Diverged Alleles of the Anopheles gambiae Leucine-Rich Repeat Gene APL1A Display Distinct Protective Profiles against Plasmodium falciparum

Functional studies have demonstrated a role for the Anopheles gambiae APL1A gene in resistance against the human malaria parasite, Plasmodium falciparum. Here, we exhaustively characterize the structure of the APL1 locus and show that three structurally different APL1A alleles segregate in the Ngousso colony. Genetic association combined with RNAi-mediated gene silencing revealed that APL1A alleles display distinct protective profiles against P. falciparum. One APL1A allele is sufficient to explain the protective phenotype of APL1A observed in silencing experiments. Epitope-tagged APL1A isoforms expressed in an in vitro hemocyte-like cell system showed that under assay conditions, the most protective APL1A isoform (APL1A2) localizes within large cytoplasmic vesicles, is not constitutively secreted, and forms only one protein complex, while a less protective isoform (APL1A1) is constitutively secreted in at least two protein complexes. The tested alleles are identical to natural variants in the wild A. gambiae population, suggesting that APL1A genetic variation could be a factor underlying natural heterogeneity of vector susceptibility to P. falciparum.

Introduction Most of the global transmission of human malaria due to Plasmodium falciparum occurs in Sub-Saharan Africa and is vectored primarily by the mosquito Anopheles gambiae [1]. Studies on the A. gambiae-P. falciparum pair have revealed a large genetic influence upon mosquito resistance to the human malaria parasite [2][3][4][5], although the causative variants underlying the mapped quantitative trait loci (QTLs) have not yet been resolved. Separately, a number of functional studies show that the immune system of A. gambiae protects mosquitoes against infection with P. falciparum, and multiple immune factors have been identified [6][7][8][9][10][11][12][13][14].
In A. gambiae mosquitoes from West and East Africa, genetic loci strongly linked to P. falciparum resistance traits were repeatedly mapped to the same segment of chromosome 2L, and were resolved to a ,10 Mb genetic interval [2,4,5]. The APL1 (Anopheles Plasmodium-responsive leucine-rich repeat 1) genes were identified as candidates within the resistance QTL cluster. The APL1 locus is ,18 kilobases (kb) in length and comprises three paralogs, APL1A, APL1B and APL1C [15].
The three APL1 genes encode a family of leucine-rich repeat (LRR) proteins that influence the survival of Plasmodium parasites, displaying distinct activities depending on the parasite species: APL1A protects against the human malaria parasite P. falciparum that is naturally transmitted by this vector, and APL1C against rodent malaria parasites [11,15]. The APL1C protein has been shown to form a heterodimer with another LRR protein, LRIM1, as part of a functional complex with the complement C3-like protein TEP1 [16][17][18][19]. Protein partners of APL1A and APL1B remain unknown.
The APL1 genes show a high degree of similarity in sequence and intron-exon structure, and they likely originate from gene duplication and diversification events. Each of the encoded APL1 proteins is characterized by the presence of a N-terminal signal sequence and a LRR region followed by a cysteine-rich tract, and some variants have a C-terminal coiled-coil domain. The APL1 locus comprises two additional genes also coding for LRR proteins, AGAP007034 (LRIM11) and AGAP007037 (LRIM3), which are not members of the APL1 family. Despite the overall structural resemblance with the APL1 proteins, these genes encode proteins belonging to the LRIM family [15,19,20].
In nature, APL1 genes display high genetic diversity, in a pattern consistent with adaptive maintenance of polymorphism [21]. The differential functional roles, if any, of APL1A alleles are not known. However, the mapped P. falciparum resistance QTLs that include the APL1 locus exert a strong influence on infection outcome [2,4]. The genetically mapped phenotypic effect is due to as yet unknown causative allelic variant(s) found within the QTL locus that includes APL1A, which remains a strong candidate gene. One previous report described fine genetic dissection of an A. gambiae locus, but the locus was linked to protection against the rodent parasite P. berghei, and does not correspond to the location of a P. falciparum-protective QTL [22].
Here we measured the individual phenotypes of APL1A alleles for protection against P. falciparum, using the Ngousso laboratory colony, after first validating the existence of the same alleles in the natural population. By genetic association studies using RNA interference-mediated knockdown assays, we evaluated the importance of the structurally different allelic variants of the APL1A gene against the human malaria parasite P. falciparum and observed distinct allele-specific protective profiles. Finally, in assays using cultured hemocyte-like cells we detected different subcellular localization and secretion patterns of the APL1A allelic isoforms.

Structure of the APL1 Locus
We generated overlapping PCR fragments over ,18 kb covering the APL1 locus from 20 individual Ngousso mosquitoes. The sizes of the obtained amplicons corresponded to the physical distances displayed in the A. gambiae PEST-strain reference genome [23], with the exception of fragments from two intergenic regions, the regions between APL1B -APL1A and APL1A -AGAP007037. These size differences are due to the presence of insertion-deletion variations (indels) and to the absence in Ngousso of the TA-III-Ag miniature inverted transposon (MITE transposable element) identified in PEST upstream of APL1A [15]. These results confirm that the APL1 locus in Ngousso has the same genomic organization as the PEST reference, and did not reveal the presence of any additional copies of the APL1 genes ( Figure 1).
Three APL1A Alleles Segregate in the Ngousso Colony Each of the three APL1 genes was amplified with specific primers flanking the coding region to avoid cross-amplification of the other paralogs due to their high similarity (Table S1). Sequencing of the diploid PCR fragments indicated a maximum of two alleles in each PCR reaction. We chose two homozygotes per allele and sequenced the cognate coding region. The structurally most polymorphic of the paralogs is APL1A, with three alleles: APL1A 1 , APL1A 2 and APL1A 3 ( Figure 2, and Supplementary Information S1), following the previous nomenclature of gene name with allele superscript [15,21]. All three Ngousso APL1A alleles present a similar overall structure, but show diagnostic indel polymorphisms ( Figure S1). The diagnostic differences allowed us to develop a genotyping assay for the APL1A alleles. The APL1A 3 allele is structurally nearly identical to the APL1A of the PEST reference.
The presence of allele specific differences in APL1A 2 and APL1A 3 leads to premature stop codons that result in proteins without the coiled-coil domain ( Figure 3, and Supplementary Information S1). The alleles, APL1A 1 and APL1A 2 , exist also in the G3 colony and in field caught mosquitoes from Mali [15,21]. In addition, we found four APL1A alleles in a sample set of wild mosquitoes carrying the standard chromosomal form 2La + /2La + from Burkina Faso (manuscript accepted [24]) and three of these alleles were structurally identical to Ngousso alleles APL1A 1 , APL1A 2 and APL1A 3 , overall indicating that the APL1A alleles observed in Ngousso are not colony artifacts or geographically limited forms. Instead, there appears to be a limited repertoire of allelic forms with major structural consequences for the encoded APL1A proteins.
Alleles of APL1B and APL1C in Ngousso and Haplotype Organization of the APL1 Locus Three APL1B allelic variants exist in Ngousso. The APL1B 1 and APL1B 2 alleles display only single nucleotide polymorphism (SNP) differences, and the encoded proteins correspond to the APL1B protein in PEST. The APL1B 3 allele, however, has a premature stop codon due to a point mutation at the beginning of the second exon (2L: 41268220), resulting in a 21 amino acid product comprising essentially just the signal peptide. We confirmed the existence of this truncated coding sequence variant in genome sequences from multiple samples from the Ngousso colony.
For the APL1C gene, the major allelic differences are due to variations in the copy number and composition of the encoded consensus amino acid sequence Pro-Ala-Asn-Gly-Gly-Leu (PANGGL) repeat in the N-terminal region of APL1C. The PANGGL repeat or its variants is encoded by all APL1C alleles and is also present in APL1A 2 but absent in the other APL1A alleles and all APL1B alleles. This amino acid repeat motif is not found in any other Anopheles genes, and its function is unknown.
In order to detect linkage between alleles, we sequenced the complete coding sequence of APL1B and APL1C from two mosquitoes homozygous for each of the three APL1A alleles (n = 6 for complete locus sequence). Alleles of the paralogs are linked to form three stable haplotypes in Ngousso ( Figure 3). Thus, each APL1A allele is consistently part of a specific combination of APL1B and C alleles. For example, the APL1A 3 -bearing haplotype also encodes the highly truncated APL1B 3 protein along with a complete APL1C protein. The haplotype structure of the locus in   wild mosquitoes is unknown, but the founding effect of colonization should in general produce more extensive haplotype structure than exists in nature.
The APL1A 2 Allele is Highly Protective against P. falciparum Infection We previously showed that APL1A displays a protective function against the human malaria parasite P. falciparum in A. gambiae Ngousso using gene silencing experiments [11]. Here, we examined the relative contribution of the three APL1A alleles to host protection. The high sequence similarity between the three APL1A alleles does not permit the design of double-stranded RNAs (dsRNAs) for allele-specific knockdown. Consequently, for gene silencing we injected a single dsRNA targeting all three APL1A alleles ( Figure S5) followed by infection challenge and individual mosquito genotyping to test association of infection status with APL1A allele. We used gene silencing of APL1A to specifically measure the APL1A effect, and isolate it from the influence, if any, of the linked alleles on the APL1-bearing haplotype. The haplotype is of unknown length, and could include the effects, possibly even contradictory in terms of phenotype, of tens or more linked alleles, including but not limited to APL1B and APL1C.
We compared the level of P. falciparum infection for each APL1A allele between the states of normal APL1A function (mosquitoes treated with control dsRNA dsGFP) and APL1A loss-of-function (treated with dsRNA for APL1A, dsAPL1A). In each of three independent replicate infections (Table S2A, Figure S2), APL1A was strongly protective, because its silencing rendered mosquitoes significantly more susceptible to P. falciparum infection by increasing infection prevalence (combined p-value for infection prevalence, 2.25e-05; Figure 4), but without affecting infection intensity ( Figure S2).
When analyzing the results by APL1A allele, only the APL1A 2 allele displayed a consistent protective effect, because the global silencing of all APL1A alleles caused a highly significant increase of infection prevalence only among carriers of the APL1A 2 allele (p = 5.07e-07; Table 1). The protective phenotype is seen in APL1A 2 carriers regardless of the other APL1A allele present in the diploid mosquito, and thus demonstrates a genetically and functionally dominant protective effect of the APL1A 2 allele against P. falciparum infection. There was insufficient sample size to test whether the APL1A 2 /APL1A 2 homozygous genotype is even more protective than the APL1A 2 allele alone.
No statistically significant allele specific effect on the infection intensities was detected in the mosquitoes of the three replicates injected with dsAPL1A (data not shown).
In contrast, mosquitoes carrying the APL1A 3 allele showed no significant difference in infection outcome after dsAPL1A treatment as compared to dsGFP-treated controls (p = 0.24722), indicating that function of the APL1A 3 allele does not underlie and is not required for the protection against P. falciparum mediated by the global silencing of APL1A. Finally, allele APL1A 1 carriers displayed a significant difference of infection in only one replicate. The combined p-value for APL1A 1 influence on infection prevalence is significant (p = 0.00059), although three orders of magnitude less than for APL1A 2 , but we regard it with caution because replicates 1 and 2, which were individually not significant, were still not significant when pooled to a single group before pvalue calculation, indicating that the significant value in the APL1A 1 combined meta-analysis was generated solely by replicate 3. Statistical power analysis indicates that sample sizes were sufficient to detect an effect of APL1 1 and APL1A 3 , if one exists. For both APL1A 1 and APL1A 3 , there is power (.0.80 probability) to detect a difference in infection prevalence of the same magnitude as the effect of APL1A 2 (,30% infection difference) in either replicate 3 alone, or by pooling replicates 1 and 2. Thus, the experiments were properly powered and the result for APL1A 3 can be considered a robust negative. For APL1A 1 , replicates 1 and 2 were robustly negative, but due to replicate 3 we regard APL1A 1 as at most weakly protective against P. falciparum. Taken together, these results show that, at least in the Ngousso colony, the APL1A 2 allele is necessary and sufficient to explain the protective effect of APL1A against P. falciparum, while APL1A 3 plays no part in protection, and APL1A 1 has no, or at most weak, effect.

Role of Ngousso APL1A Haplotypes in P. falciparum Infections
We used the APL1A allele genotyping assay to test for an effect upon infection of the entire linked haplotype, rather than APL1A alone as tested in the dsRNA silencing assays. In three independent replicates (with n = 300 mosquitoes per replicate), mosquitoes were infected with P. falciparum in the absence of the dsRNA treatment used above (Table S2B, Figure S3). Despite the large sample sets, no significant association was observed between APL1A haplotypes as alleles or genotypes (data not shown). In each replicate, the observed genotype distribution was in Hardy-Weinberg equilibrium and each replicate had sufficient statistical power (.0.8 probability) to detect differences in infection prevalence of $20% between mosquitoes carrying different APL1A alleles, thus substantiating a robust negative result for haplotype effect. Sample sizes of haplotype combinations were too small ( Figure S4) to test for genotype effects (data not shown).

APL1A Alleles Encode Proteins with Different Subcellular Localization and Patterns of Secretion
The large structural differences between predicted APL1A protein isoforms ( Figure 3) likely have functional consequences, which could explain the different protective profiles against P. falciparum infection. It is not possible to raise APL1A isoformspecific antibodies due to high peptide sequence similarity between the three structural variants. Thus, to examine protein function, we transfected the A. gambiae hemocyte-like cell line 4a-3A [25] with constructs expressing either APL1A 1 , APL1A 2 or APL1A 3 bearing a C-terminal V5-tag. Immunostaining of transfected cells with anti-V5 monoclonal antibody (mAb) showed that the weakly protective APL1A 1 isoform exhibited a diffuse distribution throughout the cytoplasm, while the protective APL1A 2 and non-protective APL1A 3 isoforms were essentially localized within large vesicle-like structures ( Figure 5A).
Comparison of protein processing by immunoblotting of washed cells and culture medium revealed that APL1A 1 was constitutively secreted into the culture medium, while APL1A 2 and APL1A 3 were retained in the cell cytoplasm ( Figure 5B). The differential constitutive secretion is even more striking given that all proteins carry a classical N-terminal peptide signal sequence [21]. Under non-reducing conditions, APL1A 1 formed two high molecular weight complexes, consistent with predicted sizes of homodimer and heterodimer, in addition to the monomeric form expected at 76 kDa ( Figure 5B, left panel). In contrast, APL1A 2 and APL1A 3 only formed complexes with sizes consistent with homodimers. Their inability to form heterodimers with other partners may be consistent with the absence of a C-terminal coiled-coil domain, and, more importantly, with the absence of a cysteine residue corresponding to the Cys-562 of APL1C (AGAP007033), which is responsible for the disulfide-bridged complex formed between LRIM1 and APL1C ( Figure 5C) [18,25]. These results suggest that the secretion of APL1A proteins may be dependent on domain interactions with binding partners. Interestingly, because these alleles are all found in nature, the observed polymorphism for the presence of a cysteine in APL1A equivalent to the APL1C C562 and consequent variation in protein partner interaction must be adaptive for APL1A function in the organism.

Discussion
Natural populations of A. gambiae display such high genetic diversity and low linkage disequilibrium that it will be difficult to carry out genetic association studies using the wild population [26,27]. This is particularly true for highly polymorphic genes, such as the P. falciparum-protective APL1A, coupled with the inherently noisy phenotype of P. falciparum infection [21]. Instead, it may be more efficient to perform initial variant discovery in a defined context that captures limited natural variation, followed by validation in the field.
In the current work, we show that the Ngousso colony has captured a repertoire of natural allelic variants of the APL1 paralogs that result in structurally different proteins. The structurally most polymorphic paralog is APL1A, with three different alleles in Ngousso. The three APL1A alleles exist in nature, demonstrating that they are not laboratory colony mutations. Two alleles, APL1A 1 and APL1A 2 , are also found in the A. gambiae G3 colony [15,21], which is much older than Ngousso and was sampled from a different geographic location in Africa [28]. Thus, the same APL1A alleles found in Ngousso have existed in nature over both space and time.
A. gambiae colonies are expected to segregate longer haplotypes than found in nature due to small effective population size and the founder effect, and we confirmed the haplotype structure of the APL1 locus in the Ngousso colony. Consequently, in order to specifically measure the relative contribution of the three APL1A alleles to protection against P. falciparum, we globally silenced all variants of the APL1A gene, determined individual phenotypes for P. falciparum infection, and genotyped mosquitoes for APL1A using a semi-automated assay. When APL1A is silenced, the APL1A 2 allele accounts for essentially all of the dsAPL1A effect on P.  (1) Chi-square p-value comparing GFPkd versus APL1Akd for each experiment (EXP.). (2) For each allele, the p-value of the 3 experiments were combined by meta-analysis. This table presents the results from three independent APL1A gene-silencing assays (shown in Figure 4). From each mosquito both APL1A alleles were determined using the APL1A-RFLP test.  Immunofluorescence analysis of 4a-3A hemocyte-like cells transfected with plasmids encoding V5-tagged APL1A 1 , APL1A 2 and APL1A 3 . Cells were stained with Hoechst 33342 to label nuclei (blue). Staining with a mouse anti-V5 mAb followed by anti-mouse Alexa 488-conjugated IgG (red) indicates that APL1A 1 exhibits a diffuse pattern in the cytoplasm whereas APL1A 2 and APL1A 3 are essentially localized in vesicles. Pictures were taken under equivalent exposure conditions. B) Immunoblot analysis of cells (C) and culture medium (M) of the 4a-3A hemocyte-like cell line transfected with plasmids encoding V5-tagged APL1A 1 , APL1A 2 and APL1A 3 under reducing (R) and non-reducing (NR) conditions. Immunoblots were probed with a mouse anti-V5 mAb, protein quantities on the different blots are not comparable. APL1A 1 is secreted in the culture medium as at least two protein complexes under non-reducing conditions, whereas APL1A 2 and APL1A 3 are retained in the cell cytoplasm and form only one protein complex under non-reducing conditions. Estimated sizes of monomeric APL1A forms including V5-tag are: 76 kDa (APL1A 1 ), 60 kDa (APL1A 2 ) and 51 kDa (APL1A 3 ). C) Amino acid alignment of cysteine-rich regions of the three Ngousso APL1A proteins and A. gambiae PEST APL1C. Numbers correspond to the amino acid positions. Blue stars represent stop codons, cysteine residues are highlighted in red. The cysteine residue corresponding to the position 562 in the APL1C sequence (APL1C C562 , black arrow) described by Povelones et al. [18], is involved in the disulfidelinked complex formed between LRIM1 and APL1C and is only conserved in APL1A 1 . The same cysteine is also referred to as APL1C C551 in the APL1C protein published by Baxter et al. [34]. falciparum infection prevalence. The APL1A 1 allele has weak or no effect, and APL1A 3 has no influence on infection outcome. We did not detect an association between the haplotype bearing the APL1 locus and infection outcome, but the haplotypes have captured linked variation for an unknown number of genes, including ones that may be able to influence infection phenotype.
The considerable structural differences of APL1 proteins, particularly of APL1A, suggest that the isoforms may interact with distinct protein partners. This hypothesis is supported by the different migration profiles of APL1A isoforms under nonreducing conditions. The APL1C protein forms a complex with LRIM1 via conserved cysteine residues, and this complex can bind, through the coiled-coil domain, to certain TEP proteins [17,18]. Interestingly, APL1A 2 and APL1A 3 , which lack a coiledcoil domain, form only one type of dimer under non-reducing conditions and are retained in the cytoplasm of hemocyte-like cells, whereas APL1A 1 , which has a coiled-coil domain and forms at least two different dimers, is secreted from the cell. The observed dimers of APL1A 2 and APL1A 3 show a size consistent with homodimers, though we cannot rule out that they could be heterodimers with unknown binding partners of similar molecular weight.
The mechanism underlying APL1A 1 secretion remains elusive and it is not clear whether the secretion pattern is related to protection against Plasmodium infection. The behavior of APL1A alleles in culture does not necessarily translate to what happens in vivo. For example, absence of secretion of APL1A 2 and APL1A 3 may indicate that these alleles are not correctly folded in 4a-3A cells or that epitope tagging impairs their secretion. In addition, the APL1A 3 allele lacks the Cys-Cys pair common to the entire LRIM family of LRR proteins (highlighted in Figure 5C) and this motif has been experimentally shown to be important for LRIM1 folding and secretion [18]. This suggests that APL1A 3 may not be correctly folded and secreted even in vivo. Moreover, if APL1A secretion is dependent on interaction with binding partners, the putative APL1A 2 and APL1A 3 binding partners may not be expressed in 4a-3A cells. An alternative hypothesis could be that APL1A 1 , APL1A 2 and APL1A 3 are secreted through distinct mechanisms, and that the variants participate in resistance against Plasmodium in different ways. In vivo, APL1A 1 might be constitutively secreted, while APL1A 2 and APL1A 3 secretion might be stimulated by immune elicitors or other factors. The potentially constitutive presence of APL1A 1 in the hemolymph, if associated with a long protein half-life, could then explain why gene silencing experiments did not show a consistent protective effect for the APL1A 1 allele. Finally, another hypothesis is that APL1A 2 functions as a dominant negative inhibitor of a factor required for P. falciparum development. Further work will be necessary to distinguish between these possibilities.
Our findings highlight the need to consider allelic variation of putative immune factors, including the spectrum of alleles segregating in a colony, in studies aimed at deciphering their function. We also demonstrate the utility of performing a discovery step in a recent colony that segregates a defined set of allelic variants that can be confirmed to exist in nature. The long haplotypes in colonies limit resolution, so in general it may not be meaningful to directly test genotype-phenotype association in colonies, because the allele genotypes are just a proxy for the longer haplotype. Rather, it is best to distinguish the effects of the candidate gene effects, and the influence of the carrying haplotype, either by allele-specific silencing when possible [22], or if the alleles offer insufficiently distinct dsRNA targets as for APL1A, by global gene silencing with allele-specific genotyping as we did here. The information gained should then permit the design of statistically well-powered tests of small numbers of candidate variants in the natural population, where resolving power is greater due to the low linkage disequilibrium. Based on the current findings, it would be interesting to directly compare phenotypic outcomes of APL1A 1 , APL1A 2 and APL1A 3 carriers in the wild population after challenge with P. falciparum.

Mosquito Rearing
The A. gambiae s.s. colony Ngousso was established with mosquitoes captured in Yaounde, Cameroon in January 2006 [29]. The mosquitoes are of the M molecular and Forest chromosomal form, fixed for the standard 2La chromosomal inversion. The colony was reared at the CEPIA mosquito production facility at the Institut Pasteur under standard rearing conditions at 26uC and 80% relative humidity, under a 12 h light/ dark cycle as described elsewhere [29].

P. falciparum Gametocyte Culture and Ngousso Infection
Parasite culture and experimental feedings with the P. falciparum isolate NF54 were done as previously described [11]. Briefly, P. falciparum NF54 was cultured using the automated tipper-table system of Ponnudurai [30] implemented in the CEPIA mosquito facility of Institut Pasteur. Fourteen days after initiating the parasite subculture and prior to each infection experiment, gametocyte maturity was assessed by testing exflagellation of male microgametes. Gametocytaemia and proportions of mature male and female gametocytes were determined on Giemsa stained slides (Table S2).
For an infectious blood meal, 10 ml of the gametocyte culture were then centrifuged at 2000 rpm, and the cell pellet was resuspended in an equal volume of normal type AB human serum. The infected erythrocytes were added to fresh erythrocytes in AB human serum and transferred into a membrane feeder warmed to 37uC. Female mosquitoes (4-5 days old) were allowed to feed for 15 min, unfed mosquitoes were removed and only fully engorged females were maintained on 10% sucrose solution for further analysis.
Each infection experiment replicate was done with a new generation of Ngousso females and a new gametocyte culture of P. falciparum NF54.

Analysis of Infection Phenotypes
As infection phenotypes we analyzed prevalence and intensity. Infection prevalence is the fraction of mosquitoes carrying at least one oocyst, while parasite intensity is the number of oocysts per mosquito determined only in the subset of mosquitoes with $1 oocyst.
To determine infection phenotype, midguts of bloodfed females were dissected 7-8 days post-infection, stained in 16PBS buffer with 0.4% mercury dibromofluorescein (Sigma) and the number of oocysts per midgut was determined using a light microscope. Carcasses of the dissected mosquitoes were immediately transferred individually into a fresh tube and stored at 220uC until DNA extraction. Amongst the infection experiments for the association study we analyzed only those trials where infection prevalence was between 35 and 55% (Table S2B).
In infection experiments with dsRNA injection, we retained only those trials where infection prevalence of mosquitoes treated with the control dsRNA, dsGFP, satisfied this criterion (Table  S2A). Differences in infection prevalence were analyzed using the Chi-Square test. For intensity-related infection, non-parametric statistical tests were used, including the Mann-Whitney Rank Sum Test and the Kruskal-Wallis ANOVA on ranks, excluding mosquitoes with zero oocysts. Three independent replicate infections were performed and data were pooled prior to statistical analysis.

DNA Preparation
Genomic DNA was extracted from individual female mosquitoes by homogenizing in 100 ml DNAzol (Invitrogen, CA, USA) using a disposable pestle, essentially following the manufacturer's protocol.

PCR Reactions
The APL1 paralogs correspond to the VectorBase identifiers APL1A (AGAP007036), APL1B, (AGAP007035) and APL1C, (AGAP007033). PCR primers were designed based on the sequence of the A. gambiae PEST reference genome or on available Ngousso sequences. Primers were designed to bind in the exon regions of the APL1 genes. Genomic DNA of a single mosquito was used for PCR and sequencing reactions. For evaluation of size differences of PCR fragments, amplification reactions were performed in a final volume of 20 ml using Taq DNA polymerase (Invitrogen). PCR cycles were as follows: a denaturation step at 94uC for 3 minutes, followed by 40 cycles of 94uC for 30 sec, 62uC for 45 sec, 72uC for 3 minutes and a final extension of 72uC for 10 minutes. Primers and their positions are given in Table S1.
DNA amplification for sequencing was done using AccuPrime SuperMixII (Invitrogen). Primers carried a 59-extension for sequencing with the universal forward (-21 M13, 59 TGT AAA ACG ACG GCC AGT 39) or reverse (M13reverse, 59 CAG GAA ACA GCT ATG ACC 39) primers [31]. Final reaction volume was 50 ml and cycling conditions for amplification used denaturation at 94uC for 3 minutes, followed by 40 cycles at 94uC for 30 sec, 62uC for 45 sec and 72uC for 3 minutes and a final extension step of 72uC for 10 minutes.
Sequencing PCR products were sequenced using ABI Big Dye Terminator v.3.1 Cycle Sequencing kit (LifeTechnologie) and an ABI Prism 3730 DNA Analyzer (Applied Biosystems). The sequences were assembled using CodonCode Aligner (CodonCode Corporation). Heterozygous peaks were identified manually. Sequences were analyzed using eBioX/eBiotools [32]. Coordinates of features including lengths of the indels are based on the PEST reference sequence. APL1 sequences were deposited into Genbank under accession numbers JX292981 to JX292986.

Development of a RFLP Assay to Determine the APL1A Alleles in Ngousso
We developed a high-throughput Restriction Fragment Length Polymorphism test (RFLP) to facilitate the identification of the APL1A alleles present in Ngousso mosquitoes. This assay takes advantage of the allele specific indel patterns in the APL1A genes. These indel patterns are stable in the Ngousso colony, distinguish the APL1A allelic variants and provide information on the haplotype of the APL1 region.
The APL1A-RFLP test is based on the PCR fragment spanning the complete APL1A coding-region. The full-length amplicon is generated with two fluorescent primers labeled with different fluorophores and then cleaved with PstI. Resulting fluorescentlabeled restriction fragments, which span the indel bearing regions at the 59 and 39 end of the APL1A alleles ( Figure 2) are distinguishable by size and color. Fragment separation and sizing were done on an ABI Prism 3730 DNA Analyzer. In combination, the two sized, end-labeled fragments provide information about the presence or absence of the PANGGL repeat region and determined the patterns of the small indels in the 39region of the APL1A gene based on alterations of the restriction fragment length. The APL1A-RFLP distinguishes the three allelic APL1A variants present in the Ngousso laboratory colony and all combinatorial genotypes of the APL1A alleles were found in the colony.

Double-stranded RNA Synthesis and Injection
Double-stranded RNAs were synthesized from PCR amplicons using the T7 Megascript Kit (Ambion) as described previously [11]. 500 ng of dsAPL1A and dsGFP in a maximum volume of 207 nl were injected into the thorax of cold-anesthetized 1 day-old A. gambiae females using a nano-injector (Nanoject II; Drummond). Mosquitoes were challenged with P. falciparum parasites 4 days after dsRNA injection.

Gene Knockdown Verification
The efficiency of transcript knockdown was monitored 4 days after dsRNA injection. cDNA synthesis was performed by using M-MLV reverse transcriptase and random hexamers (Invitrogen). In each case, 500 ng of total RNA was used in triplicate assays. The triplicates were pooled and the mixture was used as template for PCR analysis. Gene knockdown verification was performed as described in Mitri et al. 2009 [11].

Insect Cell Culture and Transfection
A. gambiae derived 4a-3A hemocyte-like cells were cultured in monolayer at 27uC in Insect Xpress medium (Lonza) supplemented with 5% foetal bovine serum (GIBCO BRL) and 50 mg/ml gentamycin (Sigma). The three APL1A alleles were amplified from selected mosquitoes with known APL1 haplotypes by PCR using the following primers flanking the coding regions of each gene: The amplicons were cloned into a dual His and V5-tag insect expression vector (pAc5.1 V5/His, Invitrogen) or a dual Strep and V5-tag insect expression vector (pAc5.1 V5/Strep). The pAc5.1 V5/Strep vector is a variant of the pAc5.1 V5/His insect expression vector that was modified to replace the His-tag by a Strep-tag. 4a-3A cells were co-transfected with the expression vector and the selection vector pCoBlast Puromycin in a ratio 9:1 with Cellfectin II reagent (Invitrogen) according to the manufacturer's protocol. Three days after transfection, antibiotic selection was started with 6 mg/ml puromycin (Invitrogen). After one month the cell population was tested for expression.

Western Blotting
To analyze the secretion pattern for tagged proteins, the culture medium was collected and spun for 10 min at 800 rpm to remove floating cells and large debris. Adherent cells were washed in 16PBS and collected by scraping in 10 mM Tris pH 8. Proteins from culture medium and cells were extracted in XT sample buffer (Bio-Rad), heated at 95uC for 5 min and separated on 4-12% Criterion SDS-PAGE gels (Bio-Rad). Reduced samples were prepared by adding XT reducing agent (Bio-Rad) before heating. After protein transfer to PVDF membrane, immunoblots were blocked for 1 hour in 5% nonfat dry milk, probed with a mouse mAb anti-V5 antibody at 1:5000, followed by 1 hour with horseradish peroxidase-conjugated anti-mouse IgG secondary antibodies (Promega) at 1:10 000. The detection step was performed using the Pierce chemiluminescence system (Pierce) following the manufacturer's instructions.

Immunofluorescence Assays
Transfected cells were allowed to grow in an 8-well Lab-Tek chamber slide system (Thermo Scientific) for one hour. The culture medium was then removed, the cells were washed in 16PBS, fixed in 16PBS with 4% paraformaldehyde for 30 min at room temperature, and permeabilized in 0.2% Triton X-100 and 5% Bovine Serum Albumin in PBS 1X for 45 min at room temperature. Cells were incubated with a mouse mAb anti-V5 antibody at 1:500 in 16PBS containing 10% foetal bovine serum for 1 hour at 4uC, followed by 1 hour with Alexa Fluor 594conjugated anti-mouse IgG secondary antibody at 1:2000 and Alexa Fluor 488-conjugated phalloidin (Molecular Probes) at 1:500 in PBS 1X. Nuclei were stained with Hoechst 33342. After mounting in SlowFade Gold antifade reagent (Molecular Probes), samples were observed at 1006magnification using a Leica DM 5000 B fluorescent microscope.

Supporting Information
Figure S1 Alignment of genomic regions coding for APL1A, APL1B and APL1C alleles in Ngousso mosquitoes. Nucleotide sequences have been aligned using ClustalW. Prettyplot (EMBOSS package) has been used for boxing and colouring with plurality = 4 to calculate the consensus. Red colour indicates nucleotides identical to the consensus. Start codons of the APL1 alleles are highlighted in blue and the stop codons in green. The deletions-insertions described in Supplementary Information S1 (Structure of APL1 alleles in Ngousso) are highlighted in yellow. Stars indicate borders of the nucleotide sequences used in the polydot blots of Figure S5. The 59 end position is common to all alleles and corresponds to the first nucleotide of the start codon. At the 39 end a black star indicates the end of the fragment for APL1A and APL1B alleles and red star the end of the APL1C alleles. Localizations of the oligonucleotides used to generate the dsRNA are indicated by arrows. (PDF) The histogram shows the allele composition (in percentage) of Ngousso females in three independent infection experiments (Inf1, Inf2 and Inf3). Numbers below the x-axis correspond to the alleles APL1A 1 (1), APL1A 2 (2) and APL1A 3 (3), respectively. B) Observed APL1A genotype frequencies. The histogram shows the APL1A genotype composition (in percentage) of all Ngousso females from the three infection experiments (Inf1, Inf2 and Inf3) analyzed in figure S4A. Numbers below the x-axis correspond to following genotypes: APL1A 1 /APL1A 1 (1-1), APL1A 1 /APL1A 2 (1-2), APL1A 1 /APL1A 3 (1-3), APL1A 2 /APL1A 2 (2-2), APL1A 3 /APL1A 2 (3-2) and APL1A 3 /APL1A 3 (3-3). (PDF) Figure S5 Polydot plot of APL1 genes and their allelic variants. Polydot software [2] has been used to perform pair wise comparisons of all Ngousso (Ng) APL1 alleles or their encoded proteins in order to illustrate their high degree of similarity. The word size used for the plots is indicated on top of the graphs. It corresponds to the length of the fragment that should have an exact match in both sequences used in the comparison. A: Polydot plot of APL1 alleles. For the comparison of all APL1 alleles in Ngousso, the corresponding genomic regions have been extracted, beginning at their start codon and up to the position of the stop codon of the longest allele of each gene, i.e.: APL1A (2L:41270938.41270940), APL1B (2L:41266619.41266621) and APL1C (2L:41257877.41257879). Red boxes highlight comparisons between alleles of the same gene (APL1A, APL1B of APL1C) and therefore show the alignments with the highest identity level. For APL1B and APL1C the nucleotide identity is high over the complete gene while for APL1A the differences between the alleles are visible: APL1A 1 and APL1A 3 display a nearly perfect diagonal while the APL1A 2 allele harbors a different 59end due to the repeat region which is similar to the APL1C alleles. It should be noted that the most divergent part on this plots between APL1A genes and APL1B/APL1C genes is the 39 end where the dsRNA has been choosen (see Figure S1C for more details). B: Polydot plot of APL1 protein variants. The protein sequences of all APL1 variants were used in the comparison. To permit the comparison with the very short APL1B 3 allele, its premature stop codon has been ignored, resulting in an artificial in silico product, which was extended to the same length as the other APL1B proteins. C: Polydot plot of APL1 genomic regions against the nucleotide sequence of the APL1A dsRNA regions. Oligonucleotides used for the synthesis of the dsAPL1A are positioned at (2L: 41271198.41271218) and downstream of the APL1A 1 stop codon (2L:41270907.41270930). For this comparison the corresponding sequence of the 39end of each APL1A allele was extracted and compared to the sequence of the PCR fragments of the entire APL1 genes. To illustrate the specificity of the dsRNA used in the knock-down experiments, two different word sizes were used. With a word size of 20, corresponding to the size range of RNAi fragments, exact matches can only be found with the APL1A genes. When applying less stringent conditions by reducing the word size to 8, still no obvious homology with the APL1B and APL1C genes can be detected, demonstrating the specificity of the dsAPL1A used in the knock-down experiments. (PDF) Table S1 Oligonuclotides used to amplify APL1 genes and genomic regions from A. gambiae Ngousso. (PDF)