Introduction

Blood-feeding insects impose a substantial burden on human and animal health through their role as disease vectors. In particular, mosquito species that feed on human blood pose an enormous public health threat by transmitting numerous pathogens such as dengue virus, Zika virus and malaria, which together kill more than one million people per year [1,2]. Human exposure to pathogens transmitted by mosquito vectors is determined by vector behaviors such as: (1) propensity to feed on humans relative to other animals (anthropophily) and (2) preference for living in close proximity to humans, as reflected by biting and resting inside houses (endophily) [3]. These traits are known to vary within and between the Anopheles mosquito species that transmit malaria [3]. It has been demonstrated since the earliest days of malaria transmission modeling that the degree of anthropophily in vector populations is strongly associated with transmission intensity [4]. Furthermore, the extent to which vectors feed and rest inside houses is a critical determinant of the effectiveness of current frontline control strategies including Long-Lasting Insecticide Treated Nets (LLINs) and Indoor Residual Spraying (IRS), which selectively kill mosquitoes that bite and rest indoors [1].

Vector species with a relatively broad host range, like Anopheles arabiensis, are thought to be better able to persist in areas of high indoor insecticide use. This is because they are more likely to avoid feeding and resting in areas protected by insecticides. For example, several studies in East Africa have shown dramatic declines in the abundance of the highly anthropophilic species An. gambiae s.s. relative to An. arabiensis as LLIN usage has increased [5–13]. Similar changes in vector species composition in response to LLINs have been reported outside of Africa, including in the Solomon Islands where the highly endophagic and anthropophilic An. punctulatus has been nearly eliminated by LLINs whereas the more exophilic An. farauti remains [14]. Given the importance of mosquito feeding and resting behavior to the effectiveness of malaria control and transmission, there is an urgent need to understand the underlying biological determinants of these behaviors and their short and long term impact on the effectiveness of the existing frontline interventions.

Environmental heterogeneity is known to have a substantial influence on several important vector behaviors [15], including host choice and resting behavior [3]. For example, a recent study in southern Tanzania reported that the proportion of blood meals taken from humans by An. arabiensis fell by over 50% when at least one cow was kept at a household [16]. The resting behavior of mosquito vectors in this study was also highly associated with proximity to livestock; the proportion of An. arabiensis resting indoors fell by ~50% when cattle were present at the household [16]. While these studies demonstrate that the environment can influence An. arabiensis behavior, far less is known about the influence of mosquito genetics on these behavioral phenotypes. An early study by Gillies [17] experimentally investigated the potential heritability of host choice behavior in An. gambiae s.l., and showed these vectors significantly increased their preference for cattle hosts (relative to humans) within a few generations of selection. Other early work demonstrated an association between the 3Ra chromosomal inversion and An. arabiensis found in cattle-sheds [18], and between the 2Rb chromosomal inversion and human-feeding [19]. Understanding the genetic basis for host choice behavior is essential for elucidation of the co-evolutionary forces that stabilize the transmission of vector-borne diseases, and may enable the development of genetic markers that could be used for rapid quantification of the degree of anthropophily in vector populations as is required to estimate transmission risk and plan vector control [20].

In the dengue and zika vector, Aedes aegypti, allelic variation in the odorant receptor gene Or4 has been linked to human-biting preference [21]. However, to date, no ortholog for AaegOr4 has been identified in Anopheline mosquitoes [22], and no direct functional links between genetic mutations in African malaria vectors and behaviors that influence transmission potential have been identified [3,23–25]. As the genera Aedes and Anopheles diverged before the emergence of humans (~150MYA) [26], anthropophily likely evolved independently in these species and may involve distinct mechanisms. Developing the ability to track mosquito behaviors such as biting time [27], degree of anthropophily [3], and resting behavior [28] will help improve vector surveillance and anticipation of whether the effectiveness of control measures are being eroded by mosquito behavioral adaptations [29]. Shifts in mosquito behavior that reduce their contact with interventions, termed behavioral avoidance, may be a significant threat to the long-term goal of malaria elimination [30]. Thus, understanding the genetic contribution to these phenotypes is a critical first step toward effective mosquito control in the future.

Due to the role of An. arabiensis in maintaining residual malaria transmission across much of sub-Saharan Africa [8,13,31], we conducted a field survey to elucidate the genetic basis of host preference and resting habitat choice in this phenotypically variable species. This is the first application of both whole genome sequencing and a population-scale assessment of chromosome inversion frequencies to test for associations between Anopheles mosquito behavioral phenotypes and genotype. As a proxy for host preference and resting behavior, we identified which host each field-collected mosquito fed upon and noted whether it was collected indoors or outdoors. We sampled hundreds of individuals to overcome phenotypic plasticity due to environmental factors (i.e. incomplete penetrance). Assaying these phenotypes in nature is important because lab experiments do not always translate to the field. However, we cannot be certain which choices a given mosquito had before collection, beyond the knowledge that humans and livestock were present in the village of collection. We also cannot conclude causation based on association alone, as other factors unrelated to host preference may be influencing allele frequencies among subpopulations (e.g. insecticide resistance). These limitations should be kept in mind when phenotypes are described in this study.

Results

Analysis of host choice

We analyzed the blood meals from 1,731 An. arabiensis females that were captured resting indoors or outdoors in 3 villages in Tanzania. Specific hosts were identified using a multiplex genotyping assay performed on DNA extracted from female abdomens (see Materials and Methods). The relative frequencies of different host species in blood meals varied by site, but cattle was the most abundant blood source detected in all three. Lupiro had a significantly higher proportion of human-fed mosquitoes (24%; P <0.0001, Fisher exact) compared to Minepa (7%) and Sagamaganga (11%; Fig 1 and S1 and S2 Tables). Mosquitoes that tested positive for more than one host were rare (<5%; Fig 1). To investigate temporal and spatial variation of host choice, mosquitoes were collected from several households throughout a period of 2 years (S9 Table). A backward elimination model selection approach using a Generalized Linear Mixed Model (GLMM) was used to investigate whether host choice was impacted by different environmental factors. Livestock presence at the household level, season (dry or wet), village and trapping location (indoors or out) were included into a maximum model as fixed effects, while collection date and household were added as random effects (S10 Table). The final model showed that livestock presence at the household level and trapping location (indoor or outdoor) were associated with the frequency of human fed mosquitoes found. The proportion of human fed An. arabiensis was inversely correlated with the presence of livestock (P<0.0001, Coeff = -1.92; GLMM, S11 Table). The frequency of human fed mosquitoes was also higher in indoor vs outdoor collections (P = 0.0083, Coeff = -0.7349; GLMM, S11 Table).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Relative host choice between villages.

This figure describes the results of bloodmeal analysis of An. arabiensis collected from: Lupiro, Minepa, and Sagamaganga (SAGA). We detected multiple hosts in <5% of individuals. Different combinations of mixed host bloodmeals were pooled and shown as “Multiple”. A few chicken bloodmeals were also detected at each site.

https://doi.org/10.1371/journal.pgen.1006303.g001

Genetic structure

To test for the existence of genetic structure within our set of 48 sequenced genomes, individuals were partitioned by genetic relatedness using a Principle Component Analysis on all SNPs (PCA; see methods). Using this approach, we observed 3 discrete genetic clusters (Fig 2A). Genome-wide F_ST in sliding windows between individuals in each PCA cluster revealed that the clusters can be explained by distinct combinations of 3Ra and 2Rb chromosome inversion states (Fig 2B). Using a novel inversion genotyping assay (see Materials and Methods), we determined the 2Rb and 3Ra inversion states for individuals represented in each PCA cluster (2Rb_3Ra): left = bb_a+, middle = bb_++, and right = b+_++. There was an enrichment of cattle-fed mosquitoes among bb_++ individuals (P < 0.001; Fisher Exact with Freeman-Halton extension).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Genetic variation explained by the 2Rb and 3Ra inversions.

a) Genetic structure was assessed using genome-wide SNP data for individual An. arabiensis females using a PCA analysis. Three discrete PCA clusters were observed. Red = human-fed and blue = cattle-fed. There is an enrichment of cattle-fed individuals in the middle PCA cluster (P < 0.001; Fisher Exact). (b) To reveal differentiated genomic regions underlying the distinct PCA clusters (left, middle, and right) we plotted F_ST for each chromosome in 100kb windows with 20kb steps between the PCA clusters. The outside PCA clusters differed at the 2Rb and 3Ra inversions (orange), left versus middle PCA clusters differed at 2Rb only (green), and right versus middle differed at 3Ra only (black). Stars indicate the position of SNPs chosen for the inversion genotyping assay.

https://doi.org/10.1371/journal.pgen.1006303.g002

Testing for associations between inversion state and host choice

To explore the relationship between the 3Ra and 2Rb inversion state and host choice, we developed and employed an inversion genotyping assay. In brief, we selected SNPs near the inversion breakpoints (Fig 2B) with extreme F_ST values between genomes grouped by distinct 3Ra or 2Rb inversion states. We then genotyped samples for our 11 inversion diagnostic SNPs (6 for 3Ra and 5 for 2Rb) in parallel using the Sequenome iPLEX platform (see Materials and Methods). We associated genotype information with the standard or inverted arrangement of the 2Rb and 3Ra inversions by genotyping karyotyped samples (S5 Table). In total, we genotyped 363 bloodfed females from the village of Lupiro for inversion state (S6 Table). The samples were composed primarily of human-fed (37%) or cattle-fed mosquitoes (36%; S7 Table). The 2Rb and 3Ra inversion frequencies were within Hardy-Weinberg (HW) expectations for all samples (P = 0.55 and 0.90, respectively). However, the 3Ra inversion was outside of HW among dog-fed individuals (P = 0.02; N = 40, S7 Table). Only four 3Ra homozygotes were observed (N = 363); three fed on dog and one fed on human. The frequency of the 3Ra inversion in Lupiro ranged from 7.94% in cattle to 16.67% in pig-fed mosquitoes. The 2Rb inversion ranged from 81.06% in human to 95% in dog-fed specimens (Table 1). We focused on three major comparisons to test for a relationship between inversion state and host choice: 1) cattle-fed versus human-fed, 2) human-fed versus non-human-fed, and 3) cattle-fed versus non-cattle-fed. After correcting for multiple tests (significant p-value threshold = 0.017), there was no evidence for an enrichment of the standard arrangement of 3Ra (3R+) in cattle-fed mosquitoes compared to human-fed (P = 0.047, χ²; N = 263; Table 1) and no relationship was detected between 3Ra and human-fed versus non-human-fed mosquitoes (P = 0.553, χ²; N = 263; Table 1). However, a significant enrichment of the standard arrangement of 3Ra (3R+) was observed in cattle-fed versus non-cattle-fed (P = 0.007, χ², N = 363; Table 1).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. 3Ra and 2Rb Inversion frequencies by host.

https://doi.org/10.1371/journal.pgen.1006303.t001

Discussion

In this study, we investigate the genetic basis of host choice and resting behavior in An. arabiensis using whole genome sequencing and a novel chromosomal inversion genotyping assay. We did not detect a genetic component (“SNP heritability”) for resting behavior (endo- versus exo-phily). This could be due to substantial “behavioral plasticity” in this phenotype [33,34]. A genetic component for host choice was detected through analysis of genome-wide SNP data. Population-scale inversion genotyping revealed an association between the standard arrangement of 3Ra (3R+) and cattle-fed An. arabiensis. Identifying functional alleles underlying host choice in An. arabiensis has relevance because this species has become the dominant malaria vector in many parts of East Africa, where insecticide use is common [13,35–37]. We highlight two intriguing candidate genes within the 3Ra, including the odorant binding protein Obp5, and the odorant receptor Or65. Obp5 is prominently expressed in female antennae and is significantly overexpressed in female versus male heads [38]. Thus, Obp5 may be involved in host seeking behavior. Obp5 is also significantly overexpressed in non-bloodfed females compared to those who have taken a blood meal in the previous 24 hours [39], further implicating its importance in host seeking behavior. We also detected allelic variation in Or65 between 3Ra inversion arrangements. This gene is a “highly tuned” odorant receptor, that has been shown to be responsive to 2-ethylphenol, a compound found in animal urine [40]. This analysis resulted in some compelling candidate genes, which may be involved in host detection and host discrimination. Controlled host preference assays with distinct genotypes are needed to test for functional links between these candidate genes, among others, and host choice.

“SNP heritability” provides an estimate of the correlation between phenotype and genome-wide SNP profile [32]. A strength of this metric is its robustness to detect genetic components for complex phenotypes that are influenced by many small-effect mutations, which may be the case for host choice in An. arabiensis. In this study, we collected mosquitoes that were blood-fed and resting either indoors or outdoors to assess the genetic basis of host choice and resting behavior. Each phenotype is complex and may be affected, at least in part, by innate preference and the local environment, including the availability of hosts and indoor resting sites. Despite our inability to control for environmental heterogeneities in the field, the SNP heritability analysis detected a genetic component for host choice. Due to the low LD (~200bp) across the genome of this species [41], larger samples sizes (e.g. 100–1000) are needed to rigorously quantify the SNP heritability of host choice, and potentially identify additional candidate genes. Larger sample sizes may also uncover a genetic component to resting behavior, which we did not detect here but cannot rule out. Previously, high inversion polymorphism has been detected in An. arabiensis in Nigeria with some inversions showing changes in frequencies linked to different geographical areas [42]. Hypothesizing there is a functional link between 3Ra and host choice, changes in inversion frequencies could be driven by a higher relative fitness for cattle-biting genotypes in areas with high LLIN usage and/or lower relative fitness of cattle-biting mosquitoes in areas with low cattle density.

This analysis of the genetic basis of host choice in An. arabiensis revealed an association between 3R+ and cattle-feeding. Previously, indirect associations have also been made between host choice and inversions, like the 3Ra in Ethiopia [18] and Kenya [43]. A non-random distribution of the 2Rb inversion has also been reported between human- and cattle-fed mosquitoes [19], but our study is the first to analyze paired karyotype and host choice information from each individual mosquito. Thus, our multiplex genotyping assays allowed us to directly estimate relationships between host choice and genotype in wild mosquitoes in a high-throughput and economical fashion. To ensure that our genotyping method was robust, we selected multiple SNPs near the inversion breakpoints for each inversion. It should be noted that each inversion state represents a suite of linked alleles located primarily within the inversion breakpoints. Further testing is needed to assess how well this assay would perform on An. arabiensis samples from outside our study sites in Tanzania.

The enrichment of 3R+ among cattle-fed mosquitoes provides support for a genetic component to host choice, which is consistent with the report that zoophily can be selected for [17]. The elevated frequency of the 3Ra inversion among various hosts, including pig-fed, dog-fed, goat-fed, and human-fed mosquitoes is suggestive that 3Ra individuals are less choosey. There is also an enrichment of 3Ra/a homozygotes among dog-fed mosquitoes, which is interesting because these genotypes are sufficiently rare in Tanzania that some have even postulated the presence of a recessive lethal in 3Ra [44]. The fact that all other species in the Anopheles gambiae species complex are fixed for the standard arrangement of 3Ra, strongly suggests that 3Ra is derived [45]. Thus, one possible explanation for the observed results is that 3R+ is the ancestral state and alleles therein facilitate specialization on cattle. A loss-of-function mutation in one or more of these genes could then have been acquired early on in the haplotype representing the inverted arrangement of 3Ra, resulting in an expanded host range. This hypothesis is consistent with behavioral heterogeneities and 3Ra frequencies across Africa. For example, An. arabiensis is reportedly more anthropophilic in West African countries like Burkina Faso and Mali [46,47], where the frequency of 3Ra is very high (~40–60%; [48–51]) compared to East African populations, like our tudy area in Tanzania (~12% or less), and others [18,43,52–54]. The diversity of host feeding behaviors among species in the An. gambiae complex, including extreme host specialists (e.g. An. gambaie s.s.) and those with wider host ranges (e.g. An. arabiensis with 3Ra), make this a fascinating system to study the evolution of host choice.

While we provide strong evidence for a role of allelic variation within 3Ra underlying An. arabiensis host choice, the effect size (i.e. relative contribution to the phenotype) is unclear. Controlling for environmental variation is likely to be very important when choosing fully representative samples for each phenotype. For example, a human-fed mosquito may provide a more meaningful representation of host preference if there is an abundance of alternative hosts nearby (e.g. cattle). Some populations of An. arabiensis persistently bite people despite being surrounded by cattle [55]. This highlights the importance of integrating genetic analyses into the wider ecological context in which behavioral phenotypes can be expressed. In a field study like ours, the host choice phenotype measured is a product of both the availability of different host species, and a mosquito’s innate preference for them. We attempted to account for ecological factors like cattle and human host availability, but each host may not have been equally accessible to mosquitoes due to factors such as the use of bednets by humans, insecticides on cattle, or other barriers to choice that are not perceptible to observers. While field studies are invaluable first steps to detect genetic components to important phenotypes, more detailed experimental manipulations will be required to confirm the role of alleles within the 3Ra inversion on blood feeding behavior. For this, we advocate laboratory or semi-field assays in which groups of mosquitoes from each 3Ra inversion state are given a direct choice between different host types in a controlled environment.

This is the first study to report a genetic component to host choice behavior in the major malaria vector An. arabiensis and we link this behavior to allelic variation between the 3Ra inversion states. Mosquitoes that had fed on cattle were significantly more likely to have the presumably ancestral 3R+ inversion. Given that human feeding is essential for malaria transmission, these results may help identify specific markers for assessing the transmission potential of vector populations, and how their behavior evolves in response to control measures, such as insecticide treated nets, which selectively kill mosquitoes attempting to feed on people. This association and the introduction of a novel inversion genotyping assay may be a valuable tool for future malaria vector surveillance. For example, tracking the frequency of the 3Ra inversion in An. arabiensis may elucidate the emergence of behavioral avoidance (e.g. shifting toward zoophily) so countermeasures can be implemented. A better understanding of the genetic basis for host choice in An. arabiensis may also improve vector control if cattle-biting mosquitoes can be genetically engineered and released in the population, having an effect similar in concept to zooprophylaxis [56].

Materials and Methods

Supporting Information

Acknowledgments

We thank Julia Malvick at the Veterinary Genetics Laboratory for her assistance in iPLEX SNP genotyping and three anonymous reviewers for their comments on the previous version of this manuscript. We also thank the residents of Lupiro, Minepa and Sagamaganga for their collaboration during field sampling.