Study site
The study was conducted on eight farms in Dangur district (woreda), in Metekel zone, Benishangul-Gumuz region, Ethiopia (Fig. 1). Four of these farms were large-scale farms (larger than 100 hectares [ha]) and four were small scale farms (less than 100 ha). The farms cultivated a range of crops, including sesame, green grams, cow peas, and sorghum. The area has a single rainy season beginning in May that continues until October. The altitudes of the farms range between 751 to 1155m above sea level. The worker housing on these farms was generally of poor quality, consisting of wooden framed houses with grass or iron sheeting walls.
All data were collected between July and December 2017. The managers of two farms (one small, one large) where mosquito collections were made in July did not wish to continue in the following months, so these farms were replaced with two other farms based on their similarity in location, size and proximity to the breeding habitats with the previous ones. At each farm, two shelters were chosen, and collections were made inside and outside these shelters during each collection. One collection was made in each site each month, resulting in 16 indoor and 16 outdoor collections made each month. The collections were made between July and December (6 months), resulting in a total of 96 indoor and 96 outdoor collections.
Mosquito collection
Mosquitoes were collected through human landing collection (HLC), the current gold standard for measurement of human biting activities of mosquitoes and entomological inoculation rates [10-12]. HLCs were chosen to have a collection method that reliably estimated human-vector contact as well as the necessity of a single method that could be used indoors, outdoors, and in night-time agricultural field work sites. HLCs involved the collection of mosquitoes on humans sitting with legs exposed during the collection hours. Each collector was provided with a flashlight, an aspirator to catch biting mosquitoes and one netting-topped polystyrene cup for each 1-hour catch-session. The mosquito holding cups were labelled with the name of the farm, shelter number, time-session and site of collection. These human collectors caught mosquitoes attempting to bite their exposed legs, and kept the mosquitoes sorted by hour of collection so that biting times could be determined. One collector sat indoors and the other one sat outdoors of the same shelter collecting the mosquitoes. The indoor and outdoor collectors exchanged site every hour to reduce biases due to differential attractiveness of the collectors to the mosquitoes. The collectors who conducted human landing collections were locally hired and trained. No personal information was collected about the collectors. The collectors were provided with prophylaxis (mefloquine [Lariam®]) to protect them from getting malaria or treated if they were diagnosed with malaria [13]. The collections were conducted over the course of 12 hours (18.00h – 06.00h).
Collections were made at the worker shelter camps in each farm which were inhabited by the workers.
Additionally, outdoor human landing collections were conducted in sites next to workers involved in night time work activities in the fields, such as the harvesting of sesame pods. One human landing collection was made in each farm between the months of September and December 2017, resulting in a total of 32 outdoor collections in the fields. Mosquitoes were killed and identified morphologically using appropriate identification keys [13]. They were then stored in 1.5ml Eppendorf tubes with silica gel before laboratory analysis.
Laboratory analysis of mosquitoes
Almost all of the mosquitoes morphologically identified as belonging to the Anopheles gambiae complex were identified using standard polymerase chain reaction (PCR) to determine the species [14]. In brief, genomic DNA was mixed with the following primers in a 25 μL reaction: AR (5′-AAGTGTCCTTCTCCATCCRA-3′; specific for An. arabiensis), AG (5′ CTGGTTTGGTCGGCACGTTT-3; specific for An. gambiae s.s.), QD-b (5′-AGTGTCCAATGTCTGTGAAG-3′; specific for Anopheles quadriannulatus species B or Anopheles amharicus) and UN (5′ GTGTGCCCCTTCCTCGATGT-3′; common for all species). Amplification reactions contained 1 μL of DNA, 1.5 mM MgCl2, 10 mM Tris–HCl (pH 8.4), 50 mM KCl, 0.1% Triton X-100, 200 μM of dNTPs (Amersham, Buckinghamshire, UK), 25 pmol of primers AR, AG, QD-b and UN and 0.25 U of SilverStar DNA polymerase (Eurogentec, Seraing, Belgium). Amplified PCR products were visualized on 2% agarose gels, stained with ethidium bromide. An An. arabiensis strain from the Sekoru colony, maintained at the Vector Biology and Control Research Unit, Tropical and Infectious Diseases Research Centre of Jimma University (Jimma, Ethiopia), was used as a positive control.
A subsample of mosquitoes was also analysed to determine whether sporozoites were present, using established methods [15]. Briefly, heads and thoraces of mosquitoes were separated from the abdomen and were homogenized. ELISA plates were coated with a capture monoclonal antibody. Following aspiration to remove un-adsorbed capture antibody, plates were incubated with blocking buffer to prevent non-specific binding in subsequent steps. The blocking buffer was removed by aspiration and the mosquito homogenate was added to the plates. After 2 hours, samples were aspirated and horseradish peroxidase-linked monoclonal antibody was added. This was then aspirated before the peroxidase substrate solution, ABTS, was added. Absorbance values at 405nm were obtained 30-60 minutes later using an ELISA plate reader. Positive reactions were those with an absorbance value of greater than two times the average absorbance values of negative control samples.
Behaviour-adjusted patterns of human exposure
To understand the risk to humans for infective mosquito bites indoors and outdoors, the biting times of An. arabiensis were compared with the outdoor and indoor times of humans as collected by Tadesse et al. [16]. The proportion of the population indoors each hour was multiplied by the number of mosquitoes collected indoors to estimate the numbers of bites that would have occurred in the absence of any personal protection measure such as LLINs. The same procedure was repeated for mosquitoes biting outdoors. The sum of all hours represented the number of An. arabiensis bites one person might expect to receive. To estimate the exposure that one person using a LLINs between the hours of 21:00 and 6:00 h would receive, the number of mosquito bites expected each hour was multiplied by 0.063, a figure used by Seyoum et al. [17] (derived from two studies in Tanzania) to estimate the number of bites that would be received, even when using a net.
Data analysis
Data were entered into Excel (Microsoft office 2007) for cleaning and data summary. Count data were analysed using generalized linear mixed effects models (glmer – function) [18] with R statistical software version 2.14.2 including the contributing packages MASS, lme4, glht, multcomp (alpha = 0.05) [19]. The biting time was included in the model as a fixed factor and collection round and site were included as random factors in the biting time analysis with Poisson distribution. The over-dispersion between data points that remained after adjustment for all other factors was adjusted by creating a random factor with a different level for each row of the data set.
The parameter estimates of the models were used to predict the mean counts or mean proportions and 95 % confidence intervals (CI) for the different size farms by removing the intercept from the models [19]. Multiple comparisons of treatments were also calculated based on the model parameter estimates.
The entomological inoculation rate was calculated as the product of the sporozoite rate and the mean number of An. arabiensis collected per person per night.