Communication between parts of mixed viral protein groups helps understand combined vaccine design

To determine whether co-transfection mimicking bivalent mRNA expression could result in mosaic SARS-CoV-2 S heterotrimer formation, we sought to co-express hexaproline-mutated S ectodomains (S-6P) which were designed to maintain soluble S trimers in the antigenic pre-fusion form [42], from G614, Omicron and XBB S variants (Fig 1A). The ConSurf maps [43] of G614 and Omicron indicate the S2 subunits, especially the central helical domains, are highly conserved (Fig 1B). Total S expression levels were compared with equal amounts of total plasmid DNA transfected under nine conditions: Omicron only, G614 only, XBB only, Omicron-G614 co-transfection (2:1, 1:1, and 1:2), and Omicron-XBB co-transfection (2:1, 1:1, and 1:2). Western blot using an antibody targeting a linear epitope in the S2 subunit that is conserved across all three S variants showed that all conditions expressed similar levels of total S protein in the supernatant (S1 Fig). This indicates that all three S-6P variants have similar transfection efficiency and expression levels under our experimental conditions, with no significant interference from co-transfection of Omicron and G614 variants, and of Omicron and XBB variants.

To evaluate if mosaic heterotrimers can be formed and how they affect trimer stability, we prepared mosaic spike from Omicron-G614 (OG) and Omicron-XBB (OX) combinations side-by-side with control preparations of Omicron, G614, and XBB homotrimers (Figs 1C and S1). Using the optimized purification protocols and low-passage, viable Expi293F cells, the best yields from parallel purifications of all five constructs were obtained (S1 Table). The data agree with the total S expression result and show that the co-expression did not interfere the S trimer production.

Nearly 90% of the protein eluted as a single peak at 8.5–9 mL from the size-exclusion chromatography (SEC) at 4 °C, consistent with well-formed trimers with comparable yields around 4–5 mg (Fig 1D). Both G614 and XBB homotrimers reflect sharp peaks that eluted slightly earlier than Omicron-containing S trimers due to the presence of affinity tags. Omicron homotrimer, eluted later from SEC, showed a modest shoulder eluting slightly before the primary chromatographic peak. The particle hydrodynamic radius and polydispersity measured by dynamic light scattering (DLS) also indicated that Omicron homotrimer had slightly larger radius and polydispersity than other S trimers at 25 °C (S2 Table). Thermal melt experiments monitoring particle radius by DLS provided a means of measuring the stability of S assemblies. Both OG and OX mosaic trimers indicated a smaller radius than Omicron homotrimer under room temperature and physiological temperature ranges (Fig 1E). The trimeric integrity and morphology of mosaic heterotrimers were also confirmed by negative-stain electron microscopy (nsEM) images (S1 Fig). Therefore, both Omicron and XBB protomers, which possess exactly the same S2 subunit sequence, and Omicron and G614 protomers, which carry numerous mutations around the S1/S2 cleavage site and S2 subunit (Fig 1A), could form mosaic heterotrimers with limited alterations in global trimeric structure and morphology.

Furthermore, based on the yield ratios of the isolated mosaic heterotrimer in the elution to the Omicron homotrimer in the flow-through from the last His-tag purification step (, OG: 4.7 mg/1.9 mg; OX: 3.6 mg/1.2 mg), the mosaic trimers were estimated to be ~ 60% of the total trimer yield from co-transfection, close to the theoretical percentage for stochastic trimer assemblies from 1:1 co-transfection. This suggests that, in both genetically close and distant strains co-transfection, homotrimer formation did not overwhelm heterotrimer formation, and the presence of heterologous S material did not lead to noticeable misassembly or aggregation. S1 Table

Another important factor that influences immunogenicity of viral glycoproteins is the glycan profile on the displayed antigens. N-linked glycans can contribute to epitopes that antibodies target [44,45]. Additionally, in SARS-CoV-2 S, N-linked glycans on the RBD have been suggested to affect trimer open/closed state equilibria, which could modulate exposure of some epitopes [46]. To determine how mosaic S heterotrimers glycoprofiles compare with their respective homotrimers, we carried out glycan profiling for the set of mosaic and homotrimer spikes. Our results show that both N-glycosylation sites on the RBD are predominantly complex forms across all five trimeric assemblies (Fig 1F). The glycan profiles we determined in our constructs are consistent across purification batches and in agreement with previous studies on the soluble pre-fusion S ectodomain constructs [47] as well as protein nanoparticle vaccines featuring pre-fusion S or RBD antigens [31,48] that have been shown to elicit potent neutralizing antibody responses. For the rest of the N-glycosylation sites we characterized, the glycan profiles are mostly unaltered in mosaic S trimers except for the N801, which exhibited more processing towards complex forms in OG heterotrimer (S2 Fig).

Next, we carried out HDX-MS structural dynamic analysis comparing three S-6P homotrimers: G614, Omicron and XBB in order to obtain baseline structural dynamic profiles against which we could compare structural dynamics and trimer integrity profiles of the OX and OG mosaic heterotrimers.

HDX-MS analysis on G614, Omicron and XBB homotrimers revealed isolate-specific differences in conformational preferences and S2 dynamics highlighting regions of the polypeptide that were ordered differently among these three homotrimer antigens (Figs 2A, 2B and S3). In our previously reported comparison of S-2P trimers [13], peptides 388–392 and 982–990 were found to serve as effective reporters for RBD up/down configurations as increased dynamics near the C-terminus of the RBD facing the apex of central helices (peptide 388–392) and the central helical apex (peptide 982–990) were observed when S adopts an open conformation. In the present experiments, both of these peptides indicated that dynamics were greater in G614 than in Omicron, which was also significantly greater than in XBB (Fig 2C, peptide #1 and #2). When S is in its open state, the RBD-up conformation also affects dynamics in the S1 hinge region connecting RBD to the rest of the spike trimer. For example, the N-terminal peptide 516–533 connecting to the RBD, showed increased dynamics in response to RBD shifting up, while peptide 542–568, was more protected by the neighboring protomer facing towards the trimer core (Fig 2C, peptide #3 and #4). These dynamic changes all support the conclusion that the G614 S-6P trimer exhibits a greater conformational bias toward populating an open state than Omicron and XBB S-6P trimers.

Sequence variations in the N-terminal domain (NTD) and RBD, especially in the receptor-binding motif (RBM) regions among the three constructs resulted in different digestion patterns, limiting the set of homologous peptides that could provide direct comparisons in these two regions (Fig 2B). By contrast, Omicron and XBB share the same S2 sequences and digestion patterns. HDX-MS analysis revealed that the only significant dynamic differences in the S2 subunit were observed in the heptad repeat 1 (HR1) region reflecting a less open state in XBB (Fig 2C, peptide #2 and #5, and S3 Table). Residue substitutions around the S2' and HR1 region in Omicron and XBB relative to the ancestral Hu-1 G614 variant yielded differences in S2 subunit dynamics near the S2' site, where both Omicron and XBB showed increased deuterium uptake over G614 (Fig 2C, peptide #6, and S3 Table). Overall, summarizing the isolate-specific dynamics in G614, Omicron and XBB S homotrimers: G614 S adopted more open conformations with stabilized S2 subunit; Omicron S showed less open conformations than G614 but possessed greater dynamics in the S2 subunit from the HR1 domain extending to the fusion peptide-proximal region (FPPR); and XBB S indicated the least open state preferences and compromised S2 subunit dynamics. The dynamic features in G614 and Omicron S closely agree with previously reported results [49].

To understand whether the formation of mosaic heterotrimers could alter the conformational preference, structural dynamics, and epitope exposure of these antigens, we compared the HDX-MS results from homologous peptides in the mosaic S heterotrimer with those in the corresponding homotrimers. Since the homologous peptides with identical sequence from two constructs will co-elute in HDX-MS, a multi-modal mass envelope is anticipated if a given peptide behaves differently in the different protomers. However, in the cases when the separation was not great enough to deconvolute the bimodal populations into individual protomer contributions, a broadened mass envelope reflecting the average of two underlying populations is observed.

In the Omicron-XBB (OX) mosaic S trimer, the Omicron and XBB protomers mostly retained the dynamic features of their respective homotrimers (Figs 3 and S4). The dynamics of NTD and RBD from each protomer were not significantly affected by mosaic trimer formation as can be seen in the differential plot compared against the baseline Omicron homotrimer (Fig 3A). The peptide uptake profiles in the NTD of OX reflect the average of corresponding uptake levels in Omicron and XBB homotrimers (Fig 3B, peptide #1 and #2).

Homologous peptides such as the peptide spanning residues 177–191 (Fig 3B, peptide #3), where the Q183E substitution in the XBB sequence produced peptides that differ by the single amino acid substitution, could also be distinguished in the mass spectrometer and directly compared with counterparts in the homotrimers. The peptide in the Omicron protomer of the OX mosaic trimer showed the identical dynamic profile as in the Omicron homotrimer and the peptide in XBB protomer matched the XBB homotrimer peptide as well (Fig 3B, peptide #3). These examples thus demonstrated that Omicron and XBB protomers mostly retained the isolate-specific dynamic features over the immunodominant NTD and RBD domains.

Besides the local structural orderings, overall, the dynamic profile of the mosaic OX trimer was found to present as a hybrid between Omicron and XBB (Fig 3). Regions that are structurally associated with S open/closed states such as the RBD C-terminus (Fig 3B, peptide #4) and apex of the central helix (Fig 3B, peptide #5 and #6) exhibited dynamics intermediate between the two homotrimers. The conservation of isolate-specific dynamics in the NTD and RBD as well as conformational preferences indicates relatively little crosstalk and alteration in behavior of the heterologous protomer in the OX co-assembled spikes, most likely due to the commonality of their S2 subunits.

We next investigated the dynamic behavior in the Omicron-G614 (OG) mosaic heterotrimer. This combination enabled us to examine whether a mosaic S trimer with divergent S2 subunits may exhibit more prominent effects on inter-protomer packing or allosteric effects across the trimeric spike assembly that could affect their antigenicity. In the OG mosaic trimer, similar effects were observed in the RBD as in the OX case, which exhibited uptake levels intermediate between those in Omicron and G614 homotrimers (Fig 4B, peptide #1 and #2, and S5 Fig). Peptide 471–486, spanning the loop connecting the RBD core and the RBM located at the surface of RBD, is a target for neutralizing antibodies and is highly mutated during viral evolution, conferring immune escape [50,51]. The S477N, T478K, and E484A substitutions in peptide 471–486 observed in Omicron versus G614 protomers enable the homologous peptides to be distinguished, allowing direct comparison between the behavior of this peptide in the two types of protomers. In this case, the uptake level in each protomer matched the behavior seen in the respective homotrimer (Fig 4B, peptide #3). Notably, in contrast to the conformational preferences observed in the OX heterotrimer, the OG heterotrimer appeared to favor open conformations. Specifically, the peptides reporting on RBD configurations in the OG heterotrimer more closely paralleled their behavior in the G614 homotrimer, showing the dominant influence of the G614 protomer on the mosaic OG assembly (Fig 4B, peptide #1, #2 and #6, and S5 Fig). One of the exceptions to this G614 dominance was observed in the hinge region (Fig 4B, peptide #4), where the dynamic behavior of this peptide in OG heterotrimer more closely recapitulated its behavior in the Omicron homotrimer. While increased dynamics at C-terminal RBD (peptide #2) and helical apex (peptide #6) were consistent with greater exposure in an open conformation, the dynamics of the peptide 553–568 loop, which is located at the interface between RBD and a neighboring NTD, was found to be more affected by the positioning and packing of the NTD on the neighboring protomer (Fig 4B, peptide #4). The dynamic features at this site in the OG heterotrimer mirrored the behavior in the Omicron homotrimer indicating that the NTD maintained a similar position and conformation as in Omicron homotrimer, despite the RBD favoring the more open state in the same OG mosaic trimer. In the mosaic trimer, these structural motifs appear to be somewhat decoupled in their dynamic behavior.

Prominent dynamic changes in the OG heterotrimer were found in the FPPR region between S1/S2 and S2' cleavage sites (Fig 4B, peptide #5) and HR1 domain (Fig 4B, peptide #7). Notably, homologous peptides in the region 962–981 were differentiated from two protomers due to the N969K mutation in Omicron. This helix-loop-helix peptide was intrinsically more dynamic in the Omicron homotrimer than the G614 homotrimer, in agreement with a previous study [49], but the trend was reversed in the Omicron protomer and the G614 protomer when they co-assembled (Fig 4B, peptide #7). Examining the mass spectra revealed a bimodal mass envelope for this FPPR peptide in the Omicron homotrimer at the 1-minute time point (Fig 5) with a fast-uptake population in the Omicron homotrimer that could adopt a more exposed, dynamic conformation. However, the same region in the Omicron protomer in the mosaic OG trimer showed decreased uptake levels with less detectable peak broadening and suppressed bimodal spectra which mostly resembled the slow-uptake population in the Omicron homotrimer (Fig 5). This suggests that the FPPR in the Omicron protomer of the OG mosaic trimer stably maintains its ordered conformation. On the other hand, the homologous peptide in the G614 protomer showed the opposite trend, exhibiting significantly increased dynamics, in the OG mosaic trimer (Fig 5). Enhanced dynamics in the G614 protomer of the OG mosaic trimer was reflected by the presence of novel dynamic conformations in FPPR, that gave rise to a bimodal mass envelope at the 1-minute time point compared to the unimodal ordered conformation in G614 homotrimer (Fig 5). This phenomenon indicates that Omicron-specific mutations Q954H and N969K can allosterically induce the neighboring protomer to adopt highly dynamic conformations in the FPPR.

The crosstalk between protomers in the OG mosaic heterotrimer gave rise to novel structural and dynamic features that S homotrimers did not present. Specifically, the mosaic heterotrimer favored an open state as seen in the G614 homotrimer, with the isolate-specific orderings in NTDs and RBDs for the Omicron protomers mostly retained. The pre-fusion S helix-loop-helix region at FPPR became more ordered in the Omicron protomers with the induction of highly dynamic conformation in the G614 protomer over the same region. Despite such prominent dynamic effects in the S2 subunit, the OG mosaic trimer integrity and stability were well-maintained, as indicated by constantly protected central helical domains from HDX-MS (Fig 4B, peptide #8, and S5 Fig), thermal stability results (Fig 1E), and nsEM images (S1 Fig). Thus, the sequence differences in S2 between the ancestral G614 subvariant and prevalent Omicron subvariant were not sufficient to impact the trimer's ability to assemble but did result in modulation of the trimer's dynamic profile at positions across the spike assembly.

Previous studies have reported that over 90% of neutralizing antibodies targeted RBD and NTD [14 –16]. Retaining the structural ordering and functioning in NTD and RBD is important for assessing the antigenicity as well as vaccine efficacy. To further investigate whether the localized structural dynamic differences in the mosaic trimers have antigenic effects, bio-layer interferometry (BLI) was used to compare human ACE2 (hACE2) and antibody binding kinetics for mosaic versus homotrimer spikes.

In the dimeric hACE2 binding experiments, anti-human IgG Fc capture (AHC) immobilized Fc-hACE2 strongly bound all five constructs with K_D < 10 nM (Fig 6), in agreement with previously reported K_D values measured by surface plasmon resonance [52]. We then tested binding with monoclonal antibody (mAb) C68.13 that competes with ACE2 binding and has been reported to exhibit 2–3 orders of magnitude weaker inhibition of XBB than G614 and Omicron [53]. This allows us to selectively examine how the critical RBD epitope presentation is modulated on the mosaic heterotrimer. In our binding assays using AHC biosensors, the G614 homotrimer, Omicron homotrimer and OG mosaic trimer were found to bind C68.13 with high affinity, K_D ~ 1 nM (Fig 6). Flat dissociation curves resulted in k_off rates in lower 10^-4 levels, close to the detection limit, indicating very stable binding. The XBB homotrimer bound C68.13 with K_D ~ 95 nM, 50-to-100-fold weaker binding to this mAb than G614 and Omicron homotrimers (Fig 6). To test whether the BLI configurations with immobilized IgG and S analyte were contributing to the measured binding affinities, we also performed binding assays using Streptavidin biosensors, which enabled biotinylated S constructs to be captured. Binding to soluble C68.13 yielded the similar binding behaviors and affinities. The G614 homotrimer, Omicron homotrimer and OG mosaic trimer showed C68.13 binding affinity ~1 nM, while XBB homotrimer showed 50-fold weaker binding and OX mosaic trimer showed a moderately compromised K_D of 6.5 nM (Fig 6). Compared with the immobilized bivalent IgG, immobilized S trimers for IgG detections indicate both higher k_on and k_off values but the resulting K_D closely matched the measurements with immobilized IgG (S4 Table). These data indicate that the mode of interaction between the RBDs and antibodies was relatively unaltered by the presence of adjacent protomers with weak binding affinities.

Compared to the Class 2/3 RBD mAb C68.13, Class 3 mAb C68.61 binds to a more conserved epitope toward the RBD core regions [53]. C68.61 showed strong binding to all five constructs in both BLI configurations with immobilized IgG and with immobilized biotinylated S (S6 Fig). Slow dissociation occurred for both IgG and Fab forms of C68.61 in these binding assays (S4 Table). These results indicate that the high affinity binding of these RBD-targeting C68 mAbs, is not primarily due to avidity or bivalent binding.

In addition, we performed competitive binding assays with C68.13 and hACE2. Omicron, OX and XBB S trimers were first bound with excessive C68.13 IgG to saturate RBDs on the Omicron protomers. When hACE2 was subsequently introduced, it was able to bind to the unoccupied RBDs on the XBB protomers, resulting in a secondary BLI response. The OX mosaic heterotrimer exhibited an intermediate response between Omicron and XBB homotrimers, indicating the accessibility and binding were less affected by the presence of the adjacent heterotypic protomer (). Our BLI results align with the inferences from the HDX experiments, suggesting that mosaic heterotrimers (OX and OG) impose limited restriction on RBD dynamics and receptor/antibodies binding. S7 Fig

The gene sequences encoding SARS-CoV-2 S ectodomain with hexaproline mutations (S-6P: F817P, A892P, A899P, A942P, K986P and V987P) [42], GSAS substitution at S1/S2 site (RRAR) and C-terminal T4 fibritin foldon were cloned into pCMV plasmid. The open reading frame (ORF) sequence of Hu-1 G614 S-6P construct is the reference Wuhan-Hu-1 sequence (residues 14–1211) with the D614G mutation, C-terminal His-tag and the alterations mentioned above. Omicron BA.2 construct was assembled from the ORF in the pαH Omicron BA.2 S-2P plasmid (Addgene plasmid #184829) into pCMV plasmid and mutated with four additional prolines using HiFi DNA Assembly (NEB). XBB construct was assembled from the ORF in the pVRC8400 XBB S-2P plasmid (Addgene plasmid #160474) into pCMV plasmid and mutated with four additional prolines and a C-terminal His-tag (NEB). The ORF sequence of SARS-CoV (SARS1) S-6P (Tor2 strain [65] with F799P, A874P, A881P, S924P, K968P and V969P) was cloned into pCMV plasmid with the same vector backbone as G614 and XBB construct.

293T adherent cell line was used for optimizing co-transfection condition. Cells were split into 12-well plate (1 mL per well) at seeding density (0.1 × 10⁶ cells per well). Confluent cells (0.5 × 10⁶ cells per well) were transfected with total 2 μg DNA and 2 μL Lipofectamine 3000 (Invitrogen) based on Lipofectamine 3000 reagent protocol. DNA being transfected or co-transfected into the cell cultures under nine conditions: Omicron only, G614 only, XBB only, Omicron-G614 co-transfection (2:1, 1:1, and 1:2), and Omicron-XBB co-transfection (2:1, 1:1, and 1:2). After 2 days in the 37 °C incubator (5% CO₂), 200 μL supernatant from each well was collected for SDS-PAGE to check expression level.

Suspension Expi293F cell culture was used for spike protein expression. For a 700 mL cell culture at confluency (3 × 10⁶ cells/mL), 0.7 mg total DNA (1:1 for co-transfection) and 2.1 mL ExpiFectamine 293 reagent were mixed in Gibco Opti-MEM and added into the cell culture following ExpiFectamine 293 user guide. The transfected culture was incubated at 37 °C and 5% CO₂ with constantly shaking at 125 RPM for 5 days before harvest.

On the fifth day post-transfection, Expi293F cells were harvested by centrifugation at 2000 RCF for 30 minutes. Supernatant was vacuum-filtered through 0.45 μm αPES filters and supplemented with Tris-HCl (pH 8.0) and NaCl.

For G614 and XBB spike homotrimers purification, imidazole was added to the supernatant to reach a final concentration of 50 mM Tris-HCl (pH 8.0), 200 mM NaCl and 20 mM imidazole, same as the Ni-IMAC binding buffer composition. A 2 mL Ni-IMAC resin (Thermo Fisher Scientific) was washed and equilibrated with binding buffer and packed for a gravity flow column. Supernatant was loaded and flowed through this gravity flow column, followed by four times 10 mL (5 CV) wash with binding buffer. The bound protein was then eluted from the column with three times 6 mL (3 CV) Ni-IMAC elution buffer (50 mM Tris-HCl, pH 8.0, 200 mM NaCl, 300 mM imidazole). Elution fractions were SDS-PAGE checked, and then combined. Combined spike protein solution was buffer-exchanged to reduce the imidazole level below 1 mM and concentrated to ~1 mg/mL with an Amicon Ultra-15-mL-centrifugal filter (30 K, Millipore), followed by flash frozen in liquid nitrogen and stored in -80 °C freezer.

For Omicron, mosaic Omicron-G614, mosaic Omicron-XBB and mosaic Omicron-SARS1 trimers purification (Fig 1C), ethylenediaminetetraacetic acid (EDTA) was added to reach a final concentration of 100 mM Tris-HCl (pH 8.0), 150 mM NaCl and 1 mM EDTA, same as the Strep-Tactin XT wash buffer composition. A 3.5 mL Strep-Tactin XT 4Flow high-capacity resin was washed and equilibrated with wash buffer and packed for a gravity flow column. Supernatant was loaded and flowed through this gravity flow column, followed by four times 8.5 mL (2.5 CV) wash. The bound protein was then eluted from the column with three times 5 mL (1.5 CV) Strep-Tactin XT elution buffer (100 mM Tris-HCl, pH 8.0, 150 mM NaCl, 1 mM EDTA, 50 mM biotin). Elution fractions were combined, buffer-exchanged to reduce the biotin level below 1 mM and concentrated to a final volume below 1 mL. HRV 3C protease (Pierce) was added accordingly based on the protein solution volume for the digestion to remove TwinStrep tag overnight at 37 °C. On the next day, digested samples were spun at 15000 RCF for 10 minutes to remove possible protein aggregation pellet. The Omicron homotrimer protein could now be flash frozen in liquid nitrogen and stored in -80 °C freezer. For mosaic heterotrimer purifications, samples were flowed through a 1 mL pre-equilibrated prepacked StrepTrap XT column (Cytiva) at 1 mL/min to separate undigested protein. Flow-through was reloaded to the column for better purification results. The final flow-through containing heterotrimer with TwinStrep tag removed was collected and buffer-exchanged to the nickel column binding buffer. Samples were flowed through a 1 mL pre-equilibrated prepacked HisTrap HP column (Cytiva) at 1 mL/min, followed by 7 mL (7 CV) wash. The bound protein was then eluted from the column with 3 mL (3 CV) elution buffer. SDS-PAGE checked elution fractions were combined, buffer-exchanged to reduce the imidazole level below 1 mM and concentrated to ~1 mg/mL with an Amicon Ultra-15-mL-centrifugal filter (30 K, Millipore). The duo-affinity-column purified mosaic spike heterotrimers were then flash frozen in liquid nitrogen and stored in -80 °C freezer.

All the spike trimers were finally size-exclusion chromatography (SEC) purified before major experiments. 500 μL sample was thawed and spun at 15000 RCF for 10 minutes to remove possible protein aggregation pellet before injecting to Superdex 200 column (GE) on an AKTA Pure system (GE). With the SEC running buffer (10 mM HEPES, 200 mM NaCl, 0.02% NaN₃, pH 7.5) and flow rate at 0.5 mL/min, spike trimer was eluted at around 8.5–9 mL. The SEC purified spike trimers were concentrated to desired concentrations according to the experiment need.

Dynamic light scattering measurements were performed on a DynaPro NanoStar II (Wyatt) to characterize spike trimer integrity, homogeneity and thermal stability. SEC purified spike samples at 0.5 mg/mL were first spun down at 15000 RCF for 20 minutes to remove aggregation pellet. 5 μL of the sample was then injected into a 2-μL quartz cuvette (Wyatt). Each DLS run was measured with twenty 10-second acquisitions at 25 °C by LASER auto-attenuation. Thermal melt measurements were performed from 25 °C to 90 °C with the rate at 0.5 °C/min. Five measurements were taken and averaged at each temperature point.

For the hydrogen/deuterium-exchange (HDX) experiments, 10 μg each of Hu-1 G614, Omicron BA.2, XBB, mosaic Omicron-G614 and mosaic Omicron-XBB spike trimer was incubated in the deuteration buffer (10 mM HEPES, pH^* 7.5, 85% D₂O, Cambridge Isotope Laboratories, Inc.) at 22 °C for three different timepoints: 3, 60, 900 seconds (five replicates for 3 seconds and four replicates for 60 and 900 seconds). All exchanged samples were immediately mixed with an equal volume of ice-chilled quench buffer containing 8 M urea, 200 mM tris(2-chloroethyl) phosphate (TCEP) and 0.2% formic acid (FA) to pH 2.5, transferred into an autosampler vial with metal cap and flash frozen in liquid ethanol. Samples were loaded to the LC-MS system by an automating platform and analyzed by a Synapt G2 mass spectrometer (Waters) with the settings described in previous publication [66]. Samples were in-line digested into peptides by an immobilized pepsin column (2.1 × 50 mm) and loaded onto CSH C18 trap cartridge (Waters) with loading buffer (2% acetonitrile (ACN), 0.1% trifluoroacetic acid (TFA)) at 800 μL/min. Peptides were then separated by a UPLC CSH C18 column (1.0 × 50 mm, 1.7 μL, Waters) with a 15-min linear gradient from 3% to 40% buffer B (buffer A: 2% ACN, 0.1% FA, 0.025% TFA; buffer B: 99.9% ACN, 0.1% FA) at a 40 μL/min flow rate.

Samples were also run on a Orbitrap Ascend Tribrid mass spectrometer (Thermo Fisher) to obtain MS/MS information for both determining N-glycosylation forms and identifying pepsin digested peptides by Byonic and Byos (Protein Metrics). Peptide mass spectra were further confirmed on DriftScope (Waters) and identified with specific retention time and drift time. HDExaminer (Sierra Analytics) was used to parse the mass spectrometer datasets and calculate deuterium uptake of each peptide identified by the specific retention time and drift time.

For all the SDS-PAGE gels using Coomassie Blue staining, 11 μL spike samples were mixed with 1 μL DTT (0.5M) and 4 μL 4 × NuPAGE LDS loading dye (Invitrogen) before loading to 4–12% NuPAGE Bis-Tris gel (Invitrogen) wells. SDS-PAGE gels were run at 120 Volts for 10 minutes and 150 Volts for 60 minutes in MES SDS running buffer (Invitrogen), followed by gel collection and PageBlue (Thermo Scientific) staining.

For SDS-PAGE gels proceeding to western blots, 5 μL spike samples were mixed with 1 μL DTT (0.5M) and 2 μL 4 × loading dye, loaded to the SDS-PAGE gel wells, and run at 120 Volts for 80 minutes. The peptides on the gel were then transferred to the Immobilon-FL PVDF membrane (Millipore) using 10% methanol supplemented NuPAGE transfer buffer (Invitrogen) at 30 Volts for 2 hours. All membranes were blocked in 5% non-fat milk overnight, followed by incubation in primary antibody, respectively, in room temperature for 1 hour. Primary antibody against spike S2 subunit (1A9, mouse, mAb, Invitrogen, MA5-35946) was used in 1:2000; against Strep-tag (mouse, mAb, GenScript, A01732S) was used in 1:2000; against His-tag (mouse, mAb, Invitrogen) was used in 1:2000; against SARS-CoV spike S1 subunit (mouse, mAb, Sino Biological, 40150-MM10) was used in 1:2000 tris-buffered saline-Tween (TBST) solution. Secondary antibody anti-mouse IgG was used in 1:5000 5% non-fat milk blocking buffer in room temperature for 1 hour. The membranes were imaged on the Odyssey M-XS imaging system (LI-COR Biosciences) using 700 nm channel and optimized exposure intensity.

Bio-layer interferometry was used to measure binding kinetics between spikes and hACE2 receptor or neutralizing antibodies. Octet anti-human Fc capture (AHC) biosensors (Sartorius) were pre-incubated in assay binding buffer (10 mM phosphate-buffered saline (PBS), pH 7.4, 0.1% bovine serum albumin (BSA), 0.02% Tween 20) for at least 15 minutes. Spike samples were 2-fold serial diluted with binding buffer from 200 nM to 3.1 nM. Ligand samples were also prepared with binding buffer at 2 μg/mL concentration for dimeric hACE2-Fc and 1 μg/mL for antibodies. Octet Streptavidin (SAX) biosensors (Sartorius) were used to immobilize 15 μg/mL biotinylated spike for antibody detection. In this assay setup, C68 IgGs were 2-fold serial diluted with binding buffer from 100 nM to 25 nM and Fab was 2-fold serial diluted from 200 nM to 50 nM. The binding assay was performed on an Octet BLI system (Sartorius) with 60-second ligand loading, 180-second association and 180-second dissociation processes. The competitive binding assay was performed with 60-second ligand loading, 360-second association with 100 nM C68.13 IgG solution, followed by another 360-second association with 100 nM C68.13 and 100 nM hACE2-Fc solution. BLI kinetics data were processed by ForteBio Data Analysis software (v11.0) with 1:1 binding model fitting.

SEC purified spike trimers were diluted to 20 ng/μL for negative-stain electron microscopy (nsEM). Carbon film 300 mesh copper grids (Electron Microscopy Sciences) were glow discharged for 30 seconds in a vacuum chamber. 3 μL of each diluted spike sample was added onto the grid (carbon side), incubated for 1 minute before blotting off with a piece of folded filter paper. The grid was then incubated in one drop of nano-W (Methylamine Tungstate, Nanoprobes) for 1 minute, followed by blotting off excess stain solution to leave a thin layer. The air-dried grid was saved in the grid box and imaged on a FEI Tecnai G2 Spirit TEM (120 kV) with a Gatan Ultrascan 4000 CCD detector. Data collection was performed using Leginon with an image pixel size of 1.6 Å.

Statistical analysis was done for pairwise comparison between HDX-MS deuterium uptake levels. Technical replicates (n = 5 for 3-second, n = 4 for 1-minute and 15-minute time points) were prepared and run for all five constructs during HDX-MS. Two-tailed t-test was used to report the statistics. All uptake plots include error bars indicating the standard derivations from (n > 3) replicates. Error bars less than 1.5% were masked by the plotted data points on the plots.

Protein structures including ConSurf maps and HDX heatmaps were visualized and exported using PyMOL 3.0 (Schrödinger). Figures were prepared in Adobe Illustrator (v29.2.1).