Our results demonstrate that variations in vaccine efficacy can be explained by the magnitude and duration of the RTS,S-induced anti-CSP antibody response. In turn, immunogenicity depends on age, pre-existing immunity, co-administration of other vaccinations, dosing regimen and adjuvant system. Children aged three months to five years had significantly higher vaccine-induced anti-CSP antibody titres than infants younger than three months, who in turn had higher vaccine-induced anti-CSP titres than adults. Also, the associations between pre-vaccination anti-CSP antibody titres and peak titres were negative in infants, positive in older children and non-significant in adults. Infants are thought to have less well developed capacity for immunological responses than older children , and our data also suggest that immunogenicity in infants is reduced further where maternal exposure to malaria was high. This might be attributable to passive immunity inhibiting the development of new immunological responses, as has been described for hepatitis B vaccination . In contrast, the positive association between pre-vaccination anti-CSP titres and peak anti-CSP titres in children and adults suggests improved immunogenicity in subjects with pre-existing naturally-acquired immunity . A positive association was also observed between pre-vaccination anti-HBs titres and peak anti-CSP titres in children, a result previously reported in a cohort of Gabonese children . Hypothetical mechanisms underlying this association include: (1) HBs-primed B cells expressing anti-HBs antibodies capturing the RTS,S antigen and ensuring efficient presentation for CSP-specific T-cell priming; and (2) HBs-specific CD4 memory T-cells providing more rapid T-cell help to CSP-specific B cells upon stimulation by the HBs antigen in RTS,S .
Linear regression models without a random effect for trial site predict a statistically significant association between peak anti-CSP antibody titre and co-administration status (76% reduction, P <0.001, Additional file 2: Table S2). However, this effect may be attributable to inter-trial site variation and was not statistically significant when random effects by trial site were incorporated in the regression models (68% reduction, P = 0.095, Additional file 2: Table S1). The cohort of infants receiving RTS,S/AS02D co-administered with EPI vaccines in Tanzania  had substantially lower immunogenicity than a comparable cohort of infants vaccinated with RTS,S/AS02D without co-administration in Mozambique : 87 (95% range: 1 to 572) EU/mL versus 211 (95% range: 6 to 1,008) EU/mL, respectively. Statistically significant reductions in immunogenicity due to co-administration have been observed in vaccine trials of infants receiving pneumococcal  and Haemophilus influenza type b  vaccines. Hence, in context, an impact of co-administration is possible despite the lack of statistical significance.
We found anti-CSP antibody titres to be associated with protection against infection and episodes of clinical malaria [29,30], with the estimated dose–response relationship predicting increasing efficacy with increasing antibody titre across multiple trial sites (Figure 2). In particular, three of the four Prentice criteria  are satisfied: (1) RTS,S vaccination affects outcome (infection or clinical malaria); (2) RTS,S affects surrogates (anti-CSP antibodies); and (3) surrogates affect outcome (anti-CSP antibodies are associated with protection). The fourth criterion, that conditional upon the surrogates outcomes are independent of vaccination status, is less easily satisfied by the diverse data from the phase 2 trials. This equates to whether anti-CSP antibody titres can be used to predict vaccine efficacy. The comparison between observed and model predicted vaccine efficacies in Figure 2d,e suggests the criterion is satisfied, although there is likely to be an important role for CSP-specific CD4+ T cell responses which has not been accounted for. The association between anti-CSP antibody titres and protection from infection is consistent with data from other vaccine studies [31,32], as well as data from mouse models where RTS,S-induced anti-CSP antibodies have been shown to inhibit sporozoite invasion . There was no evidence for a threshold anti-CSP antibody titre above which large increases in efficacy are achieved . The smooth, albeit non-linear, shape of the dose–response curve is consistent with RTS,S having the profile of a leaky vaccine, but with substantial variation in efficacy.
The use of anti-CSP antibodies as the sole marker of the RTS,S-induced immune response constitutes a potential limitation of this analysis, as RTS,S-induced CSP-specific CD4+ T cells have been shown to be associated with protection from P. falciparum infection and episodes of clinical malaria . However, analysis of immunological data from a challenge trial of RTS,S  suggests that anti-CSP antibody titres play a dominant role in protection from infection . Furthermore, the magnitude of the RTS,S induced antibody and cell-mediated responses are correlated , so anti-CSP antibody titres may act as markers for cellular immunity as well as antibody-mediated immunity. Efficacy against clinical malaria has also been shown to correlate with peripheral blood monocyte-to-lymphocyte ratios before vaccination . Total anti-CSP immunoglobulin G (IgG) responses were measured which poses another limitation as the duration and effectiveness of the RTS,S-induced antibody responses may depend on IgG subclass [38,39].
By combining our model for the decay of anti-CSP antibody titres, the estimated relationship between antibodies and efficacy, and a model for the acquisition of immunity to clinical malaria , we were able to demonstrate associations between the magnitude and duration of efficacy on a number of covariates. Most striking is the dependence of efficacy against clinical malaria on estimated transmission intensity and EPI vaccine co-administration status. Increased exposure resulted in lower initial efficacy against clinical malaria in the first year of follow-up, as well as a shorter duration of protection. At longer term, the combination of decaying anti-CSP antibody titres in the vaccinated cohort and increased naturally-acquired immunity in the control cohort resulted in waning of the efficacy of RTS,S to zero or below, over a duration of follow-up dependent on transmission intensity (Figure 3). The prediction of higher incidence of clinical malaria in the vaccine cohort compared to the control cohort during extended follow up in high transmission settings is consistent with observations of the incidence of malaria in a high exposure cohort in Kenya .
The variation in the characteristics of the phase 2 trials poses a potential limitation to this analysis. Differences in factors such as study populations, malaria transmission intensity, adjuvant formulation and co-administration of other vaccines complicate the analysis. For example, data on adults vaccinated with RTS,S/AS02 and followed for infection  are combined with data on infants vaccinated with RTS,S/AS01 and followed for clinical malaria . Despite these difficulties, there is an advantage to such diversity of data as it allows systematic comparison between participants receiving different vaccination regimens at varying ages. This is in contrast to data from ongoing phase 3 trials which have more standardised vaccination regimens in more homogeneous populations [1,2]. Comparison of the varied participants in phase 2 trials with participants in phase 3 trials may provide explanations for the lower immunogenicity and efficacy observed in infants compared to children, suggesting roles for the co-administration of EPI vaccines and reductions in the immunogenicity of RTS,S-induced responses due to interference by maternally-acquired antibodies. Knowledge of the varied determinants of vaccine efficacy will allow identification of sub-populations in which RTS,S will be most effective or cost-effective . Finally, the results presented in this analysis will need to be confirmed against individual level data from participants in the ongoing phase 3 trials once this becomes available [1,2].