Introduction A malaria vaccine is urgently needed [1,2]; however, the rational development of such a vaccine has suffered from a lack of knowledge of the relevance of experimental models [3]. This has resulted in a highly unsatisfactory situation in which each hypothesis derived from such models is investigated in lengthy and costly clinical trials. The lack of a reliable surrogate marker of protection in humans is thus a recognized limitation to the identification and development of efficacious vaccines [4]. The role of antibodies in clinical protection against malaria erythrocytic stages has long been recognized by in vivo transfer of antibodies from protected African adults to nonprotected individuals infected with P. falciparum [5,6]. However, several decades later it remains unclear which of the many antibody specificities contained in such sera may play a critical role, and hence which of the corresponding antigen(s) may represent potential vaccine candidate(s). Identification of such antigens requires the characterization of antibody species in relation to the precisely defined medical status of individuals exposed under field conditions, and in a longitudinal manner, since malaria attacks are spread over time. The corresponding epidemiological conditions were therefore established in an area of Senegal so as to fulfil two main features: (i) to include active case detection, i.e., by daily medical visits to each individual over several years and by providing access to medical care 24 h a day, (ii) making use of improved diagnostic criteria in order to distinguish malaria from other fevers, namely the pyrogenic threshold of parasite density as defined in this particular area [7], the validity of which was confirmed independently [8,9]. Employing these criteria represents a substantial improvement in the assessment of bona fide clinical malaria episodes since, as shown below, both refractory and susceptible individuals can be accurately identified using these criteria in all age groups. Most immunoclinical studies have dealt with a single antigen in a given location, precluding any comparative assessment of the relevance of each antigen. The aim of the present study was to correlate clinical protection in an endemic population with the immune response to five leading malarial vaccine candidates that are currently underway in, or about to enter, numerous clinical trials (see list of trials at http://www.who.int/vaccine_research/documents/en/malaria_table.pdf) [1]. Four of these molecules are the targets of antibodies that inhibit red blood cell invasion, namely merozoite surface protein 1 (MSP1) [10], MSP2 [11], apical membrane antigen 1 (AMA1) [12,13], and ring-infected erythrocyte surface antigen (RESA) [14], whereas MSP3 is targeted by cytophilic antibodies inhibiting intra-erythrocytic parasite growth in a monocyte-dependent manner [15]. Methods Study Area and Collection of Clinical Data The village of Dielmo (13°45′N, 16°25′W) is localized in one of the rare areas of Senegal, West Africa where malaria is holoendemic (experiencing perennially a high level of transmission by mosquitoes, due to the presence of a permanent stream), with an average of 5.16 infective bites per week during the first 2 y of the survey [16]. The 247 inhabitants of Dielmo village were enrolled in a prospective study using to our knowledge one of the most stringent protocols of clinical follow-up ever applied in the field and consisting of daily surveillance by medical staff (present 24 h/d, 7 d/wk) in order to identify and to analyse all episodes of morbidity [16]. The field set-up was designed and tested over 1 y before the actual study was conducted (e.g., questionnaires used for daily surveillance were written in three languages, and the reliability of responses were systematically addressed). Each villager was visited daily at home and had the ability to consult at any time one of the two medical doctors permanently on-site. In the event of a report or complaint of fever, headache, or vomiting, a medical examination and three thick blood films were made. One of the thick blood smears was Giemsa-stained and examined immediately on-site for the purpose of deciding on treatment. The other two slides were dehaemoglobinized, stained, and examined in our central laboratory in Dakar, using more rigorous and standardized conditions with quality control assessment [16]. The results of the latter slides were used for the present study. The criteria leading to a given episode of morbidity being attributed to malaria have been studied in detail and defined previously [7]: a malaria attack was defined as an episode of fever (temperature >38.5 °C) associated with a parasite density exceeding an age-dependent pyrogenic threshold described for this village (the parasite density threshold for each age group was determined to be 24,500 parasites/μl at ages 21 y age groups, respectively), though the majority had acquired protection. However, in adults, symptoms were of short duration and resolved spontaneously, i.e., without requiring treatment in most cases [7,16,36]. Figure 1 Means and Standard Deviations of Anti-MSP3-b IgG1, IgG2, IgG3, IgG4, and IgM Antibody Responses per Age Group Concentrations of antibodies are estimated as described in the Methods section. For each age group the mean number of malaria attacks recorded during the first year of follow-up is indicated by the shaded area. The numbers of individuals in each age group were: 5 (0–1 y), 14 (2–3 y), 17 (4–5 y), 14 (6–7 y), 12 (8–9 y), 13 (10–11 y), 11 (12–14 y), 18 (15–18 y), 13 19–21 y), 100 (>21 y), respectively. The existence of such rapidly acquired protection in very young children and, conversely, the occurrence of brief malaria attacks in adults, have seldom been reported previously, possibly due to the use of less-stringent surveillance setups. The pattern of clinical incidence in Dielmo remains similar to that described in other high-endemicity African areas [37] with a far greater number of attacks in younger individuals, e.g., 3.45 clinical attacks per year in children aged 1–5 y versus 0.1 attack per year in adults. Nevertheless, the daily surveillance of highly exposed villagers made it possible to distinguish in every age group individuals with or without malaria attacks who can therefore be referred to below as “nonprotected” and “protected,” respectively, for the duration of the 2 y follow-up period. The monthly surveys of entomological inoculations, conducted during the clinical follow-up period, showed a very high level of transmission, with an estimate of 260 infective mosquito bites per person per year, i.e., 520 parasite inoculations over the 2 y survey, with little house to house variation, leading us to exclude the hypothesis that villagers without malaria attacks were not exposed. This clinical situation was in most instances stable since only 13% of participants changed clinical status from the first to the second year of follow-up. The cases recorded are therefore well suited to the identification of immune correlates of protection and, furthermore, for the first time this analysis can be performed in an age-independent manner. Cytophilic Anti-MSP3 Antibodies Correlate with Clinical Immunity The overall prevalence of IgG and IgM antibodies against the MSP3-b epitope in the 217 individuals studied was high (97.2% and 93.1%, respectively), whereas the prevalence of IgA was low (19.3%). MSP3-specific IgE antibodies could not be detected (unpublished data). Among the IgGs, the two cytophilic classes, IgG1 and IgG3, were the most abundant (Figure 1) and they both increased as a function of age, whereas the increase was modest for IgG2 and IgM antibodies, and low for IgG4 antibodies in all age groups. Responses to other malaria antigens, including MSP1, MSP2-FC27, MSP2-3D7, RESA, and AMA-1, were also found to increase with age in this study (unpublished data) as has been observed in previous studies [38–42]. Therefore, due to this age-dependent increase, all antibodies measured showed an overall inverse relationship with the prevalence of clinical attacks (Figure 1, shaded area), a phenomenon reported in most previous studies in African settings and sometimes taken as indicating that protection is afforded by those antibodies [33,38,43–45]. The prevalence of responses to pre-erythrocytic antigens, such as CS, LSA1, and LSA3, also showed an age-dependent increase and confirmed the high degree of exposure to infected mosquitoes (see Figure S1). Antibodies against the pre-erythrocytic antigens were not included in the present statistical analysis. We first sought by stepwise regression analysis if it was possible to select a subset of antibody responses that would tend to predict the number of malaria attacks when controlling for age. We then examined the predictive value in terms of clinical protection of a positive antibody response for each IgG isotype, to each of the five molecules studied, for both each year separately and combined. A highly consistent association was observed between protection and anti-MSP3 IgG3 antibodies for each of the years studied, as well as for the two years combined, and this indication persisted even when controlling for age (age-adjusted odds ratio [OR] and 95% CI = 7.19 [2.70–22.85], p 20, p median) were considered high responders (closed circles). Individuals with antibody responses below the median (