Clinical and parasitological characteristics of the recruited subjects are presented in Table 1. From July 2006 to January 2007, 102 children with symptomatic P. falciparum malaria were recruited in three health centres (Saint-Luc, Béthesda and Cotonou hospitals) located in the urban area of Cotonou where malaria transmission is perennial, with two seasonal peaks corresponding to rainy seasons, from April to July and September to November. A survey conducted in 2000 showed heterogeneous malaria transmission in the city of Cotonou, with transmission varying from 5, 29 and 47 infective bites per person per year near the beach, in the centre of the city and in the outer-urban lagoon areas, respectively 36. The inoculation rates were undoubtedly lower during the study period, due to the introduction of effective prevention and therapy. According to clinical and parasitological features at admission described elsewhere 37, children were assigned to the severe malaria (SM, n = 65) or uncomplicated malaria (UM, n = 37) group. SM was defined as the association of fever (axillary temperature ≥ 37.5°C), presence of circulating P. falciparum ring forms and a neurologic Blantyre score < 3. Forty-two SM children presented cerebral malaria (coma duration ≥ 6 h) and eight had severe malarial anaemia (haemoglobin level < 5 g/dL). The fifteen remaining SM children combined a Blantyre coma score < 3, coma duration < 6 h as well as haemoglobin level ≥ 5 g/dL but clinical signs of anaemia. For the UM group, inclusion criteria were fever or history of fever during the last 24 hours, the presence of circulating P. falciparum ring forms and a Blantyre coma score ≥ 3. Blood was withdrawn before the administration of any drug. An anti-malarial treatment was administered in accordance with the treatment practices of each hospital.

^a Median parasite density (interquartile range IQ25-75).

^b P value of the χ² test.

^c P value of the Student's unpaired-t-test.

^d P value of the Mann-Whitney U-test.

In December 2006, 52 children with asymptomatic falciparum parasitaemia (AP) were enrolled in the primary schools of Ouidah, a semi-rural town situated 35 km west from Cotonou. The entomological inoculation rate averaged 2.05 ± 1.28 infective bites per human per 100 nights in the nearby rural area of Tori Bossito 38. AP children were non-febrile at enrolment or during the preceding 24 hours, but presented circulating P. falciparum ring forms. In two out of 52 cases, the appearance of fever in a delay of 3 days after enrolment led to the administration of antipyretics associated with an artemisinin-based combination therapy (ACT), as recommended by the National Malaria Control Programme. Another blood sample was collected from the same children one year later (January 2008).

Plasma samples from 30 healthy adults (HA) living in the rural area of Tori-Bossito, which is located halfway of Cotonou and Ouidah were used as controls of the acquired anti-malarial immunity in this area. These adults were bled in November 2008.

For each individual, a venous blood sample was collected in EDTA Vacutainer tubes. Plasmas were collected and stored at -20°C for subsequent antibody testing. Blood smears were prepared to determine blood-stage infection and parasitaemia by microscopy. Negative control plasmas were obtained from 20 healthy Europeans adults who had not been exposed to malaria (Blood bank, EFS-Rungis, France) and were used either individually or grouped into a negative control pool (NC). For a positive control (PC), a pool of plasma taken from adult donors living in Dielmo, Senegal, was used, that showed high antibody reactivity to the surface of VarO-IE 31.

This study has been approved by the ethic committee of "Faculté des Sciences de la Santé" of the University of Abomey-Calavi in Benin. For each child, a written informed consent from parents or legal guardians was obtained. The study was conducted in accordance with the Declaration of Helsinki.

Single variant Palo Alto 89F5 VarO parasites were cultivated in human O⁺ RBC. Weekly enrichment of rosettes, monthly positive selection by panning on a mouse monoclonal antibody raised to the varO-NTS-DBL1α₁ domain were done as described 31. For surface immunofluorescence assay, an aliquot of rosette-enriched 89F5 VarO parasites was incubated with the plasma sample (final dilution 1:20). Total IgG binding was detected as described 31. Surface reactivity was expressed in arbitrary units (AU) as follows: [(% IE⁺_sample - % IE⁺_NC)/(% IE⁺_PC - % IE⁺_NC)] × 100. The mean surface reactivity plus 3 standard deviations observed with plasma samples from 20 non-immune malaria individuals was used to set the positivity threshold (20 AU).

For the rosette disruption assay, an aliquot of VarO rosette-enriched parasites (20 μL, 5% parasitaemia, > 85% rosetting frequency, in complete RPMI culture medium) was incubated with plasma (final dilution 1/5) for 30 min at 37°C as described 31. Each plasma sample was assayed in duplicate. The rosetting rate was compared to a control culture in complete RPMI medium in presence of the NC pool. The positive control was PC, a pool of hyper-immune sera of adults from Dielmo (Senegal) and able to disrupt VarO rosettes 31.

The soluble recombinant NTS-DBL1α₁ domain (residues 1-487 of the predicted varO amino-acid sequence, GenBank accession number EU908205) was produced in insect cells 31. Construction of the recodoned recombinant varO-CIDRγ (residues 508 to 787) and varO-DBL2βC2 (residues 831 to 1241) domains and production of recombinant proteins will be described in detail elsewhere (Guillotte et al, in preparation). Briefly, both domains were produced from a recodoned coding sequence where all potential N-glycosylation sites were mutated. VarO-DBL2βC2 was produced in the baculovirus/insect cells system whereas varO-CIDRγ was produced in Pichia pastoris. Each domain was produced as a soluble protein with a C-terminal hexa-histidine-tag. Protein purity was evaluated by sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE) and Western blot. The protein sequence was verified by N-terminal sequencing and mass spectrometry analysis. Alike the NTS-DBL1α₁ domain 3135 the recombinant varO-CIDRγ and varO-DBL2βC2 induced varO-IE surface reactive antibodies in the mouse (Guillotte et al, in preparation). The other varO-derived DBL domains (DBL3, 4 and 5) were not available as recombinant proteins with native folding and were not used in this study.

A crude extract of VarO-IE was prepared as described 39. IgG levels were quantified in plasma diluted 1/100 essentially by ELISA as described 31 using i) 100 μL of the varO-IE crude extract or ii) 0.2 μg of recombinant protein. For total immunoglobulin-G (IgG) detection, plates were probed with an horseradish peroxidase (HRP)-conjugate goat anti-human IgG-F(ab')2 (Cappel, France; dilution 1/7,500) 31. For IgG-subclass (IgG1 and IgG3) analysis, plates were first incubated with a mouse anti-IgG1 (clone NL16; dilution 1/2,000; Skybio, England) or anti-IgG3 (clone ZG4; dilution 1/10,000; Skybio, England) and probed with an HRP-conjugate goat anti-mouse IgG (Promega, France; dilution 1/3,000). Each serum was tested in duplicate. Negative (NC) and positive (PC) controls were included in each plate. Results were expressed in arbitrary units (AU) calculated with the formula: 100 × [ln(OD _{tested plasma}) - ln(OD _NC)]/[ln(OD _PC) - ln(OD _NC)] 40. The threshold for positivity was determined for each antigen and each class/subclass from the 95^th percentile of the antibody reactivity of 20 individual plasma samples from non-immune malaria individuals. For varO-NTS-DBL1α₁ it was 43.2, 14.0 and 17.3 AU for total IgG, IgG1 and IgG3 respectively; for varO-CIDRγ it was 30.6, 47.1 and 40.5 for total IgG, IgG1 and IgG3, respectively and for varO-DBL2βC2 it was 32.4, 70.3 and 39.6 for total IgG, IgG1 and IgG3, respectively. The positivity threshold was set at 9.0 AU for total IgG to VarO-IE crude extract.

Differences in proportions were analysed using the χ² test. Differences in means were tested by the non-parametric Mann-Whitney U-test, except for age, where the Student's unpaired t-test was employed as age was normally distributed. Statview 5.0 (SAS Institute Inc., Cary, NC) was used for these calculations. The associations between antibody responses and covariates (sex, age, P. falciparum carriage at blood drawing) found to be significant in the univariate analysis were investigated by multiple linear regression analysis using STATA (StataCorp. Release 8.0). The antibody levels in 2006 and 2008 for children sampled at these 2 time-points (n = 45) were compared using a non-parametric Mann-Whitney U-test. In case of P value less than 0.20 in the univariate analysis, a multiple linear regression was performed including age as covariate, to take into account the effect of aging on the evolution of antibody levels. For all tests, P values of less than 0.05 were considered significant.

Table 1 summarizes the main characteristics of the three groups of children recruited for this study. P. falciparum asymptomatic children (AP, n = 52) were older than children with uncomplicated malaria (UM, n = 37) and children with severe malaria (SM, n = 65) (Student's t-test, P < 0.0001, for comparison between the groups). Asymptomatic children presented a lower median parasite density than children with uncomplicated or severe malaria (P < 0.001). There was no difference regarding gender, age and parasite density between SM clinical sub-groups. The control group of immune adults (HA, n = 30) comprised 15 men and 15 women, older than 20 years, who were healthy at the time of blood drawing.

Antibodies (total IgG) reacting with the VarO-IE surface were studied using a single-variant parasite culture with more than 95% IE at mature stages expressing the PfEMP1-varO adhesin. All healthy, immune adults reacted against the VarO-IE surface. Seroprevalence rate was higher in AP children (92.3%, 95% confidence interval [CI], 85.1 to 99.6) than in UM and SM children (37.8% [95% CI, 22.2 to 53.5] and 26.2% [95% CI, 15.5 to 37.8], respectively) (Figure 1A). This difference between AP children and children with clinical malaria was highly significant (χ² test, P < 0001), but there was no statistical difference between UM and SM children as well as between UM and the CM sub-group of SM children.

The level of VarO-IE surface reactive antibodies differed in the three groups of children, from nil or very low in SM (median 11.21 AU, interquartile range IQ25-75 = 7.57 - 20.25 AU), to low in UM children (median 16.71 AU, IQ25-75 = 10.85 - 33.67 AU) and high in AP children (median 61.35, IQ25-75 = 35.19 - 97.5 AU). The highest levels were observed for HA adults (median 97.71, IQ25-75 = 57.54 - 101.76 AU) (Figure 1B). The surface-reactive antibody levels did not differ between UM and SM (whole group but also CM sub-group), but differed significantly between children with clinical malaria and AP children. Multivariate analysis conducted for children only confirmed the clinical malaria-dependence of the level of VarO-surface reactive antibodies (-17.07, P < 0.0001) in spite of a small positive effect of age on antibody levels (1.91, P = 0.052), the parasite density being without effect on these observations (P = 0.12).

The VarO-rosette dissociation assay showed that few HA sera displayed anti-varO rosetting activity. Only 2 of 30 sera disrupted > 50% of the VarO rosettes of the culture, and 6 disrupted 10-50% of the rosettes. Most sera had marginal to nil rosette disrupting activity, contrasting with sera from Senegalese adults (Figure 2). None of the children sera harbouring surface-reactive antibodies (be they from asymptomatic or symptomatic children) had detectable antibodies able to disrupt VarO rosettes.

Total IgG, IgG1 and IgG3 responses to the individual varO recombinant domains were analysed by ELISA. The sandwich ELISAs for IgG1 and IgG3 detected specific responses in some children scored as negative for total IgG reactivity, likely reflecting their higher sensitivity. To rule out a technical issue with the quality of the sera, total IgG reactivity with the VarO-IE crude extract was determined.

In none of the assays did UM and SM children, as well as UM and CM children, display a statistically different response (Figure 3). Total IgG seroprevalence to the recombinant antigens and the crude extract was similarly high in HA adults and AP children. Seroprevalence to the crude extract was lower in UM children. Total IgG and IgG1 seroprevalence to the three recombinant domains was higher in AP children than in UM and/or SM children. Similar results were observed for IgG3 except for seroprevalence to NTS-DBL1α₁ which was higher in AP children than UM children but not different from SM children (Figure 3).

NTS-DBL1α₁ was the most frequently recognized domain. For total IgG, the prevalence rates among children were NTS-DBL1α₁ > DBL2βC2 > CIDRγ (paired comparisons analysed by χ², all significant: odds-ratio (OR) NTS-DBL1α₁/DBL2βC2 = 7.7 [95% CI, 3.5 to 17.1]; NTS-DBL1α₁/CIDRγ = 19.5 [95% CI, 4.5 to 84.7]; DBL2βC2/CIDRγ = 8.9 [95% CI, 3.7 to 21.2]; P < 0.0001 for each). Similar results were obtained for IgG1 and IgG3 seroprevalence rates.

When examining the number of individual varO-derived recombinant domains recognized in each group, it was clear that HA adults had a broad reactivity, with most sera (70%) reacting with each of the three domains (Figure 4). The response in AP children was narrower, as about 50% and 30% had antibodies reacting with three or two domains, respectively. In contrast, the IgG response of children with clinical malaria (UM or SM) was quite restricted, with 49% and 42% respectively with no detected seroreactivity to any of the three antigens (Figure 4). The more sensitive IgG1 and IgG3 assays reduced the percentage of seronegative children in each clinical group, but still outlined a more restricted response of the UM and SM children compared to AP children (for each Ig assay, P < 0.0001 for AP vs. UM and AP vs. SM when comparing reactivity to 0-1 and 2-3 domains).

HA adults and AP children presented similar total IgG levels, except for VarO-IE crude extract (P = 0.0004) (Table 2). NTS-DBL1α₁ was the only recombinant domain to generate lower total IgG, IgG1 and IgG3 levels in UM than in AP children (P = 0.02, P < 0.0001 and P < 0.0001, respectively). SM children had lower total IgG, IgG1 and IgG3 levels to NTS-DBL1α₁ than AP children (P = 0.002, P < 0.0001 and P = 0.01, respectively), as well as lower total IgG and IgG1 to CIDRγ (P = 0.04 and P = 0.007, respectively). The antibody levels did not significantly differ in children with UM and SM, except total IgG to VarO-IE crude extract (P = 0.03) and IgG3 to NTS-DBL1α₁ (P = 0.007) (Table 2). Similar differences were observed when considering the CM sub-group of SM children, except for CIDRγ which elicited comparable antibody levels among HA, AP, UM and CM groups.

nd: not done

^a Median values (25th-75th percentiles), including responders only, and expressed in arbitrary units (AU).

Significant differences (Mann-Whitney U-test, P < 0.05) between median values of HA and AP responders (^b), HA and UM responders (^c), HA and SM responders (^d), AP and UM responders (^e), AP and SM responders (^f), and UM and SM responders (^g).

Importantly, age and parasite density were not associated with any of the antibody level assayed in each of the three groups of children, although this should be qualified because of the limited number of responders. A multivariate analysis taking into account sex, age and parasite density confirmed higher levels of total IgG and IgG subclasses in AP compared to symptomatic children (all antigens P < 0.02, except for total IgG to NTS-DBL1α₁ [P = 0.05] and to DBL2βC2 [P = 0.06]). The observed differences did not correlate with sex and parasite density, and in some cases, the presence of an age-related increase in antibody (IgG to NTS-DBL1α₁ [P = 0.001] and to DBL2βC2 [P < 0.0001] as well as IgG1 to DBL2βC2 [P < 0.0001]) did not counterbalance the strong impact of the clinical presentation.

In each clinical group, the total IgG and IgG1 levels reactive to an individual antigen were positively correlated (Table 3). Total IgG and IgG1 were moderately related to the IgG3 levels for NTS-DBL1α₁ (Rho from 0.39 to 0.59, all P < 0.02) and for CIDRγ (Rho from 0.38 to 0.49, all P < 0.003), and somewhat stronger with DBL2βC2 (Rho from 0.42 to 0.72, all P < 0.002) (Table 3). Results were the same when considering the CM sub-group, except for total IgG levels directed to CIDRγ which were not related to IgG1 and IgG3 levels (Rho = 0.20 and 0.26; P = 0.20 and 0.10, respectively).

^a Spearman's rank correlation test (Rho; P)

Positive correlations between domain-reactive antibodies were observed in the AP group for total IgG (NTS-DBL1α₁ vs. CIDRγ: Rho = 0.63, P < 0.0001; NTS-DBL1α₁ vs. DBL2βC2: Rho = 0.49, P = 0.0002 and CIDRγ vs. DBL2βC2: Rho = 0.53, P < 0.0001) and IgG1 (NTS-DBL1α₁ vs. CIDRγ: Rho = 0.40, P = 0.004; NTS-DBL1α₁ vs. DBL2βC2: Rho = 0.46, P = 0.0006 and CIDRγ vs. DBL2βC2: Rho = 0.40, P = 0.003). In the UM and SM groups, a strong correlation was observed between total IgG levels to NTS-DBL1α₁ and DBL2βC2 (Rho = 0.74 and 0.52, P < 0.0001, respectively) but not between the other domains, whether for total IgG or for IgG1 and IgG3. The same profile of relationships was confirmed for the CM sub-group of SM children.

For all four groups, the strongest correlations involving VarO-surface reactive antibodies were observed with total IgG to NTS-DBL1α₁ (Rho from 0.38 to 0.67, all P < 0.006) (Figure 5). A similar observation was valuable for the CM sub-group (Rho = 0.37, P = 0.02). Less numerous and strong correlations were observed between VarO-surface reactive antibodies and IgG1 or IgG3 levels to varO-domains. Antibodies to the VarO-IE crude extract correlated to none of the antibody level to the individual varO-domains in the HA and AP groups (all Rho < 0.06, all P > 0.18).

Forty-five out of the 52 children with asymptomatic P. falciparum infection in December 2006, were bled again in January 2008 in order to follow the temporal evolution of their specific antibody response. Among them, six were parasitaemic for P. falciparum in 2008, with a low median parasite density (78 ring forms per microliter of blood, IQ25-75 = 36-158), i.e. were asymptomatic parasite carriers. The prevalence rate of VarO-IE surface IgG was lower in 2008 than in 2006 but not that of VarO-IE crude extract IgG (additional file 1). A similar decrease between 2006 and 2008 was observed for the prevalence rates of total IgG, IgG1 and IgG3 to the three recombinant proteins, except for IgG1 to NTS-DBL1α₁ as well as total IgG and IgG1 to CIDRγ, which remained unchanged (additional file 1). Levels of antibodies to the VarO-IE crude extract dropped in 2008 (P = 0.0007) whereas antibodies reacting with the other antigens remained stable (additional file 1).

Multivariate analysis confirmed the negative effect of time, consistent with a diminution between 2006 and 2008 of VarO-IE crude extract antibodies (-18.0, P < 0.0001) independently of a positive effect of age (3.4, P = 0.04) reflecting the ongoing acquisition if immunity. A similar observation was made for IgG1 to NTS-DBL1α₁ with a negative effect of time (-10.0, P = 0.04) dominating the positive effect of age (4.2, P = 0.03).