigenic distances compared to the monovalent vaccines. Consistent with the benefits associated with the development of multivalent vaccines for influenza virus, poliovirus, human papillomavirus, pneumococcus, and SARS-CoV-2, our findings also showed that the multivalent vaccine strategy is powerful for a significant reduction in the spread of the disease, the severity of infections, and the overall public health burden associated with COVID-19 (26). Similarly, the high levels of genetic diversity in HIV present challenges for the development of a broadly applicable HIV vaccine (27, 28). However, since the genetic variation of SARS-CoV-2 is not more diverse than that of HIV, it is expected that multivalent vaccines can induce a sufficiently wide range of neutralizing antibodies.

Throughout the history of COVID-19 vaccine development, it has been observed that additional mutations increase immune evasion, leading to breakthrough infections in individuals previously infected or vaccinated (8, 10, 29). To overcome the loss in efficacy of vaccines encoding the Wuhan-1 spike protein against Omicron strains, bivalent vaccines containing Omicron variant–matched spikes were designed and tested. Scheaffer et al. showed that the Moderna bivalent vaccines, mRNA-1273.214 (encoding Wuhan-1 and BA.1 spike proteins) and mRNA-1273.222 (encoding the Wuhan-1 and BA.4/5 spike proteins), elicited stronger neutralizing antibody responses against BA.5 Omicron variants than the monovalent mRNA-1273 (encoding the Wuhan-1 spike protein) (6). However, these bivalent vaccines and Pfizer’s BA.5 bivalent booster did not generate robust neutralization against newly emerged Omicron sub-lineages, such as BA.2.75.2, BQ.1.1, XBB, and XBB.1 (9, 10). In both studies, XBB.1 showed the greatest evasion to vaccine-induced neutralization, and the XBB.1 family eventually outcompeted the dominant lineage BA.5 and became the prevailing variant in 2023. Therefore, the COVID-19 vaccine was updated to a monovalent XBB.1.5 spike-based vaccine beginning in the fall of 2023. Chalkias et al. showed that administering the mRNA-1273.815 (XBB.1.5 monovalent) vaccine as a booster elicited neutralizing antibodies against various SARS-CoV-2 variants, including XBB.1.5, XBB.1.16, EG.5.1, and BA.2.86 (30). However, JN.1, which emerged in late 2023 and recently became one of the dominant lineages worldwide, exhibited resistance to monovalent XBB.1.5 vaccinated sera compared with BA.2.86 (31). Likewise, while the trivalent vaccine combination we developed herein effectively neutralizes the variants we tested, including EG.5.1 and BA.2.86, it remains to be determined whether it can sufficiently neutralize emerging variants. When new variants that reduce the immunogenicity of the currently licensed vaccines emerge, changes in vaccine composition and, if needed, the development of multivalent vaccines would be required.

Although the WHO declared the end of COVID-19 as a global health emergency on 5 May 2023, COVID-19 continues to cause substantial morbidity and mortality worldwide, and the WHO recommended continuing research to develop vaccines that reduce transmission and have broad applicability. To successfully create updated COVID-19 vaccines against new emerging variants, several considerations should be made.

First, it is necessary to anticipate which variant will be dominant several months in the future. As viruses continue to evolve, new COVID-19 vaccines require updates tailored to the circulating variants. Because vaccine development and manufacturing take time, WHO decides which variants to include in the seasonal influenza vaccines at least six months in advance of the next influenza season (32). COVID-19 vaccines also require a strategy to select variants to be included in the multivalent vaccines, by predicting the growth rates and prevalence of different variants, like the seasonal influenza vaccine; however, a lack of established seasonal transmission patterns for COVID-19 makes this task more challenging. Since the effectiveness of the vaccine differs depending on the accuracy of the prediction, as in the result of the seasonal influenza vaccine effectiveness, rapid and accurate prediction is essential (33, 34).

Second, it is necessary to consider developing locally tailored vaccines by using the regional evolution of dominant variants and their prevalence data. SARS-CoV-2 variant patterns vary in different geographic locations, and although there are exceptions, variant patterns in each country tend to follow the general trends of their continent (10, 35, 36). Therefore, considering not only the time horizon but also the variant pattern of each location is essential for accurate forecasting.

Third, previous infection history and vaccination status, such as vaccine type and dose, should be considered. As most of the population has been previously vaccinated or infected, multivalent vaccines are expected to be used mainly as boosters. Since we used a single administration model to determine the optimal vaccine combination in the short term, and although the multivalent vaccine as a booster could elicit broader neutralization activity than the monovalent vaccine based on results seen with the macaque booster model in this study, further studies for identifying the long-term effects of multivalent vaccine combinations in each immunological memory established models are also needed. Studies on influenza vaccines have shown that antibody levels against newly circulating influenza viruses are generally lower than those against strains that have previously circulated in an individual’s lifetime (23, 37). This back-boosting phenomenon can occur due to immune imprinting by the immune memory between related viruses with antigenic similarity. In other words, different immune repertoires may drive differences in the immune responses to subsequent variant-specific vaccine boosters. Although the short-term boosting effect is likely to be biased toward conserved epitopes, if diverse epitopes are continuously included in new vaccines, memory cells for diverse epitopes can be established and increase the potency against newly emerging variants. In addition, Fonville et al. showed that vaccination responses were greater against recent antigen clusters following vaccination with the most recently encountered antigen strain (23). These results indicate that the development of preemptive vaccines can improve vaccine efficacy as a booster.

The risk of severe outcomes increases with multiple episodes of COVID-19, and therefore, reinfection prevention strategies are needed to reduce the risk of death and disease from COVID-19. It is expected that the COVID-19 vaccine will be used and updated regularly like the influenza vaccine. Determining future booster combinations before new variants become widespread will be a key challenge to combat COVID-19.

Materials and methods

Cells.

Human embryonic kidney 293 cells expressing the tetracycline repressor (HEK293R) were used for the adenovirus-based vaccine production. HEK293T cells were used for the pseudotyped lentivirus production. HEK293T cells expressing human ACE2 (HEK293T-hACE2) were used for the neutralization assay. The cells were maintained in Dulbecco’s modified Eagle’s medium (HyClone, Cytiva, Marlborough, MA, USA) supplemented with 10% fetal bovine serum (HyClone, Cytiva, Marlborough, MA, USA) and penicillin (100 U/mL)-streptomycin (100 μg/mL) (HyClone, Cytiva, Marlborough, MA, USA) at 37 °C in 5% CO₂.

Adenovirus Vector Design and Vaccine Production.

All replication-incompetent recombinant adenovirus vectors used in this study were human adenovirus serotype 5 with E1 and E3 gene deletion and replacement of the fiber gene with a knob of human adenovirus serotype 35. The E4orf6 gene was rearranged to the E1 region to minimize the incidence of replication-competent adenovirus. All the vaccines used in this study were prepared by methods previously described (11). Vaccines against each Omicron sub-lineage were constructed by substitutions of the mutated gene of each spike protein of the relevant variants. The substitutions are listed in SI Appendix, Fig. S10. The adenovirus-vectored products used in this study were i) monovalent AdCLD-CoV19-1 BA.2.75 vaccine; ii) monovalent AdCLD-CoV19-1 BA.5 vaccine; iii) bivalent AdCLD-CoV19-1 BA.5/BA.2.75 vaccine, which is 1:1 in a vial of separately formulated AdCLD-CoV19-1 BA.2.75 and AdCLD-CoV19-1 BA.5 vaccines; iv) monovalent AdCLD-CoV19-1 XBB vaccine; v) monovalent AdCLD-CoV19-1 BN.1 vaccine; vi) monovalent AdCLD-CoV19-1 BQ.1.1 vaccine; vii) trivalent AdCLD-CoV19-1 XBB/BN.1/BQ.1.1 vaccine, which is 1:1:1 in a vial of separately formulated AdCLD-CoV19-1 XBB, AdCLD-CoV19-1 BN.1, and AdCLD-CoV19-1 BQ.1.1 vaccines; viii) bivalent AdCLD-CoV19-1 Wuhan-1/BA.5 vaccine, which is 1:1 in a vial of separately formulated AdCLD-CoV19-1 Wuhan-1 and AdCLD-CoV19-1 BA.5 vaccines; and ix) trivalent AdCLD-CoV19-1 XBB.1.5/BN.1/BQ.1.1 vaccine, which is 1:1:1 in a vial of separately formulated AdCLD-CoV19-1 XBB.1.5, AdCLD-CoV19-1 BN.1, and AdCLD-CoV19-1 BQ.1.1 vaccines.

Animal Experiments.

All the animal experiments were approved by the institutional animal care and use committee (IACUC) of Seoul National University (SNU-210409-5-1) and Korea Research Institute of Bioscience and Biotechnology (KRIBB) (KRIBB-AEC-23140) and were performed in accordance with the guidelines of the IACUC.

To examine the immunogenicity of vaccines in mice, 8-wk-old BALB/c mice were immunized intramuscularly once with 1 × 10⁹ VP of each vaccine. Blood samples were collected 2, 4, 6, 8, and 12 wk after administration. To examine the immunogenicity of vaccines in non-human primates, 4- to 9-y-old cynomolgus macaques (N = 9) were immunized intramuscularly with 5 × 10¹⁰ VP of bivalent vaccine AdCLD-CoV19-1 Wuhan-1/BA.5 as their primary vaccination. They were then divided into three groups (N = 3 for each group) based on their immunogenicity profiles that were examined at 12 wk after the primary vaccination. Macaques in each group received a booster vaccination with 5 × 10¹⁰ VP of any one of the monovalent vaccines AdCLD-CoV19-1 XBB.1.5 or AdCLD-CoV19-1 BQ.1.1 or 5 × 10¹⁰ VP of the trivalent vaccine AdCLD-CoV19-1 XBB.1.5/BN.1/BQ.1.1. Blood samples were collected before the booster administration and at 2 and 4 wk after booster. Sera were isolated from the blood samples, heat-inactivated for 30 min at 56 °C, and stored at −80 °C before use.

Pseudotyped Lentivirus and Neutralization Assay.

Neutralization against the pseudotyped lentivirus expressing SARS-CoV-2 S protein was performed using a luciferase assay. The lentivirus VSV glycoprotein gene was replaced with the plasmid expressing SARS-CoV-2 S protein; the lentiviral vector backbone carries the reporter genes for firefly luciferase and green fluorescent protein. The pseudovirus was produced by co-transducing the plasmid expressing SARS-CoV-2 S protein, the lentiviral vector backbone, and the lentiviral vector packaging plasmid into HEK293T cells. The sera from the vaccinated mice were diluted four times in a row from 50-fold to 3,200-fold and two times in a row from 6,400-fold to 12,800-fold and incubated with the pseudotyped lentivirus for 1 h at 37 °C. After 1 h, the mixture was added to HEK293T-hACE2 cells expressing human ACE2. At 72 h after the infection, luciferase activity in the cell lysates was measured using a luciferase assay kit (Promega, Madison, WI, USA) and a luminometer (Centro Tristar 3; Berthold Technologies, Baden-Württemberg, Germany). The neutralization abilities of the sera were calculated as 50% pseudovirus neutralization titer (pVNT50), which is the dilution factor in which the luciferase activity was reduced to 50% of that from the virus-only wells or dilution end-point wells.

Pseudotyped VSV and Neutralization Assay.

A neutralization assay against pseudotyped VSV expressing SARS-CoV-2 S protein was performed using a luciferase assay. Jae-Ouk Kim [International vaccine institute (IVI)] produced and supplied the VSV pseudovirus. The sera from the vaccinated mice were diluted four times in a row from 50-fold to 3,200-fold and two times in a row from 6,400-fold to 12,800-fold and incubated with the pseudotyped VSV for 1 h at 37 °C. After 1 h, the mixture was added to HEK293T-hACE2 cells expressing human ACE2. At 48 h after the infection, luciferase activity in the cell lysates was measured using a luciferase assay kit (Promega, Madison, WI, USA) and a luminometer (Centro Tristar 3; Berthold Technologies, Baden-Württemberg, Germany). The neutralization abilities of the sera were calculated as pVNT50, which is the dilution factor in which the luciferase activity was reduced to 50% of that from the virus-only wells or dilution end-point wells.

Antigenic Cartography.

Antigenic maps were constructed as previously described using the Racmacs package (https://acorg.github.io/Racmacs/, number_of_optimization = 1,000) (23, 38–40). This approach to antigenic mapping uses neutralization datasets and multidimensional scaling to position antigens (pseudoviruses) and sera in a map to represent their antigenic relationships. A grid (each square) indicates one antigenic unit, corresponding to a two-fold dilution of the antibody in the neutralization assay. Antigenic distance is measured in any direction on the map. We performed various quality assessments for the evaluation of the goodness of antigenic map fit and dimensionality (SI Appendix, Fig. S7). Maps were constructed in 1, 2, 3, 4, and 5 dimensions to investigate the dimensionality of the antigenic relationships (SI Appendix, Fig. S7 G–L). All the datasets showed good fit with only small improvements in the residual mean squared error of the maps even in two dimensions.

Phylogenetic Tree Construction.

The distance-based phylogenetic tree based on sequences of the S protein of the SARS-CoV-2 strains was constructed to assess the genetic diversity (SI Appendix, Fig. S10). Individual S protein consensus sequences were constructed using multiple sequence alignment via MEGA 11 and iTOL (https://itol.embl.de/). The phylogenetic tree was created using the maximum likelihood method based on the Jones–Taylor–Thornton (JTT) matrix-based model with 100 bootstrap replicates (41, 42).

Supplementary Material

Appendix 01 (PDF)

Data, Materials, and Software Availability

All study data are included in the article and/or SI Appendix.