Background
Efforts to control malaria rely heavily on the application of long-lasting insecticidal nets (LLINs) which are the major strategy to protect humans against bites from mosquito vectors in African homes [
1]. Rapid increases in the coverage of LLINs over the past decade have been associated with substantial declines in major African vector species [
2]. A parallel decline in malaria infection rates in people has been reported in several places, as has a decrease in malaria mortality in infants and adults [
3]. However, the widespread use of these vector control measures may be triggering changes in the ecology and genetics of mosquito populations that could threaten their continued effectiveness [
4‐
7].
Insecticide resistance is increasingly reported in areas where LLINs are widely used [
8‐
11]. There are also concerns that LLINs may be selecting for behavioural changes within malaria vectors that allow them to shift their biting to times and places where people are not protected, which can be defined as ‘behavioural avoidance’ [
7,
12‐
16]. These changes in feeding behaviours could arise either due to shifts in malaria vector species composition from dominance by highly endophilic and anthropophilic species (e.g.,
An. gambiae
s.s.) towards those with more exophilic and zoophilic tendencies such as
Anopheles arabiensis [
17]. Additionally, it has been hypothesized that selection from LLINs could generate within-species behavioural adaptations [
13,
15,
18,
19]. The ability to monitor if and how rapidly mosquito behaviour is changing in response to control measures is crucial for assessment of the continued effectiveness of LLINs and indoor residual spraying (IRS) strategies [
20‐
23].
One of the most important and widely used techniques to study the host-seeking behaviour of mosquitoes is the human landing catch (HLC) technique. This technique is regarded as the gold standard tool for sampling host-seeking malaria vectors [
24,
25]. The HLC is widely used for a range of purposes, including estimation of entomological exposure rates [
26‐
29], evaluation of vector control measures [
30,
31] and for studying mosquito vector behaviour and ecology [
16,
26‐
28,
32,
33]. Although the HLC provides a realistic estimate of the number of mosquito bites that humans are exposed to, this technique has numerous drawbacks. The most notable is ethical concerns raised by requiring the participating human subjects to expose their legs to attract mosquitoes. The aim is for participants to capture mosquitoes landing on them before they bite, but this is not always possible and could generate some risk of exposure to infection [
26,
34,
35]. To minimize exposure risk it is recommended that HLC participants use malaria prophylaxis [
36]. Whilst it has been shown this precautionary measure can reduce infection risk in HLC participants to below that experienced in the community in some settings [
36], it is unlikely to be effective in areas of high drug resistance, and/or where mosquitoes carry other pathogens (e.g., dengue, filariasis) that pose infection risks [
37]. These problems highlight the need for a more efficient, representative and ethical alternative sampling method for investigation of mosquito biting densities and behaviour.
Previous attempts have been made to develop exposure-free sampling tools for collecting indoor or outdoor biting mosquitoes. These techniques include but are not limited to the bed net trap [
38], tent traps [
39‐
41], the CDC light traps [
42], and the mosquito magnet (MM) trap [
43‐
45]. While these methods have shown promise in some settings, most have limitations that restrict their large-scale application, and/or bias collection towards mosquito species with particular phenotypes that may misrepresent the community of mosquitoes attracted to people [
46]. Recently, there has been renewed interest in exploring the use of electrocuting surfaces as a means of sampling malaria vectors [
47‐
49]. This approach was originally developed for trapping tsetse flies outdoors [
50], but later adapted to sample mosquitoes drawn towards a host odour source [
51,
52]. This trap works by placing a live host in a sealed tent and piping their odour out to an electrocuting net (E-Net) approximately 10 m away that kills mosquitoes on contact. Such E-Nets have already shown promise when used to investigate host species’ preferences and odour responses of the African vector species [
51,
52]. As a potential improvement, the use of commercially available ‘bug-zapping’ devices, which can sample insects in the immediate proximity of a host has been explored with some promise, indicating they can achieve a relative sampling efficiency of up to 50 % of the HLC in one study [
49]. However, given these devices were developed for large flies, their suitability for trapping African malaria vectors is unclear. Further work is required to develop an electrocuting trap that is optimized for malaria vectors, can meet the performance of the HLC, is suitable for use inside and outdoors, can be used safely in close proximity of humans, and is durable under field conditions. Here, a mosquito electrocuting trap (MET) was designed, developed and field-tested. This trap was custom designed to sample host-seeking African malaria vectors, with the aim of meeting all performance targets defined above.
Discussion
In this study, the potential of two electrocuting traps, the MET and CA-EG, to provide exposure-free alternatives to the HLC technique for sampling African malaria vectors was evaluated. The HLC generally collected more An. gambiae s.l. than the MET, but capture rates of An. funestus s.l. were similar between these methods. The relative sampling efficiency of the MET was reasonably high (~59 %) when used for An. gambiae s.l. outdoors, but fell to ~20 % relative to the HLC when applied indoors. In contrast, the CA-EG performed poorly relative to the HLC in both indoor and outdoor settings, for An. gambiae s.l. and An. funestus. No evidence of density-dependent sampling was observed in either electrocuting trap. Both the MET and CA-EG tended to have higher performance relative to the HLC outdoors compared to indoors, which contributed to these traps producing somewhat biased estimates of human exposure indices. While estimation of the proportion of mosquitoes caught indoors (Pi) by the electrocuting traps were similar to those estimated by HLC, there was tendency of the MET and CA-EG to underestimate (Pfl) when sampling An. gambiae
s.l., and overestimate the proportion of human exposure that occurs indoors (πi) when sampling An. funestus
s.l. On balance, the sampling sensitivity of the CA-EG was judged too low to merit further consideration as an alternative to the HLC. However, the MET showed strong promise as an alternative method for exposure-free surveillance of African malaria outdoors outside of houses.
The sampling efficiency of the MET was consistently higher for
An. funestus s.l. than for
An. gambiae s.l. Possible explanations for this include differential sensitivity of these species to electrocution. Several biological factors are known to influence the electrical conductivity of insects, including their cuticular hydrocarbon composition [
72], body size and water content. Differential electrical conductivity between mosquito species could be expected to be less consequential at higher voltages say 50,000 V as used in [
51] because this voltage would be used with lower currents. The voltage and current combination used in the MET were optimized in laboratory studies to produce a high instant kill rate (>80 %) using
An. gambiae
s.s. as a model, but may be more efficient at killing
An. funestus
s.l. A previous study using the CA-EG found that sampling efficiency varied between
An. gambiae s.s. and
An. arabiensis [
49], thus vector-specific sampling may be a common feature of electrocuting traps as has been documented with other methods, such as CDC light traps [
73].
Both electrocuting traps had higher sampling efficiency when used outside than indoors. The reasons underlying this are unknown but could be due to microclimatic variation [
74] which could modify the functioning of electrocuting traps in outdoor and indoor settings, and/or differences in how vectors host seek in outdoor versus indoor location. For example, factors such as the direction and concentration of host odours and wind movement vary between indoor and outdoor settings [
74], and could lead to differential attractiveness of the traps when used in different places. Humans conducting HLC usually bend to collect mosquitoes landing on their legs as shown in Fig.
1d, blowing carbon dioxide to the legs therefore attracting more mosquitoes when doing HLC compared to MET and CA-EG in which carbon dioxide is blown away (Fig.
1b, c). This phenomenon is expected more pronounced indoors than outdoors where wind may blow away the carbon dioxide and may therefore explain a poorer performance of MET and CA-EG indoors relative to the HLC. HLC may not therefore be a perfect indicator of mosquito-biting activities as the stated phenomenon above may bias its function. Further investigation of the performance of electrocuting traps in a broader range of ecological settings is required, including experiments that involve mechanisms to control the breath of the catchers sitting on the HLC as well as on the MET, perhaps by using a breathing tube which directs the carbon dioxide away from traps or towards the traps to increase sensitivity of both methodologies.
There were differences in the relative sampling sensitivity of CA-EG as estimated in this study compared to that reported by Majambere et al. [
49]. Whereas [
49] estimated the sampling efficiency of the CA-EG to be ~50 % relative to the HLC in indoor and outdoor locations, it was only 6–29 % in this study. One explanation could be variation in how human participants were positioned. In [
49] the human bait lay down and were covered by bed nets which were surrounded by six grid units, in this current study the humans were positioned in a sitting position using four grid units, specifically to replicate the human subject’s position in the HLC technique and thus avoid bias due to differential positioning of the hosts. Enclosing the whole human in the trap as was done by [
49] may have contributed to their higher reported performance of the CA-EG in their study compared to this study. Another difference was that the study by [
49] was conducted in Dar es Salaam where
An. gambiae
s.s. is the dominant species, compared to
An. arabiensis in the Kilombero Valley where this study was set. During preliminary laboratory optimization tests conducted during the development of the MET,
An. gambiae
s.s. was shown to be somewhat more sensitive to electrocution than
An. arabiensis. Thus, the lower performance of the CA-EG in the current study may also be due to differences in malaria vector species composition between sites.
One of the ways to make MET smaller and therefore easy to carry around would be to replace the human bait with an artificial odour delivery system. This step would additionally remove human safety concerns and significantly decrease labour. However, to be able to obtain an alternative trapping tool with sampling efficiency close to the gold standard HLC, this study tried to imitate as much as possible some of the features which make HLC superior to other host-seeking traps. Theoretically, a good host-seeking trap should represent as much as possible human exposure rates to host-seeking mosquitoes that happen in real environment. This can be most realistically achieved with the physical presence of a human close to or within the trap. Therefore, replacing the human bait from the MET would reduce accuracy of the trap because other factors than the human odour, such as visual cues and body heat, are involved in attracting host-seeking mosquitoes [
75,
76].
On a few occasions there was evidence of decreasing sensitivity of MET and CA-EG over the sampling night relative to the HLC, but this effect was not consistent between vector species, nor between indoor and outdoor settings. A reduction in the sampling efficiency of CA-EG relative to the HLC over the course of a night was reported in [
49]. This was interpreted as a sign of battery drainage through time, which reduced the electrical output. Given that a decline in the relative sensitivity of electrocuting traps was not consistently reported in this study, it is difficult to interpret the patterns of time-dependent trap performance observed here. In addition to battery drainage, other factors, such as a build-up of moisture on traps (especially as occurred in outdoor stations) may have contributed to MET’s low relative performance. The MET output voltage was checked every hour and in some cases it was shown to drop below optimal levels, especially in the later hours of the night. Additionally, there were a few occasions where traps temporarily short-circuited during experiments because opposing wires came into contact, and/or the wooden frames became moist and mildly conductive. Experiments were stopped when there was an obvious cessation of current flow, however, there could have been more minor dips occurring during sampling night that went undetected. Use of a higher-capacity battery coupled to an alarm system to notify if and when there is any dip in electrical output could resolve any issues of variable voltage output through time.
This study shows no strong evidence of density-dependent sampling in either the MET or the CA-EG. However, this study was conducted over 21 consecutive nights in the rainy season when mosquito densities were generally high. Thus, it was not possible to assess density dependence across the full range of mosquito densities that occurs between wet and dry seasons. Additionally, it is noted that the detection of density dependence in trapping studies is sensitive to the type of analysis method used [
77]. Several previous studies have assessed density dependency based on analysis of how the proportional catch rate varies with differing mosquito densities across nights [
39,
78], whereas others, including the present study, use the Bland–Altman method [
67]. The Bland–Altman method was chosen because its use of regression analysis to assess the reliability of two measures is not subject to bias inherent in the binomial, proportional catch approach. It is recommended that future studies to evaluate these trapping methods adopt a similar method so that estimations of density dependence are standardized and comparable.
For any mosquito-sampling tool to successfully replace the HLC, it must be able to give meaningful representation of key mosquito behaviours and associated human exposure risk factors. Here, three such measures were investigated that have been widely used in a number of other studies to assess both human risk and likely degree of protection from LLINs [
15,
68‐
70,
79]. One of the most direct measures of indoor exposure is the proportion of mosquitoes that bite indoors (P
i), for which comparable estimates were obtained from both electrocuting traps and the HLC. However, the proportion of mosquitoes caught when people are usually indoors (P
fl) was underestimated by the CA-EG and MET for
An. gambiae s.l., but similar to the HLC for
An. funestus. Estimates of the proportion of human exposure occurring indoors (π
i) obtained from the CA-EG and MET were similar to the HLC for
An. gambiae s.l., but overestimated for
An. funestus. This is consistent with results from a previous study [
49] where the CA-EG produced a similar estimate of P
i, but underestimated P
fl for
An. arabiensis relative to the HLC. The likely explanation for this bias is the differential sampling efficiency of the electrocuting traps when used indoors versus out. This location-dependent performance would be expected to generate biased estimates of P
fl and the proportion of human exposure predicted to occur indoors.
Historical data for the Kilombero Valley (1999) where this study took place indicates the proportion of
An. gambiae
s.l. caught indoors (P
i) estimated by HLC was 0.58 ± 0.01 [
15], which is higher than the values of 0.37 ± 0.03 (HLC) and 0.35 ± 0.03 (MET) reported here. These differences may be due to concurrent changes within the
An. gambiae
s.l. complex that have occurred over this time. Whereas most
An. gambiae s.l. were found to be
An. gambiae s.s. in 1999 [
54] to <1 % in 2009 [
15], this species represents <1 % of the
An. gambiae s.l. complex now with the remaining fraction being the more exophilic
An. arabiensis. The proportion of human exposure occurring indoors that would otherwise be directly preventable with bed net use (π
i) was estimated as 0.43 and 0.55 for
An. gambiae
s.l. and
An. funestus
s.l., respectively, using HLC in this study. Assuming that all
An. gambiae
s.l. in this study were
An. arabiensis (based on PCR results of 400 samples which showed all of them were
An. arabiensis), these estimates of π
i are low compared to that reported in western Kenya [
79], where values of 0.87 and 0.86 were obtained for
An. arabiensis and
An. funestus
s.l., respectively. A more recent study in western Kenya [
80] reported π
i values of ~0.64 for major vectors
An. gambiae
s.l. and
An. funestus, which are still higher but closer to the values reported in this setting. Another study in Dar es Salaam estimated π
i obtained for
An. arabiensis to be 0.53 [
49] which is also higher than found in this study. The consistently smaller values of π
i reported for both
An. gambiae
s.l. and
An. funestus s.l. here indicates that a lower proportion of human exposure to malaria may be occurring indoors in the Kilombero Valley than in other parts of East Africa, and highlights the particular need for interventions that can control outdoor-biting mosquitoes in this setting.
As the MET applies high voltage electricity to electrocute mosquitoes, human safety in using this trap is a priority. Two measures were taken to ensure no risk of harm to humans using these traps. First, although the MET used relatively high pulsed DC voltage (600 V DC), resistors were incorporated to limit the current to no more than 10 mA which generates a low power output insufficient to cause harm to a human who momentarily touches them [
81]. Similarly, although the CA-EG used higher voltages (800 V AC), resistors were used to limit current flow in this trap to 15 mA. A second measure can be incorporated into future versions of MET to remove even this mild risk of minor electrical sensation on contact by placing a protective barrier of non-conductive material in the inner side of the grids.