The study used wild populations of A. gambiae s.l. and natural gametocytes of P. falciparum. Data were collected during four consecutive years, from 2007 to 2010, in the rainy season (August-November). Mosquitoes were collected as stage 3 and 4 larvae at the site of Goundry (12°30' N and 1°30' W), a rural village in the Sudan-Savanna ecological zone located 35 km north-east of Ouagadougou in Burkina Faso. The landscape is modified by human activity, creating many mosquito breeding sites during the rainy season, including a dam, mud brick pits, small ponds, hoof prints, puddles, and streams. Four population groups of A. gambiae s.l. are sympatric in and around Goundry, the A. gambiae s.s. groups M, S, and Goundry, and sibling species A. arabiensis 19 28 . The susceptibility of the Goundry subgroup to P. falciparum infection was reported elsewhere 19 , and as its role as a natural vector is not yet known, samples of the Goundry subgroup were not analysed as part of the current study.

Gametocyte carriers were identified in Laye, a village situated 30 km north-west of Ouagadougou, about 60 km from Goundry. As with Goundry, Laye is situated in the Sudan-Savanna zone with one rainy season from August to November. Laye was used because it is part of the Demographic Surveillance System (DSS) used by the Centre National de Recherche et de Formation sur le Paludisme (CNRFP), the national malaria research centre for epidemiological studies. Epidemiological characteristics of the study site were recently described 29 . There is little geographic population structure of P. falciparum in Africa 30 and populations from Burkina Faso and Mali were not genetically distinct 31 , so the small distance between Laye and Goundry should have no effect.

Throughout the rainy season ~50 different breeding sites were visited twice a week and mosquito larvae were collected by the standard ladle dipping method. Ten larvae from each of the 50 larval sites were pooled to create an experimental cohort of ~500 individuals. The larval contribution per breeding site was limited in order to avoid potential bias from oversampling siblings. The larvae from a collection cycle were reared in the CNRFP insectary in the water taken from their larval pool, after filtering to remove predators and debris. As water evaporated, the larval pans were topped up with distilled water. Otherwise, standard larval rearing procedures were used. In Goundry, larval sites are shared by M and S forms and A. arabiensis, and significant partitioning of molecular forms by pool has not been detected (of 66 genotyped single-pool larval collections, 11 were comprised of M form only, seven were A. arabiensis only, which reflects the greater overall prevalence of M form and A. arabiensis, and 48 were mixtures of any two or all three groups). Thus, the collection and rearing procedures randomize and therefore control for any hypothetical effects due to differential microbial flora of larval pools, which is unlikely here in any case due to predominant larval site sharing. Emerging adults were fed on 5% glucose, and adults emerging on days 2–4 were pooled in an age-matched group for blood feeding. An age spread this small does not influence infection susceptibility 32 .

In each experimental round, a group of ten potential gametocyte carriers five to ten years old were screened for malaria parasites by standard finger-prick and Giemsa-stained blood smear. The parasite density was determined for 100 microscopic fields of the blood smear. Assuming a microscope field contains ~20 leukocytes, parasites were counted for 2,000 leukocytes and parasite density was determined for the standard number of 8,000 leukocytes/μl of blood. The screening took place the same day as the experimental infection of mosquitoes. Carriers were chosen based on two criteria, as described and used previously 29 : i) infection with only a single species of malaria parasite, P. falciparum, and ii) presence of mature gametocytes. For carriers meeting these criteria, 5 ml of venous blood was drawn into heparinized tubes for experimental feedings.

For the infection experiments, female mosquitoes were starved for 12–15 hours prior to the infectious blood meal. Freshly drawn blood was immediately transferred to artificial membrane feeders pre-warmed to 37°C as described previously 33 34 . Mosquitoes were allowed to feed for 15 min, after which unfed mosquitoes were removed. Fed mosquitoes were maintained in the insectary on 5% glucose solution. Physical removal of unfed mosquitoes is highly efficient, but blood meal was confirmed for subsamples of mosquitoes by determining Christopher’s stages of ovarian development at the time of subsequent dissection 35 . The vast majority of unfed mosquitoes (91%, n=150) were A. arabiensis. A genetic analysis comparing fed and unfed A. arabiensis using eight microsatellites on chromosome 3, analysed using Genepop as previously described 19 , showed no detectable genetic differentiation between fed and unfed A. arabiensis (Fst=0). Thus, the unfed A. arabiensis are simply a random subset of the total, and the results of genetic susceptibility to P. falciparum are valid for A. arabiensis as a whole. The presence of oocysts (infection prevalence), and oocyst number for mosquitoes carrying at least one oocyst (infection intensity) was determined seven to eight days post-blood meal by dissection of midguts, staining with 2% mercurochrome, and light microscopy. Carcasses were stored for later DNA extraction.

A quality control (QC) filter was imposed to select the most informative infections. These were defined as infection sessions with overall infection prevalence ≥30% and maximum intensity in at least one individual mosquito of ≥10 oocysts. Note that the QC filter was imposed on the entire infection session, not on individual mosquito samples. Twenty-nine infection sessions satisfied the QC, and all of the mosquitoes from these 29 sessions were used in further analysis, while the mosquitoes from infections that did not satisfy the QC criteria were not used in the analysis. Infections not meeting QC can result from technical variables or biological effects, but since it is not possible to distinguish the cause, infections below QC criteria must be assumed to be enriched for false-negative infections, and were considered to be unreliable and mostly a source of noise. The same quality cut-off has been used previously for field infections 19 34 36 . A sample of infections not meeting QC were genotyped, and the population group composition was indistinguishable from the infections that did satisfy QC (A. gambiae M form p=0.551, A. gambiae S form p=0.175 and A. arabiensis p=0.625). There is also no significant difference between the overall sample size of infections that meet the QC threshold and those that do not (p=.957). Thus, the infection QC filter reduces noise, increases data quality, and does not introduce a genetic bias.

Genomic DNA was extracted from mosquitoes using DNAzol by the supplied protocol (Invitrogen, CA, USA) and samples were genotyped as described previously 19 . Briefly, species and molecular form were typed by either or both of two widely used molecular diagnostic assays 37 38 . As described in detail elsewhere, assignment to either the M, S, or Goundry subgroups was done by Bayesian clustering methods employed by STRUCTURE, analysing microsatellite genotype data from eight microsatellites on chromosome 3 19 39 , which assigned >95% of mosquitoes to a subgroup at >80% probability. The current study analysed the canonical reproductively isolated A. gambiae s.s. M and S forms (the Endo group in 19 ) as well as sister taxon A. arabiensis.

Infection prevalence was compared between three population groups (A. gambiae M form, S form, and A. arabiensis) using the Wilcoxon Signed Rank Test. The complete sample set analysed was comprised of 29 infections that met the QC criteria described above (n=2,311 mosquitoes). In each of these 29 infections, all three of the population groups were represented by ≥1 mosquito each. A subset of 13 of these 29 infections, in which all three population groups (A. gambiae M form, S form, and A. arabiensis) were represented within the same infection by at least seven mosquitoes each were subject to additional analysis. This subset of 13 most representative infections has greater statistical power because between-infection variables were controlled, including human host factors, parasite genotype, seasonal fluctuation, and technical variation. In addition to the Wilcoxon test, the 13 infections were also analysed by chi-square test followed by the combination of p-values across all infections for a given pair-wise comparison via the Fisher method 40 . The variance of the difference in infection prevalence between M and S forms was calculated by taking the absolute value of the difference in infection prevalence between M and S form within each of the 29 infections, and then calculating the variance of these differences. Infection intensity between population groups was compared using the Wilcoxon Signed Rank Test. Regression analyses were done to examine the correlation between infection prevalence and infection intensity. These analyses were done within a population group for the 13 most representative infections (as described above). For all analyses described, p<0.05 was the threshold for statistical significance.

The study protocol was reviewed and approved by the institutional and national health ethical review board (Commission Nationale d’Ethique en Santé) of Burkina Faso (code N° 2006–032). The study procedures, benefits and risks were explained to parents or legal guardians of children and their informed consent was obtained. Children of parents or guardians who had given consent were brought to CNRFP the day of the experiment for gametocyte carriers screening. All children were followed and symptomatic subjects were treated with the combination of artemether-lumefantrine (Coartem®) according to relevant regulations of the Burkina Faso Ministry of Health.

During the four-year study period, a total of 830 children five to ten years old were screened for P. falciparum infection. The prevalence of all malaria parasites was 70.7% in the survey population and 68.20% were P. falciparum carriers. Across these infected children, 33.4% presented with P. falciparum gametocytes. A total of 110 carriers were used for infection experiments, which yielded 29 infection sessions meeting quality control criteria (see Methods). For these 29 infection sessions, the mean age of gametocyte carriers was 7.6 years and the average gametocyte density was 170.6/μl. Descriptive statistics for the 29 infection sessions are given in Table 1.

All statistics were calculated using data from 29 individual experimental infections.

^a Child age in years, ^b Trophozoite density per ml of blood, ^c Gametocyte density per ml of blood, ^d Number of mosquitoes offered a blood meal, ^e Number of mosquitoes that took a blood meal, ^f Number of mosquitoes dissected seven to eight days following infection, ^g Number of mosquitoes that died between feeding and dissection, ^h Number of mosquitoes with at least one midgut oocyst, ⁱoocyst intensity (average number of midgut oocysts in mosquitoes with ≥1 midgut oocyst).

In total, 7,400 mosquitoes were offered an infective blood meal via membrane feeding. Of these, 31.0% (2,311/7,400) successfully fed and also survived until dissection one week later for determination of infection prevalence and intensity. After genotyping for species and assignment to mosquito population group, 2,168 mosquitoes from the 29 infections were analysed for comparative susceptibility to parasite infection. The analysed samples by year were n=595, 685 and 395 and 493 for the transmission seasons of 2007, 2008, 2009, and 2010, respectively. Across all study years, the population composition was 36.0% (n=780) A. arabiensis, 12.8% (n=277) A. gambiae M form, 18.7% (n=406) A. gambiae S form and 32.5% (n=705) A. gambiae Goundry subgroup. Infection susceptibility of the Goundry group was previously analysed 19 , and is not treated further in the current study.

Mosquitoes fed on different infective blood meals are known to display wide variation in infection distribution due to numerous variables 41 42 43 . Thus, it is not statistically valid to simply pool the data across all infections and calculate a single prevalence value per population group. Instead, it is necessary to compute the infection prevalence of each population group for each independent infection, and then combine and analyse these infection prevalence values among population groups. The infection prevalence for each population group was determined for each of 29 infections, and then these 29 prevalence values were compared between all population groups. The average infection prevalence across the 29 infections was, A. arabiensis (40.1%, n=780), A. gambiae M form (40.7%, n=277) and A. gambiae S form (41.4%, n=406). Infection prevalence was not significantly different between any of the three groups (Figure 1A, Wilcoxon Signed Rank Test on 29 prevalence values per population group; A. arabiensis vs A. gambiae M form p=0.897, A. gambiae M form vs A. gambiae S form p=0.779, A. arabiensis vs A. gambiae S form p=0.690).

The result is unchanged if only the 13 most representative infections are considered. These 13 infections, which consequently have the least noise and the greatest power to detect statistical difference, display equivalent infection prevalence values among population groups: A. arabiensis (38.6%, n=356), A. gambiae M form (40.2%, n=179) and A. gambiae S form (40.2%, n=234) (Table 1 and Figure 1B, Wilcoxon Signed Rank Test on 13 prevalence values per population group A. arabiensis vs A. gambiae M molecular form p=0.635, A. gambiae M form vs A. gambiae S form, p=1.000 and A. arabiensis vs A. gambiae S form p=0.519). An alternate method of statistical analysis of the 13 infections, by comparison of individual infection rates using the chi-square test and combining of p values using the approach of Fisher, also firmly rejected the hypothesis of infection prevalence differences between any of the population groups (A. arabiensis vs A. gambiae M molecular form, p=0.996, A. gambiae M form vs A. gambiae S form, p=0.894, A. arabiensis vs A. gambiae S molecular form, p=0.920).

Examination of the 29 infections (Table 2) reveals large variation across infections, as expected, while the average infection prevalence of the three population groups within a single infection tends to be similar (Figure 2A), consistent with the statistical analysis above. The similarity of the population groups within an infection is even more evident in the 13 most representative infections (Figure 2B). M and S molecular forms were analysed further by plotting the ratio of infection prevalence in M form versus S form mosquitoes across all 29 infections (Figure 3). There is no significant difference or even a trend between M and S infection prevalence in either direction, consistent with the statistical results above. It is informative to examine the distribution of all 29 infections (open circles in Figure 3) as compared to the 13 most representative infections (filled circles in Figure 3). Of the 13 infections, almost all are tightly clustered around the null average of zero difference in infection prevalence between M and S form (9/13, ~70% with differences in infection of ≤10%). However, of the 29 infections, only nine of them (31%) show an M versus S infection difference of ≤10%, and all of these are part of the 13-infection subset. Thus, with greater representation of both molecular forms within each of the 13 infections, the measured infection difference between molecular forms becomes smaller. The study reaches the conclusion that the greater dispersion of infection difference in the 29 infections is simply noise and not actual differences between M and S form. Consistent with this observation, the 29 infections displayed a variance of the absolute difference in prevalence between M and S forms (σ=317.2) that is 2.5 times the variance of infection difference in the 13 infections (σ= 125.7). Overall, these results indicate that much of the noise in infection distribution due to extraneous variables can be excluded by making direct comparison between population groups of adequate sample size within an infection, and then combining those results over multiple independent infections using Fisher’s method (see Methods).

Infection intensity (average number of midgut oocysts in all mosquitoes with ≥1 oocyst) was also compared across population groups. For the 29 infections, the average oocyst intensity was 5.7 oocysts/midgut for A. arabiensis, and 5.2 and 5.3 oocysts for A. gambiae M and S forms, respectively (Figure 4A). Pair-wise comparisons across the 29 infections showed no significant difference in infection intensity between population groups (Wilcoxon Signed Rank Test, A. arabiensis vs A. gambiae M form, p=0.323, A. arabiensis vs A. gambiae S form, p=0.144 and A. gambiae M form vs A. gambiae S form p=0.958). For the subset of 13 infections, the results are similar (Figure 4B, A. arabiensis vs A. gambiae M form p=0.206, A. arabiensis vs A. gambiae S form p= 0.414 and A. gambiae M vs A. gambiae S form p=0.893). As with infection prevalence, there is more variation in infection intensity across independent infections, and much less variation within infections (Figure 2, see points).

In addition, the correlation between infection prevalence and infection intensity in this unique large sample set of mosquito infection phenotypes was examined (Figure 5). There is a significant trend for increased average oocyst intensity with increased infection prevalence (p<0.001, r²=0.36), and this effect is also the same for all population groups individually (A. arabiensis p=0.020, A. gambiae M form p=0.002, A. gambiae S form p=0.085).