Mosquito biting frequency and how bites are distributed among different people can have significant epidemiologic effects. An improved understanding of mosquito vector-human interactions would refine knowledge of the entomological processes supporting pathogen transmission and could reveal targets for minimizing risk and breaking pathogen transmission cycles.

We used human DNA blood meal profiling of the dengue virus (DENV) vector, Aedes aegypti, to quantify its contact with human hosts and to infer epidemiologic implications of its blood feeding behavior. We determined the number of different people bitten, biting frequency by host age, size, mosquito age, and the number of times each person was bitten. Of 3,677 engorged mosquitoes collected and 1,186 complete DNA profiles, only 420 meals matched people from the study area, indicating that Ae. aegypti feed on people moving transiently through communities to conduct daily business. 10–13% of engorged mosquitoes fed on more than one person. No biting rate differences were detected between high- and low-dengue transmission seasons. We estimate that 43–46% of engorged mosquitoes bit more than one person within each gonotrophic cycle. Most multiple meals were from residents of the mosquito collection house or neighbors. People ≤25 years old were bitten less often than older people. Some hosts were fed on frequently, with three hosts bitten nine times. Interaction networks for mosquitoes and humans revealed biologically significant blood feeding hotspots, including community marketplaces.

High multiple-feeding rates and feeding on community visitors are likely important features in the efficient transmission and rapid spread of DENV. These results help explain why reducing vector populations alone is difficult for dengue prevention and support the argument for additional studies of mosquito feeding behavior, which when integrated with a greater understanding of human behavior will refine estimates of risk and strategies for dengue control.

Dengue, a potentially lethal infection impacting hundreds of millions of human lives annually, is caused by viruses transmitted during mosquito blood feeding. With no vaccine or treatment commercially available, understanding the underlying factors linked to virus exposure is critical for developing more effective dengue interventions. We conducted a study in an endemic region of Thailand where transmission is high and children are expected to be the non-immune, amplifying portion of the host population. We examined Ae. aegypti feeding patterns and risk by matching human DNA profiles in blood-fed mosquitoes to study area residents. A small number of meals matched people from the study area, suggesting that mosquitoes feed on people moving transiently through communities. People under 25 years of age were bitten less frequently than older people. We constructed network models to explore the presence of mosquito feeding “hotspots” and detected a local market “hotspot” in one study village during the high dengue transmission season. Our results provide new details on dengue vector feeding patterns and highlight the need to conduct integrated studies of vector feeding and human behavior, and virus transmission patterns in order to better understand the dengue transmission efficiency and spread.

Citation: Harrington LC, Fleisher A, Ruiz-Moreno D, Vermeylen F, Wa CV, Poulson RL, et al. (2014) Heterogeneous Feeding Patterns of the Dengue Vector, Aedes aegypti, on Individual Human Hosts in Rural Thailand. PLoS Negl Trop Dis 8(8): e3048. https://doi.org/10.1371/journal.pntd.0003048

Editor: Andrea Bingham, University of South Florida, United States of America

Received: November 5, 2013; Accepted: June 13, 2014; Published: August 7, 2014

This is an open-access article, free of all copyright, and may be freely reproduced, distributed, transmitted, modified, built upon, or otherwise used by anyone for any lawful purpose. The work is made available under the Creative Commons CC0 public domain dedication.

Funding: This research was supported by the National Institutes of Health grant RO1 AI022119, NYC-2011-12-219, Cornell's Atkinson Center Climate Change and Disease Program, and the Research and Policy for Infectious Disease Dynamics (RAPIDD) program of the Science and Technology Directorate, Department of Homeland Security and the Fogarty International Center, National Institutes of Health. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Competing interests: The authors have declared that no competing interests exist.

Dengue is the most important arboviral diseases of humans worldwide. It occurs throughout most tropical regions. An estimated 390 million people are infected each year and approximately 96 million people suffer from clinically apparent disease annually [1], [2]. Aedes aegypti is the principal mosquito vector of the four dengue virus serotypes, lives in close association with humans, feeds preferentially on human blood [3]–[5] and has a tendency to ingest multiple blood meals during each gonotrophic cycle [4], [6] facilitating efficient transmission of human blood-borne pathogens. Although a tetravalent dengue vaccine is under development [7], a vaccine and anti-viral drugs are not currently commercially available. As a consequence, current dengue prevention programs are limited to control of the mosquito vector [8].

In this study, we used DNA fingerprinting to detect the individual human hosts from whom female Ae. aegypti took blood meals. Genetic markers have been applied to a variety of studies on mosquito feeding patterns (reviewed by Kent [9]). Coulson [10]was the first to investigate DNA fingerprinting for identifying individual human hosts. Others used a similar approach to address questions about mosquito feeding behavior and bed net efficacy [11]–[15]. In 2000, Chow-Shaffer et al. [16] used variable number tandem repeats (VNTRs) and short tandem repeats (STRs) in a pilot study to fingerprint human DNA in blood engorged Ae. aegypti collected in Thailand. De Benedictus et al. [14] applied the same approach to Ae. aegypti collected in Puerto Rico to analyze feeding patterns on 84 human hosts living in 22 houses and reported that 18% of blood meals were identified as coming from one of two people in a 36 hr time period.

Herein, we report the fine-scale details of Ae. aegypti blood feeding patterns on individual human hosts in a dengue endemic community. In order to better understand Ae. aegypti-human host interactions that underlie local DENV amplification and spread as well as human risk for infection, we used six human microsatellite markers to reconstruct blood feeding patterns of Ae. aegypti collected over multiple seasons, villages and years in west central Thailand. Our study objectives, were to (1) estimate the frequency at which Ae. aegypti bites different people in a 24 hr period; (2) determine whether human age, gender or house of residence predict the frequency at which different people are bitten and (3) evaluate the effect of advancing mosquito age on blood feeding patterns. In large outdoor field cages, we tested hypotheses concerning the impact of human host size and body position on biting behavior. In the laboratory we determined the accuracy of identifying mixed and degraded human DNA in mosquitoes that fed on more than one host. Our results indicate that frequent and heterogeneous biting by Ae. aegypti on residents and transient visitors and mosquito feeding/transmission hotspots are important entomologic features of dengue epidemiology.

Our study was conducted in four villages in northwestern Thailand: Pai Lom (16°45′N, 98°33′E) and Lao Bao (16°45′N, 98°34′E) located in Mae Pa district, 5 km north of Mae Sot in Tak Province; Mae Kasa (16°53′N, 98°37′E), located 20 km north of Mae Sot and Mae Dow (16°53′N, 98°37′E), located 20 km south of Mae Sot. Our field laboratory, a vacant village home, was located approximately 1 km from Pai Lom and Lao Bao. Descriptions of field sites, temperature and humidity during collection periods were previously described by Harrington et al. [17], [18]. Experiments were conducted during both the cool dry season (February 2000, 2001, 2002, and 2003) and warm rainy season (July 2000, 2001, and 2002). These times of the year correspond to periods of low (dry) and high (rainy) dengue transmission in Thailand [19].

Mosquitoes were collected from inside houses using CDC backpack aspirators. Aspirator cartons were placed in plastic bags on wet ice and transported to the field laboratory where mosquitoes were anesthetized with CO2, chilled and sorted by species. During February 2000 and July 2000, abdomens of engorged female Ae. aegypti were saved for DNA analysis by smearing on filter paper, drying and placing in a sterile microcentrifuge tube. Collections from January/February 2001–January/February 2003 were preserved by homogenizing abdomens in 400 µl lysis buffer (1% SDS, 50 mM EDTA, 10 mM Tris-HCL, di H2O) in individual sterile microcentrifuge tubes and transported to the University of California at Davis or Cornell University for further analysis. The right wing was removed from each female and saved for body size estimation. Forceps were sterilized and air dried between each mosquito to prevent cross contamination of samples. Legs from each mosquito were removed with clean forceps and placed in a hexane washed vial for cuticular hydrocarbon (CH) age grading [20].

After obtaining informed consent from study subjects, human DNA samples were collected by gently swabbing the inner cheek with a sterile wooden applicator stick. Four swabs were taken and swirled gently in lysis buffer. Each human sample was provided with a unique code indicating the person, village and date of collection. During each subsequent collection period from 2000–2003 we recorded which individuals were present, who left and who joined the study community. People from whom incomplete profiles were obtained on previous visits were re-swabbed. Participant data was numerically coded to protect the identity of subjects. Children were provided with vitamins and milk as compensation following collection of samples.

This research project was conducted with the approval of, and in accordance with, Institutional Review Boards (IRB) at the University of California at Davis (200210073), Walter Reed Army Institute of Research (752), Thai Ministry of Health Ethical Review Committee for Research in Human Subjects, and Cornell University (FWA00004513). All adult subjects provided written informed consent, and a parent or guardian of any child participant provided informed consent on their behalf.

Mosquito blood meals were extracted at UC Davis or Cornell University following methods described previously [14]. The amount of human DNA in samples was measured and distinguished from mosquito DNA using the Quantiblot Human DNA Quantitation system from Perkin Elmer (Wellesley, MA) following the manufacturer's instructions. Six human loci and a gender identification locus (AMEL)were amplified with PCR using Geneprint primers (Promega Corporation, Madison,WI). These loci were selected because they have been well characterized [21], have a high number of alleles, and relatively small PCR product lengths for greater detection probability through the mosquito blood meal digestion process. The AMEL gender identification locus (212 X, 218 Y bp) was employed, as well as the CSF1PO (291–327 bp), THO1(179–203 bp), TPOX (224–252 bp), D16S539 (264–304 bp), D7S820 (215–247 bp) and D13S317 (165–197 bp) loci. DNA was amplified in a DNA engine Dyad thermocycler (MJ Research, Waltham, MA). PCR products were run on 4–6% acrylamide-bis denaturing gels and visualized with silver staining as described previously [14].

After drying, gels were examined on a light box and alleles in each mosquito blood meal were assigned a number by visual comparison to a reference 100 bp DNA ladder (Promega, Madison, WI, USA). Mosquito blood meals were included in the final analysis only if amplification was successful with at least 5 of the 6 loci. All samples with incomplete profiles after a second PCR reaction were discarded.

Two independent methods were used to estimate mosquito age. The first was mark–release–recapture, as described by Harrington et al. [22]. Briefly, mosquitoes were collected as pupae from natural immature development sites in the study villages. After emergence, adult females of known age were transferred to small cardboard cartons and dusted with florescent powder (DayGlo Color Corp, OH, USA). A unique color was used for each release day. Mosquitoes were released in houses within the study community after obtaining informed consent from residents of each house. Marked mosquitoes were subsequently recaptured with CDC backpack aspirators (John W. Hock Co, Gainesville, FL USA). Collected mosquitoes were transported to the field laboratory where they were anesthetized, identified to species, and examined under a dissecting microscope for florescent dust markings. Blood engorged marked and recaptured mosquitoes were processed as described above and their age was assigned based on known days since eclosion.

A second age-grading method utilized CH analysis as described by Gerade et al. [20]. Briefly, legs were removed from each specimen using clean forceps, placed in dry n-hexane washed vials, and stored until further processing at the University of Massachusetts. Hydrocarbons were extracted and analysis of legs from each mosquito was conducted against an internal standard of octadecane as described by Gerade et al. [20].

Two experiments were conducted to determine the time intervals over which DNA could be detected in a mosquito that ingested one or two (from different people) blood meals.

Mosquito pupae were collected in water- holding containers in the villages of Lao Bao or Pai Lom during January 2001. Pupae were held in 75 mL tubes until emergence and identified to species as adults. Female and male Ae. aegypti were transferred to a large screened cage and held in the field laboratory under ambient conditions. Water was provided ad libitum for the first 2 days after eclosion. On day 3, females were allowed to feed to repletion on a human arm (one of the authors) for 15 min. Feeding time was noted and females that did not feed were discarded. Groups of 30 females were removed at 0, 6, 18, 24, 30, 36 and 42 hr after blood feeding. The experiment was replicated twice. In the field lab, abdomens of engorged mosquitoes were homogenized in lysis buffer. After extraction at Cornell University, the amount of human DNA in the blood meal was determined as described above.

In two separate multiple feeding experiments, mosquitoes were either offered two blood meals from different people with the same interval between first and second feedings (12 hr) and removed at various times following the second blood meal (experiment 1), or offered two blood meals from different people at various time intervals between first and second feedings (experiment 2) (0, 6, 12, 24 hr). One male and one female host (designated as A and B) allowed mosquitoes to bite them in this series of experiments.

In experiment 1, Ae. aegypti (Thai strain from the Mae Sot region, held for ca. 20 generations in the laboratory) were reared in an environmental chamber set with a variable temperature regime representing a typical Thai cool-dry season day (range 22–29°C, 8 degree days, 80% RH). Incandescent lighting was set to a crepuscular profile with 12 hr light:12 hr dark including 2 hr of simulated dawn and 2 hr of simulated dusk. Larvae were reared in trays to obtain medium body size (200/L water), fed 451 mg/larva of diet (1∶1 lactalbumin: brewer's yeast), and pupae were placed in cages for eclosion as described above. At 3–4 days of age, females were aspirated into individual glass test tubes (6 mL) and a piece of mesh was secured around the top.

Mesh-covered glass vials were carefully placed against the forearm of host A. Each mosquito was observed closely and blood feeding was interrupted by removing vials from the arm before mosquitoes could ingest a complete meal. Vials containing partially fed mosquitoes were covered lightly with water moistened paper towels and held in the environmental chamber for 12 hr until the mosquito was offered a second blood meal from host B. Each mosquito was observed directly to confirm feeding on host B. Any mosquitoes that did not feed on host B were removed from the analysis. After ingesting the second blood meal, one half of the mosquitoes were immediately frozen at −20°C and the other half were held for 24 hr before storing at −20°C. A second replicate was conducted following the same protocol except the human host order was reversed.

In experiment 2, mosquitoes were offered double blood meals from the two different people (A and B) at various time intervals between first and second feedings. Female Thai strain Ae. aegypti were reared as described above, placed in a 5 L cage and offered a partial meal from host A. Engorged females were divided into four subgroups of forty mosquitoes per group and allowed to feed to repletion on host B at 0, 6, 12 or 24 hr after ingesting the first blood meal. All mosquitoes were frozen at −20°C six hr after feeding on the second host (B). Mosquitoes that did not feed on both people were discarded.

Human DNA profiles were obtained with informed consent from host A and B as described above. Host DNA was extracted, amplified, and profiled as described above.

Large enclosures were constructed over vacant houses in Pai Lom as described previously [22]. Field cages encompassed an entire house and yard (∼10 m wide ×10 m wide ×4 m high).

Three-day-old female Ae. aegypti (eclosed from field collected pupae) were marked with colored dust (as described above) and released inside the field cage during July 2002. Mosquitoes were released over 4 consecutive days. A total of 353 (112, 60, 132, and 49 over days 1–4, respectively) three-day-old non-blood fed females were released inside and outside a house in the enclosure each evening. On the following day, four hosts (study authors and collaborators) entered the enclosure and remained inside for 30 min; two people were inside the house (one sitting and one lying down) and two people were outside in the same sitting or lying positions. Host location and position was rotated each day. Engorged females were collected with CDC backpack aspirators from the house and yard each day after exposure to hosts. Human DNA was extracted, amplified, profiled and matched to each participant as described above. The experiment was repeated with the same methods and 3 of the same 4 participants and one new participant during January 2003. A total of 863 (137, 75, 77, 324, 155, and 95 over days 1–6 consecutively) female Ae. aegypti were released into the enclosure as described above for July 2002.

In order to understand whether there were spatial patterns of feeding that deviated from random (e.g. “hotspots” or “cold spots”), we used our DNA fingerprinting results to build interaction networks between mosquitoes and human hosts in the villages of Lao Bao and Pai Lom during each sampling period. Each house was considered as a node. Connections were made between nodes (houses) based on mosquito blood meals, linking the house where the mosquito was collected with the house (or houses) were human host(s) lived.

We used a traditional method to characterize networks by evaluating the derived network's topological and structural properties and comparing them with those of random networks [23]. We compared the degree distribution of each network, which was based on the number of connections that a node has; i.e., its degree. The greater the number of connections, the greater the degree. For this analysis, the majority of nodes, therefore, had approximately the same degree (close to the average k of the network). To determine if there were “hotspots” or “cold spots”, we compared the degree distribution of the observed mosquito-human biting networks, with the degree distribution obtained from 999,999 Monte Carlo randomizations (to avoid artifacts) using χ2 analysis at a 5% significance level.

To test the hypothesis that mosquitoes remained in or close to the houses where they were collected and people moving from house to house were bitten [18], [24], [25], we examined the spatial autocorrelation of the relationship between mosquitoes and human hosts based on observed mosquito bites from people in each house. Blood meals where the human host house of residence matched that of the mosquito collection house were defined as “resident” meals. “Non-resident meals” were designated when mosquitoes fed on someone who did not live in the mosquito collection house.

Frequencies of blood meals matched to two different people were compared for experiments that examined time limits of host DNA detection. DNA profiles for mosquitoes and human cheek swabs were compared in a common data base. Questionable or partial profiles were re-amplified and re-run on gels for confirmation. All data were analyzed using two custom programs: Mosquito Matcher and/or Blood Match, which are available from the authors upon request. Both programs allowed matching analysis of data in two different excel spreadsheets. In this way the mosquito blood meal profiles could be matched to the human DNA profiles by village, season, and year. Match ID within Mosquito Matcher allowed matching of human DNA sets with each other to identify non-unique single profiles among the human population and non-unique double profiles in a theoretical mixed blood meal.

The frequency of single and multiple blood meals were initially analyzed by village of residence, season, year, and mosquito age with cross tabulations, χ2 test of independence and t-test. A logistic regression model was then used to model single and multiple blood feeding as a function of these variables simultaneously. Logistic regression was conducted to test the effect of village of origin, host age and sex on the probability of a person being bitten or not. A negative binomial regression model was used to test the effect of village of origin, age and sex on the number of times of a person was bitten. To analyze the effect of host age on probability of being bitten and number of times a person was bitten, age was sorted into two different types of categories. The probability of being bitten was compared among people placed in age classes by each decade from 0 years to an arbitrary upper age limit of 110. The data were also compared for people aged 0–11 representing the age group with a high probability of DENV non-immune individuals [26]. Data for the proportion of mosquitoes captured in the same house where the person they bit lived were compared across season, year and village with cross tabulations and χ2 tests of independence. A logistic regression model was then used to test the effect of all variables simultaneously.

In field cage experiments to compare the effect of body mass and position on host feeding patterns, a body mass parameter was calculated for each host by multiplying the height of the host (m) by their weight (kg). The body mass parameter was compared with feeding frequency across replicates. Regression analysis of the proportion of mosquitoes fed by body mass was performed.

Spatial autocorrelation for the network analysis was tested locally using G statistics [27]. For a distance d, a matrix of neighbor was generated and then local clustering was calculated using Gi(d). Due to differences on inter-house distance, the grain of the analysis was 10 m for Lao Bao and 5 m for Pai Lom; i.e., the variable d was increased every 10 and 5 m, respectively. The significance of G was evaluated using 9,999 Monte Carlo randomizations at a 5% significance level. We used this approach to determine whether popular daytime aggregation sites represent high biting risk to humans, such as homes with attached stores where residents frequented to purchase goods (local markets), were blood feeding hot spots.

All statistical analyses for comparison of blood meal frequency, season, village and year, as well as the effect of host body mass and host position, were performed in (SPSS Statistics 17.0, SPSS Inc., Chicago, IL). Statistical analyses on the interaction networks and local clustering were performed in R [28].

Cheek swab DNA samples were completely profiled for 676 residents from the four study villages and all study collaborators and mosquito collectors (n = 28), who periodically visited study villages. Each individual profile was unique with the exception of two identical twin boys in one village and a mother and daughter with identical profiles that lived in another village. To understand our ability to detect multiple feeding (two different people in one blood meal), we analyzed all the hypothetical combinations of two people using MatchID following the methods of DeBenedictus et al. [14]. A high percentage (85–94%) of these hypothetical combinations were unique (Table 1).

https://doi.org/10.1371/journal.pntd.0003048.t001

A total of 3,677 blood engorged mosquitoes were collected for DNA fingerprinting analysis over the course of the study. Of these specimens, we obtained complete profiles for 1,186, with the remaining samples likely too degraded to profile completely. Of the 1,186 blood meals completely profiled, 430 (36%) matched the profile of a person(s) living in a study village or a study collaborator/mosquito collector (n = 10).

Hourly temperature during the time series experiment in January 2001 ranged from 18 to 36°C, with an average temperature of 26±0.22 SE. At time