We performed a study on a subset of 179 children aged 5–17 months from the RTS,S/AS01E Phase III trial (ClinicalTrials.gov NCT00866619), described elsewhere (4): 105 children received RTS,S/AS01E and 74 children received the rabies vaccine as a comparator at study months zero (M0), one, and two. PBMC were collected at M0 before vaccination and approximately 30 days after the third vaccination dose (M3). RTS,S/AS01E-vaccinated and rabies-vaccinated children were randomly selected for this study among participants with no reported malaria episodes defined by observation of P. falciparum parasites on blood smears, identified through passive case detection during 18 months of follow-up after third vaccination dose. Of note, PBMC samples from children who had malaria cases were reserved for future correlates analyses to test the selected markers identified in this study. Samples were collected in two different African centers: Manhiça Health Research Center, Fundação Manhiça (FM-CISM, Mozambique; 120 children), and Ifakara Health Institute and Bagamoyo Research and Training Centre (IHI-BRTC, Tanzania; 59 children). The two sites had low-medium malaria transmission intensity at the time of the study (2–4). Investigators conducted all assays blinded to vaccination status.

Blood was collected in 5 ml sodium citrate (BD Vacutainer^® CPT™) tubes. PBMCs were isolated by density gradient centrifugation, cryopreserved and shipped to the Fred Hutchinson Cancer Research Center where the PBMC were thawed and stained (see Methods in Supplementary Material).

Thawed PBMC were rested in a 37°C, 5% CO₂ incubator overnight. The resting step increases the sensitivity of the assay (data not shown), probably by decreasing the stress and activation of PBMC due to the thawing process and exposure to the toxic cryopreservation agent. PBMC were stimulated with peptide pools covering the HBsAg or the CSP antigen present in the RTS,S vaccine (Table S1 in Supplementary Material). Negative controls contained 0.5% DMSO, the diluent for peptide pools, and Staphylococcal enterotoxin B was used as positive control stimulation at 1 µg/ml. Cultures were incubated 6 h at 37°C, 5% CO₂. This short incubation time increases the sensitivity and specificity of the assay to detect antigen-specific cells, avoiding non-specific and secondary immune responses. Further details are found in Supplementary Material.

Peripheral blood mononuclear cells were stained with one of two 16-color ICS panels that were designed for the study and that had previously undergone assay qualification with a formerly validated panel (17, 18). Cell staining was performed as described (17) (Supplementary Material). Antibody details can be found in Tables S2 and S3 in Supplementary Material. Data were acquired using a BD LSR II flow cytometer (BD Biosciences) directly from the plates using a high throughput sampler. We found some toxicity of HBsAg peptides, but since we could exclude dead cells, data were considered acceptable for analysis. We noted some spillover from CD154 to CXCR5 and therefore, T_FH-like cells were excluded from the analysis. Flow cytometry analysis was performed using FlowJo software (Version 9.9 Tree Star). Gating strategies for both panels are shown in Figures S1 and S2 in Supplementary Material.

The raw FCS files and manual gates were imported into the R environment (19) using the OpenCyto framework (20) and cell counts for the cell gates of interest were obtained for all stimulations and subjects.

For the analysis of the effect of vaccination on the frequencies of cells expressing the functional markers, a multivariate linear mixed effect model was fitted using logarithm-10 transformed cell frequencies (cells expressing the functional marker/total number of cells within each cell subset) as an outcome. These models are commonly used (21–24) and allow to model jointly the different cell stimulations that each subject sample underwent using random effects. The model included a random intercept for the subject effect nested within stimulation and a random slope for timepoint. In these models, observations generated by different subjects (biological replicates) are assumed to be independent. We included the following predictors in the regression models (as fixed effects): stimulation (CSP, DMSO, or HBsAg and DMSO, as appropriate), time of visit (pre- vs. post-vaccination), and vaccination status (RTS,S/AS01E vs. comparator vaccine), as well as all interactions between the three factors to allow variations of responses across subgroups of vaccinees and time. Linear combination of coefficients in these models allowed addressing specific study questions (details in Supplementary Material). Through these models, we estimated the % change in cell frequencies in RTS,S/AS01E vs. comparator vaccinees at each timepoint, the % change from M0 to M3 in each vaccine group, and the % change in RTS,S/AS01E vs. comparator vaccinees accounting for both timepoints. Statistical significance of comparisons was based on likelihood ratio tests. Antigen-specific responses were analyzed accounting for background, i.e., DMSO stimulation.

To define positivity of responses in each cell subset and subject, we identified the functional markers that were differentially expressed between the antigen (CSP or HBsAg) stimulations and their corresponding background (DMSO control stimulation) for each cell subset, subject, and at each timepoint. Thus, the proportion of cells expressing that marker was compared between the two stimulation conditions for each subject and timepoint. The statistical method used was mixture models for single-cell assays (MIMOSA) (25), using the default algorithm (Expectation–Maximization algorithm) and a false-discovery rate of 0.05%. This method was chosen because it has higher sensitivity and specificity than alternative methods such as Fisher’s exact test (25). A separate model for each functional marker, cell subset, and stimulation was fitted. Once we had defined the positivity or negativity for each functional marker, cell subset, subject, and timepoint, then the proportion of children with positive responses between vaccine groups at baseline or 1-month post-vaccination was compared using two-sided Fisher’s exact test.

Polyfunctional responses (cell subsets that express multiple functional markers) were analyzed using combinatorial polyfunctionality analysis of antigen-specific T cell subsets (COMPASS) (26). COMPASS models all cell subsets expressing functional markers simultaneously and selects the subsets most likely to have a positive antigen-specific response. The antigen-specific response is quantified by the probability of having a positive response. COMPASS functionality and polyfunctionality scores, summarizing each subject’s polyfunctional profile, were compared between vaccine groups using Wilcoxon test. The functionality score is the estimated proportion of antigen-specific cell subsets detected among all possible ones, whereas the polyfunctionality score is similar but weighted by the degree of polyfunctionality. In addition, COMPASS generates heatmaps that show the posterior probabilities for each modeled cell subset for each subject. CD107a marker was eliminated from MIMOSA and COMPASS analysis, because peptide pool stimulation was associated with a non-specific increase in CD107a compared to background, and response positivity could not be defined.

Statistical tests were considered significant at 0.05 α-level. Adjustments for multiple testing were done using Holm (27) and Benjamini–Hochberg (28) approaches. All analyses were conducted using R software. See Supplementary Material for detailed descriptions.

We used two different multiparameter ICS panels to assess cellular immune responses specific to the RTS,S/AS01E vaccine antigens. PBMC from 55 RTS,S and 41 comparator vaccinees, and 50 RTS,S and 33 comparator vaccinees were stained with antibody panel 1 (P1) and panel 2 (P2), respectively. The two panels had the same set of core markers, which allowed the exclusion of dead cells, monocytes (CD14), identification of CD4⁺ and CD8⁺ T cells (CD4, CD8 and CD3), NK and NK-T cells (CD56), the T_H1 cytokines IFN-γ, IL-2, and TNF-α, the T_H2 cytokine IL-4, the costimulatory molecule CD154 (CD40L), and the cytotoxic marker granzyme B (Figures S1 and S2 in Supplementary Material; Tables S2 and S3 in Supplementary Material). P1 additionally had CD45RA and CCR7 markers that allowed the identification of memory cell subsets, and IL-21, a cytokine related to T_FH. Panel 2 (P2) included γδ TCR for the identification of γδ T cells and several functional markers: the regulatory cytokine IL-10, the T_H2 cytokine IL-13, the T_H17 cytokine IL-17, and cytotoxicity marker CD107a. The gating strategy is detailed in Figures S1 and S2 in Supplementary Material. The common set of core markers (lineage and functional markers) allowed the pooled analysis for these markers. Males and females were similarly represented in both study sites and vaccine groups, with a proportion of 59 and 56.8% females in the RTS,S/AS01E and comparator groups, respectively.

We compared frequencies of CSP- or HBsAg-specific T cells expressing functional markers in RTS,S- and rabies-vaccinated children, accounting for the pre-vaccination frequencies in a multivariate model. RTS,S/AS01E-vaccinated children had a statistically significant increase of CSP-specific CD4⁺ T cells expressing IL-2 (217% increase), TNF-α (72.3%), and CD154 (101.4%) from baseline to post-vaccination in contrast to the rabies vaccinees (Figure 1A, i). We observed larger increases over time of HBsAg-specific CD4⁺ T cells expressing IL-2 (732.2%), TNF-α (346%), and CD154 (268.3%) (Figure 1B, i) in RTS,S/AS01E than in rabies vaccinees.

Figure 1A (ii) and Figure 1B (ii) show the frequencies of CSP- and HBsAg-specific CD4⁺ T cells after background subtraction. Higher frequencies of CSP- and HBsAg-specific CD4⁺ T cells expressing the above markers, and additionally HBsAg-specific CD4⁺ T cells expressing IFN-γ and IL-17 cells, were observed in RTS,S/AS01E vaccinees at post-vaccination compared to baseline and to comparators. No differences were detected between the two vaccine groups at baseline, or from baseline to post-vaccination in comparator vaccinees, the latter suggesting that there was no effect of naturally acquired immunity on cellular responses.

Antigen-specific T cells in memory subsets, defined by CD45RA and CCR7, were analyzed following vaccination. CD4⁺ T_CM (CD45RA⁻ CCR7⁺) and CD4⁺ T_EM (CD45RA⁻ CCR7⁻) cells recognized both vaccine antigens. After accounting for baseline frequencies, RTS,S/AS01E vaccinees had more CSP-specific CD4⁺ T_EM cells producing IL-2 and TNF-α, respectively, than comparator vaccinees (Figure 2A, i). Comparisons between vaccine groups at post-vaccination and from baseline to post-vaccination revealed additional responses: higher frequencies of CD4⁺ T_CM expressing IL-2 and TNF-α and CD4⁺ T_EM cells expressing CD154, IL-4 and IL-21, in RTS,S/AS01E than comparator vaccinees (Figure 2A, ii). Regarding HBsAg-specific responses, RTS,S/AS01E vaccinees had increased frequencies of CD4⁺ T_CM producing IL-2, TNF-α and CD154 and CD4⁺ T_EM producing IL-2, TNF-α, CD154 and IL-21 than comparator vaccinees (Figure 2B, i).

Therefore, RTS,S/AS01E vaccination induced CSP and HBsAg-specific CD4⁺ T_CM and CD4⁺ T_EM with T_H1, and additional T_H2, and T_FH functions. No overall effect of RTS,S vaccination was detected on the frequencies of terminally differentiated CD4⁺ T cells (CD45RA⁺ CCR7⁻), naïve CD4⁺ T cells (CD45RA⁺ CCR7⁺), other analyzed cell subsets (NK, NK-T like, γδ T, CD8⁺ T cells) or on other functional markers (Tables S5 and S6 in Supplementary Material).

Analysis of frequencies of cell subsets expressing functional markers in the background (DMSO, serving as the unstimulated control) revealed non-specific responses (i.e., not specific for CSP or HBsAg) upon RTS,S/AS01E vaccination (Figure 3). Accounting for baseline and responses in comparator vaccinees, RTS,S/AS01E vaccination increased the frequencies of CD4⁺ T cells producing IL-2 or TNF-α or CD154, IL-4, and CD107a (Figure 3A, i). In comparator vaccinees, a decrease in these two subsets from baseline to post-vaccination was detected. Interestingly, RTS,S/AS01E vaccination also increased the frequencies of CD8⁺ T cells producing IL-4 and CD107a (Figure 3B, i).

Since we detected an effect of RTS,S/AS01E vaccination in the background, we looked at the ex vivo antigen stimulations when not accounting for background. RTS,S/AS01E-vaccinated children had CD4⁺ T cells and CD8⁺ T cells expressing IL-4 and CD107a, as was observed for the background alone and, thus, likely reflecting non-specific responses (Figure S3 in Supplementary Material). Additionally, RTS,S/AS01E vaccination increased the frequencies of CD4⁺ T cells expressing IL-13 and CD8⁺ T cells expressing CD154 and IL-2 or TNF-α or CD154 following CSP stimulation when not taking background into account (Figure S3 in Supplementary Material). These additional responses in RTS,S/AS01E-vaccinated children were no longer significant when taking into account the background and likely reflect a non-specific effect of RTS,S/AS01E vaccination.

To assess the vaccine specificity of the antigen-specific T cell responses detected and the rate of responders to the RTS,S/AS01E vaccine (participants whose T-cells responded to stimulation), we determined the positivity of the responses for each subject comparing the frequencies of T cells expressing a functional marker in the antigen-stimulated and unstimulated control conditions for each participant using the MIMOSA statistical method. If a response is truly vaccine specific, one would expect to detect it after vaccination but not at baseline or in the comparator vaccinees. We found a significantly higher proportion of IL-2⁺, TNF-α⁺, and CD154⁺ CD4⁺ T cell positive CSP-specific responses in RTS,S/AS01E than in comparator vaccinees post-vaccination (Table 1). Response rates for these markers ranged from 24 to 30% in RTS,S/AS01E vaccinees; whereas for comparator vaccinees, there were none or very few responses (Table 1). Similarly, CSP-specific positive responses were absent or very low at baseline, supporting the specificity of the definition of positivity (Table S7 in Supplementary Material).

For HBsAg, we additionally detected more IFN-γ⁺ and IL-17⁺ CD4⁺ T cell positive responses in RTS,S/AS01E than comparator vaccinees at post-vaccination. Rates of HBsAg responders ranged from 26.7 to 76.1% in RTS,S/AS01E vaccinees (Table 1). However, a proportion of comparator vaccinees at post-vaccination and all vaccinees at M0 were also positive (Table 1; Table S7 in Supplementary Material), likely due to the fact that these children received the hepatitis B vaccine during the Expanded Program of Immunization (EPI) before the study. Most of the CSP responders were also HBsAg responders (Figure S4 in Supplementary Material).

When looking at CD4⁺ T_CM and CD4⁺ T_EM cell subsets, we found a significantly higher proportion of CSP responders in RTS,S/AS01E than in comparator vaccinees for IL-2, TNF-α, and CD154 (Table 2). Regarding HBsAg responses, RTS,S/AS01E vaccinees had higher rates of IL-2 and CD154 responses in CD4⁺ T_CM cells and IFN-γ, IL-2, TNF-α, CD154 and IL-21 responses in CD4⁺ T_EM than comparator vaccinees. Very few positive responses to CSP and HBsAg were detected in comparator vaccinees or at baseline (Table 2; Table S8 in Supplementary Material). Almost no positive responses were found in the other cell subsets analyzed (CD4⁺ T_TD, CD4⁺ T_N, NK, NK-T like, γδ T, CD8⁺ T cells, and memory CD8⁺ T cell subsets), and no significant differences were detected between vaccination groups for any of the two vaccine antigens (Tables S9–S11 in Supplementary Material).

When analyzing the pooled data for the functional markers included in both panels, RTS,S/AS01 vaccinees had significantly higher functionality and polyfunctionality scores for CSP- and HBsAg-specific CD4⁺ T cells than comparator vaccinees (Figures 4A,B). Heatmaps of posterior probabilities of responses showed CSP-specific CD4⁺ T cell responses among subsets coexpressing three markers (IL-2, TNF-α, and CD154), and two markers (TNF-α and CD154; IL-2 and TNF-α) in RTS,S/AS01E vaccinees, whereas almost no responses were detected in comparator vaccinees or at baseline (Figure 4A, ii). For HBsAg responses, more CD4⁺ T cell subsets with positive responses and with higher degree of polyfunctionality were detected in RTS,S/AS01E vaccinees (Figure 4B, ii). Besides the same triple- and double-positive CD4⁺ T cell responses recognizing CSP antigen, we also detected CD154⁺ IL-2⁺ TNF-α⁺ IFN-γ⁺ CD4⁺ T cells and CD154⁺ IL-2⁺ TNF-α⁺ IL4⁺ CD4⁺ T cells. By contrast, comparator vaccinees or all vaccinees at baseline had none or weak responses in these subsets that could be explained by background responses to previous malaria exposure or hepatitis B vaccination. Therefore, RTS,S/AS01E induced polyfunctional CD4⁺ T cells to both vaccine antigens, with a higher degree of polyfunctionality for HBsAg that included T_H1 and T_H2 responses.

When the COMPASS analysis was performed using all the functional makers included in each antibody panel separately, we identified additional highly polyfunctional CD4⁺ T cell subsets coexpressing IL-13 (for both vaccine antigens) or IL-21 (for HBsAg) in addition to IL-2, TNF-α and CD154 (Figure S5 in Supplementary Material). This further highlights the induction of highly polyfunctional CD4⁺ T cell subsets with a T_H2 and T_FH functions by RTS,S/AS01E. No polyfunctional responses were detected in CD8⁺ T cells for either vaccine antigen.