Systems serology-based comparison of antibody effector functions induced by adjuvanted vaccines to guide vaccine design

Abstract

The mechanisms by which antibodies confer protection vary across vaccines, ranging from simple neutralization to functions requiring innate immune recruitment via Fc-dependent mechanisms. The role of adjuvants in shaping the maturation of antibody-effector functions remains under investigated. Using systems serology, we compared adjuvants in licensed vaccines (AS01_B/AS01_E/AS03/AS04/Alum) combined with a model antigen. Antigen-naive adults received two adjuvanted immunizations followed by late revaccination with fractional-dosed non-adjuvanted antigen ( NCT00805389 ). A dichotomy in response quantities/qualities emerged post-dose 2 between AS01_B/AS01_E/AS03 and AS04/Alum, based on four features related to immunoglobulin titers or Fc-effector functions. AS01_B/E and AS03 induced similar robust responses that were boosted upon revaccination, suggesting that memory B-cell programming by the adjuvanted vaccinations dictated responses post non-adjuvanted boost. AS04 and Alum induced weaker responses, that were dissimilar with enhanced functionalities for AS04. Distinct adjuvant classes can be leveraged to tune antibody-effector functions, where selective vaccine formulation using adjuvants with different immunological properties may direct antigen-specific antibody functions.

Fig. 1. Adjuvants shape vaccine-induced functional antibody responses.

a The boxplots (representing medians, interquartile ranges [IQRs], minima, and maxima) show the antibody features for each vaccine adjuvant group. Groups received HBsAg adjuvanted with AS01_B, AS01_E, AS03, AS04, or Alum. Samples were profiled at day 30, 60, 360, and 390. Individuals were vaccinated at day 0, with an adjuvanted boost at day 30 and a non-adjuvanted, fractional-dose antigenic challenge at day 360. Measurements are provided as: log₁₀ MFI (mean fluorescence intensity), for the isotypes/subclasses/FcR-binding levels and C1q; as phagocytosis score, for antibody-dependent cellular phagocytosis (ADCP), antibody-dependent neutrophil phagocytosis (ADNP), and antibody-dependent dendritic cell phagocytosis (ADDCP); and as percentage of cells that are positive for each activation marker (CD107a, IFN-γ, MIP-1β), for antibody-dependent NK-cell activation (ADNKA). b Each row shows the median values and IQRs for the antibody features at one time point. The measurements were z-scored for each time point and across all adjuvant groups. c The polar plots depict the mean percentile of each antibody feature for each adjuvant group at day 390. Percentile rank scores were determined for each antibody feature across all individuals. d A partial least square discriminant analysis (PLS-DA) model was generated based on LASSO-selected features from all time points and area under the curve (AUC). Each dot represents a vaccinated subject in the PLS-scores plot. Ellipses show 75% confidence regions assuming a multivariate t distribution.

Fig. 2. Differences in correlation structures.

Correlation heatmaps were generated for each adjuvant (column) at each time point (row), using Spearman rank correlations. For each adjuvant/time point, all correlations between humoral features were calculated, with orange indicating positive and purple indicating negative correlations. ADNP antibody-dependent neutrophil phagocytosis, ADCP antibody-dependent phagocytosis by THP-1 cells, ADCD antibody-dependent complement deposition, ADNKA antibody-dependent natural-killer cell activation. ADNP antibody-dependent phagocytosis by primary neutrophils, ADDCP antibody-dependent phagocytosis by monocyte-derived dendritic cells.

Fig. 3. Dissecting differences between adjuvant clusters.

a The heatmap shows differences in the antibody features between the merged AS01_B/AS01_E/AS03 and AS04/Alum-adjuvanted vaccine groups over time. Orange tiles indicate that the feature is on average higher in subjects receiving the HBsAg vaccines containing AS01_B, AS01_E, or AS03, while purple tiles show enrichment in participants receiving HBsAg with AS04 or Alum. Significances were assessed using Mann–Whitney U tests and corrected for multiple testing using the Benjamini–Hochberg procedure. Asterisks indicate adjusted P values with *P < 0.05, **P < 0.01, ***P < 0.001. b A PLS-DA model was built based on features selected from all time points and the areas under the curves (AUCs) for the clusters AS01_B/AS01_E/AS03 and AS04/Alum. c The bar graph depicts how often antibody features were selected by repeated LASSO-based selection. The color indicates the adjuvant cluster in which the feature is enriched. The horizontal line shows the threshold of how often a feature needs to be chosen overall in order to be selected for the final set of minimal features. d The modeling approach was validated using permutation tests, for which the performance measured as classification accuracy of the actual model (using the four selected features shown in panel c) is compared to control models in a cross-validation framework. For “random features”, a fold-specific set of features of the same size as obtained by the LASSO-selection were chosen to train the model, and for “permuted labels” the modeling approaches were applied to shuffled group labels. The violin plots show the distribution of classification accuracies, for 10 repetitions and 100 permutations for the control models, and the P values indicate the median over the 10 repetitions of the exact P values obtained by permutation testing. e A co-correlate network was constructed using Spearman rank correlations. Only correlations with |r| > 0.9 to at least one of the selected features, which are highlighted in gray, are shown.

Fig. 4. Similarity in functional antibody responses across AS01_B-, AS01_E-, and AS03- adjuvanted vaccine responses.

a The heatmaps depict pairwise differences in antibody features between AS01_B versus AS01_E participant groups (top), AS01_B vs. AS03 participant groups (middle), and AS01_E vs. AS03 participant groups (bottom). Orange tiles indicate that the feature is higher in the first adjuvant group, while purple tiles show enrichment in the second adjuvant group. Significances were assessed using Mann–Whitney U tests, and asterisks indicate uncorrected P values with *P < 0.05, **P < 0.01, ***P < 0.001. b A PLS-DA model was built based on LASSO-selected features. The score plot shows high overlap between the high-response adjuvants AS01_B, AS01_E, and AS03, with the strongest overlap between AS01_E and AS03. The model only achieved a balanced accuracy of 0.23 and was not able to discriminate the adjuvants.

Fig. 5. Slight difference in humoral profiles across responses induced by the AS04- or Alum-adjuvanted vaccines.

a The heatmap shows differences in the antibody features between the AS04 or Alum-adjuvanted vaccine groups over time. Orange tiles indicate that the feature is on average higher in HBsAg/AS04-vaccinated individuals, while purple tiles show enrichment in HBsAg/Alum vaccinees. Significances were assessed using Mann–Whitney U tests, and asterisks indicate uncorrected P values with *P < 0.05, **P < 0.01, ***P < 0.001. b The PLS-DA score plot shows a slight separation between AS04 and Alum using ADNP at day 60 and the AUC for ADNP. c The modeling approach was validated using permutation tests, for which the performance measured as classification accuracy of the actual model is compared to control models in a cross-validation framework. For “random features”, fold-specific set of features of the same size as obtained by the LASSO-selection are chosen to train the model, and for “permuted labels” the modeling approaches are applied to shuffled group labels. The violin plots show the distribution of classification accuracies, for 10 repetitions and 100 permutations for the control models, and the P values indicate the median over the 10 repetitions of the exact P values obtained by permutation testing. d The bar graph depicts how often antibody features were selected by repeated LASSO-based selection. The color indicates the group in which the feature is enriched. The horizontal line shows the threshold of how often a feature needs to be chosen overall in order to be selected for the final set of minimal features. e A co-correlate network was constructed using Spearman rank correlations. Only correlations with |r| > 0.5 to at least one of the two LASSO-selected features, which are highlighted in gray, are shown.

Fig. 6. Plain language summary.

Study overview and main implications described in a manner that is understandable by a non-specialist audience.