SARS-CoV-2 spike sequence-based distance as a marker of binding antibody response to COVID-19 vaccines

Abstract

COVID-19 vaccines based on ancestral SARS-CoV-2 have proven highly effective at reducing the risk of illness, especially severe disease. Both binding and neutralizing antibodies have been demonstrated to be strong predictors of the level of vaccine efficacy (VE). Both VE and vaccine-induced antibody responses have been shown to be lower against emergent SARS-CoV-2 viruses; therefore, predicting COVID-19 VE against emergent viruses is critical for decision-making regarding the composition of new vaccines. The data needed to enable such prediction are unclear. We report on 728 individuals without prior SARS-CoV-2 infection who received primary vaccination with ancestral-virus-based mRNA and vector-based COVID-19 vaccines and who were boosted in a homologous or heterologous fashion with mRNA, vector, or protein-based COVID-19 vaccines including a bivalent B.1.351 mRNA vaccine. Post-prime and post-boost binding antibody responses were used to evaluate the extent and drivers of variability in these responses to 22 SARS-CoV-2 Spike antigens from viruses that emerged between 2020 and 2021. We evaluated how well proteomic distance between the vaccine and assay Spike antigen predicted the vaccine-induced antibody response. Following primary vaccination, antibody responses varied across Spike antigens and were, on average, 36 % lower per 10-amino acid (AA) difference between the vaccine and assay Spike antigen (95 % CI: 30 % to 43 %). The geometric mean antibody response to a given antigen was nearly perfectly predicted by the sequence-based distance of the antigen to the vaccine. Post-boost responses were less variable across antigens and weakly associated with Spike distance (17 % lower per 10-AA difference; 95 % CI: 14 % to 20 %). The high variability in binding antibodies across individuals was only partially explained by participant characteristics. Given that populations now have experienced multiple rounds of prior vaccination and infection, measurement of vaccine-induced antibody responses from representative populations will likely be needed to predict the efficacy of COVID-19 vaccines against future strains.

1. Introduction

The COVID-19 pandemic, caused by severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), has been one of the most impactful global health crises of the last century; as of July 30, 2025, it has caused more than 7 million deaths [1] including more than 1.2 million in the United States (US) alone [2]. In response, COVID-19 vaccines were developed rapidly and highly successful: randomized trials demonstrated high efficacy against COVID-19 due to ancestral SARS-CoV-2, with even higher efficacy against severe disease [3] with high efficacy confirmed in early population-based studies [4,5]. Booster vaccines were introduced, and guidelines for timing and target populations have been put forward [6,7] Despite this, uptake of booster vaccines is low in many populations and the optimal timing and composition of booster vaccines is unclear. Binding and neutralizing antibodies induced by vaccination have been established as excellent surrogates of protection against disease [8,9] and are the basis for expanding vaccination indications to new populations and for licensure of updated vaccine formulations [10,11].

However, SARS-CoV-2 is now endemic [12] and likely to cause an ongoing public health burden as new variants emerge [13], similar to influenza. Major lineages emerged in late 2020 and culminated in the Omicron variant and its descendent sub-lineages, which still dominate today [14]. The continued emergence of new variants is a key challenge: vaccines based on the ancestral variant have been found to afford reduced protection against COVID-19 caused by new variants, notably Omicron [15,16]. Emergent viruses have evaded pre-existing immune responses [17,18] and reductions in vaccine efficacy are congruent with reductions in binding and neutralizing antibodies [19]. Continued SARS-CoV-2 evolution and limited cross-variant protection remain hurdles for booster vaccine development and acceptance [20–23].

One approach to updating vaccines is to adopt a model based on influenza vaccines. The WHO Global Influenza Surveillance and Response System (GISRS) monitors influenza viruses, guides annual influenza vaccine development, and provides virus samples for vaccine production [24,25]. Of the 3–4 million specimens collected, about 10,000 are characterized for antigenic and genetic properties [24]. Antigenic characterization evaluates the immune response triggered by the currently circulating strain [26], whereas genome characterization involves monitoring how the sequence of the virus is changing over time and identifies genetic similarities to past strains. In addition, phylogenetic analysis is used to characterize the genetic differences among emergent and historical viruses [27]. Typically, twice a year these data, along with influenza surveillance, clinical studies of the previous year’s influenza vaccine efficacy, and the availability of influenza vaccine viruses are used to select strains for the following year’s influenza vaccines [26]. Leveraging this infrastructure, GISRS began tracking the global COVID-19 pandemic in 2020 and the WHO planned expanded development [28].

To apply the influenza vaccine development paradigm to SARS-CoV-2 would require understanding how well vaccines based on a particular viral strain will protect against future strains, and which data and tools are needed to predict the level of protection conferred. This is especially complex given that there are multiple COVID-19 vaccine platforms (e.g., messenger RNA (mRNA)-based, vector-based, and adjuvanted protein-based) and many individuals have been boosted at various intervals, with a range of heterologous prime-boost regimens. Further, this landscape will become increasingly complex as the portfolio of licensed variant vaccines expands.

Here, we use data from the MixNMatch Study (DMID 21–0012) [29] of 728 COVID-19-vaccinated adults in the US without prior evidence of SARS-CoV-2 infection to characterize binding antibody (bAb) responses after primary COVID-19 vaccination and homologous or heterologous booster vaccination to a variety of SARS-CoV-2 strains that emerged in 2021–2022. We profile the antibody responses across vaccine regimens to various Spike antigens and evaluate factors associated with these responses, both at an individual and population level. Specifically, we evaluate the association between the antibody responses and the proteomic sequence-based ‘distance’ of the SARS-CoV-2 Spike antigen to the sequence of the virus used in the prime and boost vaccines.

2. Materials and methods

2.1. Study design

The MixNMatch study (DMID 21–0012) was a phase 1–2 open-label, nonrandomized, adaptive-design clinical trial performed in sequential stages at 10 sites across the US, conducted by the Vaccine and Treatment Evaluation Units (VTUEs) from the Infectious Disease Clinical Research Consortium (IDCRC) with support from the Division of Microbiology and Infectious Diseases within the National Institute of Allergy and Infectious Diseases (ClinicalTrials.gov, NCT04889209). Both homogenous and heterologous COVID-19 booster vaccines were found to be safe and immunogenic [29]. The trial enrolled 761 healthy adults between May 2021 and May 2022 who received a full primary COVID-19 vaccination series at least 12 weeks earlier – two doses of mRNA-1273 (Moderna), two doses of BNT162b2 (Pfizer), or one or two doses of Ad26.COV2⋅S (Janssen) – and who reported no history of SARS-CoV-2 infection or monoclonal antibody infusion at enrollment (Fig. S1). Enrollment targeted 50 participants with approximately equal numbers in two age strata (18 to 55 and ≥ 56 years of age). Six different COVID-19 booster vaccines were provided across six different stages of the trial: 100 μg mRNA-1273 (Stage 1), 5 × 10¹⁰ virus particles dose of Ad26.COV2⋅S (Stage 2), 30 μg of BNT162b2 (Stage 3), 100 μg of bivalent mRNA-1273.211 (Stage 4; ancestral and B.1.351 Beta Spike), 50 μg of mRNA-1273 (Stage 5) and 50 μg of NVX-CoV2373 (Stage 6). Thus, in total, seventeen different combinations of primary and booster COVID-19 vaccines were studied. All participants provided written informed consent before undergoing trial-related activities.

2.2. Binding antibody response assays

Serum bAb levels against the SARS-CoV-2 Spike protein with proline modification (S-2P) and SARS-CoV-2 Spike antigens for strains that emerged 2020–2021 were evaluated with a 96-well Meso Scale Discovery (MSD) Electrochemiluminescence immunoassay analyzer (ECLIA) using a fit-for-purpose 10-plex [30]. The primary study outcome, assayed with a 384-well MSD 4-plex ECLIA, version 2, is reported elsewhere [29,31]. Responses were measured at Day 1 (enrollment; pre-boost), and Days 15, 29, 91, Day 273 (most groups) and Day 366. The analyses reported here focus on Day 1, which reflects response to the primary vaccination series, and Day 29, which reflects putative peak responses 4 weeks post-booster vaccination. Twenty-two SARS-CoV-2 Spike antigens were included in three 10-plex ECLIA panels (panels 13 and 14 for Groups 1 to 9, and panel 23 for all Groups), representing the ancestral SARS-CoV-2 (NC_045512) and viruses from successive lineages including Alpha, Beta, Eta, Iota, Kappa, Delta, Lambda, Gamma, Zeta, Theta, and Omicron (BA.1) (Table S2). Samples were run at four dilutions (1:100, 1:500, 1:2500 and 1:10,000) and the bAb level was reported as the area under the curve (AUC). Some groups (10–17) had only one version of the ECLIA assay run thus generating antibody data for a subset of eight antigens (Table S2).

2.3. Neutralizing antibody response assays

SARS-CoV-2 neutralizing antibody responses were measured at Day 1 and Day 29 using a validated pseudovirus neutralization assay [32,33]. Neutralization was assessed using Spike-pseudotyped viruses in 293 T/ACE2 cells as a function of reductions in luciferase (Luc) reporter activity. Pseudotyped lentiviruses representing the ancestral SARS-CoV-2 Spike (D614G) and the Delta (AY.4.2), Beta (B.1.351) and Omicron (BA.1, BA4/5) variants were used. Responses were measured by the serum inhibitory dilution required to achieve 50 % neutralization (ID50). Beta responses were measured in a random subset of participants (20 per group, equally distributed between age groups and among sites). ID50 values for the D614G Spike were converted to international units (IU50) per milliliter; raw ID50 values were multiplied by 0.242 to convert these to international units. ID50 values for Delta and Beta were analyzed on the raw ID50 scale.

2.4. Proteomic analysis

At Day 1, for each primary vaccination series and each antigen tested in the bAb assay, the proteomic distance between the Spike antigens in the prime vaccine and used in the bAb assay was calculated. Specifically, the “Spike distance” is the Hamming distance between the vaccine insert and the assay antigen, i.e., the number of Spike positions with different amino acids (AA) between the vaccine insert and the antigen. Also, a weighted proteomic distance was calculated to account for the physiochemical properties of different amino acid substitutions (see Supplemental Methods).

At Day 29, for the monovalent vaccine boost groups, Spike distance measures were calculated as above. Note that Spike distances to the boost vaccines would have provided the same result (up to a constant). For Groups 10 and 11 that received the bivalent mRNA-1273.211 boost, we calculated distance between the assay antigen and both the ancestral and beta components of the vaccine boost.

2.5. Statistical analysis

Analyses were conducted in R version 4.4.1.

Participants who were negative for SARS-CoV-2 by anti-nucleocapsid (anti-N) serology at baseline were included in the post-prime (Day 1) analysis. Participants who did not receive booster vaccination, had evidence of infection after enrollment before Day 29, or for whom it was not possible to ensure SARS-CoV-2 negativity before Day 29, were excluded from the analysis of post-boost (Day 29) antibody levels.

At each time point (post-prime and post-boost), antigen-specific bAb levels were evaluated for their association with Spike Hamming distance between antigen and vaccine insert. Population average bAb levels were evaluated by calculating the antigen-specific mean log₁₀ bAb AUC or geometric mean (GM), for each primary vaccine group (at Day 1) and each prime-boost vaccine group (at Day 29). To associate population-average bAb levels and Spike Hamming distances, weighted linear regression within inverse variance weighting was used to account for different numbers of bAb observations per antigen and vaccine group; the mean log₁₀ AUC was modeled with Spike distance as the predictor of interest. The Day 1 analysis adjusted for prime vaccine (mRNA-1273 vs. Ad26.COV2⋅S vs. BNT162b2) and considered an interaction between Spike distance and prime vaccine. Standard measures of explained variation (R² and partial R²) were reported. Results were back-transformed and reported on the geometric mean scale. To associate individual-level bAb levels and Spike distance, a generalized estimating equation (GEE) model with exchangeable working correlation was used to model the mean log₁₀ AUC as a function of Spike distance; robust standard errors were reported and marginal R² was used to measure explained variation [34]. The Day 1 analysis included prime vaccine and an interaction between prime vaccine and Spike distance, as well as individual-level cofactors: time since the last dose of primary vaccination (months), age, and sex at birth. For the Day 29 analyses, the log₁₀ AUC Day 1 ancestral SARS-CoV-2 bAb level was included, in addition to the prime-boost vaccine; the latter was included as both a main effect and an interaction with Spike distance. Vaccine “type” (mRNA vs. vector-based vs. protein-based) was also considered as a more parsimonious way of capturing differences among vaccine groups. Best-fitting models were selected for each type of model (population-average and individual-level), whereby interactions were only included if statistically significant based on two-sided 0.05-level Wald tests.

To enable evaluating the ability of Spike distance between antigen and vaccine insert to predict binding antibody response, the above regression models were fit using bAb data for up to 21 antigens included in the 10-plex ECLIA assay, excluding Omicron BA.1. The predictive capacity of the best-fitting model was evaluated by comparing the observed vs. model-predicted antibody responses to Omicron BA.1 and by evaluating the mean absolute error in log₁₀ AUC.

Antigenic cartography was implemented using the Racmacs package in R. The approach uses multidimensional scaling to describe variation in antibody responses to different antigens [35]. Individual-level bAb data were included for participants with responses to all 22 Spike antigens, i.e. for Groups 1–9 only.

3. Results

Primary results from the DMID-210012 “MixNMatch” study have been previously published [29,31]. Briefly, a total of 761 participants who received a primary COVID-19 vaccination series were enrolled sequentially in six stages (Fig. S1). The majority (65 %) received an mRNA COVID-19 vaccine: 29 % received two doses of mRNA-1273 (Moderna) and 37 % received two doses of BNT162b2 (Pfizer). 35 % received Ad26.COV2⋅S (Janssen) (94 % received a single dose prior to enrollment) (35).

The vast majority of MixNMatch participants were included in these analyses which restrict to those without knowledge or documentation of prior SARS-CoV-2 infection based on anti-Nucleocapsid serology: 728 (95.7 %) were included in the post-prime analysis set and 686 (90.1 %) in the post-boost analysis set (Fig. S1). Participants in the post-prime analysis set had a mean age of 51 years at enrollment and 50 % were assigned male sex at birth (Table S1). For participants who received the Ad26.COV2⋅S vaccine, a mean of 4.8 months (SD = 1.5) had elapsed since last vaccination, whereas 5.4 months (SD = 1.9) had elapsed since the prime vaccination series for those who received mRNA-1273, and 5.8 months (SD = 2.0) had elapsed for those who received Pfizer BNT162b2.

3.1. Variable binding antibody responses across individuals following primary vaccination

Antibody responses post-primary vaccination (at Day 1) were highly variable across participants (Fig. 1 A-C), with higher within-antigen variability among Ad26.COV2⋅S recipients than among BNT162b2 or mRNA-1273 recipients. Responses were generally about 1 log₁₀ higher for participants primed with mRNA vaccines vs. Ad26.COV2⋅S. Binding Ab responses tended to decrease over time since last primary vaccination, with a more pronounced decline for those who received mRNA primary vaccination. (Fig. 1D).

Fig. 1.

SARS-CoV-2 binding antibody responses post-prime. Post-prime (Day 1) responses for participants who received Ad26.COV2⋅S (A), mRNA-1273 (B), or BNT162b2 (C) primary COVID-19 vaccination. Responses to 22 different SARS-CoV-2 Spike antigens are shown, including the ancestral “WT” antigen. Within-antigen geometric mean (GM) and geometric standard deviation (GSD) are inlaid. (D) Binding antibody responses to the ancestral WT antigen post-primary vaccination post-prime by primary vaccination and weeks since the last dose of the primary series.

3.2. Variable primary vaccination antibody responses across antigens

Binding Abs induced by primary vaccination were highest to the ancestral SARS-CoV-2 variant and tended to be lower for emergent variants (Fig. 1A-C; antigens described in Table S2). Omicron BA.1 responses were nearly a full log₁₀ lower on average than responses to the ancestral SARS-CoV-2 virus. Antigenic cartography maps show these responses in a two-dimensional space, recapitulating how well the priming vaccines cover the ancestral Spike antigen and provide less coverage of the emergent variants, and how antigenically distinct the Omicron BA.1 Spike is from to the earlier variants (Fig. S2).

3.3. Inverse correlation between geometric mean antibody response and sequence-based Spike distance to primary vaccination

Spike distance between antigen and prime vaccine was variable, ranging from 0 to 25 AAs for most antigens (Fig. S3). The exception is Omicron BA.1 which had a much larger distance of 39–43 AAs across prime vaccines. Physicochemical-weighted Hamming distances were very highly correlated with (unweighted) Hamming distances (Fig. S4) and therefore all subsequent results focus on the unweighted Hamming distance metric.

When considering population average bAb levels, the geometric mean (GM) bAb level post-prime (at Day 1) to a given Spike antigen was strongly inversely correlated with the Spike distance from the prime vaccine (Fig. 2A). With each 10 AA increment increase in Spike distance there was an estimated 36 % reduction in GM AUC (95 % CI: 30 % to 43 %). This association was consistent across mRNA vs. vector-based prime vaccine regimens (p-value for interaction = 0.39).

Fig. 2.

Association between geometric mean (GM) SARS-CoV-2 binding antibody response to a given Spike antigen and Spike distance of the antigen to the COVID-19 vaccine post-prime (Day 1). (A) Association for each primary vaccination group; (B) Predicted and observed GM binding antibody response to Spike Omicron BA.1. 95 % prediction intervals are shown.

Virtually all variability in the post-prime GM bAb levels across antigens and groups was explained by the primary vaccination series and Spike distance (R² = 97 %). Both the primary vaccination series and Spike distance contributed to explaining variability in GM binding antibodies (partial R² for vaccine = 0.96, partial R² for Spike distance = 0.57).

3.4. Spike distance is an excellent predictor of geometric mean prime vaccine antibody response

We used the Omicron BA.1 bAb level to evaluate the predictive ability of the Spike distance. Spike distance and primary vaccination series provided excellent prediction of the post-prime GM antibody response to Omicron BA.1 for each primary vaccine regimen (Fig. 2B). Predicted vs. observed Omicron BA.1 GM AUCs were 403 vs. 317 among Ad26.COV2⋅S recipients, 3325 vs. 3764 among mRNA-1273 recipients, and 1759 vs 1942 among BNT162b2 recipients. The mean absolute prediction error in log₁₀ AUC was 0.044, corresponding to an approximately 11 %-fold error in AUC.

3.5. Explaining variability in individual-level prime antibody responses

We next interrogated cofactors that may explain the large between-individual variability in bAb levels following primary vaccination. The strongest cofactors were the primary vaccine regimen, time since primary vaccination, age, and Spike distance (Fig. 3). The decline in antibody response in association with increasing Spike distance differed by prime vaccine (p < 0.001 for interaction). Each 10-AA increment in Spike distance was associated with a 34 % lower GM AUC (95 % CI: 33 % to 35 %) among mRNA-1273 and BNT162b2 vaccine recipients. A more pronounced decline was observed among Ad26.COV2⋅S recipients: each 10-AA increment in Spike distance was associated with a 43 % lower GM AUC (95 % CI: 41 % to 44 %). Older age at enrollment and male sex were also associated with lower antibody responses. Each additional month since prime vaccination was associated with an 11 % reduction in GM AUC (95 % CI: 8 % to 13 %). As expected, individual-level responses had more unexplained variability than population-average immune responses: the model explained 59 % of the variability in individual-level antibody responses compared to 97 % of variability explained in average antibody responses.

Fig. 3.

Results of individual-level regression analysis of post-prime binding antibody responses. Individual post-prime (Day 1) log AUC values were modeled using a linear generalized estimating equations (GEE) model with exchangeable working correlation and robust sandwich standard errors. Geometric mean (GM) ratio per increment of a continuous factor, or, for a categorical factor, for a given factor level relative to the reference level, with 95 % confidence interval.

In accordance with the larger level of unexplained variability, the prediction of individual-level Omicron BA.1 bAb levels was less accurate than for population-average antibody responses (Fig. S5). The average absolute prediction error was 0.30 log₁₀ AUC, much larger than the 0.04 log₁₀ AUC average prediction error for population average BA.1 antibody responses.

3.6. Post-boost responses were less variable across groups receiving both homologous and heterologous COVID-19 vaccination

Binding antibody responses 28 days post-booster-vaccination (at Day 29) were high in all prime/boost groups (Fig. 4). Boosting reduced variability in antibody responses to the different SARS-CoV-2 antigens: GM levels to Omicron BA.1 were 1.7- to 5.8-fold lower than those to the ancestral SARS-CoV-2 Spike. Antigenic cartography illustrates the reduced variability in responses across Spike antigens post-booster vaccination (Fig. S6).

Fig. 4.

SARS-CoV-2 binding antibody responses post-boost. Post-boost (Day 29) responses for participants by prime vaccine (row) and boost vaccine (column). Responses to 22 different SARS-CoV-2 Spike antigens are shown, including the ancestral “WT” antigen. Within-antigen geometric mean (GM) and geometric standard deviation (GSD) are inlaid.

Lower between-individual variability was observed post-boost vs. post-prime among participants who received an mRNA vaccine prime or boost. In contrast, among participants who did not receive an mRNA prime or boost, between-individual variability post-boost was comparable to that post-prime.

3.7. Spike distance is modestly inversely correlated with the geometric mean post-boost response

The association between Spike distance and population-average bAb level post-booster vaccination (at Day 29) was modest: each 10-AA increment in Spike distance was associated with an estimated 17 % reduction in GM AUC (95 % CI: 14 % to 20 %; Table S4). The association was weaker than post-primary vaccination (estimated reduction = 36 % vs. 17 % per 10-AA difference; Fig. 5A-C). The association was consistent across mRNA prime-boost regimens and non-mRNA regimens (p-value for interaction = 0.34). The model explained 83 % of the variability in the average post-boost antibody response, with Spike distance explaining 15 % of the variability.

Fig. 5.

Association between geometric mean (GM) SARS-CoV-2 binding antibody response to a given Spike antigen and Spike distance of the antigen to the COVID-19 vaccine post-boost (Day 29). (A) Association for all prime-boost groups; (B) Association for Ad26.COV2⋅S primed groups; (C) Association for mRNA primed groups. (D) Predicted and observed geometric mean binding antibody responses to Spike Omicron BA.1. 95 % prediction intervals are shown.

3.8. Spike distance is a modest predictor of average post-boost antibody response

Geometric mean post-boost responses to Omicron BA.1 were more modestly predicted by the regression model than were post-prime responses (Fig. 5D). The mean absolute prediction error in log₁₀ AUC was 0.122, compared to the 0.044 prediction error for post-prime responses. Lower-accuracy prediction was observed for regimens with Ad26. COV2⋅S priming absent mRNA boosting.

3.9. Explaining variability in individual-level post-boost antibody responses

The cofactors most strongly associated with individual-level antibody responses to the boost vaccination were pre-boost (Day 1) bAb level, time between prime and boost, and prime-boost vaccine regimen (Fig. 6). A 1-log₁₀ higher post-prime AUC to the ancestral strain was associated with an estimated 69 % higher post-boost AUC (95 % CI: 51 % to 89 %). One additional month between prime and boost was associated with a 10 % increase in AUC post-boost (95 % CI: 7 % to 13 %). Weak or null associations were observed for age and sex. The association between Spike distance and antibody response differed across prime-boost regimens (p-value for interaction <0.0001), with the strongest association in participants who received non-mRNA prime and boost. Across antigens, each 10-AA increase in Spike distance was associated with an estimated 36 % reduction in AUC for the Ad26.COV2⋅S prime-boost regimen (95 % CI: 34 % to 38 %) and a 32 % reduction for the Ad26. COV2⋅S prime/NVX-CoV2373 boost regimen (95 % CI: 25 % to 38 %). A more attenuated association was observed for participants who received at least one mRNA vaccine (mRNA-1273 or BNT162b2) as prime, boost, or both, with estimated AUC reduction per 10 AA ranging between 11 % to 24 %. Together, the individual-level cofactors and antigen Spike distance explained 74 % of the variability in post-boost binding Ab responses, a smaller fraction than for average antibody responses for which 83 % of variability was explained.

Fig. 6.

Results of individual-level regression analysis of post-boost binding antibody responses for groups receiving monovalent COVID-19 vaccines. Individual post-boost (Day 29) log AUC values were modeled using a linear generalized estimating equations (GEE) model with exchangeable working correlation and robust sandwich standard errors. Geometric mean (GM) ratio per increment of a continuous factor, or, for a categorical factor, for a given factor level relative to the reference level, with 95 % confidence interval.

Individual-level post-boost bAb levels were more accurately predicted by the modeling than were individual-level responses post-prime (Fig. S7). The mean absolute prediction error in log₁₀ AUC was 0.19, compared to 0.30 post-prime.

3.10. Less variability in post-boost antibody responses for bivalent vaccine groups

Two groups of participants received bivalent mRNA-1273.211 boost vaccination, with ancestral and Beta components (Fig. 4). Post-boost bAb levels to the ancestral SARS-CoV-2 were similar for those who received BNT162b2 vs. Ad26.COV2⋅S prime (GM = 61,059 = 10^4.8 vs. 50,044 = 10^4.7), and similar to those in groups boosted with other mRNA vaccines (GM = 53,575 = 10^4.7 or 46,397 = 10^4.7 given mRNA-1273 boost following BNT162b2 or Ad26.COV2⋅S prime; GM = 48,564 = 10^4.7 or 42,834 = 10^4.6 given BNT162b2 boost following BNT162b2 or Ad26.COV2⋅S prime). Responses to the B.1.351 antigen, which nearly matches the Beta component of the bivalent vaccine (4-AA difference), were similarly high in the two groups (GM 43115 = 10^4.6 vs. 31,244 = 10^4.5) and again similar to those for groups boosted with monovalent mRNA vaccines (GM = 34,752 = 10^4.5or 26,285 = 10^4.4 given mRNA-1273 boost following BNT162b2 or Ad26.COV2⋅S prime; GM = 31,159 = 10^4.5 or 25,781 = 10^4.4 given BNT162b2 boost following BNT162b2 or Ad26.COV2⋅S prime).

Variability in responses across antigens was lower in the bivalent groups compared to the monovalent groups. GM levels of B.1.529 (Omicron BA.1) were lowest, but only 1.7–2.2-fold lower than responses to ancestral SARS-CoV-2. Variability in antibody responses between individuals was similar for participants who received the bivalent vaccine boost and those who received the monovalent mRNA vaccine boost.

3.11. Spike distance was weakly associated with post-boost antibody response for bivalent boost groups

Based on regression analysis of individual-level post-boost antibody responses for groups receiving the bivalent boost, the strongest cofactors of the post-boost bAb level were the post-prime bAb level, the prime-boost interval, and the Spike distance to each component of the boost vaccine (Fig. S8). A 1-log10 higher post-prime AUC to the ancestral SARS-CoV-2 Spike was associated with an estimated 38 % higher GM AUC post-boost (95 % CI: 16 % to 64 %). One month longer interval between prime and boost vaccination was associated with a 7 % higher GM AUC (95 % CI: % to 11 %). Each 10 AA increase in Spike distance to the ancestral component of the bivalent boost vaccine was associated with a 2.2 % lower GM AUC (95 % CI: 2.0 % to 2.5 %). A 10 AA increase in Spike distance from the Beta vaccine component was associated with a 1.0 % increase in GM AUC (95 % CI: 0.9 % to 1.1 %). Younger age was weakly associated with higher antibody response (AUC GMR = 0.97; 95 % CI: 0.94 to 0.99). Taken together, the model explained 45 % of the variability in the antibody response.

3.12. Binding antibody responses are moderately correlated with neutralizing antibody responses

Primary and booster COVID-19 vaccinations induced neutralizing antibody responses that were moderately correlated with the bAb levels to the same SARS-CoV-2 lineage and measured at the same timepoint (Fig. S9). Whereas neutralizing antibody responses had a large range, bAb levels exhibited a ceiling effect, likely due to signal saturation in the binding antibody assay.

4. Discussion

In this study examining a cohort previously unexposed to SARS-CoV-2 and vaccinated with a variety of COVID-19 vaccines — including mRNA, vector, and protein-based vaccines, homologous and heterologous prime-boost regimens, and monovalent and bivalent vaccines — we documented strong SARS-CoV-2-specific bAb levels to the ancestral SARS-CoV-2 Spike, and somewhat lower responses to SARS-CoV-2 variants that emerged in 2020 and 2021, with the lowest responses generated in response to the Omicron BA.1 Spike and greater breadth post-boost. Given the strong correlation between bAb levels and risk of both mild and severe COVID-19 [8,9,36,37], quantifying these responses is critical for understanding the level of vaccine efficacy conferred.

At a population level, after a primary vaccination with mRNA-1273, BNT162b2, or Ad26.COV2⋅S COVID-19 vaccines, anti-Spike bAb levels were on average 36 % lower per 10-AA increase in distance between the vaccine Spike and assay antigen Spike (95 % CI: 30 % to 43 %). This association was similar to that seen in analyses of post-COVID-19 immune responses among mRNA-1273 vaccine recipients in the COVE vaccine trial [38]. The Spike distance, together with the prime vaccine regimen, was able to nearly perfectly predict the geometric mean bAb level to Omicron BA.1. These two factors alone explained 96 % of the variability in the geometric mean bAb level. After boosting with mRNA-1273, BNT162b2, Ad26.COV2⋅S, or NVX-CoV2373, binding antibodies were higher for all homologous and heterologous prime-boost regimens. Differences in antibody geometric means across regimens and antigens were smaller as compared to prior to the boost. Spike distance was weakly associated with the average post-boost antibody response: each 10-AA increase was associated with a 17 % decrease in geometric mean anti-Spike bAb level (95 % CI: 14 % to 20 %). These findings suggest that after primary vaccination in a previously SARS-CoV-2 naïve population, proteomic sequences of emergent variants may provide excellent prediction of antibody responses to variants. In contrast, antibody responses shortly following booster vaccination are less well predicted. In a previously-vaccinated population, antibody responses likely need to be measured in order to capture any loss of antigenicity; proteomic sequences appear unlikely to suffice on their own.

Our multivariate analyses of individual-level bAb levels identified cofactors that were strong predictors of the response to primary vaccination. After adjusting for time since the last dose, bAb levels were substantially lower for the one-dose Ad26.COV2⋅S vaccine vs. two doses of mRNA vaccines, mRNA-1273 or BNT162b2. This is consistent with prior literature on immunogenicity of different COVID-19 vaccines [39–41]. Additionally, binding antibodies post-primary vaccination were lower on average among older individuals and men, also consistent with prior literature [42–44]. The effect of aging on immunity has been documented for vaccines against many other pathogens, with age-associated reductions in the size and function of the germinal center response a potential mechanism [45,46]. SARS-CoV-2 Spike antibody levels were strongly inversely associated with distance of the Spike antigen to the vaccine strain, with a larger decrement in response for antigens further from the vaccine seen among individuals vaccinated with Ad26.COV2⋅S versus mRNA vaccine. That a differential decline in antibody level by vaccine type was detected in the individual-level analysis, but not the population-level analysis, is potentially attributable to the greater power of the individual-level analysis, owing to a much larger number of observations. Importantly, even after accounting for the above factors, 41 % of the variability in individual-level bAb levels remained unexplained.

Following COVID-19 booster vaccination, the level of bAbs varied across prime-boost regimens. In particular, responses were lower in the absence of an mRNA vaccine prime or boost, as documented previously [47–49]. The increased breadth post-boost may be due to an increase and expansion of memory B cell clones [50,51]. Individuals with lower antibody responses post-primary vaccination tended to have lower responses post-boost vaccination, suggesting unmeasured individual-level factors that lead some individuals to have stronger antibody responses to the same vaccine than others. A longer interval between prime and boost was also associated with a higher antibody response post-boost, a finding consistent with other observational studies [52–55]. Older age and male sex were not found to associate significantly with lower antibody responses post-boost. Spike distance was inversely associated with individual-level post-boost bAb levels; however, this factor was a weaker predictor of the response relative to the cofactors listed above. The diminution in antibody responses with increasing Spike distance was greatest given non-mRNA prime-boost vaccination. Individual-level post-boost antibody responses were variable, and 26 % of the variability in response was left unexplained. These results suggest that there are additional factors beyond those measured here that mediate immune responses to booster COVID-19 vaccines.

We found that boosting with the bivalent mRNA-1273.211 vaccine induced bAbs to the ancestral SARS-CoV-2 Spike that were similar in magnitude to those induced by monovalent mRNA-1273 or BNT162b2 booster vaccination. Bivalent vaccination also induced similar-level responses to the B.1.351 Beta antigen included in the bivalent vaccine, as well as similar responses to Omicron BA.1. Thus, the bivalent mRNA vaccine construct had lower variability across antigens than the monovalent booster vaccine responses, suggesting the potential for improved breadth. Though the bivalent vaccine did not appear advantageous over the monovalent in terms of increased binding to the Beta antigen, bAbs against Omicron were relatively better preserved. Prior studies of the immunogenicity of mRNA-1273.211 found higher binding and neutralizing antibodies for the bivalent vs. monovalent mRNA booster among individuals primed with the mRNA-1273 vaccine; the comparable responses we saw between bivalent and monovalent vaccination might reflect small advantages in heterologous compared to a homologous booster [56].

Our study has several limitations. It is unclear how well these findings apply outside the context of the MixNMatch study. In particular, the results apply only to individuals without evidence of previous SARS-CoV-2 infection. How well the results apply to the current population, including individuals with hybrid immunity acquired through multiple infections, is unclear. Further, many individuals have now had multiple COVID-19 booster vaccinations; our results suggest boosters tend to reduce variability in responses to different antigens, but this remains to be demonstrated in populations with more than a single booster vaccination. This analysis considered antibody responses to COVID-19 boosters 28 days post-boost. It is possible that Spike distance would be a better predictor of antibody response further from the time of the boost, given the greater variation in antibody levels that would be expected. We also leveraged for this analysis a fit-for-purpose bAb assay that was unable to quantify high responses given the serum dilutions that were tested. The censoring of high antibody responses may have artificially created reduced variability in responses, especially early after booster vaccination. The associations we found between COVID-19 vaccine antibody responses and individual cofactors are subject to confounding by factors whose influence on antibody responses was not measured. Finally, we evaluated responses to the ancestral SARS-CoV-2 and variants through Omicron BA.1, but we did not measure responses to more recent Omicron sub-lineages. It is unclear how the results can be extrapolated to these new, more genetically distant sub-lineages.

As SARS-CoV-2 continues to evolve, the results of this analysis are pertinent to decision-making regarding adapting COVID-19 vaccines to include new variants. These data suggest that both sequencing of emergent viruses, as well as measurement of serum antibodies in vaccinated persons, will be necessary to predict the protection afforded by vaccination. Moreover, the antibody responses are variable, not adequately predicted by viral sequences or other cofactors, and these findings invite research investigating the factors underlying this heterogeneity.

Appendix A. Supplementary data

Supplementary data to this article can be found online at https://doi.org/10.1016/j.vaccine.2025.127738.

Data availability

Data will be made available on request.