A total of 306 samples were successfully genotyped by semi-nested PCR. Overall 524 PCR fragments were generated (Table 1). There were 247, 145 and 132 fragments assigned to the K1, Mad20 and RO33 allelic families, respectively. Based on fragment size polymorphism, 32 and 23 K1-type and Mad20-type alleles could be identified [see Additional file 1]. All RO33 fragments were of the same size. The family frequencies were 47%, 28% and 25% for K1, Mad20 and RO33, respectively. The relative proportion of the three allelic families (Figure 1) did not show significant temporal fluctuations (Pearson test, Chi2 = 14.99; p = 0.663), was not influenced by age (Fisher's exact test, p = 0.813), gender (idem, p = 0.45), β-globin type (idem, p = 0.678 for AA vs. AS; p = 0.923 AA vs. AS vs. other β-globin variants), ABO blood group (idem p = 0.688) or Rhesus blood group (idem p = 0. 390).

Many samples contained more than one Pfmsp1 block2 type. The average multiplicity of infection estimated from the number of fragments detected (estimated moi - see Methods) was 1.73 Pfmsp1 block2 fragments/sample. This figure does most probably not reflect the actual number of distinct clones present in the patient, as distinct Pfmsp1 block2 alleles yet of similar size are not taken into account and as parasites with identical Pfmsp1 block2 alleles may differ in multiple other loci across their genome.

The number of Pfmsp1 block2 fragments detected was influenced by age (Kruskal Wallis test, p = 0.0192) (Figure 2); it was highest in the 2-5 y and 6-9 y old children and lowest in the ≥ 20 y old. It was not associated with gender (Kruskal Wallis test, p = 0.670), β-globin type (idem, p = 0.482), ABO or Rhesus blood group (idem, p = 0.234 and p = 0.839, respectively) or with year of study (idem, p = 0.508).

One or more K1-type and Mad-type 20 alleles were detected in 73% and 44% of the samples, respectively, while the RO33 family was observed in 43% of the patients. For each of the three families, the infection rate was not associated with gender (Fisher's exact test p = 0.164, 0.260, 0.289 for K1, Mad20 and RO33, respectively), β-globin type (Fisher's exact test p = 0.498, 0.704 and 0.384 for K1, Mad20 and RO33 respectively), ABO blood group (Fisher's exact test p = 0.195, 0.721 and 0.467 for K1, Mad20 and RO33, respectively) and Rhesus blood groups (Fisher's exact test p = 1.000, 0.268 and 0.370 for K1, Mad20 and RO33, respectively). Seasonality did, however, have an influence (Figure 3). The infection rates of K1-types were higher and those of Mad20-types lower in the November-January period (mean no. infected bites/month ± SD = 15.42 ± 10.07) than in February-May (idem = 10.78 ± 8.54) or June-October (idem = 31.53 ± 18.14) (Fishers' exact test p = 0.011 and p = 0.005, respectively). The RO33-type infection rates tended to be lower in February-May compared to the two other periods (Fishers' exact test, p = 0.061).

Direct sequencing generated high quality sequences on both strands for 358 fragments. The 358 sequences obtained accounted for 58% (144 of 247), 62% (90 of 145), and 94% (124 of 132) of the amplified K1, MAD20 and RO33 fragments, respectively, with a fair temporal distribution of sequenced fragments [see Additional file 2].

There was a large nucleotide sequence diversity, with a total of 126 alleles. In addition to the K1, Mad 20 and RO33 families, the fourth MR hybrid family identified recently 1116, was observed (Table 2). The six RO33-type alleles (RD0-5) differed only by point mutations. In addition to the previously reported G97D polymorphism 9 [see Additional file 3] four novel positions were dimorphic (Q72E, K90N, G91D and D104N). The K1 family presented the highest diversity, with 77 distinct K1 alleles (DK1-77). Sequencing of 90 fragments assessed by nested PCR to the Mad20 resulted in identifying only 34 distinct Mad20-type alleles (DM1-34) and nine hybrid types (DMR1-8, and DMRK).

The K1- Mad20- and hybrid-types mainly differed in copy number and arrangement of sequence motifs called in the literature "tripeptide motifs", while presenting no or minimal polymorphism of the flanking family-specific sequence. Five distinct tripeptide motifs were observed in the K1 types. In addition to the known SGT, SGP, SAQ and SGA motifs (coded 1-4, respectively according to 912), a novel motif was observed (motif

7

7

Similar findings were observed for the Mad20 types alleles, which differed mainly in the number, arrangement and coding sequence of six tripeptide motifs (coded 5-9). There were two synonymous sequences coding for SGG (5 and

5

5

5

5

Sequencing showed that 22 fragments assigned to the Mad20 family by semi-nested PCR were indeed Mad20/RO33 (MR) hybrids. We are confident that these alleles are bona fide hybrids and not artifactual PCR products, as they have been observed in 14 of 22 samples where a RO33 allele could not be detected using multiple family-specific nested PCR reactions. Moreover, 7 of 22 samples where the MR allele was detected by sequencing were monoinfections (i.e. there were no two partners for template switching). This MR hybrid family was quite diverse as eight alleles were observed. Allele DMR1 had a group1 type Mad20 while alleles DMR 2-8 derived from Mad20 group 2. All DMR alleles carried the same 25-residue long, RO33-type downstream region, which interestingly was a RD5 allelic type with a G97D D104N double mutation (Table 2). A novel hybrid, DMRK, displayed a RO33-K1 hybrid sequence in the family-specific 3' region (the K1 sequence located in 3' is underlined in Table 2) [for further analysis see Additional file 4].

The large local diversity was associated with a large number of low frequency alleles in the K1 and Mad20/MR family types, contrasting with the RO33 family where a dominant RD0 allele was observed in 78% (97 of 124) of the sequenced RO33-types alleles (Figure 5A). At the population level (Figure 5B), RD0 was by far the most frequent allele, accounting for 27% of the sequenced samples (top pie chart) and 19.7% of all alleles within the village when adjusted for relative family frequency estimated by nested PCR genotyping (bottom pie chart). The second most frequent allele after adjusting for family frequency was DK65 (adjusted frequency: 4.6%). Most alleles (107 of 126) presented a less than 1% frequency in the population sample studied here. In terms of frequency, the largest contribution among the top 19 alleles came from the RO33 family.

Sequence comparison of Dielmo alleles with the RO33-, K1-, Mad20-, and MR-types alleles deposited in the database [see Additional file 3], [see Additional file 5], [see Additional file 6], [see Additional file 7], respectively, showed that only 13 of the 126 alleles from Dielmo were identical to reported sequences from other endemic settings, namely K1-types (DK5, -7, -9, -18, -59, -65 -70 and -74), two Mad20-types (DM29 and DM34), two RO-33 types (RD0 corresponding to RO33-Ghana and RD4) and one MR allele (DMR4) [see Additional file 8]. Interestingly, only four of the top 19 alleles have been reported elsewhere in the world, (DK59, DK65, DMR4 and RD0), suggesting local diversification of the parasite population.

To gain some insight into a possible positive selection on this locus regarding the level (family and/or intra-family) and the type of selection operating, Ewens-Watterson-Slatkin tests for neutrality 3839 were conducted. At the family level, i.e. grouping alleles by family type considering three families irrespective of size or sequence polymorphism, this showed a significant departure from neutrality on a yearly basis and when grouping all years together, i.e. for a 10 year period (Table 3). Thus, there was evidence for balancing selection at the family level in this setting, the observed homozygosity being lower than expected (Table 3).

We then considered the within family diversity of the K1, Mad20/Hybrids (DMR and DMRK) and RO33 alleles separately to look for evidence of selection within each family (Table 3 lower panels). Tests were performed for each year separately or for the 10 year period. Alleles were differentiated by either size polymorphism or both size and sequence polymorphism. Overall, the null hypothesis was not rejected, implying that there was no evidence for significant within-family balancing selection on the Pfmsp1 block2 locus. The results of these Ewens-Watterson-Slatkin tests need to be interpreted with caution though. These tests are based on the assumption that no recurrent mutation has occurred at the locus studied. Since the mutation rate is known to be high in minisatellite/repetitive sequences, this assumption may be violated. In other words, one cannot exclude that recurrent mutations may have occurred and in turn have artificially reduced our power to detect balancing selection acting at the intra-family level.

Within the 124 RO33 PCR fragments sampled there was no size polymorphism and six different allele sequences were identified. An alignment of 126 nucleotides for all 124 alleles contained five polymorphic sites, all of which were non-synonymous single nucleotide polymorphisms. This indicates that dN/dS is infinite. Nucleotide diversity (π = average number of differences between any two sequences) was 4.84 × 10^-3. To examine the possibility of natural selection acting on the RO33 family, Tajima's D and Fu and Li's D* and F* were calculated 4041. In view of the high number of segregating sites (N = 5), these tests are expected to show high statistical power for natural selection. No evidence for departure from neutrality was obtained, with non significant Tajima's D value, Fu and Li's D* and F* values (Table 4), thus confirming results obtained using the Ewens-Watterson test.

The sequence-specific antibody response was studied by ELISA using biotinylated MSP1 block2-derived peptides bound to streptavidin-coated plates that overall represented a fair coverage of the sequence diversity observed in the village [see Additional file 9]. We recorded as seropositive any individual reacting with one or more peptide. Seroprevalence was analysed at the village level using an archived cross-sectional study conducted at the beginning of the 1998 rainy season, to which 85% of the villagers had contributed. We recorded as seropositive any individual reacting with one or more peptide. Overall, seroprevalence was 25% (62 of 243 sera analysed). Seroprevalence increased with age and reached 40.5% in adults (Figure 6). Confirming previous observations in this setting 2627, all anti-block2 IgGs were exclusively IgG3 [see Additional file 10]. No anti-block2 IgM was detected.

The frequency of recognition of each allelic family mirrored the frequency distribution of the family types within the parasite population (Figure 7A). The antibody reaction was family-specific and usually restricted to one family, with 73%, 23% and only 4% of the positive plasma reacting with one, two and three allelic families, respectively (Figure 7B), consistent with our previous survey in this village 27. Figure 7C shows that antibody response to pools 1 and 2, derived from the adjacent block 1 domain and block 3 respectively, was rare. No immunodominant region was identified within block2. Antibodies to the repeats were detected alongside antibodies to the family-specific N- or C-terminus block2 sequences. Interestingly, when scrutinizing the response at the individual peptide level in the context of the allelic sequence diversity within this village, it was clear that antibodies reacted with motifs displayed by the vast majority of the alleles observed in the village, namely either frequent/universal di-motifs (such as motif 31 in the K1 family) or family-specific unique sequences. Indeed, 24 of 26 villagers with antibodies to K1-type peptides reacted with sequences present in 74 or more of the 77 observed K1 alleles. Similarly, 16 of 16 responders to Mad20-type peptides reacted to sequences present in 32 or more of the 34 observed alleles.

In addition to the family-specific antibodies, some villagers had sequence-variant specific antibodies, namely reacted with only one of sibling peptides while others reacted with multiple sibling peptides displaying sequence variants. For example, within the group of sibling peptides derived from the N-terminus of Mad20 block2 (peptides #04, 13, 25, 11 and 29), some villagers reacted with one peptide (#29), whilst others reacted with two (#29 and 04 or 29 or 11), but none reacted with all five peptides. Likewise for the group of sibling peptides derived from the K1 block1/block2 junction (peptides #46, 61 and 74), some villagers reacted with one (#61), two (#61 and 74) or all three peptides. This suggests that sequence variation indeed translates into antigenic polymorphism. Whether antibody reaction with multiple sequence variants reflects serologic cross-reaction or accumulation of distinct antibody specificities is unclear.

To look for a putative association between the presence of anti-MSP1 block2 antibodies and protection against clinical malaria during the subsequent high transmission season, we mined the database for the occurrence of clinical attacks within the 5-month period following the 1998 cross-sectional blood sampling studied above. The 243 individuals experienced a total of 266 clinical malaria attacks (mean = 1.09, 95%CI: 0.88-1.30). The number of clinical malaria attacks experienced per individual varied from 0 (140 individuals) to 7 (1 individual). Recordings of the entomological inoculation rate indicated a mean of 170 infected bites/person during this time period. Twenty-nine percent of the seronegative individuals (with no detected anti-MSP1 block2 antibodies) experienced a clinical attack during that period, compared with 15% of individuals with anti-block2 antibodies. Using a Poisson regression model, the crude estimates of the Incidence Rate Ratio (IRR) of malaria attacks associated with the presence of antibodies to one allelic family or ≥ 2 families (no antibodies as reference group) were 0.55 (95%CI: 0.38-0.80) and 0.21 (95%CI: 0.08-0.58), respectively (P < 0.0001). In a multivariate Poisson regression analysis, this association was independent of haemoglobin type or ethnic group. However, it was confounded by age, i.e. within the age groups, there was no significant association between the incidence of clinical malaria attacks and the number of MSP1 block2 allelic families recognized.

To study the impact of novel infections during the transmission season on the humoral response to MSP1 block2, we investigated the fingerprick blood samples collected from 25 seropositive individuals throughout the high transmission season.

By the end of December 1998, namely five months after the cross-sectional sampling, the anti-MSP1 block2 antibody level was reduced by ≥ 2-fold in 15 subjects (59%), had varied less than 2-fold in 9 individuals (36%) (typical profiles are shown in Figure 8 upper and middle panel, respectively) and was ≥ 2-fold higher in one individual (Figure 8, lower panel). Importantly, when a change was observed, it concerned the intensity of the reaction but not its specificity. In other words, responding individuals usually reacted with the same pool(s) and within the pool(s) with the same individual peptide(s) before and after the transmission season. In none of the studied individuals were novel antibody specificities stably acquired during that time period, despite an elevated infection rate.

We then carried out a follow up of the antibody responses in villagers who experienced clinical malaria during the 5-month transmission season, using archived fingerprick sera collected monthly, and when available, sera on the day of the clinical malaria episode. Transient fluctuations were observed, with in some cases boosting of a pre-existing response (see a representative example in Figure 9A), in others a decrease in antibodies (idem Figure 9B) or evidence of a short-lived response (idem Figure 9C). This was also observed in children experiencing multiple clinical episodes during that same time period (idem Figure 9D). In nine out of 10 subjects in whom peripheral blood parasites collected at diagnosis of the clinical malaria episode were genotyped, the three allelic families were detected, and one individual harboured only 2 allelic families. In all 10 cases, infection with an allele against which there was no evidenced pre-existing response did not elicit any long lasting novel antibody specificity.

To analyse antibody patterns over several years, we used archived systematic blood samples collected during the longitudinal survey. Confirming a previous study in this village 27, once acquired, the response to MSP1-block2 was essentially fixed over time. A typical example is shown in Figure 10, where a 6-year follow-up was carried out on child 01/13, starting at 6 months of age. The child had been exposed to a mean of 200 infected bites each year over the six years. A single peptide pool was recognised by this child from the age of 2.5 years onwards (Figure 10A). The intensity of the signal fluctuated subsequently, including a drop during malaria attacks [e.g. the 2/11/98 blood sample, which was collected on the day a malaria episode was diagnosed presented a lower signal than the preceding (23/10/98) and following (4/12/98) samples], but nevertheless there was a progressive increase with cumulated exposure. Analysis of fine specificity on the individual constituents of peptide pool 11 showed the same pattern for all positive samples collected from this child with recognition of peptides # 46, 61 and 74, namely of the K1-specific block1-block2 junction (Figure 10B). The occurrence of clinical malaria episodes in this child resulted in temporarily reduced signals (hence antibody levels), but was not associated with stable acquisition of any novel specificity.

Dielmo, located in Sine Saloum, Senegal, is a village of approximately 250 inhabitants, where malaria is holoendemic. In 1990, the entire village population was enrolled in a longitudinal prospective study described in detail elsewhere 60. The main vectors in the village are Anopheles gambiae s.s. and An. funestus 59. Informed consent was obtained from each adult participant and from parents or legal guardians of each child at the beginning of the study and was renewed on a yearly basis. Individuals could withdraw from the study at any time. Each year the project was reviewed and approved by the Joint Ministry of Health and Pasteur Institute Surveillance Committee. The retrospective analysis has received ethical clearance from the National Ethical Committee of the Republic of Senegal.

We studied here 336 samples collected from mild malaria episodes selected from the existing collection of frozen blood samples and analysed for drug resistance markers 61. The sampling strategy was as follows: From a list of approx 3,400 samples collected longitudinally during a malaria episode, samples were chosen for molecular analysis so as to survey the largest possible panel of villagers. Since in this hyperendemic setting the heaviest clinical malaria burden is in the <10 y olds and since some children are more susceptible than others 62, we needed to avoid iteration bias due to the increased susceptibility of some individuals. This reduced the risk not only of over-representing certain genotypes to which some individuals might be more susceptible than others, but also of overestimating polymorphism, because each of the successive clinical malaria attacks experienced by one person is caused by "novel" parasites 48. We therefore set an interval of >3 years between two samples from the same individual, with the further restriction that no person could contribute with more than three samples in all. The number of samples studied each year is indicated in Table 1. Of the 336 samples selected, 306 were genotyped for Pfmsp1 block2. They originated from 229 villagers, with 159, 63 and 7 villagers contributing once, twice and three times, respectively, to the panel of Pfmsp1 block2-positive samples studied here. This included 120 males and 109 females [159 and 147 males and females, respectively, among the panel of samples studied]. The mean age ± standard deviation at the time of blood sampling was 11.5 ± 13.36 years (minimum = 0.3 y, maximum = 89.7 y, median = 7.3 y, Q25-Q75 = 3-13.9 y). [Mean age of males 12.08 y median age 6.9 y; mean age of females 10.82 y; median 7.9 y]. There were 275, 5, 29 and 1 samples from villagers with an AA, AC, AS and SS haemoglobin type (six of unknown type). The samples were from 31 out of the 34 village compounds.

Frozen blood samples were thawed and extracted with phenol-chloroform 48 and stored at -20°C until use. Pfmsp1 block2 was amplified by semi-nested PCR in a 50 μL reaction volume containing 5 μL DNA, 50 mM KCl, 1.5 mM MgCl₂, 10 mM Tris-HCl pH 9.0, 200 μM dNTP, 5 U Taq Polymerase (Amersham Pharmacia), 1 μM of each primer. For the first PCR, the conserved primers used were Fmsp1uf 5'GAAGATGCAGTATTGACAGG and Fmsp1ur 5'CATTAATTTCTTCATATCCATC. A first denaturation step at 96°C for 5 min was used, followed by 25 cycles of denaturation at 96°C for 1 min; hybridization at 64°C for 90 sec, extension at 72°C for 45 sec with a final extension at 72°C for 7 min. The semi-nested PCR was carried out using a forward family specific primer and the reverse conserved primer Fmsp1ur under the same conditions for 25 cycles except that the annealing was done at 68°C. The sequence of the forward family-specific primers was k1ff 5'ATGAAGAAGAAATTACTACAAAA, mad20ff 5'GAAGGAACAAGTGGAACAGC and ro33ff 5'TACTCAAGTTGTTGCAAAGC. PCR products were analysed on 1.5% Nusieve:agarose gels (1:3). The size of the bands was evaluated using a 100 bp DNA ladder (Bio-Rad) as size markers. Alleles were classified in 10 bp bins. A Pfmsp1 block2 genotype could be generated for 306 of the 336 samples. Of the 30 negative samples, one had a poor DNA quality (negative PCR for five loci tested), but the other 29 generated PCR products for other loci (Pfcrt, Pfdhfr-ts and microsatellite loci). Whether the failure to amplify Pfmsp1 block2 was due to polymorphism within the primer sequence or a lower sensitivity of the reaction as compared to the other loci is unknown. These DNAs were excluded from the analysis.

In the case of mixed infections where different alleles belonging to the same family were detected by size polymorphism, the bands of different size were excised from the agarose gel, re-amplified with specific primers to recheck the allele type.

PCR products obtained by semi-nested PCR using family specific forward primers were directly sequenced. All Pfmsp1 block2-derived PCR products were purified using polyacrylamide P-100 gel (Bio-Gel, Bio-Rad, 150-4174) on 96 well plates equipped with a 0.45 μm filter (96 well format, Millipore,1887, ref MAHVN4550). The purified product was quantitated by comparing it with DNA quantitation standards (Abgene^® QSK-101) after electrophoresis on 1.2% agarose gel. The sequencing reaction contained 2 μl of PCR product (≥ 20 ng), 1.25 μL 5× Buffer, 1.5 μL BigDye v3.1, 2 μL of 2 μM primer in a 10 μL final volume. Amplification was performed in a GeneAmp9700 (Applied Biosystem) [1 min at 94°C followed by 35 cycles of (10 sec at 96°C, 5 sec at 50°C and 4 min at 60°C), and held at 4°C. The products were then precipitated and sequenced on both strands using an ABI^® prism 3100 DNA analyzer as described [61]. There were a few cases where sequencing of the excised band proved not possible because of ambiguity in base calling, probably reflecting mixture of alleles with similar size. These samples were discarded from the analysis. We retained in the analysis only sequences where base calling was non ambiguous and the signal accounted for more than 95% of the signal for each individual base.

False recombinant alleles can be generated during PCR as a result of template switching, when long amplicons are generated, namely Pfmsp1 blocks 2-6, with cross-over sites identified in the distal part of block 3 and in block 5 63. To reduce the risk of this potential pitfall, short regions were amplified (i.e. upstream from the identified cross-over sites), with PCR anchored in conserved regions but relatively close to the junction with polymorphic sequences. Second, we verified absence of undesired amplification from deliberate mixture of reference alleles using the semi-nested PCR strategy The PCR fragments were sequenced and all were perfect match with the reported sequence.

Accession numbers are as follows: [Genbank: EU032016-EU032159, EU032160-EU032227, EU032227-EU032246, EU037095, EU032250-EU032276 and EU032248] for the DK, DM, RD0, RD1-5, DMR and DMRK sequences, respectively.

Pfmsp1 block2 alleles deposited in Genbank were retrieved by repeated blasting using each individual 9-mer nucleotide sequence observed in K1-type or Mad20-type alleles and the full length RO33-type block2 sequence. In addition, K1 alleles reported by Tetteh et al 15 originating from Zambia were included. The curation indicated by Miller et al 8 was included when needed. The various alleles were aligned using ClustalW and curated manually. Redundant alleles were discarded. This resulted in overall 59 distinct K1-type [see Additional file 5], 52 Mad20-type [see Additional file 6], four RO33-type [see Additional file 3] and nine MR-type alleles [see Additional file 7]. The alleles from Dielmo were compared to the reported alleles for the structure of the microsatellites: frequency of the individual tripeptide motifs, overall number of repeats, numbers of each individual tripeptide and combinations thereof (dimers, trimers and tetramers).

Allele distribution was analysed using the Ewens-Watterson-Slatkin (EWS) tests 3839. The test was applied considering a family as a single allele (i.e. grouping all alleles from that family together) or by considering individual alleles within each family independently. Individual alleles were then classified 1) by size and nucleotide sequence polymorphism or 2) by size polymorphism alone. Ewens-Watterson tests were performed using the software Pypop 64. Nucleotide diversity within the RO33 family was analysed using Tajima's D test 40 and Fu and Li's test 41 from DnaSP version 4.0 software developed by Rozas et al 65.

Archived sera, collected throughout the longitudinal follow up were used. Seroprevalence was studied using 243 plasma (i.e. 95% of the village population) collected during a cross-sectional survey conducted on 2-3 August 1998 at the beginning of the rainy season (27, 25, 26, 40 46 and 79 in the 0-2 y, 3-5, 6-8, 9-14, 15-24 and ≥25 y age groups, respectively). A subset of 25 sera collected in December 1998 from individuals whose August 1998 scored positive for antibodies to one or more MSP1-block2 derived peptides was analysed. A follow up of ten individuals during the 1998 rainy season was carried out using the monthly fingerprick blood samples collected on a systematic basis together with a fingerprick sample collected on diagnosis of clinical malaria when available. The entomological inoculation rate during the August-December 1998 period, assessed as described 59, was 170 infected bites/person. In addition, archived sera from children, collected longitudinally during the survey were used to follow the acquisition of antibodies over a period of several years.

A set of 82 15-mers derived from MSP1 block2 tripeptide repeats and the family-specific flanking region was synthesized by Chiron Mimotopes Pty. Ltd. (Clayton, Victoria, 3168, Australia). There were 34, 31, and 12 K1- Mad20- and RO33-specific sequences. In addition, 5 peptides derived from the junction with block1 were used. The peptide sequences are described in Table 5. The peptides represented the tripeptide combinations observed in Dielmo for the K1 and Mad20 families [see Additional file 9]. These peptides were synthesized with an N-terminal biotin group separated from the peptide sequence by a SGSG spacer and with an amidated C-terminus. All peptides were soluble. A similar set of peptides was used to explore the humoral response in Dielmo villagers in previous studies 2627. Based on these results, which showed a restricted specificity, and in view of the limited volume available for several sera, we first screened individual sera using 16 peptide pools (4-6 peptides per pool as described in Table 5) and in a second step analysed the reactivity of the positive sera on individual peptides from each positive pool. ELISA was performed on streptavidin-coated plates with either pools of 0.1 nM each biotinylated peptides or 0.5 nM biotinylated peptide adsorbed in each well as described 27. We checked with control mouse sera and individual human positive controls that peptide dilution within the pool of peptides did not modify the outcome of specificity analysis. Human plasma was tested in duplicate at a 1:500 dilution and bound IgG or IgM was measured using horseradish peroxidase-conjugated goat F(ab')2 to human IgG Fc (γ) or to human IgM Fc (μ) (Cappel, Organon-Technica, Turnhout, Belgium). Optical density (OD) was measured on an Emax reader (Molecular Device) at 450 nm. Control wells without peptide were used to check for potential anti-streptavidin antibodies. The wells that gave a signal twice the OD value of the wells without peptide were considered positive. IgG subclass analysis was performed as described 27.

This was done based on the data gathered during the longitudinal survey protocol and available in the database. Daily clinical surveillance was carried out over the August-December 1998 follow-up period, as described 6066. Each villager was visited at home for clinical surveillance and blood films were made in case of fever. The protocol included the notification of all febrile episodes to the medical staff and the controlled use of anti-malarial drugs. A malaria attack was defined as an association of symptoms suggesting malaria with parasitaemia above an age-specific threshold as described 6667. An anti-malarial drug cure was administered by the medical staff in all cases of malaria attacks. Procedures to estimate association with protection have been described 565768. In brief, the number of clinical malaria attacks experienced during the follow-up was analyzed as the dependent variable in a Poisson regression model, using the number of days of presence in the village as exposure variable. The association between the incidence of clinical malaria attacks and independent variables, i.e. presence of antibodies to allelic families, age, haemoglobin type or ethnic group, was tested.

Yearly distribution of the 524 PCR fragments by allelic family was analysed by Pearson Chi2 with the assumption that the alleles co-infecting the same individual were independent. Allelic family distribution by gender, age, Hb type, ABO group, Rhesus group and by month was analysed by Fisher's exact test. The allelic family infection rate (percentage of infected individuals harbouring one or more alleles from that family) by gender, β-globin type, ABO or Rhesus blood group, by age (0-1 y, 2-5 y, 6-9 y, 10-19 y and ≥20 y) and by season in the year was analysed by Fisher's exact test. For the analysis of seasonality, the year was divided into three periods based on the rains, the vectors present and the entomological inoculation rate. The mean entomological inoculation rate was 32, 140 and 39 infected bites/person/year in February-May (dry season), June-October (rainy season), and November-January, respectively.

The estimated multiplicity of infection was first analysed using a zero-truncated Poisson regression model, with the assumption of a constant probability to detect an additional allele in a homogeneous carrier population. The mean predicted estimated moi was 1.193 allele/infected individual. The predicted distribution was calculated, grouping the classes with estimated moi ≥ 4 and did not differ from the observed one (51.6% vs. 51.9%, 29.4% vs. 31%, 15.0% vs. 12.3%, 3.9% vs. 3.7% for observed vs. predicted estimated moi 1, 2, 3 and ≥4, respectively (Chi2 test, 3 df ≥ 2.53, p = 0.47). Estimated moi distribution by age group (0-1 y, 2-5 y, 6-9 y, 10-19 y and ≥20 y), gender, Hb type, ABO group, Rhesus blood group, year, month of the year and season was analysed by non parametric Kruskal-Wallis test.