Relative Contributions of Genetic and Non-Genetic Factors on Mosquito Microbiota

The commensal microbiota of mosquitoes impacts their development, immunity, and competency, and could provide a target for alternative entomological control approaches. However, despite the importance of the mosquito/microbiota interactions, little is known about the relative contribution of genetic and non-genetic factors in shaping the bacterial communities of mosquitoes. We used a high-throughput sequencing based assay to characterize the bacterial composition and diversity of 665 individual eld-caught mosquitoes, as well as their species, genotype at an insecticide resistance locus, blood meal composition, and the eukaryotic parasites and viruses they carry. We then used these data to rigorously estimate the individual effect of each parameter on the bacterial diversity as well as their relative contribution to the microbial composition.


Results
Overall, multivariate analyses did not reveal any signi cant contribution of the mosquito species, insecticide resistance or blood meal to the bacterial composition of the mosquitoes surveyed. Infection with parasites and viruses only contributed very marginally and the main driver of the bacterial diversity was the location where each mosquito was collected which explained roughly 20% of the variance observed.

Conclusions
This analysis shows that, when confounding factors are taken into account, the sites where the mosquitoes are collected is the main driver of the bacterial diversity of wild-caught mosquitoes, although further studies will be needed to determine how the speci c components of the local environment affecting the bacterial composition.

Background
Aedes, Culex and Anopheles mosquitoes can transmit eukaryotic parasites and viruses to humans and are responsible for devastating diseases such as malaria, lymphatic lariasis, dengue, West Nile, and Zika that affect hundreds of millions of people worldwide and cause hundreds of thousands of deaths annually [1]. Successful disease elimination campaigns have focused on interrupting transmission of these diseases by targeting their mosquito vectors. Thus, in the early twentieth century, vector control approaches primarily relied on the use of chemicals and larviciding tools (i.e., petroleum oils and larvivorous sh) to eliminate larval and adult stages of the mosquito [2,3]. Environmental and human health concerns brought by the persistent use of chemicals like Dichlorodiphenyltrichloroethane (DDT) over the years have partly led to the increased use of pyrethroids, which are safer alternatives, in vector control measures [3]. In the last three decades, entomological control strategies based on pyrethroidtreated bed nets and indoor residual spraying have been extensively used in the ght against vector-borne diseases with considerable success in reducing disease burden [4][5][6]. However, wide-spread use of these approaches, combined with exposure to agricultural pest control chemicals, have led to the emergence and rapid spread of insecticide-resistance alleles in many areas [7,8]. In addition, several populations of mosquitoes have modi ed their behaviors (e.g., the host biting time [9,10] and location [11,12] or the host species preference [13,14]) upon exposure to insecticides. This acquisition of chemical and behavioral resistance to insecticides threatens the advances made in control of mosquito-borne diseases and highlight the need for alternative measures.
One alternative to chemicals is to leverage biological agents to control mosquito populations (sometimes referred to as biological controls). For example, Bacillus thuringiensis, a spore-forming bacterium with larvicidal characteristics, is used extensively against disease-transmitting insects as well as agricultural pests [15]. In recent years, the use of Wolbachia as a biological control agent has also gained momentum, with studies revealing its ability to promote pathogen interference and to reduce the life span of mosquitoes [16][17][18]. Furthermore, recent ndings have highlighted how modi cations of the bacterial communities present in the midgut of mosquitoes could decrease or inhibit transmission of pathogens.
For example, studies have demonstrated the importance of gut microbiota in mosquito larval development and shown that bacteria are required for Aedes mosquitoes to survive to the adult stages [19,20]. Similarly, elimination of native microbiota resulted in delayed growth in Anopheles larvae [21]. Functional studies on adult stage mosquitoes showed that the gut microbiota can increase the resistance of mosquitoes to human pathogens by modulating the mosquito innate immune response [22,23] or directly through production of anti-pathogen molecules from speci c microbial species [24,25]. Overall, these studies demonstrate the potential of microbiota manipulations for inhibiting pathogen transmission and/or reducing vector competence.
However, while these laboratory and eld studies highlight the role of the mosquito midgut micro ora in regulating the development and transmission of human pathogens, very little is known about the factors that shape the diversity of the bacterial composition in wild mosquitoes. Some studies have suggested that mosquito collection location is associated with the composition of the microbiota [26,27] while others showed that the microbial composition differed between mosquito species, even when they are closely related and collected at the same location [28], or reared under the same conditions [20]. In addition, there is evidence that resistance to pesticides can be mediated by symbiont bacteria [29,30], and therefore it is possible that genetic resistance to insecticides could also in uence the microbial composition (e.g., resistance would allow exposure of the gut microbiota to insecticides and alter the composition of the microbiota). Lastly, the source of the blood-meal has also been shown to impact gut microbiota composition, with mammalian blood-meal source altering the gut bacterial composition of adult mosquitoes [31]. However, since these studies typically addressed only one of those factors at a time (without correcting for confounding effects), the relative contribution of each of these parameters to shaping the midgut micro ora composition of wild-caught mosquitoes remain unclear.
Here, we analyzed 665 individual Anopheles mosquitoes collected in Guinea and Mali. We characterized the microbial composition of these mosquitoes and screened them for a large variety of eukaryotic parasites and viruses. We also characterized the species of all mosquitoes, genotyped them at a major insecticide resistance locus and examined the source of their last blood meal. We then tested how these genetic and non-genetic factors in uenced the bacterial composition, the species richness and the bacterial diversity, and simultaneously estimated the relative contribution of these factors to the mosquito microbiota.

Sample Collections
Mosquitoes were collected in Guinea by Human Landing Catches (HLC) and Pyrethrum Spray Catches  Table 1 for details of sites/villages mosquitoes were collected and type of collection method used. Mali Mosquito DNA Extraction DNA was extracted from the body section of each mosquito using Chelex-100 (Bio-Rad Laboratories, Hercules, California) protocol. Brie y, dissected individual mosquito sections were placed in separate 1.5 mL tubes containing deionized water. Pipette tips were used to grind each sample in the tube. Each sample was further homogenized in PBS (1X)/1% saponin solution, shaking gently for 20 minutes. The homogenates were then incubated at room temperature (25 °C) overnight. The tubes were then centrifuged at 20,000 x g for 2 minutes and supernatants were discarded. After washing with PBS (1X), each pellet was resuspended in deionized water and 20% Chelex-100 resin solution. This mixture was placed on a heating block for at least 10 minutes and stirred every 5 minutes. DNA was eluted after a nal spin for 1 minute.
PCR ampli cation and sequencing DNA extracted from individual mosquitoes (and 95 water control samples) were ampli ed separately with each primer pair using the Promega GoTaq DNA Polymerase with the following conditions: initial denaturing step at 95 °C for 2 minutes followed by 40 cycles of 95 °C for 30 seconds, 50 °C for 30 seconds and 72 °C for 30 seconds, and a nal extension of 5 minutes. Only 35 cycles were used to amplify bacterial 16S rRNA. All PCR products generated from one DNA sample (i.e., individual mosquito) were pooled together and reampli ed in a second PCR to add a unique barcode and the Illumina sequencing adaptors [35]. Finally, all barcoded products from all mosquitoes were pooled and sequenced simultaneously on Illumina HiSeq 2500 to generate 300 bp paired-end reads [35]. synthesis is carried out at 37 °C for 60 minutes. The cDNA products are then used as templates for PCR ampli cation and sequencing as described above.

Microbiota assessment
After discarding reads that did not carry the 16S rRNA primers or the expected barcode sequences, each read was assigned to a given sample. The resulting demultiplexed FASTQ les were then analyzed in DADA2 [36] (v1.6.0) by rst trimming low-quality bases at the end of each read pair (> 250 bp for forward, > 210 for reverse reads) using the following parameters: maxN = 0, maxEE = 2, truncQ = 2. Dereplication was done by combining identical reads and assigning the number of reads belonging to each unique read (derepFastq). Next, the dereplicated data was analyzed with the dada core sample inference algorithm followed by the merging of read pairs that overlap by at least 12 bases (mergePairs). An amplicon sequence variant table (ASV) table was constructed with makeSequenceTable for all samples and chimeras were removed using removeBimeraDenovo. For taxonomic assignment, Silva (v128) [37] was used as a training set (using assignTaxonomy and addSpecies) to create taxonomy data. Finally, the R package phyloseq (v1.25.2) was used to combine the ASV table, taxonomy data, and sample metadata for downstream microbiome data analyses. Samples with less than 5,000 reads were discarded from further analysis, as they likely represent low-level cross contamination. PCoA with Bray-Curtis and weighted UniFrac distance matrices were calculated in phylosEq. Adonis analyses were performed to investigate the statistical signi cance of observed differences in PCoA plots. Adonis calculates signi cance via P value with R 2 representing the variation explained by sample groupings (i.e., collection site, mosquito species, kdr-w genotype, blood-meal status, and infection status).
Eukaryotic parasites, viruses, and blood meal composition assessment First, reads that did not contain the exact barcode and primer sequences were discarded. The rest of the reads were assigned to a given sample according to their unique barcode sequence. In order to remove low-quality bases and sequences that were likely primer dimers, each sequence was searched for the forward and reverse primers and trimmed after the reverse primer (if both primers are found). Sequences where the forward primer was found but the reverse primer was missing were left untrimmed. Untrimmed sequences that belong to a primer with an expected amplicon length of < 300 bp were trimmed from 50 bp from the 5' end for further quality ltering. Afterwards, ltered paired-end sequences from each read pair were merged using FLASH to generate a consensus sequence of the overlapping region. Each correctly merged sequence was trimmed of both ampli cation primers (forward and reverse) and kept only if it was longer than 90 bp. Then all concatenated sequences ampli ed with the same primer pair (from all samples) were compared to each other and a single copy of each unique sequence was kept (while the number of times it is observed in each sample was recorded). Unique sequences that were observed less than 10 times across all samples were removed as they likely represent instances of sequences carrying errors [35]. The remaining unique sequences were compared against all DNA sequences annotated on the NCBI nr database using BLAST [35]. We then retrieved the taxonomic information of the most similar sequence(s) if it had at least 70% identity over the entire sequence length. Finally, we summarized the parasite and virus species identi ed, the percent identity (i.e., similarity to the most similar sequence(s) on NCBI), and the number of reads observed in each sample.
When evaluating the blood meal composition of each mosquito, the same procedure was applied but only samples with at least 1,000 reads were considered (to avoid including possible cross-contamination or sequences mis-assigned to one sample due sequencing errors in the barcode) [38].
kdr genotype (L1014F) and species determination (S200 × 6.1) Reads ampli ed from the kdr and S200 × 6.1 primer pairs were processed as described above, with sequences assigned to their speci c sample, ltered for quality, and merged with FLASH.
For the L1014F locus, the top two most abundant unique sequences in each sample (Seq1 and Seq2) were considered for further analysis. If the abundance of Seq2 was greater than 35% of the sum of Seq1 and Seq2 abundance, the sample was considered heterozygous Seq1/Seq2 while if this ratio was smaller than 35%, the sample was deemed homozygous for Seq1. Only samples with at least 1,000 reads were analyzed.
For the S200 × 6.1 locus, the most abundant sequence for each sample was compared against DNA sequences on NCBI as described above. For samples with greater than or equal to 1,000 reads, the species level taxonomy was retrieved and the Anopheles species as well as the total read count were summarized. A primer targeting the COX1 gene (CulicCox1) was used to identify An. nili species in samples as it was not able to be determined by the S200 × 6.1 locus [34,39] Results Bacterial composition of Anopheles mosquitoes from West Africa We extracted DNA from 665 individual Anopheles mosquitoes collected in Guinea (N = 584) and Mali (N = 81). To characterize the micro ora of each mosquito, we ampli ed and sequenced the V2 variable region of the bacterial 16S ribosomal RNA genes (see Methods for details). We obtained a total of 8,467,703 sequences derived from 760 samples (665 mosquitoes and 95 extraction controls). On average, each mosquito samples yielded 11,730 sequences compared to 5,984 sequences on average per extraction control. We assigned these sequences to 21,527 amplicon sequence variants (ASVs, analogue of operational taxonomic units [36]), representing 37 phyla including Proteobacteria (6,692 ASVs accounting for 64% of all reads), Firmicutes (26%), Actinobacteria (6%) and Bacteroidetes (2%) (Fig. 1).
To investigate differences in bacterial composition among mosquitoes, we calculated β-diversity estimates using weighted UniFrac and Bray-Curtis dissimilarity matrices. Principal coordinate analyses (PCoA) conducted using Bray-Curtis dissimilarity or weighted UniFrac distances showed that the microbial composition separates mosquitoes into distinct clusters ( Fig. 2A and Additional le 2: Figure  S1A, respectively). These clusters appeared to group mosquitoes collected in the same sites (Fig. 2), and this observation held true when we restricted our analyses to mosquitoes only collected from sites in Guinea ( Fig. 2B and Additional le 2: Figure S1B). The details of all the ASVs identi ed and their taxonomy are provided in Additional le 3: Table S2.

Assessment of mosquito species and kdr mutation
We simultaneously genotyped the same mosquitoes at loci informative of their species and insecticideresistance status by high-throughput sequencing (see Methods).
Out of 665 mosquitoes, 551 (82.9%) were successfully genotyped for the S200 × 6.1 [34] and cox1 [39] loci. We primarily used the S200 × 6.1 locus to identify the species of each mosquito as i) this locus was more robustly ampli ed and sequenced than the cox1 locus (with an average read count of 2,917 and 1,181 per mosquito, respectively) and ii) provided clearer taxonomic resolution (with, for example, 233-234 (mean of 233.67) nucleotides differentiating the sequences from Anopheles gambiae s.s. from those of Anopheles coluzzii, compared to 0-4 (mean of 1.71) nucleotide differences using the cox1 locus) (Additional le 4: Table S3). However, the S200 × 6.1. locus systematically failed to yield sequences for some mosquitoes that were identi ed at Anopheles nili using the cox1 sequences. Overall, we identi ed that the mosquitoes analyzed belonged to ve Anopheles species. 404 mosquitoes (74.5%) were identi ed as Anopheles gambiae s.s., while the remaining mosquitoes consisted of Anopheles coluzzii (61 mosquitoes or 11.3%), Anopheles melas (with 57 mosquitoes or 10.5%), Anopheles arabiensis (8 mosquitoes or 1.5%), and Anopheles nili (7 mosquitoes or 1.3%) (Fig. 3). We also identi ed 5 mosquitoes that were heterozygous for the S200 × 6.1 locus and likely represented F1 hybrids of An. gambiae s.s. and An. coluzzii species. The species distribution varied extensively between locations, with An. gambiae s.s. accounting for more than 90.00% of all mosquitoes collected in ve out of six locations in Guinea, while An. melas was the most abundant species (79.2%, 57/72) in Boffa, a coastal region in western Guinea, and An. coluzzii predominated in Bandiagara, Mali (86.3%, 44/51) (Fig. 3).
Pyrethroid resistance is often due to a point mutation in the voltage gated sodium channel gene, described as knockdown resistance (kdr) [33]. 550 (82.7%) of the mosquitoes were successfully genotyped at this locus (with an average coverage of 2,436 reads per mosquito). In Guinea, with the exception of mosquitoes collected in Boffa, most mosquitoes (> 92.6%) were homozygous for the kdr-w (L1014F) alleles that is associated with resistance to pyrethroids [33] (Fig. 4). In Boffa, where An. melas is the predominant species, most mosquitoes were homozygous for the wild-type allele (L1014L). In Mali, the distribution was more heterogeneous, with roughly equal proportions of mosquitoes homozygous for the wild-type, resistant allele or heterozygous. Across mosquitoes, the genotype at the kdr-w locus correlated almost perfectly with the mosquito species, with An. gambiae carrying primarily L1014F alleles while An. arabiensis and An. melas were essentially wild-type. Only An. coluzzii showed high proportion of both alleles (Additional le 5: Figure S2). The details of all the genotypes and sequences ampli ed from each mosquito are provided in Additional le 6: Table S4.

Determination of the blood meal composition
To characterize the composition of the last blood meal of each these mosquitoes, we used the same DNA extract to amplify and sequence a short fragment of the mammalian mitochondrial 16S rRNA gene. 133 mosquitoes yielded > 1,000 reads and were considered blood fed in later analyses. 126 mosquitoes carried human DNA, 14 mosquitoes cow DNA, and 2 sheep DNA ( Fig. 5 and Additional le 7: Table S5).
Nine mosquitoes fed on more than one mammalian host species (Fig. 5). The blood meal composition differed between sites with, for example, 12 mosquitoes (20.1%) from Kankan that fed, at least partially, on cow while mosquitoes from all other sites, in Mali and Guinea, fed almost exclusively on human.

Identi cation of eukaryotic parasites and viruses from individual mosquitoes
Finally, we determined whether each mosquito carried a eukaryotic parasite and/or arbovirus using a sequencing-based assay recently developed in our laboratory [35]. After PCR ampli cation, and sequencing of DNA extracted from individual mosquitoes, we identi ed DNA sequences from eukaryotes and viruses from 127 mosquitoes, with on average, 1,876 reads supporting each identi cation in each mosquito. Nine out of 95 extraction controls (water samples that have been processed simultaneously) were also positive for one or more parasites, but with an average of 54 reads per parasite (Additional le 8: Table S6). The low read counts in those water controls could be explained by low-level crosscontamination during extraction, or sequence mis-assignment due to sequencing errors in the Illumina index sequences [35].  [40]. Interestingly, from the fourteen samples for which Bos indicus (cow) was identi ed with greater than 500 reads, we detect Theileria species in twelve (86%), suggesting the tick-transmitted parasite may have been ingested by these mosquitoes during their last blood-meal. Seven mosquitoes yielded a DNA sequence identical to Loa loa and to several other larial worms (and were further characterized as deriving from Mansonella perstans after sequencing of longer ampli cation products, M. Cannon. personal communication). Several DNA sequences were closely related to known parasites of mosquitoes themselves, belonging to microsporidia [41] (e.g., Parathelohania sp.), mosquitotransmitted nematodes (e.g., Setaria sp. [42,43]), as well as two recently discovered Anopheles aviviruses (e.g., Anopheles avivirus variant 1 and variant 2) [44] ( Table 2). The details of all parasite sequences ampli ed and their taxonomic information for each mosquito are provided in Additional le 8: Table S6.

Evaluation of the factors in uencing microbial composition of wild mosquitoes
The characterization of the mosquito species, insecticide-resistance genotype, blood meal status and infection from the same mosquitoes that have been examined for their microbial diversity enables a rigorous assessment of the relative contribution of these factors to the microbial composition. Note that to avoid possible biases introduced by sample storage or DNA extraction, we restricted this analysis to mosquitoes collected in Guinea that were all processed identically and simultaneously. We implemented an analysis of variance [45] to simultaneously test the contribution of each factor, while accounting for the others (multivariate analysis). The geographical location of the samples explained most of the variation in microbial composition (Adonis test, R 2 = 0.200, ρ = 0.001), whereas the mosquito species (Adonis test, R 2 = 0.004, ρ = 0.208), kdr-w genotype (Adonis test, R 2 = 0.006 ρ = 0.438), and feeding status (Adonis test, R 2 = 0.003, ρ = 0.173) were not statistically associated with the bacterial composition. The mosquito infection status was signi cantly associated with the microbial composition but had a very marginal effect (Adonis test, R 2 = 0.015, ρ = 0.001) ( Table 3). To further investigate the relative roles of geography and species that are confounded in this dataset, we examined PCoAs of the bacteria composition, restricting the analysis to i) all An. gambiae collected in seven sites (Additional le 9: Figure  S3A) and ii) mosquitoes from all Anopheles species collected in Boffa (Additional le 9: Figure S3B). Together, these analyses validated the results of the statistical testing and con rmed that geographical location of the mosquitoes had a much greater in uence on the bacterial composition than the species of the mosquitoes.

Discussion
The importance of the mosquito microbiota on vector biology and pathogen transmission has been well recognized, with several studies demonstrating the role of endogenous bacteria on the vector's development [19,20], immunity [22,23], and competency [46]. However, few studies have examined the factors that shape the bacterial composition of mosquitoes and most of those used laboratory-reared mosquitoes to assess in uences on bacterial communities [20,31,47]. This latter limitation could be especially problematic since bacterial diversity of wild-caught Anopheles mosquitoes has been shown to be greater than that of mosquitoes reared in the laboratory [48]. In addition, these studies typically focus on testing the in uence of a single factor on the bacterial composition without accounting for confounding factors. In this study, we examined the microbial diversity of 665 individual wild-caught Anopheles mosquitoes collected in six sites in Guinea and one site in Mali. For each mosquito, we characterized their species, kdr-w genotype, blood-meal status, and infection with various eukaryotic parasites and viruses. We then simultaneously estimated the relative contribution of each of those genetic and non-genetic factors on the microbial composition of the mosquitoes. In this analysis, the mosquito collection site accounted for ~ 20% of the variation in bacterial composition, whereas the other factors showed marginal or non-signi cant contribution (Table 3).
Our ndings are consistent with previous studies that showed that collection site was a major contributor to the microbial diversity of eld-caught Anopheles mosquitoes [26,[49][50][51]. For example, Muturi et al. found that sampling site has a strong effect on microbial composition and diversity, even when examining nine different mosquito species [27]. Note however that "collection site" in our study summarizes many parameters. Thus, the in uence of the "collection site" on the microbial composition could re ect the effect of differences in larval habitats, ora the mosquitoes rely on for nectar feeding, and/or local population differences. In this regard, it is worth noting that the adult mosquito bacterial composition has been shown to vary depending on the larval breeding sites and the bacterial composition of these aquatic habitats [52]. In addition, sugar source appears to have a pronounced in uence on the vectoral capacity of An. sergentii mosquitoes [53] and impacts the microbial composition of laboratory-reared adult mosquitoes [54]. Future studies using a denser, more local sampling of wild-caught mosquitoes, will be required to better understand the individual contribution of these local parameters.
On the other hand, our analyses provided new insights on the role of other factors on the microbial composition of mosquitoes. We did not observe any signi cant contribution of feeding status on microbial variation of the wild mosquitoes. This nding contrasts with a previous study that described that the bacterial diversity of Aedes aegypti mosquitoes fed on human, chicken, or rabbit blood was signi cantly lower than that of newly emerged unfed mosquitoes [31]. This discrepancy could re ect differences between mosquito species/genera, or more likely, differences between wild-caught mosquitoes (that might have had prior blood meals) and laboratory-reared mosquitoes with less variable microbial composition. Similarly, our study did not reveal any signi cant contribution of genetic factors (i.e. mosquito species, kdr-w genotype) on mosquito microbial variation. These observations contrast with a previous study that described distinct bacterial compositions in two species of Culex mosquitoes collected from the same site and with identical larval aquatic environment [28], possibly due to differences among Culex species in their larval feeding habits [55]. The L1014F kdr allele frequency was reported high or near xation in Kankan and Kissidougou sites of Guinea and low in Boffa in a previous study [56], consistent with our ndings here. In theory, selection of insecticide-resistant mosquitoes could also select for bacterial species with adverse effects (i.e. in uence on disease transmission). Here, we did not see evidence of L1014F resistant allele in uence on adult mosquito microbial composition. Note however that resistance allele genotypes and mosquito species only represent a small fraction of the genetic factors that could impact the mosquito microbiota and that, given our observation that the collection site is strongly associated with the bacterial composition, it would be interesting to further investigate whether genetic diversity is associated with the microbiota of mosquitoes [57].
Interestingly, we observed a marginal but statistically signi cant association between infection status (infected n = 127 vs. non-infected n = 513) and the mosquito microbial composition. Modi cation of insect gut microbiota by parasitic [58] or viral [59] infections has been demonstrated in few studies. Pathogenic or non-pathogenic (e.g., insect-speci c viruses) species could be involved in crosstalk with insect metabolism pathways or immune system to in uence the microbiota [60]. It is worth noting that, due to the low infection rate with parasitic and viral species we found in the mosquitoes (< 5.0%), we assessed the in uence infection has on the microbiota using an aggregate of all the parasite and viruses we detected (as opposed to individual parasite and virus species) and it is possible that the effect of one organism on the micro ora might be diluted down and undetected once analyzed together with other parasites and viruses that have no in uence on the bacterial communities. For example, Theileria sp. are transmitted by ticks and unlikely to be viable in mosquitoes and therefore, have probably little to no contribution on the in uence of infection on the mosquito microbiota. Future studies assessing the direct in uence of some of the parasites found in abundance in this study (e.g., Parathelohania sp., Microsporidium sp.,) and the recently discovered virus (Anopheles avivirus) could further elucidate the tripartite relationship between the mosquito, microbiota, and mosquito infecting agents.

Conclusions
In summary, this study provides a comprehensive assessment of the microbial composition and diversity of 665 wild mosquitoes and a simultaneous examination of the relative contribution of ve different mosquito factors on microbial variation. This approach enables rigorous estimation of the importance of these factors to shaping the bacterial composition, while correcting for their often confounding effect.
Our results highlight the prominent role of the mosquito collection site and, to a lesser extent, the parasitic and viral infection, on shaping the bacterial composition of wild-caught mosquitoes. These ndings provide a solid foundation to implement further investigations and examine the speci c components of the environment (e.g., bacterial communities of the larval habitats, source of nectar, genetic diversity) shaping the microbial composition of wild mosquitoes and the mechanisms mediating these effects.

Consent for publication
Not applicable.

Availability of data and materials
All data generated or analyzed during this study are included in this published article and its additional les.

Competing interests
The authors declare that they have no competing interests.  Average relative abundance of bacterial phyla from each mosquito collection site in Guinea and Mali.
Bacterial species from the Proteobacteria phylum are the most abundant followed by Firmicutes and Actinobacteria. Less than 1% abund. represents phyla that make up <1% of all bacteria.  Mosquito species diversity across collection sites in Guinea and Mali. Hybrid represents samples identi ed as heterozygous for An. gambiae and An. coluzzii at the S200X6.1 locus. Numbers above each bar represent the total number of mosquitoes that species is determined, per site.

Figure 4
Distribution of L1014F mutation (kdr-w) in mosquitoes collected across Guinea and Mali. Numbers above each bar represent the total number of mosquitoes that kdr-w genotype is identi ed, per site.