Samples used in this study were obtained in March 2002 from a longitudinal study conducted in Dodowa, in which 352 children aged three to 10 years (in the active phase of acquiring immunity to malaria), were enrolled in a study, whose original aim had been to assess the role of cytokine regulation and immunity to malaria. Dodowa is a semi-rural town in the Dangme West District of the Greater Accra Region of Ghana, about 50 km from the capital Accra and is an area of moderate and stable malaria transmission with a seasonal peak. Bed net coverage in this area was low, about 10% 27. The study was approved by the Noguchi Memorial Institute's Ethical Review Board. After obtaining consent from parents, blood was obtained from each child and plasma stored at -20°C until use. Malaria episodes were detected using both active and passive surveillance implemented over a period of nine-months, spanning the entire malaria transmission season. Clinical and parasitological information was captured using a standard questionnaire. Each child was visited once a week and the child and the parents or guardians were asked about symptoms of malaria since the last visit and whether she/he had received any anti-malarial treatment. The child was then given a physical examination and the body temperature measured. Children with a history of fever within 48 hours and/or axillary temperature equal to or above 37.5°C had a rapid test for malaria parasitaemia using OptiMAL™ (DiaMed, FLOW Inc. Portland, Oregon) and then thick and thin blood films were prepared for microscopy for estimating the parasite density which was used later in the more specific malaria case definition employed for data analysis. Only children with measured or reported fever, and with a positive rapid test were treated with chloroquine (in accordance with the then prevailing national malaria treatment policy) and in the case of severe symptoms, the child was referred to the hospital. At monthly intervals, blood smears from finger pricks were obtained from all children irrespective of symptoms to estimate the prevalence of asymptomatic malaria infections. The parents of the children were instructed to report to the field team if the child had any symptoms of disease at any time. The qualifying case definition for malaria in the data analysis was reported fever and/or a measured temperature equal to or above 37.5°C, with parasitaemia ≥ 5,000 parasites/μl of blood; this case definition has been found to have 90% sensitivity and specificity in the study area 282930. Positive and negative control plasma used in ELISA measurements were obtained from adult Liberians and Danes respectively.

AMA1 was from the Pichia pastoris expressed ectodomain of Plasmodium falciparum FVO strain comprising amino acids 25–545 31 (Donated by A Thomas, Biomedical Primate Research Centre). GLURP was an Escherichia coli recombinant protein containing the conserved non-repeat N-terminal region (amino acids 25–514) called R0 32 (Donated by M. Theisen, Statens Serum Institut). MSP1₁₉ was a Baculovirus antigen of the C-terminal region of the merozoite protein surface 1, produced in insect cells infected with a recombinant Baculovirus containing a synthetic G-C enriched PfMSP1 gene (Palo Alto allele), coding for 43 N-terminal MSP1 precursor residues and 16 amino acid residues upstream of the "classical" MSP-1₁₉ (NIS---FCS) 33 (Donated by S. Longacre, Institut Pasteur). The MSP3 antigen used in this study was a long synthetic peptide called LR55 (amino acids 181 – 276) of the merozoite surface protein 3 34 (Donated by M. Theisen, Statens Serum Institut). All the antigens were provided through the AIA Project.

Specific isotype and IgG subclass levels against GLURP, MSP1₁₉, MSP3 and AMA1 were measured using indirect ELISA according to the AIA standard ELISA protocols 91011. All antigens tested were optimized and shown to be stable for at least three weeks, when antigen-coated plates and serum/plasma dilutions are refrigerated. The subclass specific reagents used were selected on the basis of low cross reactivities among themselves. To control for inter-assay and day-to-day variations in the standardized ELISA procedure, three-fold serial dilutions of reference standard reagents (IgG, IgM and IgG1 to IgG4) were directly coated on each ELISA plate (Maxisorp Nunc, Denmark) at a start concentration of 1,000 ng/ml (100 μl/well). OD values for the test samples were converted into antibody units with the standard reference curves generated for each ELISA plate using a four parameter curve-fit Microsoft Excel-based application. Samples were re-tested if the coefficient of variation between duplicate absorbance values were higher than 15% and plates were also re-tested if the R-square value of the standard curve was less than 97%. The reference standards, PBS buffer blank, positive and negative control plasma pools that were included in each ELISA test plate allow for the determination of failed assay runs. The AIA ELISA procedure used in this study is described in detail elsewhere 10.

Clinical data were double entered using Microsoft Fox Pro and immunological data using Excel. STATA version 9.2 (Statcorp, Texas) was used for statistical analysis. Children were considered to have a clinical malaria episode if they had parasitaemia of ≥ 5,000 parasites/μl, with a measured temperature ≥ 37.5°C or a history of fever in the last 48 hours 63536.

For each antigen, Poisson regression was used to investigate the association between the levels of antibody measured at baseline and the incidence rate of the first (or only) episode of clinical malaria. The level of total IgG, IgM and each IgG subclass, were analysed for each antigen in turn. Antibody values were transformed to log base 2, so that the rate ratio represents the ratio of malaria incidence corresponding to a doubling of antibody level. To investigate whether the relationship between malaria incidence and antibody level was nonlinear, a likelihood ratio test was used to compare the fit of the model when antibody level was included as a categorical or a continuous variable. When there were zero antibody values, indicating levels below the detection limit, the zero (left censored) values were assigned a nominal value equal to half the smallest measured value for that variable. If the proportion of zero values was large, the variable was treated as categorical with the reference category containing the zero values and the positive values divided into three equal groups. A likelihood ratio test was used to determine the P-value for the association with malaria incidence. Age at enrolment was considered to be an important potential confounder, and was included in the regressions as a factor with categories defined by quintiles. To model seasonality in malaria incidence, the calendar month of surveillance was included in the models as a factor. To construct a parsimonious model using all the immunological variables, firstly a model was produced for each antigen; in this model each IgG subclass, total IgG and IgM were candidates for inclusion provided the P-value for association with malaria incidence was 0.1 or less when considered individually. Variables were then removed from the model if the P-value for the likelihood ratio test was more than 0.1, provided removal did not change coefficients of variables in the model by more than 10%. In a second stage, the variables included in these models were candidates for inclusion in a final model derived in a similar way. Baseline parasitaemia was not considered as a potential confounder, but the interaction between each immunological variable and the presence of parasitaemia at baseline in their effects on malaria incidence was examined. A consequence of using a more specific case definition in the analysis than was used to decide treatment during the study is that children could have received anti-malarial treatments during the period they are considered at risk in the analysis. To explore the impact of these drug treatments, a time dependent variable was defined, to allow for a reduced risk of malaria for a period of 28 days after each drug treatment, which was included as a covariate in the Poisson regression model. LOESS smoothing was applied in R software to plot antibody levels in relation to age. Spearman's rank correlation test was used to assess associations between antibody levels and age.

Of the 352 children recruited for the original study, eight were lost to follow-up immediately after the baseline blood sampling. Of the 344 children followed up, sixty four (19%) had at least one episode of malaria (53 children had one episode, nine had two and two had three episodes). The incidence rate of malaria in the study cohort was 0.35 attacks per child per year (Table 1). The risk of clinical malaria decreased with increasing age 37. Sixty-six percent of the children had asymptomatic parasitaemia at baseline. Parasites were predominantly P. falciparum (95%) and the prevalence of parasitaemia measured each month was roughly constant (ranging from 50% to 65%). The incidence of clinical malaria, however, varied during the survey period, rising gradually from March to May, peaking in July and then decreasing until November (Figure 1).

Several studies in malaria endemic regions have shown increasing antibody levels with age. This pattern is more pronounced the greater the intensity and duration of malaria transmission. In this study, the relationship between antibody levels and age were assessed for antibodies against MSP1₁₉, AMA1, GLURP and MSP3. The levels of IgG and IgM to MSP1₁₉, MSP3, AMA1 and GLURP increased with age (Spearman correlation coefficient (r_s) 0.21 – 0.45; p < 0.001, Figure 2). IgG levels to AMA1 however, gradually increased until six years of age, and then leveled off. For the four antigens tested, IgG1, IgG2 and IgG3 significantly increased with age (r_s, 0.12 – 0.36; p < 0.03) with the exception of IgG2 to AMA1. There was no evidence that the level of IgG4 was associated with age for any of the antigens. Like IgG to AMA1, IgG3 to MSP3 steadily increased with age until seven years of age, then leveled off (Figure 3).

The levels of both IgG and IgG1 were highest for AMA1 and lowest for MSP3, whereas IgG levels to MSP1₁₉ and GLURP were comparable (Figures 2 and 3). The levels of IgG3 were higher in AMA1 than the comparable levels in GLURP, MSP1₁₉ and MSP3 (Figure 3). In general, the levels of cytophilic IgG1 and IgG3 were higher than those of non-cytophilic IgG2 and IgG4, IgG4 levels being the lowest.

Total IgG to MSP3, MSP1₁₉, and GLURP, and IgM to all four antigens tested (MSP3, MSP1₁₉, GLURP, and AMA1) were associated with reduced malaria incidence in crude analyses (Table 2). The incidence of both clinical malaria and antibody levels were associated with age, age is therefore a potential confounder, and it is important to adjust for its effects. After adjusting for the effect of age, there was evidence of a significant association between total IgG to MSP3, MSP1₁₉, GLURP and reduced risk of malaria (IgG to MSP3: rate ratio 0.69 (95%CI 0.53, 0.90) P = 0.01; IgG to MSP1₁₉: 0.75 (0.61, 0.92) P = 0.01; IgG to GLURP: 0.79 (0.64, 0.98) P = 0.04); and IgM levels to AMA1, MSP3, MSP1₁₉ were also significantly associated with reduced risk of malaria (Table 2).

A large proportion of the measurements of IgG4 to MSP3 and AMA1 and IgG2 to AMA1 were zero (left-censored) values and so these variables were treated as categorical variables (Table 3). In the crude analysis, IgG1, IgG2 and IgG3 to MSP3, MSP1₁₉, GLURP and also IgG4 to MSP3, were associated with a reduced risk of clinical malaria (Table 3, 4), but after adjustment for age, only one of these variables, IgG1 to MSP1₁₉, remained significantly associated with malaria incidence (rate ratio 0.89 (95%CI 0.80, 0.99), P = 0.04, Table 4).

When all immunological variables were considered simultaneously, only two variables were independently associated with reduced malaria incidence, IgG1 [(0.80 (0.66–0.96), p = 0.018)] to MSP1₁₉, and IgM [(0.48 (0.32–0.72), p < 0.001)] to MSP1₁₉ (Table 5).

The estimated rate ratio for the effect of antimalarial treatments given to children with parasitaemia < 5000/uL was 0.25 (95%CI 0.06–1.05), indicating these children had a substantially reduced risk of being found positive with parasitaemia >= 5000/ul during the 28 days following the treatment. But the estimates of rate ratios for other variables in the model were unchanged when this variable was included in the model suggesting treatment effects did not cause a bias in the estimation of the effects of immunological variables.