Development and evaluation of mosquito-electrocuting traps as alternatives to the human landing catch technique for sampling host-seeking malaria vectors

Field experiments were conducted at Lupiro village (−8.38 S, 36.67 E) located in the Kilombero Valley of southern Tanzania. This village is situated in a malaria-endemic region where the most recent estimate of entomological inoculation rate (EIR) was 33.9 infectious bites per person per year [53]. Historically An. gambiae s.s. was the most abundant member of the An. gambiae s.l. species complex in the Kilombero Valley [54]. However in conjunction with the increasing coverage of insecticide-treated bed nets (ITNs) in this area over the last decade [55, 56], An. gambiae s.s. has almost virtually disappeared and its sibling species An. arabiensis now constitutes >98 % of the species complex in most areas [57‐59]. Anopheles funestus is the only other important vector species in the area [60].

Three different trapping methods were used in this experiment: the HLC, a MET developed in collaboration between the Ifakara Health Institute (IHI) and the University of Glasgow, and a commercially available ‘bug zapper’ device (PlusZap™ model ZE107 PZ40W [61]; defined as the CA-EG in this study) which is sold for domestic electrocution of insects [49]. The MET consists of four 30 × 30 cm panels connected together to make a square trapping box. On each panel, a mesh grid was made by placing ~1 mm thick (stainless steel) wires parallel to one another, at a spacing of 5 mm. Adjacent wires were supplied with opposite electric charge (positive next to negative) from a common positive or negative electric terminal. Wires were fixed into a wooden frame, with the four wooden frames being attached together to make the trap (Fig. 1a). The space between adjacent wires was set at 5 mm because this was deemed to be sufficiently small to prevent a mosquito from flying through, but ensure a mosquito made contact with both a negative and positive grid resulting in electrocution. Anopheles gambiae s.l. wing size, which was used to decide the spacing between the grids, was estimated from [62]. To collect mosquitoes using a MET, a person sits with their lower legs positioned inside the trapping box (Fig. 1b). This set-up imitates the HLC with the intention of making the MET as closely efficient to the HLC as possible. The CA-EG trap, with one of its four panels having surface dimensions of 68 × 24 cm (Fig. 1c), electrocutes flying insects on contact with the electrified surface, which is made up of grid wires placed about 8 mm apart. The adjacent wires of the electrocuting grid are connected to positive and negative terminals at ~800 V alternating current (AC). Four panels of the CA-EG were joined to form a square structure (Fig. 1c) into which a human sitting on a chair placed their legs. Mosquitoes which were attracted to bite the person on the legs were electrocuted on contact with the electrified grids. In both MET and CA-EG, the remaining part of the catchers were not surrounded by the electric grid but were protected by netting.

This experiment was conducted using a series of experimental huts designed to imitate the typical design of local houses in the study area [63]. Experimental huts dimensions were 6.5 m long × 3.5 m wide × 2 m high, with a 20-cm wide gap between the top of walls and roof to simulate the open eaves found in most local houses. Trapping stations were set up inside each hut and at an associated outdoor point approximately 10 m away. For outdoor stations, tents were used to provide roofing that protected the traps and collectors from rain (Fig. 1). A 3 × 3 Latin square design was used in which each of the three trapping methods was randomly assigned to one of three experimental huts on each night. On each night, collections were conducted at paired indoor and outdoor trapping stations. Over consecutive nights, the three trapping methods were trialled at each hut to complete a full rotation in 3 days. Seven rounds of trapping were conducted over 21 trapping nights between March and May 2012. The first two rounds were conducted in a group of three experimental huts defined as A, B and C (at site 1), and the remaining five rounds were conducted in a different group of experimental huts (defined D, E and F) which were situated approximately 200 m from the first.

Trapping was conducted from 19.00 to 07.00 hours. During each hour, volunteers spent 45 min passively sitting in a trap (MET or CA-EG) or actively collecting mosquitoes (HLC, Fig. 1d), with the remaining 15 min used as a break. Collectors moved to a different trap or position every hour throughout the night to minimize bias due to variation in their relative attractiveness to mosquitoes. At the end of each hour, MET and CA-EG traps were checked and trapped mosquitoes were removed by mouth aspirators or forceps and placed in labelled cups. On the following morning, mosquitoes from all the three trapping methods were sorted using morphological keys to identify their genera and gender. Female mosquitoes visually identified as belonging to a malaria vector group (An. gambiae s.l. or An. funestus s.l.) were individually stored in Eppendorf tubes with silica gel. Anopheles gambiae s.l. were later analysed using the polymerase chain reaction (PCR) technique to identify their species identity [64].

All statistical analyses were carried out using the R statistical software version 2.15. Generalized linear mixed models (GLMMs) [65] were used to assess variation in mosquito vector abundance between trap types. Mosquito abundance data were highly overdispersed and thus modelled as following a negative binomial distribution using the generalized linear mixed model automatic differentiation model builder (glmmADMB) package [66]. Here, trap type was fitted as the primary main effect of interest, and experimental night and hut as random effects. The relative sampling efficiency of the novel trap types relative to the HLC was estimated by computing the ratio of the predicted nightly abundance of vectors from these statistical models.

To test whether there was any systematic increase or decrease in the sampling efficiency of the CA-EG or MET relative to the HLC over the course of a night (e.g., perhaps due to battery decline), a model was constructed in which the fraction of the hourly catch occurring either in CA-EG or MET (e.g., ‘novel trap’/(‘novel trap + HLC’) was defined as the response variable, and trapping hour (defined as being ‘1’ on the first hour, and increasing through the night to 12 as the last hour) fitted as a continuous fixed effect, with experimental night added as a random effect. Here, the proportion of mosquitoes caught by a novel method (CA-EG or MET) out of the total caught from this method and the HLC, was modelled using a GLMM using a logit link function.

Additional analysis was conducted to test whether the relative performance of the novel trap types was density dependent. Density dependence was investigated using the Bland–Altman method which assesses the reliability of two measures via regression analysis of the relationship between their difference and their mean [67], where non-linearity in this relationship indicates density dependence (see Additional file 1). Values of R _adj ² obtained from these analyses can be interpreted as an estimate of the proportion of deviation from perfect linear correlation due to density dependence rather than random error (with a high value indicating support for density dependence). The precision of the R _adj ² estimate was gauged by estimating its 95 % confidence interval as the 2.5th and 97.5th centiles from 10,000 bootstrap replicates.

Finally, analyses were conducted to assess if the three focal trapping methods varied in their prediction of key mosquito vector behaviours and their related human exposure outcomes [68]. The predictors of malaria vector type of behaviour that were analysed here are the proportion of mosquitoes that were caught feeding indoors (P_i), the proportion of mosquitoes that were caught feeding when most people were indoors (P_fl), and the proportion of human exposure that occurs indoors (π_i) [49, 68‐70]. The proportion of mosquitoes that were caught indoors (P_i) was calculated by dividing the total number of mosquitoes caught indoors by the total number caught outdoors and indoors over 12 h of the night: I_19→07 h/(I_19→07 h + O_19→07 h) [70], where I and O, respectively, represent mosquitoes collected indoors and outdoors, and subscripts indicate the start and the end time of the sampling period. The calculation of P_fl and π_i requires definition of the period of the night when most people (>50 %) are expected to be indoors and sleeping. This time period was previously estimated for the community living in Lupiro village as 21.00–05.00 [71]. Therefore, the proportion of mosquitoes caught when most people were likely to be indoors (P_fl) was calculated as follows: (I_21→05 h + O_21→05 h)/(I_19→06 h + O_19→06 h) [70]. The proportion of human exposure that occurs indoors (πi) was calculated by dividing the number of mosquitoes caught indoors during the period that most people are inside (21.00–05.00) by itself plus the number of mosquito caught outdoors outside of the sleeping hours (I_21→05 h)/(I_21→05 h + O_19,20,06 h) [70]. Binary estimates of P_i, P_fl and π_i were estimated using GLMMs with a binomial distribution and a logit link function [65]. In these models, trap type was fitted as a fixed effect, and experimental night as a random effect.

Ethical approval was obtained from the Institutional Review Board of the Ifakara Health Institute (Reference number IHI/IRB/A.50) and the Medical Research Coordination Committee of the National Institute for Medical Research, Tanzania (Reference number NIMR/HQ/R.8a/Vol. IX/801.) All volunteers recruited in this work were given informed consent forms with details of the procedures, explanation of their right to withdraw at any time, potential risks, and mitigation plan. All participants read and signed the forms before taking part in the work. All participants were provided with malaria prophylaxis, Malarone (250 mg atovaquone and 100 mg proguanil hydrochloride, GlaxoSmithKline) before and during the experiments to prevent malaria infection.

Approximately 3.5 times more An. gambiae s.l. (N = 5559) were collected than An. funestus s.l. (N = 1543, Table 1), with more of both species being sampled outdoors than indoors (Fig. 2). The sampling sensitivity of traps varied between indoor and outdoor environments for both An. gambiae s.l. (trap × location interaction: χ ₂ ²  = 253.4, p < 0.001) and An. funestus s.l. (trap × location: χ ₂ ²  = 9.0, p = 0.003). Regardless of location (indoor vs out), the HLC consistently sampled significantly more An. gambiae s.l. than either the MET (outdoors: z = 4.10, p < 0.001; indoors: z = 7.89, p < 0.001) or the CA-EG (outdoors: z = 16.00, p < 0.001, indoors: z = 11.99, p < 0.001, Fig. 3a, b). There was significant variation between the electrocuting traps, with the MET catching significantly more An. gambiae s.l. than the CA-EG both indoors (z = 4.89, p < 0.001, Fig. 3a) and outdoors (z = 12.4, p < 0.001, Fig. 3b). Based on these results, the sampling efficiency of the MET relative to HLC for An. gambiae s.l. was estimated to be 59 % outdoors, and 21 % indoors (Table 2). The sampling efficiency of the CA-EG achieved <10 % of the HLC indoors and out (Table 2). The number of An. funestus s.l. caught per night by the HLC and MET was not significantly different either when used indoors (z = 1.71, p = 0.09, Fig. 3c) or outdoors (z = 0.58, p = 0.56, Fig. 3d). In contrast, the CA-EG caught significantly fewer An. funestus s.l. than either the HLC or MET (p < 0.001 for indoors and outdoors, Fig. 3c, d). Overall, the CA-EG had a sampling efficiency of <30 % for An. funestus s.l. relative to both HLC and MET (Table 2).

The sampling efficiency of the MET relative to the HLC remained constant across the hours of the night when used for An. gambiae s.l. indoors (χ ₁ ²  = 0.001, p = 0.98), however there was evidence of a moderate decline through time when used outdoors (χ ₁ ²  = 52.11, p < 0.001, Fig. 4a). This trend was reversed for An. funestus s.l. where the sampling efficiency of the MET relative to the HLC was observed to decline somewhat over the sampling night indoors (χ ₁ ²  = 12.42, p < 0.001, Fig. 3b), but remained stable outdoors (χ ₁ ²  = 0.76, p = 0.38). The sampling efficiency of the CA-EG relative to the HLC showed some increase across the night when used to sample An. gambiae s.l. indoors (χ ₁ ²  = 10.36, p = 0.001, Fig. 4c), but declined outdoors (χ ₁ ²  = 17.42, p < 0.001, Fig. 4c). The sampling efficiency of the CA-EG relative to the HLC for An. funestus s.l. was constant across the night both indoors (χ ₁ ²  = 0.39, p = 0.54, Fig. 4d) and outdoors (χ ₁ ²  = 2.31, p = 0.13, Fig. 4d).

In general, there was a positive association between the number of malaria vectors caught per night in the MET and the HLC, although the Pearson linear correlation coefficients were not statistically significant in all cases (Additional file 2: Table S2). A similar pattern of positive, but not always statistically significant, correlations between CA-EG and HLC catches was observed (Additional file 2: Table S2). Nightly catches were log(x + 1) transformed and plotted for further investigation of potential density dependence as evidenced by deviation from linearity. In all cases, there was much stronger support for a linear relationship between the log-transformed values of nightly catches than a curvilinear association (Fig. 5). All of the estimates of the strength of density dependency (adjusted R ²) were close to zero, but often with a wide confidence interval ranging from below zero to above 40 % in some cases (Table 3), suggesting that power to detect low-to-moderate levels of density dependence was limited. However, based on the range of mosquito densities encountered in this trial, there is no evidence to indicate the relative performance of the CA-EG or MET is density dependent when used indoors or outside.

Mosquito hourly biting activity was quite variable between nights, and revealed no obvious peaks in biting times for either An. gambiae s.l. or An. funestus s.l. (Fig. 6). All traps indicated that An. gambiae s.l. was significantly exophilic (>60 % of bites taking outdoors), while An. funestus s.l. was estimated to bite indoors and outdoors at similar rates (~50:50 split between indoor and outdoor biting, Table 4). Estimates of the proportion of An. gambiae s.l. that feed indoors (P_i) obtained from the HLC and MET were similar (z = −1.15, p = 0.25, Table 4), as were those obtained from the HLC and CA-EG (z = −1.77, p = 0.08, Table 4). Estimates of the proportion of indoor biting in An. funestus s.l. were also similar between the HLC and MET (z = 1.65, p = 0.10, Table 4), and the HLC and CA-EG (z = 1.95, p = 0.051, Table 4).

The electrocuting traps were less consistent with the HLC when used to estimate other human exposure indicators. The HLC predicted that approximately 98 % of An. gambiae s.l. attempted to feed during hours when most people would be indoors (P_fl, Table 4), which was underestimated at 68 % (z = 9.27, p < 0.001, Table 4) and 40 % (z = −12.91, p < 0.001, Table 4) by the MET and CA-EG, respectively. Predictions were less variable for An. funestus s.l., where P_fl was estimated to be ~70–75 % by the MET and HLC, respectively (z = −1.34, p = 0.18, Table 4), but underestimated as 65 % by the CA-EG (z = −2.62, p = 0.009, Table 4). It is noted that values of P_fl were underestimated in all scenarios where the novel trap type (CA-EG or MET) had a lower sampling sensitivity inside than outside. The MET somewhat underestimated the proportion of human exposure occurring indoors (π_i = 36 %) in comparison to the HLC for An. gambiae s.l. (43 %), a difference of borderline statistical significance (z = −2.04, p = 0.04). Estimates of π_i for An. gambiae s.l. as obtained from the CA-EG and HLC were indistinguishable (43–46 %, z = 0.77, p = 0.44, Table 4). Both the MET (z = 4.21, p < 0.001) and CA-EG (z = 5.23, p < 0.001) overestimated π_i for An. funestus s.l. (73–80 % Table 4) compared to the HLC (55 %, Table 4).