Study area
The study was conducted in four villages across Ulanga and Kilombero districts in south-eastern Tanzania (Figure 1). These included, Kivukoni (–8.2021, 36.6961) and Tulizamoyo (–8.3669, 36.7336) in Ulanga district, and Sululu (–7.9973, 36.8317) and Ikwambi (–7.9833, 36.8184) in Kilombero district. The area is within a low-lying river valley extending 250 km long and up to 65 km wide, interspersed with villages and farmlands. It has two rainy seasons, short rains between November and December and long rains between March and May, while between rainy seasons spans two dry seasons. Annual rainfall and temperatures vary from 1200mm to 1800mm, and 16°C to 32°C respectively [45]. Residents are mostly subsistence farmers, though some are also fishermen or owned small businesses.
During this study, typical house types in the villages were either thatch-roofed or metal-roofed (with corrugated iron sheets), and had either mud walls or brick walls, which were sometime plastered with concrete. Primary malaria vectors in this region are An. funestus and An. arabiensis, with An. funestus contributing more than 80% of current malaria transmission [44]. Culex pipiens are nuisance bitters contributing 79% of all indoor biting risk [46].
Figure 1. Map showing study villages and study households in both Kilombero and Ulanga districts, south-eastern Tanzania. Indoor-resting mosquitoes were collected multiple times from each household during the study period.
Selection and characterization of study houses
Field collection of resting mosquitoes was done inside human-occupied houses, ensuring to cover the main house types. Candidate houses were selected based on construction materials for walls (mud or bricks, with or without concrete plastering) and roofs (metal or thatch). This resulted in four classes of houses (Figure 2) commonly found in the study area, namely: i) houses with thatched roofs and mud walls, ii) houses with thatched roofs and brick walls (none of these houses had plastered walls), iii) houses with metal roofs and un-plastered brick walls, and iv) houses with metal roofs and plastered brick walls. Ceilings were uncommon and therefore excluded in this survey. All individual houses were also geo-referenced, then characterized by other attributes, namely: a) whether eave gaps were open or closed, b) number of rooms in the house, c) height of walls and d) maximum daily temperatures (°C), recorded using Tinytag® data loggers (Gemini, UK).
Prior to commencement of mosquito collections, 20 houses were purposively selected in each of the four villages upon consent by household heads. These included five houses per house type.
Figure 2. Typical house types in the study villages in rural south-eastern Tanzania. The pictures depict only outside views of the houses, and does not show actual concrete plastering of some brick walls. These four are used as representative of the different house types, but the actual sizes and shapes of individual houses was varied.
Collection of mosquitoes resting on different surfaces inside the houses
Potential mosquito resting places were identified to include: a) walls, b) roofs (underside of the roofs) and c) other surfaces such as floor, clothing, bed nets and other household items. The household items were varied but generally included furniture such as beds, tables, chairs, cupboards, wood blocks, other household items such as bicycles, and utensils, wash basins, water containers, clay pots and cooking pans. The clothing included hanging garments, curtains, sacks and bags. Actual mosquito collections were done using Prokopack aspirator [47], by trained technicians. Collections involved hovering the aspirator systematically over the surfaces and collecting all mosquitoes. Lighting was provided using hand-held flash lights. The sequence of collection between resting surfaces in each room was changed to minimize sampling biases. The collections were done for five days each week in each village, visiting 2–4 houses per day. Initially the collections were done between 6a.m. and 12p.m, from January 2019 to May 2019. Then from May to July 2019, the collections were done three times a day (in the morning (between 7:00a.m. and 8:30a.m.), evening (between 6:00p.m. and 8:00p.m.) and at night (between 12:00a.m. and 2:00a.m.)), to minimize variations associated with mosquitoes moving between different resting surfaces within the houses. Unlike the other collections done by trained technicians, the late evening and late-night collections were done by trained household members to avoid intrusion of their privacy.
In total, there were 277 house visits for indoor resting mosquito collections, including 76 visits to houses with thatched roofs and mud walls, 70 to houses with thatched roofs and brick walls, 70 to houses with corrugated iron roofs and un-plastered brick walls, and 61 visits to houses with corrugated iron roofs and plastered brick walls.
Morphological identification and processing of collected mosquitoes
Mosquitoes collected from each of the resting surfaces were placed in separate disposable cups and labelled appropriately. They were sorted by sex and taxa, then all Anopheles sorted and identified using the morphological keys [48]. Physiological status of each female Anopheles was determined as unfed, partly fed, fully fed, gravid or semi gravid. All records were kept by house, surface, house type and village.
Identification of sibling species of malaria vectors, blood meal analysis and detection of Plasmodium falciparum sporozoites in the mosquitoes
The field-collected mosquitoes were packed individually in 1.5 ml microcentrifuge tubes (BioPointe Scientific®) containing silica plugged with cotton wool. Sub-samples of An. funestus s.land An. gambiae s.l females were further analysed for sibling species, Plasmodium falciparum sporozoites and blood meal sources (if the mosquitoes were blood-fed). Sibling species identification for An. funestus s.l and An. gambiae s.l was done using PCR protocols originally developed by Koekemoer et al. [49] and Scott et al. [50] respectively. Blood meal analysis was done using ELISA tests [51], and parasite infections detected by screening for the P. falciparum circumsporozoite proteins in salivary glands of the adult females [52]. Heat-labile non P. falciparum were eliminated by boiling the ELISA lysates at 100°C for 10 minutes to remove false positives [53].
Determination of physiological ages of mosquitoes
Parity of mosquitoes was approximated following procedure described by Detinova [54] as a proxy of physiological age of mosquitoes. A subsample of non-blood fed, An. funestus and An. arabiensis, were first immobilized in a refrigerator. Under stereo microscope abdomens of anesthetized mosquitoes were dissected to extract ovaries. Ovaries were examined under compound microscope to determine whether mosquitoes had laid eggs or not.
Data analysis
Data analysis was done using open source statistical software, R version 3.6.0 [55]. Generalized linear mixed effects models (GLMM) were built using functions within the lme4 package [56] to assess: i) preferences of mosquitoes (An. funestus, An. arabiensis and Culex) for different resting surfaces and ii) relationships between various household risk factors and number of mosquitoes caught on different surfaces. Initially, the number of female mosquitoes of each species was modelled as a response variable against resting surfaces as a fixed factor. Since walls are typically the main target for insecticide spraying, they were used as reference against which other surfaces were compared.
To assess relationships between household risk factors and mosquitoes resting on different surfaces, the number of mosquitoes caught from each surface was modelled as function of: i) roof type, ii) wall type, iii) whether interior walls were plastered with cement or not, iv) eave gaps, v) number of rooms, vi) wall height and vii) daily maximum temperatures inside the houses.
In all models, households nested within villages and sampling days were used as random terms, to capture unexplained variations, and account for pseudo-replication. Poison distribution was used when fitting GLMM models, except when overdispersion was detected, in which cases, negative binomial distribution was used instead. The best fitting models were selected using Akaike Information Criterion (AIC) [57], and results presented as relative rate ratios (RR) at 95% confidence intervals. In addition, the dabestr package for estimation statistics [58], was used to depict effect sizes of differences in mean numbers (at 95% confidence intervals) of mosquitoes collected on different resting surfaces relative to walls.