Polysaccharide and conjugate vaccines to Streptococcus pneumoniae generate distinct humoral responses

Abstract

Streptococcus pneumoniae is a major cause of community-acquired pneumonia, bacteremia, and meningitis in older adults worldwide. Two pneumococcal vaccines containing S. pneumoniae capsular polysaccharides are in current use: the polysaccharide vaccine PPSV23 and the glycoconjugate vaccine PCV13. In clinical trials, both vaccines elicit similar opsonophagocytic killing activity. In contrast to polysaccharide vaccines, conjugate vaccines have shown consistent efficacy against nasopharyngeal carriage and noninvasive pneumonia overall and for some prevalent individual serotypes. Given these different clinical profiles, it is crucial to understand the differential immunological responses induced by these two vaccines. Here, we used a high-throughput systems serology approach to profile the biophysical and functional features of serum antibodies induced by PCV13 and PPSV23 at 1 month and 1 year. In comparison with PPSV23, PCV13 induced higher titers across antibody isotypes; more durable antibody responses across immunoglobulin G (IgG), IgA, and IgM isotypes; and increased antigenic breadth. Although titers measured in opsonophagocytic activity (OPA) assays were similar between the two groups, confirming what was observed in clinical studies, serum samples from PCV13 vaccinees could induce additional non-OPA antibody-dependent functions, including monocyte phagocytosis and natural killer cell activation. In a multivariate modeling approach, distinct humoral profiles were demonstrated in each arm. Together, these results demonstrate that the glycoconjugate PCV13 vaccine induces an antigenically broader, more durable, polyfunctional antibody response. These findings may help explain the increased protection against S. pneumoniae colonization and noninvasive pneumonia and the longer duration of protection against invasive pneumococcal disease, mediated by PCV13.

INTRODUCTION

Two types of vaccines against Streptococcus pneumoniae disease are in current use in adults. The first, the 23-valent, adjuvant-free pneumococcal polysaccharide vaccine (PPSV23), contains 25 μg each of unconjugated polysaccharide antigens from 23 serotypes. Although its effectiveness has been demonstrated for invasive pneumococcal disease outside of early childhood, it has many limitations (1, 2). It does not provide protective immunity in children under 2 years of age (3, 4), has relatively short duration of immunity (5, 6), and has not been demonstrated to protect against carriage at any age (7). Efficacy against vaccine serotype pneumonia has not been demonstrated. Despite being studied in several randomized controlled trials (RCTs), efficacy against clinical or all-cause pneumonia in older adults has only been demonstrated in a single trial in Japan (8, 9).

The 7-, 10-, and 13-valent pneumococcal conjugate vaccines (PCV7, PCV10, and PCV13) and two PCVs recently licensed by the Food and Drug Administration for adults, PCV15 and PCV20, comprise an alternative vaccine strategy. These vaccines use only 2 to 4 μg of each polysaccharide antigen, conjugated individually to diphtheria cross-reactive material (CRM₁₉₇) carrier protein, and are adjuvanted with aluminum phosphate. This strategy promotes T cell help in generating immunity. PCV7, PCV10, and PCV13 have demonstrated robust protection against invasive and noninvasive disease including pneumonia, acute otitis media, and nasopharyngeal carriage among children; further, PCV13 protected against adult vaccine-type invasive disease and bacteremic and nonbacteremic pneumonia in an RCT in The Netherlands (10–13). Although no RCT has directly compared the efficacy of PCV vaccines with that of PPSV23 in adult populations, one study in Louisville, KY, among persons aged 65 or older reported vaccine effectiveness of 71% for PCV13 against vaccine-type pneumonia (14), whereas a second study reported effectiveness of 17% for PPSV23 against pneumonia for PPSV23-unique serotypes (15). Similar paired data from Germany and the United States reported effectiveness against all-cause pneumonia in older adults of 12 and 8.8% for PCV13 compared with 3 and 1% for PPSV23 (16–18).

Mechanistic studies of S. pneumoniae immunity have traditionally focused on opsonophagocytic activity (OPA), which measure bacterial killing mediated by the binding of serum antibodies, opsonization of complement, and killing by neutrophil phagocytosis. Immunogenicity measured by OPA is used for pneumococcal vaccine licensure in adults (19). However, there is no established OPA threshold for protection nor has the clinical relevance of antibody-mediated functions other than OPA been thoroughly examined. Modest but statistically significant increases in total immunoglobulin G (IgG) and OPA titers have been noted at 1 month after vaccination in PCV13 vaccinees compared with PPSV23 vaccinees (20–22). Yet, the disconnect between the major differences in effect size for clinical outcomes and minor differences in quantitative functional and total antibody concentrations suggests that our understanding of the immunologic mechanisms of protection may be incomplete.

Although the immune mechanisms underlying the apparent clinical differences between PCVs and PPSV remain unclear, differences in humoral responses may play a role. Asymptomatic nasopharyngeal carriage serves as the primary reservoir for S. pneumoniae transmission and also precedes pneumococcal disease states (23). Thus, prevention of carriage at the mucosa, by localization of mucosally relevant antibody isotypes, differential downstream effector functions, or differences in anamnestic responses and immune memory, may be one possible mechanism underlying differential clinical responses. In this regard, most carbohydrate antigens are unable to bind major histocompatibility complex class II and thus generate T cell–independent humoral responses, which are traditionally thought to be of low affinity, short-lived, unable to generate effective memory B cells, and restricted to isotypes such as IgG2 and IgM that opsonize effectively but have weaker Fc receptor (FcR)–mediated recruitment of innate immune effector cells (24, 25). Thus, the conjugate strategy used by PCV13 may generate T cell–dependent responses that are higher affinity or use more highly functional antibody subclasses.

To begin to understand the mechanisms underlying the differential clinical responses to the two vaccine platforms, we conducted a descriptive study aimed at defining the differences in the polysaccharide-specific antibody responses induced by each vaccine. We profiled the humoral immune response of adults 60 to 64 years of age randomized to a single dose of either PCV13 or PPSV23 (20). Our approach relied on a systems serology platform, which uses a suite of antibody-profiling assays to capture both the biophysical and functional characteristics of antigen-specific antibodies (26). Two collections of serotypes were evaluated. To broadly interrogate vaccine-induced responses across serotypes, antibody isotypes, subclasses, and downstream FcR binding were measured against a panel of S. pneumoniae polysaccharide antigens. This included serotypes that cause substantial disease in adults despite the herd protection induced by pediatric pneumococcal conjugate vaccine introduction (1, 3, 7F, 19A, and 19F), the single serotype present in PCV13 but not PPSV23 (6A), and two serotypes that led to lower antibody concentrations compared with PCV7 in pediatric PCV13 licensing trials (6B) or to markedly lower post–second versus post–third dose infant response (6B and 23F) (27, 28). To investigate downstream antibody functions, we performed a panel of antibody-dependent functional assays to measure antibody-dependent cellular phagocytosis (ADCP), neutrophil phagocytosis (ADNP), dendritic cell phagocytosis (ADDCP), complement deposition (ADCD), and natural killer (NK) cell activation (ADNKA). These assays were performed targeting the two most common prevalent serotypes in adults (19A and 3) and one serotype unique to PCV13 (6A). Together, these studies create a comprehensive profile of humoral responses to the two vaccines.

RESULTS

PPSV23 and PCV13 induced modest differences in IgG and OPA titers

We performed comprehensive humoral profiling of serotype-specific antibody responses among a cohort of individuals aged 60 to 64 years who received either PCV13 (n = 40) or PPSV23 (n = 40) who were randomly selected from a larger noninferiority trial comparing the two vaccines (20). Prior studies, including the larger study that included this sample set, have shown modestly increased IgG and OPA titers induced by PCV13 compared with PPSV23 (20–22). We first asked whether we could confirm these findings in this smaller subset of individuals. IgG concentrations against a panel of pneumococcal polysaccharides were newly measured using Luminex for all 40 individuals. OPA titers against a smaller panel of pneumococcal polysaccharides in this subset of individuals were gathered from previously published data available at 1 month only (20) in 17 individuals per serotype.

Serotype-specific IgG concentrations rose in response to both PPSV23 and PCV13 at 1 month compared with prevaccine titers (Fig. 1). Across all serotypes, concentrations of IgG at 1 month were similar between the two groups but higher in the PCV13 group for serotypes 19F (P = 0.0263) and 23F (P = 0.0103). Pn6A, a serotype in PCV13 and not in PPSV23, was included as a positive control. Similarly, slightly increased OPA titers against 6B (P = 0.0312) and 23F (P = 0.0446) were observed in PCV13 vaccinees compared with PPSV23 at 1 month after vaccination (fig. S1). Thus, as previously noted in the larger trial (20), the subset of individuals included here exhibited a response to vaccination across both vaccine platforms with increased serotype-specific IgG concentrations and OPA titers.

Fig. 1. IgG responses are generated by PCV13 and PPSV23 vaccination.

The Luminex platform was used to assay antigen-specific responses in IgG against serotypes 1, 3, 6A, 6B, 7F, 19A, 19F, and 23F between PPSV23 (yellow) and PCV13 (blue) vaccinees before vaccination (prevacc; n = 10 for each group), at 1 month (n = 20 for each group), and at 1 year (n = 20 for each group). MFI, median fluorescence intensity. Mann-Whitney U tests were used to compare vaccination arms at each time point. *P < 0.05, **P < 0.01, and ****P < 0.0001.

Longitudinal differences were observed using systems serology profiling of serum from recipients of PCV13 versus PPSV23

Beyond titer and opsonization, antibodies can protect against pathogens by selectively leveraging the innate immune system by differential isotype, subclass, and Fc glycoform selection (29). To gain a broad understanding of overall qualitative differences in the humoral immune response that could explain clinical differences across the vaccines, we therefore next used a panel of “systems serology” assays to comprehensively characterize pneumococcal specific antibodies (26). We measured responses against S. pneumoniae polysaccharide antigens from serotypes 1, 3, 6A, 6B, 7F, 19A, 19F, and 23F and the CRM carrier protein, capturing IgG subclass, IgA subclass, IgM, and Fcγ receptor (FcγR)–and Fcα receptor (FcαR)–binding antibody concentrations across the vaccine groups. In addition, we performed a panel of antibody-dependent functional assays to measure ADCP, ADNP, ADDCP, ADCP, and ADNKA targeting serotypes 3, 6A, and 19A.

Systems serological analysis of vaccine-induced antibody profiles highlighted an increase in antibody responses across antibody isotypes, subclasses, and FcR binding, present in both vaccine groups 1 month after vaccination and declining by 1 year (Fig. 2A). Specifically, distinct responses were noted vaccinees across most of the measured antibody features, as well as in the functional assays (ADCD, ADNP, ADCP, and ADNKA). Moreover, responses appeared qualitatively more persistent in PCV13 vaccinees at 1 year after vaccination (Fig. 2A). These data point to previously unappreciated differences in the vaccine-induced functional humoral immune response induced by PCV13 versus PPSV23.

Fig. 2. Systems serology identified differential responses to PCV13 and PPSV23.

(A) The heatmap represents all data generated, including Ig titer of IgG subclasses, IgA subclasses, and IgM; Fcɣ and α receptor binding; and antibody (Ab)–dependent functions such as ADCD, ADCP, ADNP, ADNKA, and ADDCP (PPSV23, n = 10; PCV13, n = 10. Each row represents an individual at a given time point (tp). Raw data for isotypes, subclasses, and ADCD were log₁₀-transformed, and all data were then z-scored. (B) Data for each assay measuring a given antigen-specific antibody feature were z-scored across all tested samples PPSV23, and changes in z score between prevaccination and 1-month or 1-year postvaccination time points were calculated for all individuals (PPSV23, n = 10; PCV13, n = 10). The heatmap represents the median change in z score for the indicated group and antibody feature. Statistical significance comparing change from prevaccination time point was measured using the Wilcoxon signed-rank test followed by BH correction for multiple comparisons. *q < 0.05 and **q < 0.01. (C) Representative plots are shown for response changes to serotype 19A in PPSV23 (yellow, n = 10 and PCV13 (blue, n = 10) vaccinees over the study period. Corrected q values are shown. mo, month; yr, year.

To gain a more granular understanding of the differences between the vaccines at 1 month and 1 year after vaccination, vaccine-induced immune responses were plotted in relation to preexisting baseline pneumococcal antibody profiles (Fig. 2B, with representative plots for serotype 19A shown in Fig. 2C). At 1 month after vaccination, both vaccines induced robust IgG responses, marked by high concentrations of IgG2. In addition, elevation of the mucosally relevant IgA isotype across serotypes was observed for both vaccines, including the induction of both IgA1 and IgA2. However, PCV13 broadly demonstrated greater magnitude of responses in IgG subclasses IgG1, IgG2, and IgG3 than PPSV23, as well as greater magnitude of IgA and IgM responses. In PCV13 vaccinees only, antibodies binding FcγR2A, FcγR2B, and FcγR3B remained elevated from the prevaccination time point (q < 0.05) across all serotypes.

At the 1-year time point, both groups exhibited a decline in the number of antibody features that nevertheless remained increased from their prevaccination baseline concentrations. IgG2 against multiple antigens remained elevated in both groups, with persistent elevation of IgG1 across multiple serotypes uniquely present in PCV13. Both groups maintained some increased IgA1 and IgA2 responses, and only PCV13 showed any persistent elevation of IgM at this time point. Similarly, binding of FcγR2A, FcγR2B, and FcαR persisted in both groups. Together, these data indicate that both vaccines generated responses with similar isotype and subclass selection profiles including IgG2, IgA1, IgA2, and IgM; however, differences between vaccines were observed including increased IgG1 and IgG3 responses, broader binding of FcRs, and persistence of measured responses up to 1 year after vaccination in PCV13 vaccinees.

Differential maturation of IgG responses was induced by PCV13 and PPSV23 vaccination

Although polysaccharide antigens alone typically generate low-affinity T cell–independent responses, glycoconjugate vaccines are well established to elicit T cell help, generating higher-affinity, long-lived, class-switched antibody responses to encapsulate bacteria (30). Prior work in macaques has suggested that conjugate vaccination generates increased breadth of antibody response (31). However, the antigenic breadth of responses to PCV13 and PPSV23 remains unexamined in humans, as do differences in avidity and class switching. We thus investigated whether these properties differed between vaccinees in this cohort.

To quantify the antigenic breadth of responses in each individual, we created a breadth score. This score was defined as the number of shared serotypes, from 0 to 7, for which an individual had a given antibody response above the median of the full cohort (Fig. 3A). PCV13-exclusive antigens CRM and serotype 6A were excluded from this analysis, as they would amplify responses in PCV13 exclusively. At 1 month after vaccination, a third of PCV13 vaccinees had a maximum breadth of 7 across antibody features, with statistically significant increases in IgM breadth compared with PPSV23 vaccinees (P = 0.0014). At 1 year, the increased antigenic breadth in PCV13 vaccinees persisted, with statistically significantly higher IgG1 (P = 0.0034) and IgG4 (P = 0.0157) breadth in PCV13 vaccinees compared with individuals who received the PPSV23 vaccine.

Fig. 3. PCV13 and PPSV23 responses demonstrate differential antigenic breadth and class switching.

(A) A breadth score was calculated for every individual, indicating the number of shared antigens from 0 to 7, for which that individual had antibody responses that were above the median for the whole cohort. CRM and 6A were excluded from this analysis. Each row indicates an individual (PPSV23, n = 20; PCV13, n = 20), with breadth score indicated by color on heatmap. The mean breadth score for each group is shown above the heatmap. A Mann-Whitney test was used to calculate difference in breadth score between groups at each time point. *P < 0.05 and **P < 0.01. (B) The IgG2 avidity index was measured for each individual (PPSV23, n = 20; PCV13, n = 20) and serotype. Luminex was used to measure the IgG2 concentrations with and without the presence of a chaotropic agent, 7 M urea. Avidity index was calculated as ratio of MFI in the presence of urea to that without urea. P values were calculated using a Mann-Whitney test and those <0.05 are shown. (C) Serotype-specific antibody concentrations of IgG1, IgG2, IgG3, IgA1, and IgM were measured by Luminex (PPSV23, n = 20; PCV13, n = 20). Spearman correlations were then calculated within each vaccine group at 1-month and 1-year time points. The heatmaps represent Spearman correlation coefficients between all indicated antibody and isotype concentrations for PPSV23 vaccinees at 1 month and 1 year (left) and PCV13 vaccinees at 1 month and 1 year (right).

We next hypothesized that CD4⁺ T cell help could improve antibody affinity maturation and improve binding to pneumococcal antigens. Thus, we next evaluated whether we could detect differences in antibody avidity between the two vaccine groups using an established method for measuring antibody avidity (32). We defined avidity index as the fraction of antigen-antibody binding that remained after incubation with a chaotropic agent, 7 M urea (Fig. 3B). Because the predominant vaccine response was in IgG2 response for both vaccines, we focused on the avidity of IgG2 responses at 1 month after vaccination. A wide range of high- and low-avidity responses were present across individuals, with limited differences across most shared serotypes, with the exception of elevated serotype 3–specific avidity in PCV13-immunized individuals.

CD4⁺ T cell help plays a critical role in antibody class switch recombination (33). Thus, we lastly explored the antibody isotype and subtype selection profiles within each vaccine group, by constructing correlation networks across measured isotypes and subclasses targeting all tested serotypes within each vaccine group (Fig. 3C). Although both vaccines generated an IgG2-dominant response (Fig. 2B), this analysis revealed distinct underlying isotype and subclass correlation structures across the vaccine responses (Fig. 3C). For PCV13, the dominant IgG2 response correlated closely with IgG1 but correlated negatively with IgG3 and IgM at both 1 month and 1 year. Increasing evidence suggests that IgM to IgG class switching occurs sequentially, beginning at the start of the IGH locus with IgG3 and then to IgG1, IgG2, and IgG4 (34, 35). Thus, this finding suggests class switching out of IgM and IgG3 and into later subclasses. This relative exclusion of IgG3 persisted 1 year after immunization with the additional exclusion of IgM. Conversely, PPSV23 immunization induced a diffuse humoral immune response across most serotypes and across antibody subclasses and isotypes at both 1 month and 1 year after immunization, with no evident signatures of coordinated or structured class switching. Collectively, these findings suggest a unique presence of robust class switching in PCV13 vaccinees. Thus, despite the elevated IgM and IgG3 responses observed at a univariate level after PCV13 vaccination, these data point to a coordinated and progressive shift from IgM and IgG3 toward IgG1 and IgG2 responses, marking a functional maturation of the humoral immune response using the conjugate vaccine, representing a possible signature of CD4⁺ T cell help–driven selection of more functional subclasses. Together, these data demonstrate antigenically broader pneumococcal serotype coverage over time in PCV13 vaccinees, in collaboration with highly functional antibody subclass switching from early IgM and IgG3 responses to robust IgG1- and IgG2-mediated immunity. We were unable to identify major differences in avidity of the antibodies induced by PCV13 compared with PPSV23.

OPA is differentially driven by IgG subclasses in PCV13 and PPSV23 serum samples

OPA is used as the regulatory criterion for licensure of adult PCVs. Given the differences in antibody isotype, subclass, FcR binding, and functional differences induced by PCV13 versus PPSV23, we next aimed to determine whether OPA activity was driven by similar or distinct antibody subpopulations between the vaccine groups. To determine which antibody subclasses were likely to be primarily responsible for OPA, we performed Spearman correlations between previously obtained OPA results (20) and vaccine-induced antibody isotype and subclass titers (Fig. 4, A and B). As very limited OPA data were available for the prevaccination and 1-year time point, only the 1-month time point was included in this analysis. IgG titers were tightly linked to OPA activity across both PCV13 and PPSV23, in line with previously published reports using enzyme-linked immunosorbent assay (ELISA) (36). However, the underlying IgG subclasses that correlated with OPA titers were distinct between the two vaccines. IgG2 was linked to OPA activity across both vaccines, consistent with high IgG2 induction after both vaccines (Fig. 2B). However, IgG1 responses were tightly linked to OPA only in PCV13-immunized individuals, whereas IgG3, IgG4, and IgM were linked to OPA in PPSV23-immunized individuals. FcR binding activity, particularly to the activating FcγR2A and inhibitory phagocytic FcγR2B, correlated tightly with OPA activity across both vaccines, highlighting the critical role of specific FcRs in driving OPA activity. These correlations raise the possibility that complex combinations of antibodies and differences in subclasses may drive OPA activity in PCV13 versus PPSV23, but that responses converge on the use of the same FcR, which may be key to driving OPA.

Fig. 4. PCV13 and PPSV23 differentially drive OPA and non-OPA antibody-mediated functions.

(A) Spearman correlations were calculated between OPA GMTs and each antigen-specific isotype/subclass concentration or binding of FcR at the 1-month time point for all individuals across all tested antigens (serotypes 1, 3, 6A, 6B, 7F, 19A, 19F, and 23F), for PCV13 (n = 14) and PPSV23 (n = 19). Heatmaps represent the correlation coefficients between OPA GMT and the indicated antibody response. Correlation coefficients and P values are indicated for significant features. *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001. (B) Representative dot plots indicate correlations for all antigens between OPA GMTs and concentrations of IgG, IgG1, IgG2, and FcgR2A binding for each respective serotype, for PPSV23 and PCV13 vaccinees at 1 month. (C) Spearman correlations were calculated at the 1-month time point between OPA GMTs and the indicated antibody-mediated functional assays for tested antigens (serotypes 3, 6A, and 19A) for PCV13 and PPSV23 groups. Heatmaps represent the correlation coefficients between OPA GMT and the indicated functional response. Significant correlations have correlation coefficients and significance indicated: *P < 0.05, **P < 0.01, and ***P < 0.001. (D) Representative dot plots are shown, indicating correlations between OPA GMTs and antibody-mediated functions (ADCD, IFN-γ positivity of NK cells, ADNP, and ADDCP), for equivalent antigens in samples isolated from PPSV23 and PCV13 vaccinees at 1 month. Antibody-mediated functional assays were performed only on the three antigens shown.

PCV13 and PPSV23 serum samples have distinct, non-OPA antibody-mediated functions

Given the similar OPA titers observed across the two vaccines, we next aimed to determine whether the vaccines drove distinct additional antibody effector functions. The ability of each vaccine to induce ADCD, ADNP, ADCP, ADDCP, and ADNKA was assessed against serotypes 3, 6A, and 19A. We asked whether the generation of effective OPA responses correlated to other, non-OPA antibody-mediated effector functions. We correlated OPA titer values to ADCD, ADCP, ADNP, ADDCP, and ADNKA at the 1-month time point (Fig. 4, C and D). Both PCV13 and PPSV23 demonstrated correlations between OPA titer and ADCD and ADNP. This is in accordance with the mechanisms proposed to be captured by the OPA assay, which relies on antibody-mediated opsonization by complement followed by phagocytosis by a neutrophil cell line (37). Although PPSV23 showed a weak correlation with interferon-γ (IFN-γ) release by activated NK cells, the PCV13 group uniquely demonstrated correlations between OPA across antibody-mediated functions, including ADCP, IFN-γ release by activated NK cells (a measure of ADNKA), and ADDCP. Although these cell types are not present in OPA assays, these additional functional signatures may represent additional immune mechanisms that may be selectively coordinated with OPA activity in PCV13 vaccinees that may contribute to protection in vivo. Thus, this finding indicates that robust antibody responses elicited by PCV13 are reflected not only by OPA titers but also by the generation of other immune functions in multiple innate immune cell types that may be relevant to antimicrobial activity.

Because PCV13 overall induced modestly increased titers compared with PPSV23 (Fig. 2), we confirmed that the observed differences were not driven by titer alone. For each vaccine group and time point, we correlated data from antibody-mediated functional assays to total IgG for the same antigens (fig. S2). Comparison of titers and individual FcR binding and Fc effector functions across the vaccines highlights the strong linear relationship between titers and the binding of FcRs, particularly FcγR2A and FcγR2B, but not Fc effector functions across the vaccines, highlighting the qualitative, and not strictly quantitative, impact of vaccination on shaping antibody effector functions.

Generation of distinct humoral profiles in PCV13 and PPSV23

Having identified a variety of differences in antibody functions and coordination within the humoral responses generated by the two vaccines, we lastly aimed to define whether these vaccines induce distinct humoral signatures across shared antigens. To avoid model overfitting, we initially used a least absolute shrinkage and selection operator (LASSO) to identify the minimal number of antibody features that could resolve differences across the two vaccine groups. The minimal features were then used to determine whether the groups exhibited unique antibody profiles using a partial least squares discriminant analysis (PLS-DA). CRM and serotype 6A, unique to PCV13, were excluded from this analysis to avoid artificial separation of the vaccine groups, but an analysis that includes 6A is shown in fig. S3.

Clear separation in antibody functional profiles was observed across PCV13 and PPSV23 vaccinee pneumococcal antibodies at 1 month (Fig. 5A) and 1 year (Fig. 5B) after vaccination. Nearly no overlap was observed in antibody profiles. Only four and seven features were necessary to fully resolve profiles for PCV13 versus PPSV23 at 1 month and 1 year, respectively. Univariate comparisons shown in Fig. 5 (C and D) illustrate the distinguishing components in antibody profiles induced across the two groups.

Fig. 5. Multivariate analysis separates PCV13- and PPSV23-induced humoral profiles.

(A and B) To model the multivariate differences in vaccine profiles induced by each vaccine, a least absolute shrinkage and selection operator (LASSO) was used to select the minimal discriminatory features between PCV13 and PPSV23 groups. Multivariate modeling was then performed by partial least squares discriminant analysis (PLS-DA) at 1-month (A) and 1-year (B) time points (n = 20 per group). Cross-validation accuracy was 63.9 and 68.5%, respectively. Loading plots on the right indicate relative contribution of the indicated features to the x axis. (C and D) Univariate plots for the LASSO-selected features are shown for 1 month (C) and 1 year (D) for PPSV23 and PCV13 vaccinees. P values were calculated by Mann-Whitney tests. (E and F) Correlation networks were created, indicating all measured features that correlated to LASSO-selected features with absolute Spearman ρ > 0.7 and BH-adjusted P < 0.05 at 1 month (E) and 1 year (F). LASSO-selected features are indicated with triangles. Color indicates relative enrichment of each feature in PCV13 (blue) or PPSV23 (yellow). CRM and serotype 6A were excluded from LASSO selection or multivariate modeling but have been included for reference in correlation networks.

At 1 month, all the discriminating features were enriched in the PCV13 vaccine group, most prominently including IgG-driven features including IgG3 against serotype 19F, ADCP against 19A, and binding of FcγR2A to antibodies targeting 23F, which each contributed independently to discriminating the PCV13 and PPSV23 vaccine profiles (Fig. 5A). Elevated breadth of IgM also emerged as enriched in the PCV13 group at this time point. Conversely, at 1 year, two features emerged as uniquely enriched in the PPSV23 group: IgA2 against Pn19F, and binding of FcγR3B to Pn19A-specific antibodies (Fig. 5B). However, most differentiating features continued to be enriched in the PCV13 arm. These again included IgG-driven features including breadth of IgG1 response and ADCP against 19A. However, IgM targeting serotypes 1, 6B, and 23F also emerged as enriched in the PCV13 group.

The LASSO algorithm is designed to select a minimal set of features that account for the overall variation across groups, and it will exclude features that correlate highly with those that are selected. Therefore, to fully capture the spectrum of antibody features differentiating the two groups, correlation networks were constructed across all individuals for the LASSO-selected features at 1 month and 1 year (Fig. 5, E and F, respectively). The PCV13 carrier protein CRM and unique serotype 6A were included for this analysis although were not part of the LASSO PLS-DA models. At 1 month, the time point at which all selected features were enriched in PCV13, responses were clustered largely by isotype. IgG3 specific to 19F correlated with IgG3 against 19A; similarly, breadth of IgM correlated with IgM targeting 19F. Serotype 23F generated a unique cluster of correlated responses in total IgG, IgG2, FcγR2A, and FcγR2B, suggesting that this serotype may be generating a distinct pattern of responses from other serotypes. Last, ADCP against serotype 19A correlated with responses to the carrier protein CRM. At 1 year, the most prominent network, enriched in PCV13, included IgM responses against serotypes 1, 6A, 6B, and 23F. The predominant feature enriched in PPSV23, FcγR3B binding to 19A-specific antibodies, correlated to FcγR3B binding to antibodies against serotypes 1, 6A, 19A, and 23F, suggesting persistence of antibodies able to interact with this neutrophil-predominant FcR in this vaccine group.

Together, these data indicate that the measured antibody features against shared antigens alone are able to differentiate the profiles induced by these two vaccines. Further, beyond the differences identified in univariate comparisons (Fig. 2C), this multivariate approach reveals durable increases in humoral responses to PCV13 across serotypes, including the IgG1 subclass, recruitment of FcγR binding, and antibody-mediated innate immune effector functions. IgM, a mucosally relevant isotype, emerges as of particular importance in PCV13-driven profiles at 1 year. These data highlight differences in immunity elicited by these distinct vaccines that go beyond quantitative measures of IgG and OPA titers, and they may begin to explain the differences in the clinical impact of these vaccines across populations.

DISCUSSION

PPSV23 and PCV13 are effective against invasive pneumococcal disease; however, only PCV13 has been shown consistently to generate a distinct clinical effectiveness profile with increased durability of responses and reduction of nonbacteremic pneumonia and nasopharyngeal carriage. Despite this, only modest differences between PPSV23 and PCV13 have been reported from immunogenicity studies measuring antigen-specific IgG concentrations and OPA titers. This disconnect suggests that unrecognized immune pathways differentiate clinical responses to these two vaccines. Understanding the differences in humoral responses induced by the two vaccines could provide critical insights into pneumococcal immunity and guide the most effective use of these vaccines.

Here, we use systems serology to comprehensively profile the pneumococcal specific humoral immune response generated by these two vaccines across seven shared serotypes and one PCV13-unique serotype, 6A. This work reveals several differences between the two vaccines. Overall, a more robust humoral immune response was elicited by PCV13 compared with PPSV23, with increased durability of responses at 1 year. Although titer-based responses to both vaccines were dominated by IgG2, a subclass frequently induced by many carbohydrate antigens (24), univariate and multivariate analyses revealed subtle differences in other relevant antibody features. PCV13 induced slightly increased responses across highly functional IgG subclasses IgG1 and IgG3, as well as mucosally relevant isotypes IgA and IgM. PCV13 antibody responses additionally had increased antigenic breadth, with evidence of effective class switching, in line with prior studies of memory B cells in vaccinated macaques (31). Although these observed differences were quantitatively small, they were functionally relevant, as evidenced by the ability of IgG1 to drive OPA uniquely in PCV13 vaccinees, as well as the emergence of non-OPA antibody-mediated functions such as ADCP and ADNKA in PCV13 vaccinees. Thus, this systems approach enabled the identification of small but potentially immunologically important changes in humoral immunity.

The differences in vaccine-induced humoral profiles identified here may provide important insights into the mechanisms underlying the observed differences in clinical protection between PCV13 and PPSV23. First, our findings implicate differences in vaccine-induced mucosal immunity. The fact that both vaccines appear to be effective against invasive pneumococcal disease, whereas PCV13 is more effective against carriage and nonbacteremic pneumonia, implicates differential protection at the nasopharyngeal and respiratory mucosa. Secretory IgA is traditionally thought to be the major isotype relevant to mucosal immunity, one which S. pneumoniae subverts by cleaving one subclass, IgA1, with a protease (38). However, there is a growing appreciation for roles of IgM and IgG at the mucosal surface, particularly in the lower respiratory tract (39). In our study, antipneumococcal IgM, IgG, and IgA2 had more robust and more durable responses in PCV13 versus PPSV23. Further work will be necessary to determine how these identified humoral differences may directly affect functional protection from S. pneumoniae carriage and disease.

In addition, the observed clinical differences may be caused by non-OPA antibody-mediated functions both at the mucosa and systemically. Our data indicate that PCV13 can differentially induce antibody-mediated functions that are not captured by OPA titer but that may nevertheless be relevant to protective immunity. PCV13 generated antibodies that more effectively recruited FcγR2A, important for phagocytosis, as well as a broader array of FcR binding at both 1 month and 1 year. Moreover, PCV13 generated an array of increased antibody-mediated effector functions. These included not only ADNP and ADCD, functions that contribute to measurement of OPA, but also those involving other innate effector cell types, including ADCP and ADNKA. NK cells have been implicated in protective immunity against other streptococcal species and are the only fully mature innate effector cells in early life (40, 41). Further, published studies suggest that not only neutrophils but also monocytes, which were recruited by PCV13-induced antibodies, are essential for clearance of S. pneumoniae colonization (42, 43). Thus, these non-OPA functions, including recruitment of monocytes and NK cells, may underlie clinical differences between PCV13 and PPSV23 across the life span.

Unlike PCV13, PPSV23 has been observed to have short duration of clinical immunity requiring subsequent booster doses (5, 6) despite declines in OPA titers for both vaccines. The factors determining the duration of protective immunity are poorly understood and vary across antigens (44); however, T cell–independent responses, such as those presumed to be generated by PPSV23, and in contrast to T cell–dependent responses generated by conjugate vaccines, are traditionally thought to be short-lived (30). Our data extended only until 1 year after vaccination, and thus, we were unable to formally compare the rates of decay of humoral responses to the two vaccines for longer periods. However, even at the 1-year time point, PCV13 vaccinees uniquely maintained an array of antibody responses and FcR binding, increased antigenic breadth, and binding of complement above prevaccination values. Further, the humoral profiles of the two vaccine groups remained distinct even at the 1-year time point. These findings suggest that the antibody response generated by PCV13 differs qualitatively from that generated by PPSV23 in ways that may improve immune duration for the former.

We believe the findings in this study to be broadly relevant to other adult age groups, but the conclusions of the current study are constrained by the limited age range of 60 to 64 years. Further, a critical difference between PPSV23 and PCV13 is the relative lack of immunogenicity of PPSV23 in young children (3, 4). The reasons for the lack of robust humoral immunity to T cell–independent antigens in this age group and the distinct mechanisms of protective immunity that may be exploited instead have not been well elucidated. Although the current study was not designed to address the distinct antimicrobial mechanisms elicited in children by the two vaccines, concentrations of circulating IgG2 and complement concentrations are established to be relatively low in young children (45, 46). The data generated here in adults suggest that protection by PPSV23 relies on IgG2-mediated complement deposition, whereas PCV13 can more effectively recruit other subclasses and antibody-mediated cellular effector functions. This finding may contribute to the relative lack of pediatric efficacy of PPSV23 in this age group. Future studies in vaccinees of various age groups will be necessary to address age-dependent correlates of immunity and strategies to maximize immunity across populations.

Although our study focused specifically on deeply profiling vaccine differences in the humoral response, including downstream effects, this work does not address differences in vaccine-induced innate or cellular immune responses, particularly CD4⁺ T cells. Comprehensive work in both mice and humans has suggested that CD4⁺ T cells play important roles in mediating immunity against pneumonia and invasive pneumococcal disease and may be the major mediator of clearance of colonization (47, 48). The generation of T cell adaptive immune responses thus provides another mechanism by which glycoconjugate vaccines may induce differential clinical efficacy. Further studies are ongoing to elucidate roles for cell-mediated immunity in protection by PCV13 compared with PPSV23.

The different humoral profiles identified here indicate differences induced by the presence of a glycoconjugate in PCV13. However, PCV13 also differs from PPSV23 in the presence of an aluminum phosphate adjuvant. Emerging studies have demonstrated that adjuvants can alter the magnitude, durability, and functionality of antibody responses. Vaccination of rhesus macaques with aluminum-adjuvanted HIV Env induced greater antibody-dependent IFN-γ release by NK cells and, to a lesser extent, increased IgG titers and ADCP (49). In contrast, vaccination of humans with aluminum-adjuvanted H5N1 avian influenza vaccine did not induce robust antibody effector functions (50), suggesting that adjuvant-mediated changes may differ across species and across antigens. Moreover, more recent data suggest that distinct adjuvants can shape antibody-effector functions, marked by robust OPA induced with AS03 and robust NK cell activation induced by Toll-like receptor 7–recruiting AS37 in the context of a severe acute respiratory syndrome coronavirus 2 nanoparticle vaccination in nonhuman primates (51). Serotype-specific IgG titers induced by PCV13 with and without adjuvant have been shown to be almost identical (52). However, whether antibody isotype or subclass, FcR binding, and effector functions are potentially altered by adjuvant selection has not been closely examined in the context of pneumococcal vaccination. Thus, it is plausible that the observed differences in antibody profiles are due not only to the presence of a carrier protein but also to the presence of adjuvant, pointing to future opportunities to further tune and improve vaccine induced immune responses through the addition of distinct adjuvants.

Most assays were performed using only seven serotypes—six shared between PPSV23 and PCV13 and one unique to PCV13 alone—and functional assays were performed on only three. In the current work, we aimed to focus on specific serotypes of interest while addressing the analytic challenges of studying a collection of structurally similar polysaccharide antigens. The vaccine-induced responses identified here to each serotype correlate highly, but not perfectly, with each other, indicating serotype specific differences in the induced antibody Fc-profiles. The small number of serotypes studied in functional assays, however, limits our ability to assess correlations across antibody-dependent functions. Thus, it is not possible to fully extrapolate the current findings to other relevant serotypes. In future work, we aim to comprehensively investigate vaccine-induced responses to a wider array of serotypes within each vaccine, as well as the induction of cross-reactive humoral profiles to nonvaccine serotypes of clinical interest.

Important limitations of this study include the narrow age range of participants (60 to 64 years), the relatively short 1-year duration of follow up, and the limited number of pneumococcal serotypes assayed; these limitations make it difficult to determine whether the findings here apply to other populations, time frames, and serotypes. In addition, the observed in vitro differences between vaccine responses must be tied to clinical protection in vaccinated individuals. In further studies, we therefore aim to expand upon the parameters of the current study and further investigate the mechanisms of antibody-mediated protection from pneumococcal disease in vivo.

This study begins to describe insights into differences in the humoral responses generated by PCV13 versus PPSV23. In particular, PCV13 induced durable, robust, antigenically broad responses across antibody isotypes and subclasses that generated downstream antibody effector functions beyond those traditionally captured in OPA assays. Whether and to what degree these findings underlie critical clinical differences in vaccine efficacy and durability will need to be evaluated in future studies. The results presented here point to the need for further immunological dissection of the distinct antibody subpopulations and effector functions that may be most critical for the control and elimination of the pathogen. Ultimately, these findings provide insights into candidates for immunologic correlates of clinical efficacy and may guide the development of future vaccination strategies against S. pneumoniae.

MATERIALS AND METHODS

Study design

Samples analyzed in the study originated from individuals enrolled in a parallel-group, randomized, active-controlled, modified double-blind multicenter trial (NCT00427895) to evaluate the immunogenicity of PCV13 versus PPSV23 for the 12 common serotypes contained in PCV13 as measured by serotype-specific OPA titers 1 month after vaccination, as previously described (20). In brief, individuals in the original trial were age 60 to 64 years with no history of vaccination with any licensed or experimental pneumococcal vaccine. Healthy participants and those with stable preexisting underlying chronic conditions (such as cardiovascular, pulmonary, renal, and liver diseases including alcoholic liver disease and alcoholism, and diabetes mellitus) were included, and immunocompromised individuals were excluded. Individuals in the larger study were 53.5% female in the PCV13 group and 60.9% in the PPSV23 group; 96.4% had white race in PCV13 group versus 94.2% in PPSV23 group; and individuals had a mean age of 61.8 years in the PCV13 group and 61.7 years in the PPSV23 group (20).

Serum samples were eligible for analysis if sufficient volume (1.0 ml) was available from the prevaccination study visit, 1 month (29 to 43 days) after vaccination, and 1 year (351 to 379 days) after vaccination. Individuals were randomly selected from eligible samples, for a total of n = 20 from each vaccine arm at 1-month and 1-year time points and n = 10 from each arm at the prevaccination time point (total, n = 40 individuals and 100 specimens). These numbers were chosen on the basis of assay feasibility and the ability of previous systems serology studies to successfully identify distinct multivariate humoral signatures. Multivariate models have been cross-validated to avoid overfitting and selection bias and to ensure that the samples have the power to generalize a pattern. Demographics for the 40 individuals included in the current study are not available as the small cohort size may compromise privacy.

Residual serum samples from the original study were stored at Pfizer Inc. according to the study protocol and the terms of the informed consent. These specimens were deidentified, heat-inactivated, and blinded at Pfizer Inc. and then shipped to the Ragon Institute of Massachusetts General Hospital, Massachusetts Institute of Technology and Harvard. Samples remained blinded to individuals conducting experiments until all data had been collected, quality-controlled, and integrated. The current study was reviewed by the Partners Human Research Institutional Review Board and determined not to meet the definition of human subject research.

Assessment of OPA

OPA titers were measured in the original study of immunogenicity (20). Because of the large sample volumes required to perform OPA assays, OPA data for most participants were only available for the 1-month time point, and assays were not performed across all individuals. Data were included here for n = 17 participants (n = 8 for PCV13 and n = 9 for PPSV23) against serotypes 1, 3, 6A, 7F, and 19A and for a nonoverlapping 16 participants (n = 6 for PCV13 and n = 10 for PPSV23) against serotypes 6B, 19F, and 23F.

Measurement of antigen-specific antibody titers by isotype and subclass

A customized, multiplex Luminex assay was used to measure antigen-specific antibody responses for multiple isotypes and subclasses. Vaccine antigens tested included poly-l-lysine (PLL) conjugated polysaccharides from S. pneumoniae serotypes 1, 3, 6A, 6B, 7F, 19A, 19F, and 23F and the carrier protein CRM (provided by Pfizer). An equal mixture of influenza antigens from HA1(B/Brisbane/60/2008) and HA1(H1N1)(A/New Caledonia/20/99) (Immune Technology Corp) was used as a positive control, and recombinant polyhistidine (His) tagged ebolavirus glycoprotein minus the transmembrane domain (EBOV GPdTM, Mayflower Bioscience) was used as a negative control. All antigens were coupled to carboxylate-modified microspheres (Luminex Corp.) by covalent N-hydroxysuccinimide (NHS)–ester linkages by 1-ethyl-3-(3-dimethylaminopropyl)carbodiimide hydrochloride (EDC; Thermo Fisher Scientific) and sulfo-NHS (Thermo Fisher Scientific) as per the manufacturer’s instructions.

Specificity of the Luminex platform was confirmed using mouse monoclonal antibodies targeting each serotype. Assays were optimized over a dilution curve to ensure selection of a dilution within the linear range of the assays. A 1:1000 dilution was selected to maximize the dynamic range across control samples and to capture the area under the curve for the full range of dilutions tested. Diluted serum samples at 1:1000 were incubated with pooled microspheres for 2 hours at room temperature and then washed three times with phosphate-buffered saline (PBS) with 0.1% bovine serum albumin (BSA) and 0.05% Tween. Secondary detection reagents included phycoerythrin (PE)–conjugated goat antihuman IgG, IgG1, IgG2, IgG3, IgG4, IgM, IgA1, and IgA2 (SouthernBiotech). For FcR binding, human FcγR2A, FcγR2B, FcγR3A, FcγR3B, and FcαR (Duke University Protein Production Core) were biotinylated using BirA (Avidity) and conjugated to streptavidin-PE (PhycoLink). All secondary incubations were performed over 1 hour at room temperature. Median fluorescence intensity (MFI) for each bead region was measured using an iQue Screener PLUS (Intellicyt). All samples were assayed in duplicate, and values were averaged. For measurements of avidity, samples were incubated with PBS with 0.1% BSA and 0.05% Tween with or without 7 M urea for 20 min and then washed three times before incubation with secondary detection reagents. Avidity index was calculated as the ratio of MFI in the presence of urea treatment to that without urea treatment (32).

Antibody-dependent cellular phagocytosis

THP-1 phagocytosis of beads coated with PLL-conjugated polysaccharides was performed as previously described (53). Briefly, PLL-conjugated polysaccharides from 3, 6A, and 19A were biotinylated with 50-fold excess biotin with EZ-Link NHS long-chain biotin (Thermo Fisher Scientific) following the manufacturer’s instructions and then adsorbed onto 1-μm fluorescent neutravidin beads (Invitrogen) at a 1:1 (micrograms:microliters) ratio of biotinylated polysaccharide to beads. Ten microliters of a 1:100 suspension of antigen-coupled beads were added to each well of a 96-well plate along with equal volume of serum diluted 1:50, and plates were incubated for 2 hours at 37°C and then washed with PBS. A total of 25,000 THP-1 cells (human acute monocytic leukemia cell line, American Type Culture Collection, RRID:CVCL_0006) were added and incubated at 37°C for 18 to 20 hours. Cells were fixed with 4% paraformaldehyde solution (Santa Cruz Biotechnology) before data acquisition. Phagocytosis was measured by iQue Screener PLUS (Intellicyt). Phagocytic scores were calculated as (percent bead-positive cells) × (geometric mean fluorescence intensity)/10,000. Specificity of the assay was confirmed using control monoclonal antibodies targeted to each serotype. Each sample was assayed in two independent technical replicates and averaged.

Antibody-dependent neutrophil phagocytosis

Neutrophil phagocytosis was evaluated as previously described (54). Briefly, as described for cellular phagocytosis, PLL-conjugated polysaccharides for serotypes 3, 6A, and 19 were biotinylated and coupled to 1-μm fluorescent neutravidin beads. Ten microliters of a 1:160 dilution of coupled beads was opsonized with 10 μl of serum diluted 1:25 at 37°C for 2 hours. Whole blood was collected from healthy donors, red blood cells were lysed with ammonium-chloride-potassium (ACK) lysis buffer (Quality Biological), and primary leukocytes were isolated by centrifugation and washed in PBS. A total of 50,000 isolated leukocytes were added per well and incubated for 1 hour at 37°C. The cells were then stained with Pacific Blue anti-human CD66b antibody (10 μg/ml; clone G10F5; BioLegend, catalog no. 305112; RRID:AB_2563294) and fixed before measurement and analysis on the iQue Screener PLUS (Intellicyt). Cells were gated on CD66b⁺ and phagocytic scores calculated as above. Specificity of the assay was confirmed with relevant control monoclonal antibodies. Two healthy leukocyte donors were used as biological replicates for each sample and assayed in parallel, and replicates were averaged.

Antibody-dependent dendritic cell phagocytosis

PLL-conjugated polysaccharides for serotypes 3, 6A, and 19A were coupled to carboxylate- modified 1-μm fluorescent microspheres (Thermo Fisher Scientific) at a ratio of 2 μg of antigen to 1 μl of beads by covalent NHS-ester linkages by EDC (Thermo Fisher Scientific) and sulfo-NHS (Thermo Fisher Scientific) per the manufacturer’s instructions. Primary monocytes were isolated from fresh blood using CD14 microbeads (Miltenyi) and differentiated for 1 week in MoDC medium (Miltenyi). Ten microliter of a 1:200 antigen-coupled bead suspension was opsonized with 10 μl of a 1:25 dilution of serum for 2 hours at 37°C and then washed with PBS. Differentiated dendritic cells were added at 40,000 cells per well and incubated for 4 hours at 37°C. Cells were then stained with fluorescein isothiocyanate anti–human leukocyte antigen-DR isotype (HLA-DR) at 5 μg/ml (clone L243; BioLegend, catalog no. 307604; RRID:AB_314682), PE-Cy7 anti-CD86 at 1 μg/ml (clone 2331; BD Biosciences, catalog no. 561128; RRID:AB_10563077), and allophycocyanin (APC)–Cy7 anti-CD83 at 2 μg/ml (clone HB15e; BioLegend, catalog no. 305330; RRID:AB_2566393), washed, and fixed with 4% paraformaldehyde solution. Data were acquired using an iQue Screener PLUS (Intellicyt). Measured outputs included MFI of HLA-DR, CD86, and CD83. Specificity of the assay was confirmed using control monoclonal antibodies specific to each serotype. Two leukocyte donors were used as biological replicates in parallel, and replicates were averaged.

Antibody-dependent complement deposition

Antibody-mediated complement deposition was measured as previously described (26). PLL-conjugated polysaccharides for serotypes 3, 6A, and 19 was biotinylated and coupled to 1-μm fluorescent neutravidin beads. Ten microliters of a 1:100 bead suspension was opsonized with 10 μl of serum diluted 1:5 in PBS for 2 hours at 37°C in 96-well plates. Lyophilized guinea pig complement (Cedarlane) was resuspended in ice-cold water and then diluted 1:50 in veronal buffer with 0.1% gelatin (Boston BioProducts) added. Diluted complement (200 μl) was added to the opsonized beads and incubated for 20 min at 37°C. Beads were then washed with 15 mM EDTA in PBS and stained with fluorescein-conjugated anti-guinea pig complement component C3 at 30 μg/ml (MP Biomedicals, catalog no. 855385). Samples were washed and analyzed on the iQue Screener PLUS (Intellicyt). MFI of each sample was used as complement score. Each sample was assayed in three independent technical replicates performed on different days, and replicates were averaged. Specificity was confirmed using relevant control monoclonal antibodies for each serotype.

Antibody-dependent NK cell activation

Antibody-dependent NK cell activation and degranulation were measured as previously described (26). ELISA plates (Thermo Fisher Scientific, NUNC MaxiSorp) were coated with PLL-conjugated polysaccharides (3 μg/ml) of serotypes 3, 6A, and 19A overnight at 4°C and then blocked with 5% BSA in PBS for 2 hours at 37°C. One-hundred microliters of serum diluted 1:10 was added to each well and incubated for 2 hours at 37°C. NK cells were isolated from buffy coats from healthy donors using the RosetteSep NK Cell Enrichment Kit (STEMCELL Technologies) and rested in interleukin-15 (1 ng/ml) at 37°C overnight. Isolated NK cells were premixed with a cocktail of anti-CD107a PE-Cy5 at a 1:5 dilution (clone H4A3; BD Biosciences, catalog no. 555802; RRID:AB_396136), brefeldin A at 0.15 μg/μl (Sigma-Aldrich), and GolgiStop as per the manufacturer’s instructions (BD Biosciences), then added to the wells, and incubated for 5 hours at 37°C. Cells were stained for surface markers using anti-CD56 PE-Cy7 at a 1:10 dilution (clone B159; BD Biosciences, catalog no. 557747; RRID:AB_396853), anti-CD16 APC-Cy7 at a 1:10 dilution (clone 3G8; BD Biosciences, catalog no. 557758; RRID:AB_396864), and anti-CD3 Alexa Fluor 700 at 5 μg/ml (clone UCHT1; BD Biosciences, catalog no. 558117; RRID:AB_397038), then fixed, and permeabilized using the FIX & PERM Cell Permeabilization Kit (Thermo Fisher Scientific). Cells were stained for intracellular markers using anti–macrophage inflammatory protein–1β (MIP-1β) PE at 4 μg/ml (clone D21–1351; BD Biosciences, catalog no. 550078; RRID:AB_393549) and anti–IFN-γ APC at 1 μg/ml (clone 25723.11; BD Biosciences, catalog no. 340449; RRID:AB_400425). Fixed cells were analyzed by the iQue Screener PLUS (Intellicyt). NK cells were gated as CD3⁻ and CD16/56⁺ cells. MFI of MIP-1β, IFN-γ, and CD107a was measured. Two healthy NK cell donors were used to assay each sample in two independent assays on different days, and values from these biological replicates were averaged. Specificity of the assay was confirmed using serotype-specific control monoclonal antibodies.

Statistical analysis

All data were compiled quality-controlled and compiled in data file S1 before analysis. Data points were excluded only if an inadequate number of events were captured, according to the predetermined criteria for each assay. Data analysis was performed using GraphPad Prism and R (4.0.2). Comparisons between the prevaccination time point of 1 month and 1 year were all performed using Wilcoxon signed-rank test, allowing pairing of samples from the same individuals, followed by Benjamini-Hochberg (BH) correction for multiple comparisons. Comparisons between vaccination arms for all assayed antibody features, where pairing was not present, were performed using Mann-Whitney U test followed by BH correction. Fold changes and enrichments were calculated using median values.

Multivariate classification models were built to discriminate humoral profiles between vaccination arms. Before analysis, all data were normalized using z scoring. Feature selection was performed using LASSO. Classification and visualization were performed using PLS-DA. Model accuracy was assessed using 10-fold cross-validation. In each run, samples were stratified into training and test subsets, ensuring that both vaccination groups were represented in each set. For each test fold, LASSO-based feature selection was performed on logistic regression using the nine subsets designated as the training set for that fold. LASSO was repeated 100 times, and features selected at least 90 times of 100 were identified as selected features. PLS-DA classifier was applied to the training set using the selected features, and prediction accuracy was recorded. Selected features were ordered according to their variable importance in projection score, and the first two latent variables of the PLS-DA model were used to visualize the samples. These analyses were performed using R package “ropls” version 1.20.0 (55) and “glmnet” version 4.0.2. Co-correlates of LASSO-selected features were calculated to find features that can equally contribute to discriminating vaccination arms. Correlations were performed using Spearman method followed by Holm correction. The co-correlate network was generated using R package “network” version 1.16.0 (56). In addition, orthogonal analytical methods were used to ensure that the signatures were consistent independent of the analytical method.

Data and materials availability:

All data associated with this study are present in the paper or the Supplementary Materials. R code for LASSO PLSDA models can be found at DOI 10.5281/zenodo.6581786.