Concentration and avidity of antibodies to different circumsporozoite epitopes correlate with RTS,S/AS01E malaria vaccine efficacy

RTS,S/AS01E has been tested in a phase 3 malaria vaccine study with partial efficacy in African children and infants. In a cohort of 1028 subjects from one low (Bagomoyo) and two high (Nanoro, Kintampo) malaria transmission sites, we analysed IgG plasma/serum concentration and avidity to CSP (NANP-repeat and C-terminal domains) after a 3-dose vaccination against time to clinical malaria events during 12-months. Here we report that RTS,S/AS01E induces substantial increases in IgG levels from pre- to post-vaccination (p < 0.001), higher in NANP than C-terminus (2855 vs 1297 proportional change between means), and higher concentrations and avidities in children than infants (p < 0.001). Baseline CSP IgG levels are elevated in malaria cases than controls (p < 0.001). Both, IgG magnitude to NANP (hazard ratio [95% confidence interval] 0.61 [0.48–0.76]) and avidity to C-terminus (0.07 [0.05–0.90]) post-vaccination are significantly associated with vaccine efficacy. IgG avidity to the C-terminus emerges as a significant contributor to RTS,S/AS01E-mediated protection.

Post-vaccination adjusted increase in NANP and C-terminus CSP IgG concentrations induced by RTS,S/AS01E as compared to comparator vaccinees. The increase of IgG is adjusted by the effect of variables that impacted antibody concentrations (significant in the interaction with vaccination). Adjusted p-values comparing significance of logarithm of RTS,S/Comparator ratios were estimated through mixed models. Error bars represent 95% confidence intervals. M=month.      Correlation analysis between baseline (M0) and post-vaccination (M3) IgG concentrations (log 10 -transformed EU mL -1 ) and between M0 and M3-M0 difference in IgG concentrations. In all subjects, M0 vs M3 correlations: a CSP NANP, 5-17 months old children (n = 276 matched M0 and M3); b CSP NANP, 6-12 weeks old infants (n = 246); c CSP C-terminus (C-term), 5-17 months old children (n = 279); d CSP C-term, 6-12 weeks old infants (n = 249), with the Spearman correlation coefficients (rho

Correlation with protection from first clinical malaria episode of Avidity Index (AI) given IgG concentration of CSP NANP (a) and C-term (b).
Kaplan-Meier curves with IgG (log 10 EU mL -1 ) and AI (log 10 AI) stratified in tertiles (t3 = highest, t2 = moderate, t3 = lowest). P-values were obtained through log-rank tests.

Season malaria transmission
Ref.
Low or no

Supplementary Methods
Subject and sample selection. Individuals considered eligible for the study were those from the according to protocol (ATP) MAL067 cohort who had 75 µl of M3 plasma/serum sample stored. The ATP criteria included receiving 3 doses of vaccine, and having a blood sample collected approximately 30 days after vaccination (M3). In Kintampo (n= 311, 148 children and 163 infants) and Bagamoyo (n=272 children) all eligible children were selected, and in Nanoro, 445 participants (199 children and 246 infants) were randomly selected (Table 1 and Supplementary Figure 1). Nearly all subjects in Kintampo (100%) and Bagamoyo (99%) had pre-vaccination (M0) samples available, while in Nanoro, only 41% subjects, all infants, had M0 samples.
Subject follow up. Subjects were followed up by passive case detection (PCD) starting 14 days from the date of sample collection to ascertain that antibody responses measured would be predictive of a future malaria episode and not markers of past malaria infection. The follow-up for clinical malaria events in this study was defined for the subsequent 12 months, during which subjects were visited monthly to also assess whether they were still living in the study area and participating in the study. After the 12 months of follow up, subjects were censored. Reasons for early termination were recorded.
ELISA data pre-processing. ELISA data were captured using Magellan software (TECAN, Switzerland). The standard curves were estimated using a 4-parameter logistic (4PL) regression and were fitted after averaging the optical density (OD) of the standard points replicates. To provide EU mL -1 data in the absence of an international standard, the curves were pre-calibrated at IAVI-HIL by determining an average reciprocal of dilution for each monoclonal antibody that provided an OD=1 (12,114 EU mL -1 and 6,966 EU mL -1 for CSP C-term and NANP, respectively). Sample IgG concentrations (EU mL -1 ) were interpolated from the 4PL standard curve using the OD of the lowest dilution in the linear section of the curve (under the upper limit of detection). The replicates of the test samples were averaged and used as the test sample measurement. Quality control for each plate was based on positive, negative, and blank controls. A plate was considered to fail based on: i) thresholds for blank and negative controls and Westgard rules for positive controls; ii) expected standard curve characteristics. A sample was considered to fail based on: i) replicates variability; ii) high IgG titre requiring additional dilution to fall on the linear section of the 4PL curve (above 1:7,200).
Using a uniform distribution, the non-quantitated ELISA concentration data were imputed using values between limit of detection (LOD) and limit of quantification (LOQ), provided by IAVI-HIL. To account for the correlation between avidity concentration and ELISA concentration data, the imputed values of the avidity samples were generated given the ELISA values, imputed or observed (whether quantitated or not) by fitting a model with the OD of the non-quantitated samples. This model was fitted using generalized least squares models (GLS). Using the original ODs we maintained the correlation structure between ELISA and avidity data, and by fitting a GLS model we accounted for the heteroscedasticity of the data. The avidity ODs were fitted using as a predictor the OD of the imputed (or observed) ELISA after converting the imputed (or observed) concentration in OD (using a randomly selected standard curve from all assayed plates). To convert the imputed avidity OD into concentration we used the same standard curve used to convert imputed ELISA concentrations to ODs. The GLS models were antigen-specific, type of non-quantitated data-specific (i.e., whether it was avidity or not) and were validated by inspecting the residuals distribution and comparing the correlation structure between the original OD of the data and the imputed one (Supplementary Tables 4 and 5). Tables contain information on how many values were imputed and how many measured.
Immunogenicity data analysis. Mixed linear models were fit including as outcomes M3 and M0 anti-NANP and M3 and M0 anti-C-term CSP IgG concentrations, as well as vaccination, time point, and the appropriate interaction. Necessity of including a random slope for changes over time was assessed during model building through exploratory analysis (trajectory plots) and statistical tests, i.e., ANOVA test comparing two models: i) those with random intercept and random slope, and ii) those with only random intercept. The main mixed model equation, without any other adjusted covariate, used in the analysis was expressed as follows: ∼ ! + ! * 3 + ! * , + ! * 3 , + ( / ) Where ( / ) defines the random intercepts per subject and random slope M3-M0. Using this notation we were able to estimate the concentration, for example, for the M3-RTS,S cohort as ( ! + ! + ! + ! ) and for the M0-RTS,S cohort as ( ! + ! + ! ), being the concentration difference between both timepoints M3-M0 for RTS,S cohort equal to ! . After linearity of associations with continuous covariates was evaluated, parametric generalized linear models (GLM) were fit and non-linear terms were included as b-splines with degrees of freedom estimated for the spline term in the generalized additive models (GAM) model 1 . In models including relevant covariates, an adjusted effect of vaccination was evaluated and the impact of these covariates on immunogenicity was assessed through interaction terms. Covariates were retained in models based on statistical significance association with outcomes, and on their impact on the correlate coefficients.