After 16 successive generations of continuously feeding on P. berghei-infected mouse blood (exposed A. gambiae lines A and B) or normal non-infected mouse blood (non-exposed control line C), we found no significant differences in mortality, size or other visible characteristics between the exposed and control lines. The mortality ranged from 15%–50% (Additional File 1 Table), and death mostly occurred between 24 and 48 h after feeding. However, the permissiveness to Plasmodium infection at the pre-mature oocyst stage differentiated substantially between the exposed lines and the control line.

After 13 generations, differences at the infection phenotype could already be detected between the continuously exposed and the control lines (Fig. 1a). The exposed lines showed a general tendency toward a lower permissiveness to Plasmodium, with a few deviations for some generations and lines (Fig. 1). The levels of infection of generations 13, 14, and 15 of the exposed line B were significantly lower than those of the control line (43%, 74%, and 64%, respectively). Generation 16 displayed only a 27% decrease in infection level (Fig. 1b and Additional File 3 Table). For the same generations (13–16), the exposed line A showed a reduction in the number of oocysts of 10%, 44%, 14%, and 57%, respectively (Fig. 1b and Additional File 3 Table).

The statistically insignificant (27%, 10% and 14%) reduction in Plasmodium permissiveness observed for some of the generations of the exposed lines could be due to a number of factors, such as small differences in feeding behavior and the general nutritional and health status of the mosquitoes that can be greatly influenced by rearing conditions such as larval densities. Due to the labor intense nature of rearing several independent mosquito colonies only a limited number (15–40) of mosquitoes were assayed for oocysts number. This could partly also have contributed to this trend of insignificance. However, the exposed lines never displayed a higher permissiveness to Plasmodium than that of the control line in any of the assayed generations.

The differences observed in infection phenotype between control and exposed mosquitoes suggest that continuous exposure to Plasmodium-infected blood and infection reduces the permissiveness of mosquitoes to Plasmodium at the pre-mature oocyst stage. This reduction could be attributed to differences in the transcription levels of genes that can control Plasmodium infection. Therefore, we sought to determine the effect of continuous malaria exposure on the global transcriptome of naïve non-blood fed mosquitoes which had been raised on P. berghei infected blood (exposed) for thirteen and more consecutive generations. For this purpose, we used microarray analysis to compare mRNA abundance between RNAs from naïve 4-days-old female mosquitoes of the exposed lines (generations 13, 15, and 16 of line A and generations 14 and 15 of line B) and the non-exposed control line (generations 15 and 16) that had been raised on non-infected blood.

Fig. 2 shows the functional class distribution of those genes that showed significant differential expression between naïve non-blood-fed mosquitoes of the exposed lines (generations 14 and 15 of line B and 13, 15 and 16 of line A) and the control line (pooled generations 15 and 16). In all, continuous exposure to malaria-infected blood and infection caused a significant change in the constitutive transcription level of 2.39% of the total predicted transcriptome (16,148 genes, according to the version 35.2 annotation); 0.83% (124) of these transcripts showed a higher level of expression in the exposed lines, and 1.56% (230) had a higher expression level in the non-exposed control line (Fig. 2, Additional Files 4 and 5 Tables). Among the putative immune genes, only 7 were expressed at higher levels in the exposed lines, while 26 were expressed at higher levels in the non-exposed control line (Fig. 2, Additional Files 4 and 5 Tables). A similar pattern was observed for genes encoding cytoskeletal and structural components, with 7 showing a higher expression in the exposed lines and 18 a higher expression in the control non-exposed line. For the other functional gene groups, no such differences in expression levels were observed; for instance the digestion-related genes group (33 vs 49) and the genes with putative roles in metabolism (5 vs 6). To validate the robustness of the microarray results, we analyzed eight genes from the immunity-, redox/stress-, and transport-related functional groups by qRT-PCR (from Fig. 2, Additional Files 4 and 5 Tables), comparing the expression of these genes between naïve adults of the two exposed lines to the control line. The cDNA templates were normalized to the A. gambiae S7 gene, and the -fold differences were calculated as described in Experimental Procedures. These analyses showed a highly significant correlation (Pearson coefficient = 0.91; best-fit linear regression, R² = 0.84; slope of the regression line, m = 0.85) between the qRT-PCR and the microarray log₂-transformed values (Fig. 3 and Additional File 6 Table).

The majority of the differentially expressed genes have no known function, and 45 of these showed a higher expression in the exposed lines, while 115 had a higher level of expression in the control line (Additional Files 4 and 5 Tables). This expression pattern is similar to that of putative immune gene group (7 vs 26) (Pearson's Chi-square p = 0.415) and may therefore indicate an infection-related functional bias for many of these unknown genes. In the assays reported here, gene expression was monitored in the entire mosquito, as opposed to specific tissues and cell types that may be more or less affected by and involved in limiting infection. It is likely that some gene expression signatures related to decreased permissiveness to Plasmodium infection in the exposed lines are specific for particular tissues and were not identified in our assays because of a dilution effect related to our use of total mosquito RNA. Nevertheless, the assays showed consistent differences in expression of specific functional gene groups that are likely to relate to the observed differences in infection phenotype between the exposed and control lines. Gene expression was also compared between generation 10 and 16 of the control line with the regular insectary colony of the same mosquito strain to test whether the differential gene expression patterns could have resulted from a potential genetic drift of the control line, rather than the exposure to Plasmodium. These assays did not identify any significant overlap with the genes that were differentially expressed between the control and the exposed lines (data not shown), and thereby confirm the impact of exposure to Plasmodium as the selective pressure responsible for the observed differences.

A notable number of immune genes showed elevated expression in the non-exposed lines (Fig. 2 and Additional File 4 Table). These genes belonged to a variety of functional classes, such as pattern recognition receptors, signal amplification cascade components, and immune signaling pathway components. The pattern recognition receptors group is dominated by thioester-containing protein (TEP) and fibrinogen domain immunolectins (FBN). Interestingly, five of these putative immune proteins, TEP1, FBN9, infection responsive putative short peptide 1 (IRSP1), IRSP2, and Gram-negative bacteria binding protein (GNBPB1) have been shown in RNAi gene-silencing assays to possess anti-Plasmodium activity [4,22-24]. TEP1 is involved in phagocytosis of bacteria and anti-Plasmodium defense at the stage of ookinete invasion of the mosquito midgut [22,23]. FBN9 belongs to the fibrinogen domain immunolectin gene family, which is represented by at least 61 FBNs in the A. gambiae genome [4,25,26]. IRSP1 and IRSP2 are putative short secreted peptides of unknown function [4]. IRSP1 has homology to the salivary gland secreted peptide gVAG [27]. GNBPs are thought to participate in the immune response of Anopheles against Plasmodium infection [2,28,29] and were first isolated from Bombyx mori through their capacity to bind to the surface of Gram-negative bacteria [30]. In D. melanogaster, a GNBP has been linked to the activation of the TOLL pathway, together with a peptidoglycan recognition protein (PGRP) [31,32]. The silencing of GNBPB3 and GNBPB4, as well as GNBPB1 in the mosquito, significantly increases the number of oocysts in the midgut after P. berghei infection (Dimopoulos lab, unpublished data). The majorities of the other putative immune genes have never been tested for anti-Plasmodium activity but belong to functional classes with members that are known to be involved in the defense against Plasmodium. For instance, leucine rich repeats (LRRs) are found in a variety of mosquito immune genes, such as the TOLL-like receptors (TLRs) and the anti-Plasmodium factors LRIM1, LRRD7, and APL1 [4,24,33,34]. Several A. gambiae CLIP domain serine proteases have been shown to be responsive to bacterial and Plasmodium infections [4,25,33]. Finally, transferrin also showed a lower level of expression in the exposed lines. Several studies have previously shown that transferrin is induced upon challenge [25,35]; this protein seems to be implicated in limiting bacterial growth by sequestering iron [35].

Putative immune genes that showed higher constitutive expression in the exposed lines than in the control line (Fig. 2, Additional File 5 Table) included TEP15, FBN30, CED6, cecropin 3, PGRPLC1, C type lectin MA1 (CTLMA1), and a TOLL-like precursor. Of these genes only cecropin has been experimentally linked with anti-Plasmodium activity [36]. CED-6 is an adaptor molecule that acts in a signal transduction pathway that specifically mediates the recognition and engulfment of apoptotic cells in C. elegans [37]. The four-fold higher expression in the naïve adults of the exposed lines suggests that its regulation was strongly affected by the adaptation to Plasmodium exposure. The A. gambiae antimicrobial peptide cecropin is active against numerous Gram-negative and Gram-positive bacteria, as well as several species of filamentous fungi and yeasts, and when is injected into Anopheline mosquitoes previously infected with a variety of Plasmodium species, sporogonic development is disrupted by the abortion of the normal development of oocysts [38-40]. The peptidoglycan recognition protein PGRP has been shown to be involved in TOLL and IMD (immunodeficiency) pathway activation in D. melanogaster and to be induced after Plasmodium infection [33,41]. Two C-type lectins (CTL4 and CTLMA2) have been shown to protect the ookinetes from destruction in the midgut epithelium [24].

Eight redox/stress-related genes showed differential expression in the exposed and non-exposed colonies. Two cytochrome P450 and two gluthatione-S-transferases (GST) were expressed at higher levels in the non-exposed line, and two other cytochrome P450 and two DNAJ protein genes were expressed at higher levels in the exposed lines (Fig. 2, Additional Files 4 and 5 Tables). Cytochrome P450, one of the major enzymes involved in the detoxification of harmful compounds, plays a major role in the mosquito defense system; it has been linked to DDT and pyrethroid resistance in mosquitoes and other insects [42,43]. GSTs are involved in the detoxification of endogenous compounds, such as peroxidates lipids, and in the metabolism of xenobiotics. GSTs have been reported to be differentially expressed in resistant and susceptible strains of malaria [44,45]. Both cytochrome P450 and GSTs have been shown to be regulated after microbial challenge [25].

The cytoskeletal and structural component class of genes displayed a general trend toward lower expression in the exposed lines; the major gene families involved included actins and cuticle proteins (Fig. 2, Additional Files 4 and 5 Tables). In addition to serving as a structural component of the exoskeleton, cuticle proteins play a role in wound healing and are also expressed in hemocytes, one of the major immuno-competent cell types in invertebrates [46,47]. Cuticular proteins have also been shown to participate in the non-self recognition of E. coli by interacting with its surface [48]. Previous studies have shown infection-responsive regulation of several different cuticle proteins after challenge with either bacteria or Plasmodium [4,25]. Actin, microtubules, and other cytoskeletal proteins have been shown to play a major role in the ookinete's traversal of the midgut epithelial cells and the repair of the midgut epithelium after ookinete penetration [49,50].

Two trypsin genes were found to be differentially expressed: Trypsin 1 displayed elevated expression and Trypsin 7 displayed lower expression in the exposed lines (Fig. 2, Additional Files 4 and 5 Tables). Trypsins are known to influence the infectivity of malaria parasites in the mosquito midgut, and for some Anopheles-Plasmodium species combinations, trypsins can stimulate or inhibit midgut invasion by the ookinetes [51-53]. This difference in gene expression may also be related to the influence of Plasmodium-infected blood on mosquito feeding behavior [54]. Among the putative transport-related genes that showed lower expression in the exposed lines was an apolipoprotein D precursor (APOD), the major component of insect lipid transport and a component of the mosquito's immune system and anti-Plasmodium defense [55]. Silencing of an APOD has been shown lower the resistance of Anopheles to Plasmodium infection [4].

Genes in the diverse functional group are associated with a variety of functions, and those belonging to the unknown group did not display significant homology to genes of known function (Additional Files 4 and 5 Tables).

The fact that significantly larger number of immune genes, including anti-Plasmodium factors, was expressed at elevated levels in the non-exposed line suggests that continuous Plasmodium exposure may also lead to a divergence in the infection-responsive regulation of these genes. Therefore, we compared the expression patterns of four anti-Plasmodium berghei factors, TEP1, LRIM1, SPCLIP1 and IRSP1, between the exposed and non-exposed lines in response to P. berghei infection at 24 h after feeding (infected), the time at which the ookinetes invade the midgut epithelium [4,23,24]. A comparison was also done for naïve non-blood fed mosquitoes for the same genes. Due to the limited number of available mosquitoes in each generation, these assays used infected mosquitoes from different but consecutive generations to those used for the microarray expression analyses of naïve mosquitoes. Mosquitoes were collected from generations 11 and 13 for assays on naïve expression levels and generations 10 and 12 for evaluation of infection-responsive expression (Fig. 4 and Additional File 7 Table). The constitutive expression levels of the immune genes TEP1, LRIM1, and SPCLIP1 were in most cases higher in the naïve mosquitoes of the non-exposed line than those of the exposed lines (Fig. 4). Infection with Plasmodium caused a stronger induction of TEP1, LRIM1, and SPCLIP1 in the exposed lines than in the non-exposed control line: an increase of more than two-fold for TEP1, four-fold for LRIM1, and close to two-fold for SPCLIP1. IRSP1 did not follow the same pattern and displayed lower expression levels in both non-exposed and exposed mosquito lines.

Three independent lines (two exposed or experimental, A and B, and one non-exposed or control, C) of the A. gambiae Keele strain (~400 adults per line) were maintained at identical conditions on sugar solution at 27°C and 70% humidity with a 12-h light/dark cycle, according to standard rearing procedures [63]. The exposed lines A and B were reared on GFP P. berghei-infected Swiss Webster mice [64] for 16 consecutive generations; the control line was reared on uninfected Swiss Webster mice for the same number of generations. Plasmodium infections were performed according to [29]. Briefly, GFP P. berghei parasites were passaged in Swiss Webster mice, and parasitemia was determined by Giemsa-stained blood films. The exflagellation events were controlled at averagely 2 to 3 per 20× field. For each feeding experiment, the two exposed lines were fed on the same infected mouse/mice. When two mice were required for one experiment, mice with similar infection levels were used and feedings were done by placing the mice with head part on one cage and tail part on another cage. Every 5 minutes, the positions were flipped to ensure ingestion of equal parasite numbers. The fed mosquitoes from either exposed lines or the control line were firstly kept at 19°C for about 2 days before transferred back to normal conditioned insectary. For colony maintenance, the unfed female mosquitoes from either exposed lines or the control line were removed from the cage after blood feeding, and only fully engorged females were used for eggs laying to produce the next generation of progeny. The larvae hatched ~2–3 days after eggs laying, and the pupae were collected ~10–12 days later; they matured to adult mosquitoes after an additional ~1–2 days. The mosquitoes were maintained on sugar solution, and 5- to 6-day-old mosquitoes were blood-fed (with infected or uninfected blood) and reared in this manner for 16 generations.

For analyses related to mortality, infection phenotype (oocysts number in the midgut), and the expression of immune genes after infection, a pool of mosquitoes from the control line (~80 adults) from each generation was also fed on the same infected mice as were the two exposed lines. Two mice with similar infection levels were laid on the cages with head/tail on two different cages, and the positions of the mice were rotated every 2 to 5 minutes between the cages. The mosquitoes that had fed on infected and naïve blood were firstly kept at 19°C [65] for about 2 days before transferred back to normal conditioned insectary, and their mortality was determined 3–4 days after feeding (Additional File 1 Table). The transfer of mosquitoes to 27°C at two days after feeding did not affect P. berghei development as indicated by the formation of mature oocysts at 9 days after feeding. The majority of mosquito mortality occurred between 24 to 48 hours after feeding on infected blood as previously described [1]. Control line fed on the non-infected blood showed no significant mortality. After 7–9 days, midguts were dissected (~15–20) to determine the level of infection, as measured by oocysts numbers (Fig. 1 and Additional File 3 Table). For RNA isolation, adult females (~20 per line) from all three lines were also collected in each generation before feeding on either infected or non-infected blood (naïve) and after feeding on infected blood (infected). All the assays were done in triplicate.

RNA was extracted from whole mosquitoes using the RNeasy kit (QIAGEN). Quantification of RNA was performed using a Biophotometer spectrophotometer. RNA samples were treated with Turbo DNase and reverse-transcribed using Superscript II (Invitrogen) with oligo dT₂₀. Real-time quantification was performed using the QuantiTect SYBR Green PCR Kit (Qiagen) and ABI Detection System ABI Prism 7300 (Applied Biosystems, California, USA). All PCR reactions were performed in triplicate. The specificity of the PCR reactions was assessed by analysis of the melting curves for each data point. The ribosomal protein S7 gene was used for normalization of the cDNA templates and to calculate the threshold values (Ct). The -fold differences in expression level were calculated by the standard E^ΔΔCt [66] method. Primers used for all qRT-PCR (Fig. 3 and Fig. 4) are listed in Additional File 2 Table.

The Low RNA Input Fluorescent Linear Amplification Kit from Agilent Technologies was used to synthesize Cy-3 and Cy-5 labeled samples from 2 microgram RNA according to the manufacturer's instructions. RNA from the control line generations 15 and 16 were pooled and used as one control sample (labeled with Cy-3-dUTP fluorescent nucleotides) to hybridize to the microarray against the five exposed lines samples (exposed line A from generations 13, 15, and 16, and exposed line B from generations 14 and 15) and labeled with Cy-5-dUTP fluorescent nucleotides. Microarray hybridizations were performed as previously described [4], and all assays were done in triplicate. Spot intensities were measured with a GenePix 7000 autoloader scanner (Axon Instruments) using a 10 μm pixel size. The laser power was set to 60%, and the PMT was adjusted to maximize the effective dynamic range and minimize pixel saturation. Images were inspected manually using GenePix Pro 6.0 software (Axon Instruments, Union City, CA), and any spots that were covered with hybridization artifacts were removed and not included in further analysis. Express Converter software was used to convert the Gene Pix results files to MEV files and be analyzed by using TIGR-MIDAS software, which is available online from TIGR [67]. The data sets were filtered using a signal cut-off intensity of 100 to remove low-intensity/poorly hybridized spots from the analysis. Loc-Fit normalization (LOWESS) was performed independently for each data set. The TMEV software, available online [67], was finally used for SAM analysis with a 5% false discovery rate (FDR). Only transcripts that had signal values above the log₂ cutoff values of ± 0.807 were used for further analysis as previously described [4,68] (Fig. 2, Additional Files 4, and 5 Tables).