Background
The feeding locations and the biting times of individual
Anopheles spp. could potentially confound assessments of their role in local malaria transmission [
1,
2]. There is evidence that in Kenya and elsewhere in Africa, primary vectors and other potentially important secondary malaria vectors do not feed exclusively within houses [
1,
3‐
14] and that significant levels of vector exophagy, feeding outdoors, can occur at times when the human population is still outdoors [
5,
7,
11‐
13,
15,
16]. Malaria eradication has recently returned to the global health agenda for the first time since the failure of the Global Malaria Eradication Programme (GMEP) of the 1950s and 1960s [
17‐
20]. The development of insecticide resistance, and the exophily and exophagy of
Anopheles species (resting and feeding outdoors) are thought to be among the key contributors to the failure of the original p rogramme [
21] which relied heavily on indoor residual spraying (IRS) with DDT. It has, therefore, been suggested that any future campaign to achieve eradication, still less elimination, may fail if the lessons learnt from the collapse of the GMEP are forgotten or ignored [
20,
22].
Today, vector malaria elimination plans are heavily reliant on the use of long-lasting insecticide treated nets (LLINs) and IRS, both of these being strategies that are theoretically less effective against the malaria vectors that are fully or partially exophilic or exophagic [
23]. Successful malaria control is threatened by the emergence of physiological, biochemical or behavioural adaptations within the vector population in response to the use of insecticide [
24,
25]. IRS and LLINs require direct contact between the mosquito and surfaces carrying sufficient levels of insecticide to kill or repel the vector. Pre-existing or adapted feeding and resting behaviour may reduce or negate this contact [
19].
The feeding behaviour and circadian rhythms of
Anopheles are genetically determined [
26,
27], with the former being linked with inversion polymorphisms [
26]. There is an added complication of intraspecies variation, where mosquitoes of the same species but different homokaryotypes react to identical environmental conditions in different ways [
26]. There has been some debate surrounding the importance of pre-existing exophilic and exophagic
Anopheles populations when planning control efforts [
1,
19,
28‐
30]. Whilst the occurrence and mechanisms of insecticide resistance over the last century have been well documented in African
Anopheles populations [
21,
25,
31], the extent to which the emergence of population-wide vector behavioural change in response to control methods, known as ‘behaviouristic resistance’, affects the use of nets and IRS remains unclear. This can only be established by observing vector population behaviour in the field and there is a lack of basic pre-intervention baseline studies [
12,
25,
31‐
34].
The time of feeding in both endophagic and exophagic populations may also be of critical importance if it occurs in the hours outside of LLIN use [
16,
28,
30,
35‐
38], particularly in areas where nets are the main control intervention used [
1].There have been reports of net and IRS use leading to a reduction in indoor biting or resting, and a shift to exophagic behaviour, earlier feeding times or feeding on different hosts [
10,
39‐
48]. In Kenya, a pronounced reduction in endophily was observed in the vectors
Anopheles gambiae sensu stricto (s.s.) and
Anopheles funestus sensu lato (s.l.) and a shift in host preference from humans to other mammals after 5 years of bed-net use [
44]. Similarly, host choice change in
An. funestus s.l. was observed by Githeko et al. following use of permethrin-impregnated eave-sisal curtains [
49]. In Benin,
An. funestus s.l. populations exhibited increased exophagy and a shift in feeding times after LLIN introduction and demonstrated a shift to diurnal feeding in a recent study in Senegal [
50,
51]. For these species complexes, this could be due to a change of the sibling species composition, rather than a behavioural change of a single species
per se, as some members demonstrate higher zoophagy and exophagy than others. This was demonstrated in Kenya where following mass net distribution the
An. gambiae s.s. population decreased and the remaining sibling species
Anopheles arabiensis, demonstrated higher exophagy and zoophagy [
52]. In Tanzania, substantial reduction in the indoor resting and a small increase in the exophagic behaviour of
An. gambiae s.s. was recorded after the introduction of pyrethroid-impregnated bed nets in one study village [
39]. It should also be noted, that these changes are not universal, a recent study in Kenya noted that late night vector feeding behaviour still persisted in areas 10 years after bed net distribution [
53].
Human behaviour may also influence the extent of human-vector contact. Entomological studies carried out in Zambia and Tanzania incorporated the proportion of the human population indoors but not asleep and those indoors and asleep under an LLIN, in order to calculate the protective efficacy of bed nets [
37,
38,
45]. The methodology of these studies provides a useful insight into the true protective efficacy of bed nets when both human and vector behaviours are combined but are partially limited, as they do not estimate the area-wide effects on the vector population that universal coverage of LLIN can offer [
54].
The World Health Organization recommends that adequate baseline information is collected in an area before residual insecticide is used [
55]. Without a good understanding of the baseline entomological situation, the emergence of true behavioural adaptations will be difficult to detect. This concern has led to a call for regular monitoring of vector feeding behaviour as control programmes are expanded [
37]. Regrettably, as noted by Smits et al., vector control is susceptible to a reduction in supervision and evaluation when activities have been in place for some time [
4]. Success is more likely if control efforts are designed to adapt to changing local conditions [
4]. Without a baseline vector dataset it is difficult to identify the emergence of behaviouristic resistance, and the accuracy of malaria transmission models used to plan future control efforts will be compromised [
56‐
58].
This study aimed to assess the behaviour of exophagic or partially exophagic malaria vectors in Rachuonyo South, western Kenyan highlands, over different seasons, and to assess the level of exposure to Anopheles bites that individuals experience when not protected by an LLIN. Using vector exposure calculations, the protective efficacy of nets was calculated for this population.
Discussion
In common with the previous work carried out in Zambia and Tanzania to determine the protective bed net efficacy, this study highlights the importance of integrating human behaviour into the assessment of human-vector contact in relation to malaria transmission [
16,
37,
38,
45]. Despite predominantly endophagic primary vectors in this region, the overall P* was low at 51% (95% CI 50–53%) and this may be explained by exposure occurring indoors at times of the evening before nets are used which equates to 31% of total mean daily exposure. This is substantially lower than the bed net efficacy using similar methods reported from rural Tanzania [
37], but higher than that reported from urban Tanzania where
An. arabiensis is predominantly exophagic [
45]. In the present study, 90–95% of vector exposure was calculated to occur within the house if LLINs were not used, which is similar to levels reported for
An. funestus s.l. in Zambia [
38] and the results of a study of matched surveys of human and mosquito behaviour from Burkina Faso, Tanzania, Zambia, and Kenya [
91]. The use of LLINs in the present study reduced an individual’s exposure from 1.3 bites per night to 0.47 bites per night. In agreement with a recent study carried out in Western Kenya the majority of exposure occurred indoors [
53], an estimated 65% of mean daily exposure occurred during sleeping hours, indicating that nets still may offer personal protection in an area of low transmission.
The two primary vector species An. funestus s.l. and An. arabiensis were both active inside and outside from 18:30 onwards, two-and-a-half hours before the mean time local residents reported going to bed. When studying mosquito activity outside times when individuals are likely to be asleep, the peak hours of biting varied between species, but universally very little activity occurred during the early evening (17:30–18:29) and morning (05:30–06:29). The latter may be due to the low dawn temperatures in this area, but the former may have been influenced by the heat and light intensity in the hours before dusk. During the times studied, An. funestus s.l. demonstrated a distinct bimodal pattern of indoor feeding activity, with the first increase in biting activity between 18:30 and 20:30 followed by a second at 21:30 and 22:29. Although there was no evidence that these periods differed in intensity (p < 0.05), they were both significantly higher than the preceding or interim hours (p < 0.05).
The residents of this area reported that 90% used nets, greater than that previously recorded in Kakamega in the western Kenyan highlands (56%) [
92], or by the Malaria Indicator Survey in 2010, 61% [
62]. However, the former survey was conducted in a different area with a different ethnic populations. Furthermore, the area of the current study was a research site where active health teams had been working for the past 2 years and data were collected during a year of mass LLIN distribution with prolonged marketing campaigns to increase awareness and adherence. Net use recorded in the present study may not reflect wider patterns of bed net use.
It is important to note that this study, in common with previous work [
16,
37,
38,
45], did not estimate the area-wide effects on the vector population that may result from universal coverage of LLIN [
54]. It has been shown that mass distribution will reduce transmission of principally endophagic vectors by reducing the reservoir of disease [
16]. The P* estimated here may be an underestimation as it does not include any potential community-wide effects.
Anopheles funestus s.l. was the most abundant primary vector species trapped in the area throughout the year with an indoor MBR of 0.15–1.2 and an outdoor MBR of 0.13-1.2 bites per person per night. Similar findings were reported from lowland areas in Nyanza Province [
93].
Anopheles funestus s.s. is considered the anthropophagic exception in a complex of zoophagic species [
94], so it is likely that the
An. funestus s.l. in this study contain other morphologically identical members of the complex. Work continues to genetically sequence the full set of anophelines caught to confirm species identities. Alternatively, it is possible that the LLIN and IRS use in this area has induced this species to seek alternative hosts. Such phenotypic, plastic feeding behaviour has been observed in
An. gambiae s.s., which can demonstrate zoophilic behaviour in field conditions if their preferred human hosts are not readily available [
95]. This shift from anthropophagy to zoophagy was noted in Kenyan
An. funestus s.l. populations in response to permethrin-impregnated eaves-sisal curtains [
42] but again no data were given as to the sibling species of the complex.
Anopheles arabiensis was also present in the study site, with a peak MBR of 0.12 bites per person per night. This is not consistent with either the historical distribution of this species or recent work carried out in the Nandi hills, where
An. gambiae s.s. females were more prolific than
An. arabiensis [
72,
96]. However, these findings do align with the observations of Ndenga et al. who surveyed larval breeding sites above 1,500 m in neighbouring Western province, where
An. arabiensis represented a third of the
An. gambiae s.l. larvae collected [
74].
Anopheles arabiensis is found at high densities in lowland Nyanza and it is therefore conceivable that this species has encroached upon the neighbouring highland fringe areas, filling the niche left by
An. gambiae s.s., which was selectively targeted by local control efforts [
41,
44,
52,
68]. It is possible that the distribution of
An. arabiensis may have always included highland areas, with this species being overlooked by those studies that predominantly used indoor traps that do not target outdoor-resting and feeding species [
74].
EIR estimates were higher than those previously reported for similar areas of western Kenya [
49,
63]. Ndenga et al. reported an EIR of 0.2–1.1 in highland areas of the neighbouring district Kisii Central and in Kakamega (neighbouring province) and Githeko et al. recorded a peak EIR of 12.8 from comparable elevations in Kakamega [
49,
63]. Those studies may have underestimated the EIR as they used pyrethrum spray catches, which will not trap endophagic and exophilic
Anopheles that are infected but exit the house early. Furthermore, in the current study, the site was specifically selected due to high
P. falciparum prevalence and incidence and high indoor-resting densities of anopheline mosquitoes. Within this area of higher transmission, only houses within quadrants that contained breeding sites were selected, and thus the EIR from the present study could be interpreted as that of a transmission ‘hotspot’ [
97]. In common with studies that used methods other than human landing catches (HLC) to estimate EIR [
98], the present study did not include an estimation of outdoor transmission and thus potentially overestimated the total exposure an individual will experience throughout the year. In addition to these limitations, it is also possible that the EIR may be overestimated. This study did not include steps to limit false-positive CSP-ELISA results by reanalysing the homogenate therefore it is possible that false-positives were included in the EIR estimate [
99].
Across all Anopheles species trapped, there was evidence (p < 0.05) that females carrying eggs were 4.5 times more likely to feed indoors, potentially presenting a higher transmission risk indoors as these mosquitoes are older than nulliparous females. However, unfed parous females without eggs are used as a proxy for older females and were more likely to bite outdoors (p < 0.05) and, conversely, younger nulliparous females were more likely to feed indoors (p < 0.05). Therefore, the number of gravid females caught in traps indoors may reflect the recruitment of the female indoor-resting population that are attracted to the CDC-light trap during egg development.
The findings of this study support the hypothesis that the levels of both LLIN and IRS coverage are currently not sufficient to interrupt transmission in this setting. IRS should be an effective control tool in a region where the majority of exposure occurs inside the house and should complement the use of LLINs if biting occurs before times of net use and/or the observed exophagy is also accompanied by indoor-resting behaviour. IRS was and is still implemented in Rachuonyo district but coverage at the time was not universal, with 38% of houses sprayed in the previous 12 months [
62]. Improving the coverage of the current IRS campaign may be more effective, but if conducted poorly it may also encourage the development of insecticide resistance. Therefore, as the majority of exposure is currently occurring indoors, measures to bar entry to
Anopheles may be a cost-effective option to complement existing interventions. These could include the use of ceilings, window and door screens, measures that have successfully reduced the number of
Anopheles indoors both historically and in experimental hut trials [
100,
101].
An important limitation of the present study is the use of light-traps outdoors. Light-traps have been in use since the early part of the 20th century, and have been used widely in a variety of transmission settings, including Africa [
56,
82,
102]. These traps work on the principle that the mosquito is drawn into the ‘dazzle zone’, at which point the fan mechanism sucks them into the trap [
78,
102]. The exact mechanics of this process and the extent to which it is species-specific are not well understood [
102,
103]. The type and size of catch may be influenced by a number of factors, including the species of mosquito [
78], the model of trap and the wavelength of the light used [
102] and whether the strength of illumination can be kept constant. Indeed, it is reasonable to assume that the traps used during the present study could not achieve a uniform level of illumination throughout the night.
Light-traps have several practical advantages: they are commercially available which aids standardisation [
104], they are easily accepted by communities within study sites [
105] and they have low running costs. A number of experiments have been carried out to establish whether light-trap catches correlate well with those from HLC and some studies have indicated that light-trap catches of
Anopheles have relatively high sporozoite rates [
103‐
105]. Other studies have reported no significant difference between sporozoite rates from light-traps and HLC, with a corresponding similarity in parity rates between these trapping methods [
106‐
108]. With a lack of standardisation between studies, there appears to be no definitive evidence to indicate whether light-traps, with or without human bait, can catch the anthropophagic vector population.
It has been claimed that CDC light-traps cannot be used outdoors [
109], yet this appears to be based on limited evidence. The small number of studies that assessed HLC with light-traps hung outside tended to place the light-traps directly under the eaves of houses [
110,
111], either with an accompanying light-trap inside the same house [
110,
112] or with no accompanying human bait [
110,
113]. Costantini et al. (1998) did hang CDC light-traps away from houses, under a thatched rain shelter with human bait, but found no correlation between its catch and that of HLC when comparing
An. gambiae s.l. However, when
An. funestus numbers were compared there was a density-dependent correlation between the catch of the outdoor HLC and the CDC light-trap [
114]. The authors concluded that outdoor traps were not effective but acknowledged that this was based on a limited data set [
114]. Overgaard et al. (2012) used a CDC light-trap with a UV bulb outdoors but with no human bait and reported a correlation between the numbers of
An. gambiae s.l. and
An. melas trapped by the two light-traps. The authors did, however, express some doubts about the practicality of using light-traps outdoors with such low numbers and such high levels of variability between catches [
110]. Currently, there is insufficient evidence to definitively dismiss the use of light-traps outdoors as a means of collecting anthropophagic
Anopheles. Where HLC is not available, light-traps remain one of the few viable trapping methodologies not designed solely to catch the resting
Anopheles population, and may represent a useful tool to catch the vector population.
The present study contributes to the knowledge of both primary and secondary vector species dynamics in the fringe area of the western Kenyan highlands. The existence of predominantly exophagic potential secondary vector species such as An. coustani and An. demeilloni should be an important consideration when planning future control efforts, as they are likely to be overlooked during campaigns targeted at the primary vector species that feed indoors during sleeping hours. These species have the potential to maintain low levels of transmission in this area. It is therefore vital that entomological surveillance should be carried out on a regular basis in this area and in other regions of unstable malaria transmission targeted for malaria control or future malaria elimination.