Vertebrate-Aedes aegypti and Culex quinquefasciatus (Diptera)-arbovirus transmission networks: Non-human feeding revealed by meta-barcoding and next-generation sequencing

Abstract


Introduction
Mosquitoes (Culicidae) are a diverse taxonomic group found worldwide except Antarctica [1].Most mosquito species require a blood meal for egg maturation, and this behaviour plays a central role in the transmission of pathogens that cause millions of deaths annually.The most common arboviral vector, Aedes aegypti (L.) often takes more than one blood meal during a single egg maturation cycle [2], while Culex quinquefasciatus also transmits numerous viruses through its feeding habit.These feeding requirements elicit anthropophilic and zoophilic behaviours that result in complex blood-feeding patterns [3].Understanding these interactions by characterizing mosquito blood meals is an important tool in unravelling host-vector relationships and associated diseases transmission dynamics [4].Furthermore, analysing the vector host range by estimating the frequency at which vector populations feed on particular hosts can be used as a proxy for assessment of human exposure and in estimation of vectorial capacity [5].
Various techniques have been employed for the analysis of mosquito blood meals such as ELISA or precipitin tests [6,7], and PCR-based approaches [8].Apart from the PCR-Sanger sequencing, such methods are often limited to identifying only a certain protein or region of a gene from a specific host, rendering them amenable to only absence/presence detection of a priori identified blood meal hosts [3].Even with PCR-based methods, results can be limited due to a lack of primers that amplify a wide range of vertebrate hosts [9].Furthermore, when PCR amplification is successful, mixed-feeding events can be missed by direct Sanger sequencing [10].High-throughput next generation sequencing technology can overcome this barrier, but to our knowledge this application has only been applied to Anopheles spp.[3].
The mitochondrial cytochrome c oxidase subunit I (COI) gene has been widely adopted for species identification because it allows species identification through the assignment of a Barcode Index Number [11,12].It has the additional advantage that reference sequences are available for more than 700,000 animal species, far broader coverage than for other mitochondrial genes, such as cyt b [9] and 16S rRNA.DNA metabarcoding is a rapidly evolving method for biodiversity assessment of bulk samples using the COI gene and next-generation sequencing.It has been used in diverse applications including biodiversity monitoring, reconstruction of paleo-communities, and animal gut contents [13].Given the importance of correctly identifying the preferred vector when examining host-vector contact [a concept expressed by the Human Blood Index (HBI)], which is defined as the proportion of insects in a vector population with human blood [14], we investigated the host feeding pattern of Ae. aegypti and Cx.quinquefasciatus using a DNA metabarcoding approach at sites near Reynosa, Mexico, a USA-Mexico border city.In addition, we tested for the presence of key arboviruses which are known to circulate in the region.

Ethics statement
The present study was approved by the Bioethical Committee of the Faculty of Medicine at the Universidad Auto ´noma del Estado de Me ´xico (Approval No. 011/2015) to JGEF.

Study sites and sample collection
Mosquitoes were collected at sites in Pedro J. Me ´ndez neighbourhood (26˚1' 3.576" N, 98˚16' 30.18"W), Reynosa, Tamaulipas, in Northern Mexico (Fig 1).The neighborhood is semiurban characterized by poor infrastructure.Intrastreet roads are unpaved, and houses are mostly uncompleted or unplastered.Waste management practices are poor and streets are often flooded after rainfall.The population of the area is about 7,031 people over a ca.6,016 square/meters comprising an estimated 1,750 households.
Sampling was carried out in the rainy (May to July 2018) and dry (September to November 2018) seasons in 16 households, although some received limited collections due to difficulties in access.Four traps were used for collection in each household.Host-seeking mosquitoes were collected using Biogents Sentinel 1 traps (Biogents AG, Germany); one trap was placed outside each home on a weekly basis for a 24-hour period.Gravid mosquitoes were collected using the Autocidal Gravid Ovitraps (AGO) traps (Springstar) that were checked weekly [15].Indoor resting and host-seeking populations were collected using a CDC backpack aspirator (John W. Hock Company, No. 1412.01,Gainesville, FL) for about 3-min; outdoor resting and blood-engorged mosquitoes were collected using the back-pack aspirator that collected all resting mosquitoes inside a 1 m 3 wooden box opened on one side with both the outside and the inside painted red.Collected mosquitoes were preserved on ice in the field and transported to the laboratory, where they were stored at -80˚C until processing.Sample identification and sorting were carried out on a BioQuip chill plate (BioQuip, USA).Mosquito morphological identification was conducted using standard keys [1,16].Female mosquitoes were further classified into different feeding stages and only engorged mosquitoes of Sella scale 2 (freshly fed) and 3 (late stage fed) were processed for the blood meal analysis [17].Engorged mosquitoes were placed individually into nuclease-free 1.5 mL micro-centrifuge tubes, labelled with an identification number and stored at -80 0 C for further processing.
The study area was residential and the objective of the study is to understand vector behavior in this residential area.The household selection was random and largely dependent on permission granted by house owners as well as availability to allow entrance on their properties to check traps and retrieved mosquitos.Mosquito collection was done outdoors in the yard of these properties using AGO traps, Biogents Sentinel 1 traps (BGS1), and the resting boxes.Only the backpack aspiration was done indoors.Also, the nature of the study area, where all of the houses shared fences, provided opportunity to sample mosquitoes within a 50-meter radius of each outdoor trap, as this is within the flight range of mosquitoes.Furthermore, with a sampling duration of 5 months, we have a strong reason to believe there is enough representation of the mosquito population in this area.We did not attempt to quantify the behavior of the household occupants and how much time was spent inside versus outside.However, anecdotally, we observed a significant proportion of their activities, including cooking, grilling, and evening family gatherings, were done outside.Permeability of houses to mosquitoes as well was not measured.Given mosquitoes can have a flight range of up to 100 m post-feeding [18] and species of Aedes and Culex have 2-5 km flight range [19], we do not think the spatial distribution of the households might have significantly impacted the bloodmeal feeding pattern observed.

DNA extraction and purification
Blood-fed Ae. aegypti and Cx.quinquefasciatus were sent to the Canadian Centre for DNA Barcoding, Centre for Biodiversity Genomics, University of Guelph (Guelph, ON, Canada) for DNA extraction and sequencing.The frozen mosquitoes were incubated in liquid nitrogen for one minute to ensure that the blood meal did not thaw and leak during removal of the abdomen.The abdomens were removed with forceps and placed directly into 60 μL of lysis buffer [700 mM guanidine thiocyanate (Sigma), 30 mM EDTA pH 8.0 (Fisher Scientific), 30 mM Tris-HCl pH 8.0 (Sigma), 0.5% Triton X-100 (Sigma), 5% Tween-20 (Fluka Analytical), and 2 mg/mL of Proteinase K (Promega)], which was then incubated for 18 h at 56˚C with gentle shaking.DNA purification employed the glass fiber technique described by Ivanova et al. [20].The lysate was mixed with two volumes (120 μL) of binding mix [3 M guanidine thiocyanate, 10 mM EDTA pH 8.0, 5 mM Tris-HCl pH 6.4, 2% TritonX-100, and 50% ethanol] and applied to a well of a 96-well silica membrane plate (PALL), 700 μL at a time.The plate was centrifuged at 5000g for 5 min after each application.The membrane was washed once with 180 μL of protein wash buffer [1.56 M guanidine thiocyanate, 5.2 mM EDTA pH 8.0, 2.6 mM Tris-HCl pH 6.4, 1.04% Triton X-100, and 70% ethanol] and centrifuged at 5000g for 2 min.The plate was then washed a second time with 750 μL of wash buffer [50 mM NaCl (Fisher Scientific), 0.5 mM EDTA pH 8.0, 10 mM Tris-HCl pH 7.4, and 60% ethanol] and centrifuged at 5000g for 5 min.Centrifugation was repeated a second time to remove residual wash buffer.All traces of wash buffer were removed by incubating the plate at 56˚C for 30 min.To release DNA from the silica membrane, 40 μL of elution buffer [10 mM Tris-HCl, pH 8.0, pre-warmed to 56˚C] was applied directly to the membrane and allowed to incubate at room temperature for 1 min.DNA was eluted from the membrane (into a clean plate) via centrifugation at 5000g for 5 min.A negative control (lysis buffer only) was processed in parallel with the samples.

PCR amplification and sequencing
Host DNA was prepared for deep sequencing using a two-stage fusion primer approach [13] with vertebrate-specific primers.Two microliters of DNA extract was added to a PCR reaction consisting of 6.25 μL of 10% D-(+)-trehalose dihydrate (Fluka Analytical), 2.0 μL of Hyclone ultra-pure water (Thermo Scientific), 1.25 μL of 10X Platinum Taq buffer (Invitrogen), 0.625 μL of 50 mM MgCl2 (Invitrogen), 0.125 μL of each 10 μM primer, 0.0625 μL of 10 mM dNTP (KAPA Biosystems), and 0.060 μL of 5U/lL Platinum Taq DNA Polymerase (Invitrogen) for a total reaction volume of 12.5 μL.The primers were a cocktail of BloodmealF1_t1, BloodmealF2_t1, VR1_t1, VR1d_t1, and VR1i_t1 (S1 Table ), which were tailed with M13F and M13R sequences and targeted a 185 bp fragment of the COI barcode region.The Blood-mealF primers were designed in this study to amplify DNA from common vertebrates used by mosquitoes, not to amplify insect DNA including mosquito, and to amplify short fragments (<200 bp) given the nature of degraded DNA in blood meals.Thermocycling consisted of an initial denaturation at 95˚C for 2 minutes, 60 cycles of 95˚C for 40 seconds, 56˚C for 40 seconds, and 72˚C for 30 seconds, and a final extension at 72˚C for 5 minutes.PCR products were visualized on a 2% E-gel (Invitrogen) to confirm amplification, and then diluted two-fold with sterile water.
The diluted products were then used as a template for a second round of PCR using M13F primers tailed with Ion Xpress universal molecular identifier (UMI) tags and the Ion Torrent "A" sequencing adapter, and M13R primers tailed with the Ion Torrent trP1 sequencing adapter (S1 Table ).Reaction components for the second round of PCR were identical to the first and thermocycling consisted of an initial denaturation at 95˚C for 2 minutes, 5 cycles of 95˚C for 40 seconds, 45˚C for 40 seconds, and 72˚C for 30 seconds, 35 cycles of 95˚C for 40 seconds, 51˚C for 40 seconds, and 72˚C for 30 seconds, and a final extension at 72˚C for 5 minutes.The products from the second round of PCR were pooled in equal volumes and purified by mixing 400 μL of pooled product with 400 μL of purification beads (Aline Biosciences, Woburn, MA, USA) for a ratio (v:v) of 1:1 (beads:DNA).The mixture was incubated at room temperature for eight minutes to allow the DNA to bind to the beads, after which the beads were pelleted on a magnet.The supernatant was discarded, and the pellet was washed three times with 1 mL of freshly prepared 80% ethanol and then air dried for 10 minutes.The purified product was eluted from the beads by resuspending the dried pellet in 200 μL of sterile water, pelleting the beads, then carefully transferring 180 μL of the supernatant to a clean 1.5 mL tube.The purified product was quantified using a Qubit 2.0 fluorometer and adjusted to 26 pM with sterile water.The 26 pM library was prepared for sequencing and loaded onto a 530-chip using an Ion Chef automated platform (Thermo Scientific) with standard 400 bp chemistry.Deep sequencing was performed on an Ion Torrent S5 using 850 flows.

Bioinformatics analysis
Following sequencing, raw sequence reads were automatically assigned to samples (via the UMI tags) by the Torrent Browser suite.The reads were filtered based on a minimum quality value of QV20 and a minimum read length of 225 bp using Sickle (github.com/ucdavisbioinformatics/sickle).Adapter and primer sequences were removed using CutAdapt [21].Since the forward primer should be readily visible in the reads, any reads in which the forward primer was not recognized were discarded to ensure that only high-quality reads were included in the final dataset.Additionally, trimmed reads were only retained if they were between 150-200 bp.Reads passing these filters were collapsed into unique sequences (http:// hannonlab.cshl.edu/fastx_toolkit/index.html) while retaining original read counts.The collapsed sequences were then clustered into operational taxonomic units (OTUs) using 97% identity [22], and OTU sequences were retained only if they were less than 220 bp in length and composed of at least 100 reads.Each OTU sequence was used to query (BLAST) a custom database composed of global vertebrate COI sequences downloaded from BOLD.The resulting BLAST hits were filtered to retain only those with a minimum match of 95% identity and 100 bp of coverage between the queried sequence and a reference sequence.Furthermore, identifications were only considered reliable if supported by at least 50 original reads.All raw data has been submitted to the SRA under accession PRJNA600226.

Animal surveys
In order to estimate the relative abundance of avian species in the trapping areas, a bird survey was conducted as described by Brugman [17] in eight points approximately 50 meters apart between 08:00 and 10:00 hrs.when the avian population will still be relatively more active.Observations were made at each point for 10 minutes, and other observed vertebrates, including horse and dogs, were also noted.We also conducted individual household censuses of vertebrates in the trapping area using a questionnaire.An adult in each household was asked for the number of dogs, cats, chickens, and other animals in the household.Where possible, study personnel conducted headcounts of the animals for each household.The combined total number of animals for each species recorded was used to estimate the abundance of common vertebrates in the trapping area [17].

Viral extraction and detection by qRT-PCR
Non-engorged Ae. aegypti mosquitoes were tested for arboviruses including the four serotypes of DENV, ZIKV, and CHIKV.Mosquito pools were processed according to standard procedures [23].In brief, each mosquito pool was homogenized in 600 μL of Hank's buffer solution (Gibco cat.#14170112) with a single bead and centrifuged for 5 minutes at 18000 g.RNA extraction was carried out using 100 μL of the supernatant with a MagMax CORE Nucleic Acid Purification Kit (ThermoFisher, Waltham, MA) following the manufacturer's instructions.For a subset of samples we quantified nucleic acid recovered using a spectrophotometer (Epoch, BioTek Instruments, Inc.).DENV was tested for using the PathID Multiplex one step (Applied Biosystems) and Qiagen QuantiTect probe reverse transcriptase polymerase chain reaction (RT-PCR) kits with the primers DENV_F, DENV_R1-3, DENV_R4, and DENV_P probe described by Alm et al. [23].DENV strains used as positive controls were provided by the University of Texas Medical Branch.We added 5 μL of RNA to a 20 μL master mix containing 12.5 μL RT-PCR buffer, 0.1125 μL each of the primers, 0.2 μL of DENV_P probe, 0.25 μL enzyme, and RNase free water.Cycling conditions were as follows: reverse transcription cycle at 50˚C for 30 minutes, followed by an RT-enzyme inactivation cycle at 95˚C for 20 minutes, then 45 cycles of 95˚C for 3 seconds and 60˚C for 30 seconds.CHIKV and ZIKV were tested for using a multiplex RT-qPCR described by Lanciotti et al. [24,25].Briefly, 5 μL of RNA was added to a master mix containing 0.3 μM of each probe, 2 μL of each of the four primers, 2.5 μL of enzyme mix and RNase free water.Zika PRVABC59 RNA lysate and CHIKV R80422a [23] RNA lysate received from the CDC were used as positive controls.Cycling conditions were as follows: reverse transcription cycle at 48˚C for 10 minutes, followed by an enzyme inactivation cycle at 95˚C for 10 minutes, then 40 cycles of 95˚C for 15 seconds, and 60˚C for 1 minute.All samples with Ct values >38 were considered negative [22,23].Cx. quinquefasciatus was tested for WNV using the primers WNENV-forward (TCAGCGATCTC TCCACCAAAG) and WNENV-reverse (GGGTCAGCACGTTTGTCA-TTG) and the probe WNENV-probe as described by Hamer et al. [26].

Data analysis
We estimated the forage ratio (FR), defined as the frequency at which a mosquito selects a vertebrate host over other available vertebrate hosts, by dividing the observed frequency of blood meals by the expected frequency of blood meals of a given species [27,28].FR = s/a where s = the percentage of female mosquitoes containing blood of a particular host, and a = percentage of the total available host population represented by that particular host (as determined by the animal survey).A FR of 1.0 indicates mosquitoes are feeding on hosts in equal proportion to availability, whereas values >1.0 indicate over-utilization and values <1.0 indicate under-utilization.The human blood index (HBI) was calculated as the sum of human blood meals for a mosquito species divided by the sum of engorged mosquitoes of a particular species with successful host identification [29].
Virus screening results were analysed using the Bayesian algorithm in the Katholi´s poolscreen program as previously described [30].The 95% exact upper limit confidence interval (95%-ULCI) for point prevalence of arbovirus infection in both vector species populations were estimated.

Mathematical modeling
Investigating the transmission efficiency and risk of human exposure to Ae. aegypti-transmitted viruses are essential steps to developing successful intervention strategies.Thus, vectorial capacity (VC) that incorporates blood feeding behaviour and vector longevity along with vector competence was calculated to determine the intensity of dengue virus transmission with the following equation [31]: where n = extrinsic incubation period; b = vector competence; p = survival rate; and h = blood feeding rate.
We applied the model of Chan and Johansson [32], which relates the extrinsic incubation period (EIP) with temperature as a covariate.The EIP (n, in days) was estimated as a function of temperature (T) using the following equation: n T ð Þ ¼ 600ð0:3=2pÞ 0:5 exp À 0:3 TÀ 5: Vector competence (b) was calculated by using the equation of The survival rate, p, can be expressed by e −μ [34].Besides, according to the experiments by Yang et al. [35] in female Ae.aegypti over the temperature range of 10.54˚C< = T< = 33.41˚C, a 4 th order polynomial function can be used to describe the relationships between the temperature and μ as follows:

Forage ratio and human blood index
Based on our animal survey, each household contained an average of 3.6 (± 0.45 SEM) human inhabitants, and the study area contained an average of 79 humans, 65 dogs, 48 cats, 37 chickens, and 3 pigs (S4 Table ).Forage ratios for Ae.aegypti and Cx.quinquefasciatus were calculated with host availability estimated from these results.The highest forage ratio of Ae. aegypti was for dogs (4.34), which was nearly twice as high as the second highest forage ratio for humans (2.16; Table 2).The highest forage ratio of Cx. quinquefasciatus was also for dogs (5.64), followed by Virginia opossum (Didelphis virginiana; 4.83) and turkey (4.83;Table 3).
The human blood index (total number of human blood meals divided by the total number of engorged females with a confirmed result) for Ae.aegypti and Cx.quinquefasciatus was 39.3% and 4.1%, respectively.

Mosquito arbovirus RNA testing
A total of 2,941 female Ae.aegypti were collected during the study.Out of these, 118 pools were made from 2,545 mosquitoes with an average of 18.4 (± 1.73 SEM) mosquitoes per pool.These were tested and none was found positive (point prevalence = 0) for DENV 1-4, ZIKV and CHIKV using qRT-PCR (positive controls returned positive results, indicating that the assay functioned as expected).Similarly, of the 3,536 female Cx.quinquefasciatus specimens, 37 pools, representing 1,057 mosquitoes, was tested for WNV (using qRT-PCR), and all were found to be negative (point prevalence = 0).The associated 95%-ULCI for prevalence of any of the viruses examined in Ae. aegypti was 0.75/1,000 (0.075%) mosquitoes examined, while the associated 95%-ULCI for prevalence of WNV in Cx. quinquefasciatus was 1.79/1,000 (0.18%) mosquitoes examined (Table 4).

Mathematical modelling
According to the findings of Rodrı ´guez-Pe ´rez et al. [36], six people were IgM-positive and seroconverted from a total of 77 examined people, therefore the Incidence/S H = 6/(77-6).The prevalence of dengue virus infection should be varying with time.We assumed that it is a constant due to the absence of such data.Similarly, the ratio of the number of new cases to the number of humans is also assumed to be constant.We derived the ratio of the number of humans to the number of mosquitoes.Namely, 1,527 landing counts in 231 households; then assuming on average four persons per household, yields N H /N M = 231 � 4/1527.The infection rate of DENV in Ae. aegypti is 0.47% (3/633) as from Rodrı ´guez-Pe ´rez et al. [36].The biting rate (a) as from Garza-Herna ´ndez et al. [37], where a = 1/5, the human biting rate for DENV in Mexico.PLOS NEGLECTED TROPICAL DISEASES Furthermore, the blood feeding rate h be (105/1106) � (57/145) = 3.7% (S2 Table ).Here, 105 is the number of fed females and 1,106 is the total caught number of Ae. aegypti.According to Figs 2 and 3, 57/145 = 39.3% is the proportion of human blood meals identified.Then 105 � 57/145 is the number of mosquitoes that will bite human among the total 1,106 mosquitoes.Therefore, 3.7% is the proportion of mosquitoes that will bite humans among the total mosquito population.
Using the equation of vectorial capacity, we estimated VC of Ae. aegypti for DENV in Reynosa which ranged from 112 to 172 from temperatures ranging between 22 and 30˚C, respectively (Fig 4).

Discussion
Information on vector-host interactions and associated disease transmission dynamics obtained through the identification of blood meal origin in mosquitoes is important for disease control [4].

PLOS NEGLECTED TROPICAL DISEASES
The acquisition of a blood meal is an essential component of the vectorial potential of any mosquito vector.Also important is whether the female mosquito preferentially feeds on animals or humans, particularly in estimating the basic reproduction number associated with the transmission potential of the system under study.This, in turn, is closely linked to vector control strategies.In this context, the occurrence of epidemics involving viruses transmitted by Ae. aegypti have increased in the Americas.In the present study, we observed an overall relatively low density for Ae.aegypti in residential neighbourhoods of Reynosa, relative to Cx. quinquefasciatus.An earlier study in the Nuevo Amanecer neighbourhood in the eastern tip of Reynosa city, reported five host-seeking Ae. aegypti per household [36], a stark contrast to 20 females/household reported in a dengue-endemic tropical country of Thailand [38].Although it is arguable that a low abundance of Ae. aegypti might reduce disease transmission, there is limited knowledge as to how this influences the ability of Ae. aegypti to feed on humans in highly urbanized areas of Northern Mexico, and how proportional human feeding might drive virus transmission potential.
Aedes aegypti is considered a highly anthropophilic species [39][40][41][42]; however, we found that 61.7% (129/209 examined; S3 Table) of individual Ae.aegypti with a blood meal result had fed on non-human hosts.A similar study conducted on the opposite side of the US-Mexico border in South Texas recently found a similar result, with 70% of Ae. aegypti feeding on non-human hosts [43].In the current study, the proportional feeding of Ae. aegypti on domestic dogs (54.2%) was almost twice that of the second most common host, humans (28.0%).This was also similar to the findings of the study north of the US-Mexico border where dogs were responsible for 50% of the Ae.aegypti host identifications [43].However, the blood meal analysis methodology of the Olson et al. [43] study was PCR-Sanger sequencing which is different to the meta-barcoding next-generation sequencing approach used in the current study, making the results difficult to compare.Equally surprisingly is the host feeding pattern observed in Cx. quinquefasciatus, with domestic dog as the dominant host species.This finding was unexpected because prior studies have shown that Cx. quinquefasciatus is strongly ornithophilic [8,44].Estimating the relative density of wild and domestic animals is difficult, and although our study found that chicken was the second most fed on host by proportion, forage ratio analysis revealed dog, Virginia opossum, and turkey as the most over-utilized hosts.The domestic dog was over-utilized by Cx. quinquefasciatus even more so than Ae.aegypti, with a 20-fold (79.2%) increase over the second most abundant host, chicken (3.7%).Our forage ratio analysis was based on a survey of animals present in these communities.Given that we were attempting to sample a combination of wild and domestic animals, we used a combination of a 'census', by counting all animals observed during point counts as well as a 'survey', in which a subsample of the animals present was estimated.We asked the homeowners for the number of animals in each household.The forage ratios are based on the relative abundance of these vertebrates, which would be difficult to compare wild animals (such as birds) to domestic animals (such as dogs) given the difference in survey methodologies.The interpretation of the high forage ratio on dogs is difficult to interpret for this reason.Prior published studies report that Culex quinquefasciatus are ornithophilic [45] so the difference compared to the current study could be due to host availability.The Olson et al. [43] study conducted in South Texas found that only 22% of the Cx.quinquefasciatus blood meals came from dogs while 67% were from chicken and 9% were from wild birds.In the South Texas low-income residential study sites, dogs made up 31% of the vertebrates surveyed in the neighborhood and chickens made up 17% [43].In the current study, dogs accounted for 11.8% of the vertebrates and chickens accounted for 6.7%, although the survey in the current study included wild birds which were not included in the South Texas study.Considering the slightly different survey designs, the availability of dogs versus chickens appear similar in the low-income communities north and south of the US-Mexico border.Mixed feeding patterns were also observed in common hosts: 57 (39.3%)Ae. aegypti out of 145 individuals examined fed on humans (30 individuals contained single feeding with human blood and 27 individuals contained mixed feeding with human blood; Figs 2 and 3).Very limited studies have reported a feeding pattern in Ae. aegypti where humans represent less than half of the blood meals (but see [34]).To our knowledge, this is the first study using next-generation sequencing for the blood meal analysis of Ae. aegypti, making it difficult to compare these results to prior published studies.Prior studies on Ae. aegypti blood meal analysis often utilized precipitin or ELISA assays, however, ELISA has been reported to have less specificity with assays involving human, dog, and cat blood meals [46].Additionally, PCR and direct sequencing of the amplicon could miss blood meals containing more than one vertebrate species.Therefore, previous studies likely underestimated the frequency of Ae. aegypti feeding on multiple vertebrate species.Additionally, multiple feeding on different individuals of the same species likely occurs frequently, but our methodology would not have detected these observations.Although relatively still expensive in resource-limited settings, advances in next generation sequencing techniques and higher taxonomic coverage in reference databases have enabled the exploration of the meta-barcoding approach for a more reliable analysis of mosquito blood meals.
Anopheles gambiae was shown to exhibit polymorphism for host preference [47], and histological examination revealed that nearly one-half of all engorged Ae. aegypti females in Thailand contained more than one blood meal, while about one-third of engorged females in Puerto Rico contained multiple blood meals [41].Although the results of the present study cannot be explicitly compared with these earlier studies [41,47] because constraints of methodology applied; nevertheless, nearly one-quarter (38/145) of Ae. aegypti blood meals were mixed and accurately identified by next-generation sequencing technology.Females of Ae. aegypti have been shown to take blood meals 2-3 times per gonotrophic cycle, seeking and rapidly taking another meal, two to five hours after taking the first one [41,48].This probably explains the high proportion of mixed dog-human feeding observed in the present study.
A plausible explanation for the low anthropophily observed in Ae. aegypti could be the relatively low human density compared to dogs.Only three of the households surveyed for the animal census were without a pet dog, and that was nullified by the number of stray dogs roaming the streets of the Pedro J. Me ´ndez neighbourhood.This explanation is corroborated by the observation that Cx. quinquefasciatus, a species ubiquitously classified as ornithophilic [8,44], also fed proportionally more on dogs than avian hosts.304 (78.3%)Cx. quinquefasciatus out of 388 individuals examined fed on dogs (254 individuals contained single feeding with dog blood and 50 individuals contained mixed feeding with dog blood; Figs 2 and 3).Indeed, the relative availability of vertebrate hosts can profoundly influence the feeding patterns of host seeking mosquitoes [49].Although, the estimated abundance of humans in the study area (14.4%) was higher that observed for dogs (11.8%;Table 2), it is significantly lower when compared with the relative abundance of all non-human hosts estimated.Host selection pattern in Cx. quinquefasciatus has been suggested to follow the domestic environment [8], which is expected to be predominantly non-human in this study.
Aedes aegypti evolved as an effective vector as part of forest-dwelling mosquito group with resting preferences on tree tops and breeding preferences in accumulated water in tree holes.Domestication of these species via the replacement of natural larval habitat with artificial container habitats in the urban landscape resulted in the adaptation to the human environment, and selected for highly endophilic and endophagic behaviour [41,50].However, Ae. aegypti has spread throughout much of the tropical and subtropical regions of the world, with very unique urban landscapes and very different degrees of humans as viable hosts.For example, housing quality and characteristics are known to influence the ability of mosquitoes to enter homes [15].Additionally, continuous application of insecticides to households using ultra-low volume adulticides, likely further limits the endophilic and endophagic behaviour in certain regions.Although not explicitly investigated in the current study, fewer mosquitoes were collected indoors using the CDC backpack aspirators, as a proxy for indoor collection, than any of the other traps placed outside (S2 Table ).Elizondo-Quiroga et al. [8] also observed a higher density of mosquitoes when collected outdoors than indoor in Nuevo Leon.Likewise, a twoyear surveillance study of Ae. aegypti in Texas showed that outdoor collections in low-income neighbourhoods were almost eight times that of indoor collections, with this ratio increasing to as high as 15 times in middle-income communities [15].A small change in host defensiveness has been predicted to drive a larger change in host choice [49].Mosquito blood meal sources often occurring in aggregation are sometimes density-dependent, and anthropophagous insects tend to select for less defensive hosts, a host choice selection mechanism reinforced through limited learning [41,49].Selection for host preference after insecticide application has been reported in Anopheles gambiae, with accompanying behavioural change from indoor to outdoor resting, and feeding on cattle and other non-human hosts [47].
The neurological behaviour underlying mosquito feeding preferences is not well understood, and few efforts have been made to understand the genetic basis [41,47].It is not audacious to speculate that host selection and preference could be regulated by as yet unidentified genes, which have undergone positive selection in response to conditions that threaten survival.The low incidence of human feeding of Ae. aegypti observed in the present study is significant with regards to disease transmission dynamics and control.In 1970, Hess & Hayes [51] proposed that the concept of zooprophylaxis, wherein non-human mammals available as hosts can reduce bites to humans by dengue vectors, thus reducing the arbovirus transmission potential.This occurrence has been demonstrated in Kenya and Uganda where non-human host feeding by Ae. aegypti was considered responsible for this species being a poor vector of yellow fever virus [52].The suggestion to genetically modify vectors in order to eliminate the chemical signals that promote human-feeding preference has not been seriously considered due to the uncertainty of potential inherent risk [41], leaving zooprophylaxis as a safe and viable option.The utilization of non-human hosts and exophilic feeding observed in Ae. aegypti populations in Reynosa and in the South Texas study [43] suggests that zooprophylaxis could be a potential control measure to consider.In places where dogs contribute substantially to Ae. aegypti feeding, treating dogs with anti-parasitic drugs would not only protect them from ectoand endo-parasites, but could also reduce vector populations and reduce the risk of human exposure to arboviruses.For example, ivermectin has been reported to be a safe, short-term insecticide useful for malaria control in areas where mosquitoes are exophagic or exophilic [53].In some cases, mosquito populations were reduced up to 80% after feeding on ivermectin treated individuals [54].We also support our argument that anti-parasitic drugs can have an effect on vector populations in the context of the same vector/virus system.For instance, Deus et al. [55] found that ivermectin induces Ae. aegypti adult mortality and decreases the hatch rate of their eggs.
Diversion of vector bites to non-human competent hosts reduces viral amplification and reduces potentially infectious bites to humans by dengue vectors, thus could be used as a control tool to limit arbovirus transmission.Although an earlier study in a similar Reynosa neighborhood isolated CHIKV from patients with dengue-like symptoms [56], and co-circulation of DENV1-3 and West Nile virus [57], all pools of host-seeking Ae. aegypti were negative for the presence of arbovirus RNA (DENV 1-4, ZIKV, and CHIKV), and all pools of Cx. quinquefasciatus were negative for WNV RNA.The explosive arrival of viruses transmitted by mosquitoes, such as CHIKV and ZIKV, into the western hemisphere has demonstrated their rapid ability to become endemic, with significant public health consequences.These viruses are of public health significance in Mexico, and are zoonotic with transmission cycles involving wild or domestic animals.Enzootic transmission cycles with wild mammal reservoirs can sustain virus populations in nature and initiate new foci of urban transmission through spill-over events.As Ae. aegypti is a vector of DENV, CHIKV, and ZIKV, while Cx.quinquefasciatus is known vector of WNV, we can hypothesize that domestic dogs can serve as a sentinel species for predicting virus activity and human disease risk in Northern Mexico, given the observation of high host utilization of dogs.Canine sero-surveillance has been suggested as a useful tool for public health risk assessment [58].A study in France and Africa demonstrated the usefulness of dogs as a sentinel for studying WNV epidemiology and circulation [11], and similar studies have been conducted to study Chagas diseases in Texas and Northern Mexico [58,59].Several studies have suggested dogs can be a sentinel for WNV [11].Thus, although dogs may be unable to achieve the infectious viremia of arboviruses, seroconversion and neutralizing antibody production increases the opportunity of detecting low level arbovirus transmission.
Here, VC of Ae. aegypti for DENV ranged from 112 to 172 depending on temperatures (between 22 and 30˚C, respectively).However, if h were 100% (i.e.all Ae.aegypti biting humans; 97% of Ae. aegypti feeding on humans was documented in Thailand and Puerto Rico) [41], VC would increase 2.5 times (ranging from 287 to 437, respectively).Hence, if the proportion of Ae. aegypti bloodmeals on humans were even higher than that reported here, the transmission of dengue viruses in this area would likely be much higher.Pooled mosquitoes would not have contained blooded individuals to avoid false positive results when modeling the VC due to undigested blood meal.The vector competence is the product of several factors, and should be varying with time.For example, the factor IM(t)/N_M(t), which represents the prevalence of dengue virus infection should be varying with time.We assumed that it is a constant due to the absence of such data.Similarly, the ratio of the number of new cases to the number of humans is also assumed to be constant.Some parameters that are unavailable were derived from literature, such as human biting rate, a, and the ratio of the number of mosquitoes to the number of humans are also estimated assuming that there are four people in one household.The survival rate varying with time is also from literature.The estimations would be more accurate with more data available.
In conclusion, the population of Ae. aegypti in Reynosa, Mexico fed on humans only 39.3% of the time, probably due to the abundance of non-human hosts in the residential neighbourhood and the low human density.The high rate of non-human blood meals for Ae.aegypti and dog blood meals for Cx.quinquefasciatus suggests that zooprophylaxis could be a control option to consider for DENV, ZIKV, CHIKV, and WNV, respectively.Nonetheless, the high number of blood meals from dogs and cats is in of itself a public health concern, due to the concomitant increase in the likelihood of zoonotic pathogen transmission (such as canine heartworm) and the potential spill-over to human populations.This study identifies the need to re-evaluate our knowledge of host utilization by Ae. aegypti and Cx.quinquefasciatus by meta-barcoding-like approach with next-generation deep sequencing technology in other regions of Mexico and the tropics in general; Determining VC´s components would be useful determinants to help tailor successful vector control programs.

Fig 1 .
Fig 1. Study site and location of traps in Pedro J. Mendez neighbourhood in Reynosa, Northern Mexico.Fig 1 is an original map, created by co-author AAA and is available for use under the CC-BY license.ESRI 2011.ArcGIS Desktop: Release 10.Redlands, CA: Environmental Systems Research Institute.https://doi.org/10.1371/journal.pntd.0008867.g001

Fig 4 .
Fig 4. The relationship between temperature and vectorial capacity of Aedes aegypti.Extrinsic incubation period, vector competence, survival rate, and vectorial capacity were calculated using models as reported elsewhere [31-34].It should be clarify that the vectorial capacity model used is for the mosquito population in general, instead of an overall model of vectorial capacity for Aedes aegypti.https://doi.org/10.1371/journal.pntd.0008867.g004

Table 1 . Average ± SEM of mosquitoes per trap per collection.
The total mosquito counts retrieved from each trap were averaged by the number of collection events that yielded the count, which translates to the number of individuals per trap per collection event.

Table 2 . Forage ratio estimates for Aedes aegypti collected in Reynosa, Northern Mexico.
� This is greater than the 145 number of Ae. aegypti with successful host identification because each mixed feeding event represented two host species.n/a = not surveyed.HBI-Human blood proportion or engorged females containing human blood.https://doi.org/10.1371/journal.pntd.0008867.t002

Table 3 . Forage ratio for Culex quinquefasciatus collected in Reynosa, Northern Mexico.
� This is greater than the 388 number of Cx. quinquefasciatus with successful host identification because of mixed host feeding, with each mixed feeding representing two host species.https://doi.org/10.1371/journal.pntd.0008867.t003

Table 4 . qRT-PCR virus screening results of the mosquito pools examined.
� The value represents the maximum likelihood point estimate and the value in parentheses represents the upper bound 95%-confidence interval.https://doi.org/10.1371/journal.pntd.0008867.t004