We next evaluated the impact of the different boost immunizations on Ab responses against the RBD, the primary target of neutralizing Abs³. The initial Wu1 S mRNA-LNP prime-boost vaccinations elicited robust and similar Ab responses against the Wu1 RBD for all cohorts on day 56 (Fig. 2A). The BA.5, BQ.1.1, and XBB.1 variants harbor multiple RBD substitutions, particularly within the class 1 and 2 epitopes at the tip of the RBD³³ (Fig. S2), promoting effective escape from neutralizing Abs¹⁸. Compared to Wu1 RBD binding, Ab binding titers were weaker against the BA.5, BQ.1.1, and XBB.1 RBDs on day 56 (Fig. 2B-D). The robust Ab binding against the BA.5, BQ.1.1, and XBB.1 S trimers (Fig. 1E-G) in the absence of strong RBD responses was at least in part driven by potent responses against the S2 region of S (Fig. S3), which is relatively conserved in SARS-CoV-2 variants³⁴ (Fig. S2). Similar to binding titers against S-6P trimers (Fig. 1D-G), Wu1, BA.5, BQ.1.1, and XBB.1 RBD binding slightly dropped on day 154 without statistically significant differences between cohorts (Fig. 2A-D). The booster immunizations increased Wu1 RBD binding on day 244 compared to day 56 and 154 levels, but no differences were detected between the four boosting immunogens (Fig. 2A). In contrast, the bivalent Wu1/BA.5 S-EABR mRNA-LNP booster elicited the highest BA.5, BQ.1.1, and XBB.1 RBD responses on day 244 (Fig. 2B-D), which were significantly higher than Wu1 S mRNA-LNP booster-induced responses against BQ.1.1 RBD (p = 0.0499) (Fig. 2C) and narrowly failed to reach statistical significance compared to Wu1 S mRNA-LNP booster-induced responses against BA.5 RBD (p = 0.07) (Fig. 2B). Compared to baseline titers measured prior to boost immunizations on day 154, the bivalent Wu1/BA.5 S-EABR mRNA-LNP immunization elicited significantly larger increases in BA.5 RBD binding than the monovalent Wu1 S mRNA-LNP (p = 0.02) and the bivalent Wu1/BA.5 S mRNA-LNP boosters (p = 0.01) (Fig. 2B), as well as significantly larger increases in BQ.1.1 RBD binding than the Wu1 S mRNA-LNP booster (p = 0.002) (Fig. 2C). While overall titers were similar on day 244, the monovalent Wu1 S-EABR mRNA-LNP booster consistently induced larger increases in Ab binding to the BA.5, BQ.1.1, and XBB.1 RBDs than the monovalent Wu1 S mRNA-LNP booster (Fig. 2B-D), although these differences failed to reach statistical significance. The monovalent Wu1 and bivalent Wu1/BA.5 S mRNA-LNP boosters elicited similar binding titers against the BA.5 and XBB.1 RBDs (Fig. 2B and D), but the bivalent booster induced slightly higher responses and a larger increase compared to day 154 titers against BQ.1.1 RBD (not statistically significant) (Fig. 2C). Overall, these results demonstrate that booster immunizations with the bivalent Wu1/BA.5 S-EABR mRNA-LNP elicited the highest Ab responses against Omicron RBDs in pre-vaccinated mice. Notably, post-boost Ab binding against Omicron RBDs remained lower than binding titers against Wu1 RBD in all cohorts, suggesting that booster responses were affected by immune imprinting induced by the primary vaccination series.

The ability of boost immunogens to elicit neutralizing Ab responses in pre-vaccinated mice was evaluated by in vitro pseudovirus neutralization assays^35,36. The primary vaccination series with Wu1 S mRNA-LNP elicited potent and consistent autologous geometric mean neutralizing titers as assessed by half-maximal serum inhibitory dilution (ID₅₀) values against Wu1 pseudovirus in all cohorts on day 56 (Fig. 3A). Neutralizing activity was >100-fold lower against the BA.5 variant and undetectable in most animals against the BQ.1.1 and XBB.1 variants on day 56 for all cohorts (Fig. 3B-D). On day 154, neutralizing titers against the Wu1 and BA.5 variants were consistent in all cohorts and slightly lower compared to day 56 titers (Fig. 3A-B). Neutralizing activity against the BQ.1.1 and XBB.1 variants was undetectable in all but one animal on day 154 (Fig. 3C-D). Booster immunizations on day 230 increased autologous neutralizing titers against Wu1 pseudovirus on day 244 compared to day 154 in all cohorts (Fig. 3A). Monovalent Wu1 S mRNA-LNP and S-EABR mRNA-LNP immunizations elicited the smallest and largest increases in Wu1 neutralization (4.2-fold versus 6.6-fold), respectively. All boost immunogens elicited increased BA.5 neutralizing activity on day 244 compared to day 56 and 154 titers (Fig. 3B). Boosting with monovalent Wu1 S-EABR mRNA-LNP induced 4.8-fold higher BA.5 neutralizing titers than a third Wu1 S mRNA-LNP immunization, but this difference did not reach statistical significance. Using the respective day 154 titers as baseline, the Wu1 S-EABR mRNA-LNP booster induced a 23.8-fold increase in BA.5 neutralization on day 244 compared to only a 6.3-fold increase detected after boosting with Wu1 S mRNA-LNP (not statistically significant) (Fig. 3B). Compared to day 56 baseline titers, the Wu1 S-EABR mRNA-LNP booster induced a significantly higher increase in BA.5 neutralization on day 244 than the conventional Wu1 S mRNA-LNP booster (16.2-fold versus 3.1-fold; p = 0.048) (Fig. S4). The Wu1 S-EABR mRNA-LNP booster also elicited 3.4-fold higher neutralizing responses than the Wu1 S mRNA-LNP booster against the Omicron BA.1 variant on day 244 (p = 0.03) (Fig. S5). Based on day 154 titers, boosting with monovalent Wu1 S-EABR mRNA-LNP induced significantly higher increases in BA.1 neutralization compared to monovalent Wu1 S mRNA-LNP (8.8-fold versus 2.5-fold; p = 0.009) (Fig. S5). Interestingly, the monovalent Wu1 S-EABR mRNA-LNP and the bivalent Wu1/BA.5 S mRNA-LNP boosters elicited similar BA.5 neutralizing responses on day 244 (Fig. 3B), suggesting that the EABR mRNA vaccine approach might reduce the need for updating boost immunogens to achieve robust Ab responses against emerging SARS-CoV-2 variants. The highest BA.5 neutralizing titers were elicited by the bivalent Wu1/BA.5 S-EABR mRNA-LNP booster, which were 12.3-fold higher than titers induced by the monovalent Wu1 S mRNA-LNP booster (p = 0.02) and 2.4-fold higher than titers induced by the bivalent Wu1/BA.5 S mRNA-LNP booster (not statistically significant). Based on titers measured on day 154, the Wu1/BA.5 S-EABR mRNA-LNP booster induced a 67-fold increase in BA.5 neutralization, which was significantly higher than the increase elicited by boosting with monovalent Wu1 S mRNA-LNP (p = 0.003).

While the BA.5 neutralizing activity induced by the monovalent Wu1 S and S-EABR mRNA-LNP boosters can likely be attributed to cross-neutralizing Abs that neutralize both the Wu1 and BA.5 variants, it is possible that the bivalent Wu1/BA.5 S and S-EABR mRNA-LNP boosters also elicited de novo strain-specific Abs that neutralize BA.5 but not Wu1. Day 244 serum samples from mice that received bivalent Wu1/BA.5 S or S-EABR mRNA-LNP boosters were depleted of Abs that bind to Wu1 RBD. Sera were passaged over a column presenting either a mock protein (HIV-1 gp120) as a control or the Wu1 RBD to remove cross-neutralizing anti-RBD Abs while retaining BA.5 RBD-specific Abs. Following depletions, residual neutralization activity was measured against the BA.5 variant. Interestingly, Wu1 RBD depletion resulted in 17- and 36-fold lower geometric mean neutralizing titers compared to mock depletion for the bivalent Wu1/BA.5 S and S-EABR mRNA-LNP boosters, respectively (Fig. S6). In both cohorts, BA.5 neutralizing titers in all but two mice dropped below ID₅₀ values of 1:120 following Wu1 RBD depletion, consistent with the neutralizing activity in these samples being primarily driven by cross-neutralizing Abs targeting the RBD induced by the primary vaccination series. The potent residual neutralizing activity observed in two mice in each cohort after Wu1 RBD depletion implies either the presence of boost-induced de novo BA.5-specific RBD Abs or neutralizing non-RBD Abs that target another region of S, e.g., the N-terminal domain³⁷ or S2³⁴.

Booster immunizations had less pronounced effects on neutralization activity against the BQ.1.1 and XBB.1 variants. Against BQ.1.1 pseudovirus, the bivalent Wu1/BA.5 S-EABR mRNA-LNP booster elicited the most potent neutralizing activity on day 244, which was significantly higher than responses induced by a third immunization with the monovalent Wu1 S mRNA-LNP booster (p = 0.0495) (Fig. 3C). The monovalent Wu1 S-EABR mRNA-LNP and bivalent Wu1/BA.5 S mRNA-LNP boosters elicited 1.8- and 2.5-fold higher BQ.1.1 titers than the monovalent Wu1 S mRNA-LNP booster, respectively, the latter of which narrowly failed to reach statistical significance (p = 0.07). Based on day 154 BQ.1.1 neutralizing titers, the monovalent Wu1 S mRNA-LNP and bivalent Wu1/BA.5 S-EABR mRNA-LNP boosters induced the smallest and largest increases in neutralizing activity on day 244, respectively (1.4-fold versus 4.7-fold; p = 0.0495). The monovalent Wu1 S-EABR mRNA-LNP and bivalent Wu1/BA.5 S mRNA-LNP boosters elicited 2.4- and 3.4-fold increases in neutralizing BQ.1.1 titers, respectively. Notably, day 244 neutralizing titers against BQ.1.1 were substantially lower than BA.5 neutralizing titers for all boost immunogens (Fig. 3B-C), consistent with previous reports^18,38. Notably, Ab binding responses against the BA.5 and BQ.1.1 RBDs were similar (Fig. 2B-C), implying that RBD binding responses are not always predictive of neutralizing activity against Omicron subvariants.

Neutralizing responses against the XBB.1 variant were slightly lower than titers against BQ.1.1 pseudovirus (Fig. 3D). Once again, the bivalent Wu1/BA.5 S-EABR mRNA-LNP booster elicited the highest XBB.1 neutralizing titers, which were 2.1-fold higher than titers induced by the monovalent Wu1 S mRNA-LNP booster (not statistically significant). Based on day 154 titers, the monovalent Wu1 and bivalent Wu1/BA.5 S-EABR mRNA-LNP boosters increased XBB.1 neutralization by 2.5- and 3.2-fold, respectively, while conventional monovalent and bivalent S mRNA-LNP boosters only induced 1.5- and 2.2- fold increases, respectively.

Notable differences were also observed in the number of mice in each cohort that reached robust neutralizing titers (defined as ID₅₀s exceeding 1:100) against Omicron subvariants following booster immunizations, titers that are likely important for the prevention of symptomatic infections³⁹. The primary vaccination series with Wu1 S mRNA-LNP elicited robust neutralizing titers against the Wu1 variant in all mice on day 56 (Fig. 3F). However, the initial Wu1 S mRNA-LNP vaccinations elicited robust titers in only 6 to 8, 1 to 3, and 0 to 2 of 10 mice against the BA.5, BQ.1.1, and XBB.1 variants on day 56, respectively. On day 154, ID₅₀s over 1:100 were measured for 3 to 5 mice in each cohort against BA.5, but for none of the animals against BQ.1.1 and XBB.1. While a third immunization with Wu1 S mRNA-LNP induced robust titers against BA.5 in only 8 of 10 mice on day 244, the monovalent Wu1 S-EABR mRNA-LNP booster induced robust titers against BA.5 in all 10 mice, including the 4 mice that did not have detectable BA.5 neutralizing activity on day 56 (Fig. 3F). As expected, robust neutralizing titers against the BA.5 variant were also observed in all mice that received the bivalent Wu1/BA.5 S or S-EABR mRNA-LNP boosters. Against the BQ.1.1 and XBB.1 variants, a third immunization with Wu1 S mRNA-LNP induced robust neutralizing titers in only 2 and 3 mice, respectively (Fig. 3F). In comparison, the bivalent Wu1/BA.5 S-EABR mRNA-LNP booster induced ID₅₀s over 1:100 in 7 and 6 mice against the BQ.1.1 and XBB.1 variants, respectively.

Overall, these results demonstrate that the EABR mRNA vaccine approach improves the neutralizing activity against Omicron subvariants induced by monovalent and bivalent boost immunogens in pre-vaccinated mice.

Several distinct Ab-binding epitopes have been defined on RBDs, including the variable class 1 and 2 RBD epitopes and the more conserved class 3, 4, 1/4, and 5 RBD epitopes^33,40,41 (Fig. S2). We determined which RBD epitopes were targeted by booster-induced Abs in pre-vaccinated mice using deep mutational scanning (DMS) of yeast display libraries derived from Wu1⁴² and XBB.1.5 RBDs⁴³. In previous studies, we showed that equimolar mixtures of monoclonal Abs targeting distinct RBD epitopes produced less pronounced DMS escape peaks, which we defined as a polyclass Ab response⁴⁴. This suggests that polyclonal serum that includes roughly equal distributions of Abs targeting multiple classes of RBD epitopes would exhibit relatively flat DMS profiles, while more defined DMS escape peaks would be observed for less diverse polyclonal serum responses containing Abs that primarily target particular RBD epitopes⁴⁴. For this study, we used DMS escape fractions to define polyclass responses. Escape fractions for each RBD mutation range from 0 (no cells with this mutation are sorted into the serum Ab escape bin) to 1 (all cells with this mutation are sorted into the serum Ab escape bin)⁴⁵. Total escape peaks represent the sum of escape fractions for all mutations at a specific site⁴⁵. DMS profiles with total escape peaks below 0.5 at all sites were classified as polyclass responses. In contrast, DMS profiles with total escape peaks of 0.5-1, 1-2, or above 2 at one or more sites were classified as weak, moderate, or strong escape profiles, respectively, indicating weak, moderate, or strong targeting of particular RBD epitopes (Table S1). DMS was performed for 6 serum samples for each cohort, which were selected based on XBB.1 neutralization activity (1^st, 2^nd, 3^rd, 5^th, 6^th, and 10^th best neutralizer from each cohort). For the Wu1 RBD library, only 5 DMS profiles were obtained due to poor sequencing quality for one sample in each cohort.

Against the Wu1 RBD library, immunizations with the monovalent Wu1 S and S-EABR mRNA-LNP boosters or the bivalent Wu1/BA.5 S mRNA-LNP booster resulted in DMS profiles consistent with moderate targeting of the variable class 2 epitope primarily inducing escape at RBD residue 484 (Fig. 4A-B). The bivalent Wu1/BA.5 S mRNA-LNP booster also elicited weak to moderate responses that mapped to the variable class 1 and class 3 RBD epitopes. In contrast, the bivalent Wu1/BA.5 S-EABR mRNA-LNP booster DMS profiles were consistent with polyclass Ab induction, suggesting a more diverse response targeting multiple RBD epitopes (Fig. 4A-B). Notably, samples from 4 out of 5 mice in this cohort exhibited polyclass DMS profiles, whereas moderate to strong escape fractions for class 1, 2, or 3 epitopes were observed in 3-4 mice in each of the other cohorts (Fig. S7; Table S1).

As expected, no class 1 and 2 DMS profiles were observed against the XBB.1.5 RBD library⁴³ since these epitopes are highly substituted in Omicron subvariants (Fig. S2). All booster immunizations induced strong class 5 escape peaks targeting RBD residues 352, 357, 396, and 468 (Fig. 4C-D). Importantly, the bivalent Wu1/BA.5 S-EABR mRNA-LNP booster induced a less pronounced overall class 5 profile than the other boost immunogens, which was primarily driven by a strong class 5 response observed in a single animal in this cohort (Fig. S8; Table S1). The other mice that received the bivalent Wu1/BA.5 S-EABR mRNA-LNP booster exhibited polyclass or weak to moderate responses against class 4 and 5 epitopes. In contrast, strong class 5 profiles were detected in 4-5 mice in cohorts that received the monovalent Wu1 S or S-EABR mRNA-LNP boosters or the bivalent Wu1/BA.5 S mRNA-LNP booster (Fig. S8; Table S1). Overall, these results suggest that the bivalent Wu1/BA.5 S-EABR mRNA-LNP booster induced a more diversified polyclass Ab response that likely contributed to the higher neutralizing activity observed against Omicron subvariants, which could result in more effective recognition of future variants.

The results from this and previous studies^19–22 demonstrated that bivalent booster immunizations improve the potency and breadth of Ab responses against Omicron subvariants compared to monovalent boosters. Serum depletion assays also showed that the BA.5 neutralizing activity elicited by the bivalent Wu1/BA.5 S and S-EABR mRNA-LNP boosters was primarily driven by cross-reactive Abs that bind to the Wu1 and BA.5 RBDs (Fig. S6). Preferential activation of cross-reactive B cells could be achieved through co-display of Wu1 and BA.5 S trimers on the surface of a cell and/or an eVLP (Fig. 1C), which could result in inter-S crosslinking between cross-reactive B cell receptors (BCRs) and neighboring homotrimeric Wu1 and BA.5 S trimers (Fig. S9). As an additional mechanism for bivalent booster-induced activation of cross-reactive B cells, we hypothesized that co-expression of ancestral Wu1 and Omicron S in the same cell could result in the formation of S heterotrimers consisting of Wu1 and Omicron S protomers and could promote intra-S crosslinking, i.e., simultaneous binding of cross-reactive BCRs to two neighboring RBDs within the same S trimer (Fig. S9). We previously postulated that intra-S crosslinking is possible for several anti-RBD IgGs based on modeling of intact IgG coordinates into single-particle cryo-electron microcopy (cryo-EM) structures of fragment antigen-binding (Fab)-S complexes^33,40, and intra-S crosslinking was observed in EM studies of IgG-S trimer complexes⁴⁶. However, it is currently unknown whether co-expression of Wu1 and Omicron Ss induces heterotrimerization, and if so, whether heterotrimer formation affects the overall structure of the S ectodomain and influences bivalent booster-induced immune responses.

To investigate whether S heterotrimers could form with different S protomers, we designed 6P versions³² of Wu1/BA.1 and Wu1/BA.1/Delta soluble heterotrimer Ss (HT2 and HT3, respectively) by co-transfection with specific C-terminally-tagged constructs. HT2 was generated by co-transfecting StrepII-tagged Wu1 S and His-tagged BA.1 S, while HT3 was generated using StrepII-tagged Wu1 S, His-tagged BA.1 S, and D7324⁴⁷-tagged Delta S (Fig. 5A). Each heterotrimer was subsequently purified using sequential affinity purification specific to each tag, confirming the inclusion of distinct protomers in the purified heterotrimers (Fig. 5A).

HT2 and HT3 S proteins were then evaluated using single-particle cryo-electron microscopy (cryo-EM) to determine whether heterotrimeric Ss resembling their homotrimeric counterparts were formed. To ensure that C-terminal tags did not disrupt the formation of trimers, we included the following controls: (1) three homotrimers with single tags (StrepII-tagged Wu1 S, His-tagged BA.1 S, and D7324-tagged Delta S), and (2) a homotrimer comprised of His- and StrepII-tagged protomers (Wu1-His/Wu1-StrepII).

Single-particle cryo-EM analysis confirmed that all control S proteins formed trimers, indicating that C-terminal tags did not disrupt trimerization (Fig. S10). Because most of the EM constructions for homotrimeric S controls exhibited preferred orientations, we determined overall conformational states (Fig. 5) for the control structures but did not fit atomic models. Structural analysis also revealed trimerized HT2 and HT3 Ss (Fig. 5B-C, Fig. S10). Although atomic models could not be built for the HT2 and HT3 trimer reconstructions because the identities of individual protomers (Wu1 or BA.1 for HT2, and Wu1, BA.1, or Delta for HT3) could not be conclusively determined at relatively low resolution, the EM densities of S heterotrimers and control homotrimers showed no major conformational differences in domain organization (Fig. 5B-C, Fig. S10). As previously seen in SARS-CoV-2 S trimer structures^48,49, we observed two trimer populations in all datasets—one with all three RBDs in the ‘down’ conformation and one with a single RBD in an ‘up’ conformation (Fig. 5D). These data demonstrate heterotrimeric S formation for soluble forms of SARS-CoV-2 S proteins. This result suggests that the bivalent S or S-EABR mRNA-LNP boost immunogens likely induce heterotrimerization, which could promote activation of cross-reactive B cells through intra-S cross-linking, a potential mechanism warranting further investigation.

SARS-CoV-2 S-EABR constructs were designed as described³¹. The EABR domain (residues 160-217) of the human CEP55 protein was fused to the C-terminus of the SARS-CoV-2 S protein obtained from the Wu1 or BA.5 variants separated by a 4-residue (Gly)₃Ser (GS) linker (Fig. S1A). These constructs contained the native furin cleavage site, 2P stabilizing mutations⁵⁵, and the C-terminal 21 residues were truncated to remove an endoplasmic reticulum retention signal⁵⁶. The D614G substitution⁵⁷, naturally present in BA.5, was inserted into Wu1 S and S-EABR constructs. To prevent endocytosis and enhance the cell surface expression of S-EABR constructs, an EPM (residues 243-290 of mouse FcgRII-B1)^31,58 was inserted upstream of the 4-residue Gly-Ser linker and the EABR domain as described³¹ (Fig. S1A).

To verify that the designed constructs induce eVLP budding, HEK293T were transiently transfected with DNA plasmids encoding Wu1 or BA.5 S-EABR using Fugene HD (Promega). After 48 hours, transfected cell supernatants were harvested, and eVLPs were purified by ultracentrifugation on a 20% sucrose cushion as described³¹. The purified eVLP fractions were 3-fold serially diluted in 0.1 M NaHCO₃ pH 9.6, directly coated onto Costar high binding 96-well plates (Corning), and eVLP-associated S protein levels were quantified by ELISA. After 1 hour blocking in Tris-buffered saline with 0.1% TWEEN 20 (TBS-T) and 3% bovine serum albumin (BSA) (TBS-T/3% BSA) at room temperature, the monoclonal anti-S2 Ab S2P6⁵⁹ was diluted to a concentration of 5 µg/mL in TBS-T/3% BSA and added to plates for 1 hour at room temperature. After washing with TBS-T, HRP-conjugated goat anti-human IgG (2015-05; SouthernBiotech) was diluted 1:8,000 in TBS-T/3% BSA and added to plates for 30 min at room temperature. After washing with TBS-T, plates were developed using 1-Step™ Ultra TMB-ELISA Substrate Solution (Thermo Fisher Scientific) and absorbance was measured at 450 nm.

Codon-optimized mRNAs encoding Wu1 S, Wu1 S-EABR, BA.5 S, and BA.5 S-EABR constructs were synthesized by RNAcore (https://www.houstonmethodist.org/research-cores/rnacore/) using proprietary manufacturing protocols as described³¹. In brief, mRNAs were synthesized by T7 RNA polymerase-mediated in vitro transcription using DNA templates encoding the immunogen open reading frame flanked by 5′ and 3’ untranslated regions and terminated by an encoded polyA tail. CleanCap 5’ cap structures (TriLink) were co-transcriptionally incorporated into the 5′ end. Uridine was completely substituted with N1-methyl-pseudouridine to lower immunogenicity⁶⁰. mRNAs were purified by oligo-dT affinity purification and high-performance liquid chromatography (HPLC) to remove double-stranded RNA contaminants⁶¹. Purified mRNAs were stored at –80 °C.

Purified mRNAs were formulated in LNP as described^31,62. In brief, 1,2-distearoyl-sn-glycero-3-phosphocholine, cholesterol, a PEG lipid, and an ionizable cationic lipid dissolved in ethanol were rapidly mixed with an aqueous acidic solution containing the mRNAs using an in-line mixer. The ionizable lipid and LNP composition are described in the international patent application WO2017075531(2017). The post in-line solution was dialyzed with PBS to remove ethanol and displace the acidic solution. Subsequently, mRNA-LNP were measured for size (60-65 nm) and polydispersity (PDI < 0.075) by dynamic light scattering (Malvern Nano ZS Zetasizer). The encapsulation efficiencies were >97% as measured by the Quant-iT Ribogreen Assay (Invitrogen).

All animal procedures in this study were performed in compliance with the U.S. Department of Agriculture’s (USDA) Animal Welfare Act (9 CFR Parts 1, 2, and 3); the Guide for the Care and Use of Laboratory Animals (Institute of Laboratory Animal Resources, National Academy Press, Washington, D.C., 2011); and the National Institutes of Health, Office of Laboratory Animal Welfare. Whenever possible, procedures in this study were designed to avoid or minimize discomfort, distress, and pain to animals. Animals were observed for signs of morbidity and/or mortality at least once daily, according to animal welfare act and the Association for Assessment and Accreditation of Laboratory Animal Care International (AAALACi). All mouse procedures were approved by the Labcorp Institutional Animal Care and Use Committee (IACUC).

7-8 week-old female BALB/c mice (Charles River Laboratories) were purchased and housed at Labcorp Drug Development, Denver, PA for immunization studies. All animals were healthy upon receipt and were weighed and monitored during a 7-day acclimation period before the start of the study. Mice were randomly assigned to experimental cohorts of 10 animals. Up to 10 mice were co-housed together, and cages were kept in a climate-controlled room at 63 to 77°C at 50±20% relative humidity. Mice were provided Rodent Diet #5001 (Purina Lab Diet) ad libitum.

2 µg of Wu1 S mRNA-LNP were administered to all cohorts of mice by IM injections on days 0 and 28. On day 230, pre-vaccinated mice received different boosters by IM injection: i) 2 µg Wu1 S mRNA-LNP; ii) 2 µg Wu1 S-EABR mRNA-LNP; iii) 1 µg Wu1 S mRNA-LNP + 1 µg BA.5 S mRNA-LNP; iv) 1 µg Wu1 S-EABR mRNA-LNP + 1 µg BA.5 S-EABR mRNA-LNP. For bivalent boosters (iii and iv), the Wu1 and BA.5 S or S-EABR mRNA-LNP were mixed prior to the IM injections. Serum samples for ELISAs and neutralization assays were obtained on indicated days by retro-orbital bleeding. One small drop of either 0.5% proparacaine hydrochloride ophthalmic solution or 0.5% tetracaine hydrochloride ophthalmic solution was applied in the eye of the mouse prior to performing retro-orbital blood collection. Eyes were alternated for each blood collection. Euthanasia was performed in accordance with American Veterinary Medical Association (AVMA) guidelines for euthanasia. Mice were individually placed in a CO2 chamber and rendered unconscious. Mice were verified as unconscious by lack of response when toe was pinched. Unconscious mice were subjected to bilateral jugular vein severance for terminal exsanguination. Mice were placed back in the CO2 chamber until death occurred. A double thoracotomy was performed as a second confirmation of death.

Soluble SARS-CoV-2 S trimers (Wu1/D614G, BA.5, BQ.1.1, XBB.1)³², RBDs (Wu1, BA.5, BQ.1.1, XBB.1), and a prefusion-stabilized S2 protein⁶³ were expressed as described^37,64. Briefly, Avi/His-tagged S trimers and His-tagged RBD and S2 proteins were purified from transiently-transfected Expi293F cells (Gibco) by nickel affinity chromatography (HisTrap, Cytiva) and size-exclusion chromatography (SEC) (Superose 6 Increase 10/300, Cytiva)^33,37,64. Peak fractions corresponding to S, RBD, or S2 proteins were pooled, concentrated, and stored at 4°C. Biotinylated S trimers for ELISAs were generated by co-transfection of Avi/His-tagged S constructs with a plasmid encoding an endoplasmic reticulum-directed BirA enzyme (kind gift from Michael Anaya, Caltech).

For heterotrimeric constructs, Wu1/BA.1 (HT2) and Wu1/BA.1/Delta (HT3) heterotrimers were expressed via the addition of C-terminal tags and co-transfection of Expi293F cells (Gibco) with specific plasmid DNA ratios. HT2 heterotrimers were produced by co-transfecting a 1:1 plasmid DNA ratio of StrepII-tagged Wu1 S and His-tagged BA.1 S constructs. HT3 heterotrimers were produced by co-transfecting a 1:1:1 plasmid DNA ratio of StrepII-tagged Wu1 S, His-tagged BA.1 S, and D7324⁴⁷-tagged Delta S constructs. HT2 was sequentially purified using StrepTrap (Cytiva), HisTrap (Cytiva), and SEC, while HT3 also underwent immunoaffinity chromatography using the D7324-specific monoclonal Ab JR-52⁴⁷ (kind gift from James Robinson, Tulane University) before SEC.

The Wu1-His/Wu1-StrepII S homotrimer was generated by co-transfecting Expi293F cells with a 1:1 plasmid DNA ratio of His-tagged Wu1 S and StrepII-tagged Wu1 S, followed by purification as described for the HT2 heterotrimer. Single-tagged S homotrimers included StrepII-tagged Wu1 S, His-tagged BA.1 S, and D7324-tagged Delta S, each purified using their respective affinity chromatography steps followed by SEC.

Grids of SARS-CoV-2 S homotrimers and heterotrimers were prepared at a protein concentration of 2 mg/mL. Prior to freezing, fluorinated octylmaltoside solution (Anatrace) was added to the S proteins at a final concentration of 0.02%(w/v). Three µL of S proteins were immediately applied to Quantifoil 300 mesh 1.2/1.3 grids (Electron Microscopy Sciences) that were freshly glow discharged with the PELCO easiGLOW (Ted Pella) for 60 seconds at 20 mA. Grids were blotted for 3 seconds using Whatman No.1 filter paper with 0 blot force, 100% humidity and room temperature, and vitrified in 100% liquid ethane using a Vitrobot Mark IV system (Thermo Fisher Scientific).

Single-particle cryo-EM datasets for SARS-CoV-2 S homotrimers and heterotrimers were collected using SerialEM automatic data collection software⁶⁵ on a 300 keV Titan Krios (Thermo Fisher Scientific) cryo-electron microscope equipped with a K3 direct electron detector camera (Gatan). Movies were acquired in 40 frames with a defocus range of −1 to −3 µm and a total dosage of 60 electrons per Ų using the 3 by 3 beam image shift pattern with 3 exposures per hole in the super-resolution mode at a pixel size of 0.416 Å. Data processing workflows are outlined in Figure S10. For all datasets, motion correction was done with patch motion correction with a binning factor of 2 and contrast transfer function (CTF) parameters were estimated with Patch CTF in cryoSPARC⁶⁶. Particles with a diameter of 100 to 200 Å were picked with blob picker, inspected, extracted, and finally classified with 2D classification in cryoSPARC⁶⁶. Ice particles, junk particles, and particles with preferred orientations were discarded, and the remaining particles were used for ab initio reconstruction of three to four volumes. These reconstructions were further refined with heterogeneous refinement in cryoSPARC⁶⁶. Depending on the state for the receptor-binding domains (RBDs) in the reconstructions, particles were combined for homogeneous and non-uniform refinements to obtain the final reconstructions in cryoSPARC⁶⁶.

ELISAs were performed as described^31,44. To analyze serum binding to S trimers, pre-blocked streptavidin-coated Nunc® MaxiSorp™ 384-well plates (ThemoFisher Scientific) were coated with 5 µg/mL biotinylated S trimers in TBS-T/3% BSA for 1 hour at room temperature. To analyze serum binding to RBDs or S2, Nunc® MaxiSorp™ 384-well plates (Sigma) were coated with 2.5 µg/mL RBD or S2 in 0.1 M NaHCO₃ pH 9.8 and incubated overnight at 4°C. Plates were blocked with TBS-T/3% BSA for 1 hour at room temperature followed by removal of the blocking buffer by aspiration. Serum samples from immunized mice were diluted 1:100, 4-fold serially diluted in TBS-T/3% BSA, and added to plates. After a 3-hour incubation at room temperature, plates were washed with TBS-T using an automated plate washer. HRP-conjugated goat anti-mouse IgG (715-035-150; Jackson ImmunoResearch) was diluted 1:100,000 in TBS-T/3% BSA and added to plates for 1 hour at room temperature. After washing with TBS-T, plates were developed using SuperSignal^TM ELISA Femto Maximal Signal Substrate (Thermo Fisher Scientific) and absorbance was measured at 425 nm. Area under the curve (AUC) calculations for binding curves were performed using GraphPad Prism 9.3.1 assuming a one-site binding model with a Hill coefficient as described⁶⁷.

Lentivirus-based SARS-CoV-2 pseudoviruses were generated as described^35,36 using S proteins from the Wu1/D614G, BA.1, BA.5, BQ.1.1, and XBB.1 variants in which the C-terminal 21 residues of the S cytoplasmic tails were removed³⁵. Serum samples from immunized mice were heat-inactivated for 30 min at 56°C. Three-fold serial dilutions of heat-inactivated samples were incubated with pseudoviruses for 1 hour at 37°C, followed by addition of the serum-virus mixtures to pre-seeded HEK293T-ACE2 target cells. After 48-hour incubation at 37°C, BriteLite Plus substrate (Perkin Elmer) was added and luminescence was measured. Half-maximal inhibitory dilutions (ID₅₀s) were calculated using 4-parameter non-linear regression analysis in AntibodyDatabase⁶⁸ and ID₅₀ values were rounded to three significant figures.

HiTrap™ NHS-activated resin (Cytiva) was coupled to purified SARS-CoV-2 Wu1 RBD or a mock protein (HIV-1 gp120) at a 10 mg/mL concentration and stored in PBS supplemented with 0.02% sodium azide at 4°C. For Ab depletions, 50 µL dry worth of RBD-coupled resin were aliquoted into microcentrifuge tubes and spun down at 13,000 x g for 2 minutes. The supernatant was discarded. The resin was washed with 100 µL of sterile PBS to remove excess PBS-azide solution. The supernatant was again discarded. 10 µL of mouse serum was then diluted in 90 µL PBS, and added to the resin. The tubes were mixed and incubated at room temperature for 10 minutes. The resin-serum mixture was spun down at 13,000 x g for 7 minutes to allow the resin to settle, and 90 µL of the supernatant was transferred into a fresh microcentrifuge tube. To ensure no resin was present in the final sample, the collection tube was spun down again, and supernatant was transferred to a fresh microcentrifuge tube. Sterile PBS was added for a final volume of 200 µL. Prior to neutralization assays, the Ab depleted serum samples were heat-inactivated at 56°C for 30 minutes.

DMS studies to map RBD epitopes recognized by serum Abs were performed in biological duplicates using independent mutant RBD yeast libraries (Wu1^42,69 and XBB.1.5 RBD⁴³ libraries were generously provided by Tyler Starr, University of Utah) as described^44,64,70,71. Briefly, yeast libraries were stained with serum Abs, labeled with a secondary Ab (115-605-008, Jackson ImmunoResearch), and then processed using a Sony SH800 cell sorter. Cells were gated to capture RBD mutants that had reduced Ab binding for a relatively high degree of RBD expression. Ab-escaped cells were expanded for DNA extraction and Illumina sequencing was done as described^44,70,71. Raw sequencing data are available on the NCBI SRA under BioProject PRJNA1067836. Escape fractions were computed using previously described processing steps^44,70,71 and implemented using a Swift DMS program available from authors upon request.

Escape map visualizations shown as static line plots, logo plots and structural depictions were all created using Swift DMS as previously described^41,44,71. In some visualizations, RBD sites were categorized based on epitope region, which were defined as class 1 (pink) (RBD residues 403, 405, 406, 417, 420, 421, 453, 455-460, 473-478, 486, 487, 489, 503, 504), class 2 (purple) (residues 472, 479, 483-485, 490-495), class 3 (blue) (residues 341, 345, 346, 437-450, 496, 498-501), class 4 (orange) (residues 365-390, 408)⁷², or class 5 (green) (352-357, 396, 462-468)⁴¹. Structural visualizations were performed using an RBD surface (PDB 6M0J) colored by the site-wise escape metric at each site as described⁴⁴.

Statistically significant titer differences between immunized cohorts of BALB/c mice (10 mice per cohort) for ELISAs were determined using analysis of variance (ANOVA) test followed by Tukey’s multiple comparison post hoc test calculated using Graphpad Prism 9.3.1. Titer differences between immunized cohorts of BALB/c mice for Wu1, BA.5, BQ.1.1, and XBB.1 neutralization assays were determined using the non-parametric Kruskal-Wallis test followed by Dunn’s multiple comparison post hoc test calculated using Graphpad Prism 9.3.1. For BA.1 neutralization titers, statistically significant differences were determined using the non-parametric two-tailed Mann-Whitney test calculated using Graphpad Prism 9.3.1.