Background
The increasing use of intradomiciliary-based control measures such as long-lasting insecticide-treated nets (LLINs) and indoor residual spraying (IRS) has shown substantial success in reducing malaria transmission in sub-Saharan Africa [
1-
3]. The success of LLINs and IRS is mainly due to their effective targeting of indoor-biting, highly anthropophilic vectors such as
Anopheles gambiae s.s. [
1,
4,
5]. However these methods are less effective at controlling vectors that bite at dusk, rest outside the home (exophilic) and feed on livestock (e.g., zoophagic) as well as humans [
6-
11]. Recently, the abundance of highly anthropophilic, endophilic vector species such as
An. gambiae s.s. has declined relative to more behaviourally plastic species such as
Anopheles arabiensis in areas of high LLIN coverage [
7,
12-
14]. Unlike
An. gambiae s.s.,
An. arabiensis will readily feed and rest outside as well as inside houses, and feed on cattle [
15-
21]. At present, few outdoor-based control measures exist to effectively target this and other vector species with exophagic behaviour. Several potential methods for controlling outdoor-biting mosquitoes are under development (e.g., outdoor-based, vector-killing stations [
22,
23], biological control [
24] and use of insecticide-treated livestock [
25-
27]), but at present there is no standard method under routine operational use. The successful implementation of all these methods would benefit from clear understanding of the ecology and behaviour of vectors outside of domestic environments [
28].
The potential use of alternative host species to divert malaria vectors away from people has long been recognized as a potential environmental strategy for the reduction of malaria transmission [
29]. This strategy, known as zooprophylaxis, is credited with playing a major role in the elimination of malaria from Europe and other temperate areas following an increase in livestock keeping [
30]. However, increasing the availability of alternative hosts such as livestock could alternatively enhance human malaria exposure (e.g. “zoopotentiation”) if the heat and odour cues emitted by animals attract a greater number of vectors to households in or near where they are kept [
31]. Also zoopotentiation could occur if the physical disturbances created by animals (e.g., puddles, hoof prints, watering sites) increases larval habitat [
32] and thus adult vector density near households. There have been relatively few investigations of the impacts of household cattle ownership on malaria exposure rates in Africa, and their results have been mixed. Whereas some studies have reported an association between livestock keeping and reduced mosquito biting rates and malaria risk [
31,
33,
34], others have found no effect [
21,
35]. In the latter case, the study was conducted in a setting where the dominant vector species was highly anthropophilic and endophilic (
An. gambiae s.s.) [
21,
35]. This may account for the absence of any zooprophylactic effect in contrast to settings where
An. arabiensis is prevalent [
33,
34]. Further investigation of zooprophylaxis within rapidly transmission settings dominated by zoophilic, exophilic vectors is thus needed to fully assess the potential of this approach.
In Tanzania, malaria is endemic in many parts of the country and is the leading public health problem [
36,
37]. The Kilombero Valley in south-eastern Tanzania experiences year-round malaria transmission due to the presence of
An. arabiensis,
An. gambiae s.s. and
Anopheles funestus [
38]. Livestock keeping within the region increased significantly over the past decade due to the immigration of pastoralists from other parts of the country (i.e. from 42,385 to 55,994 cattle in 2001–03 livestock census, (DALDO Kilombero district livestock department (2003), Brehony
et al. unpublished reports), The population of livestock kept increasing even after 2003 census, this was reflected by the increase in needs of health services to livestock in the Kilombero Valley time after time (district livestock officer, personal communication ). In parallel with these changes, the coverage of LLINs has significantly increased, 2004 had a coverage of 75% of untreated nets and 2009 with the coverage of 47% of ITNs [
12]. Concurrent with these changes the abundance of
An. gambiae s.s. has rapidly declined [
12], with
An. arabiensis now being responsible for the remaining transmission. The relative frequency of
An. arabiensis with the
An. gambiae species complex grew from 13% in 2005 [
39], to 98% in 2009 [
40]. The presence of this zoophilic vector in addition to smaller populations of the more anthropophilic vectors
An. gambiae s.s. and
An. funestus make the Kilombero Valley an ideal location to investigate the potential impact of livestock on malaria vector ecology and human exposure risk. A three-year field study was conducted here to estimate the impact of local household livestock ownership on: (1) the abundance and diversity of mosquito vectors, (2) the feeding and resting behaviour of vectors and finally (3) net malaria exposure risk to humans. Malaria exposure risk as estimated in terms of the total number of malaria-infected mosquito bites (
An. gambiae s.l. and
An. funestus s.l.) expected to be received by people sleeping indoors at night. It was hypothesized that the presence of cattle at a household could reduce human malaria exposure rates if associated with a significant change in vector behaviour towards increased feeding on cattle and outdoor resting.
Discussion
In this study, the overall abundance of malaria vectors in both host seeking and resting collections was not consistently different between households with or without livestock. The abundance of mosquito vectors found host-seeking indoors and resting outdoors was slightly lower at households with cattle in only one of three study years. In other years there was no detectable difference. However, livestock ownership was associated with differences in malaria vector species composition, resting site usage and feeding behaviour. Over most years, An. arabiensis constituted a significantly higher proportion (5-15% more) of the indoor biting and outdoor resting An. gambiae s.l. population at households with cattle. Additionally, at households where cattle were present, significantly more vectors were found resting inside cattle sheds than inside houses or outdoor resting boxes. Further, the human blood index of
An. arabiensis and An. funestus s.l. was approximately 50% lower at households with livestock than without (inside houses and outdoor resting boxes respectively). These results confirm that the local presence of alternative host species such as cattle can significantly alter the habitat and host use of mosquito vectors at the household level.
Whilst the impact of cattle on mosquito vector behaviour was pronounced, the potential for these ecological effects to influence human malaria exposure risk was unclear. Malaria infection rates in An. gambiae s.l. collected from households with livestock tended to be lower than at those without livestock. However, the statistical significance of this effect depended on how background spatial variation in mosquito infection rates was controlled for. When village-level variation in mosquito sporozoite rates was incorporated into analysis, the impact of household livestock ownership was not significant. However, when village-level effects were removed and replaced by another proxy of spatial clustering, the nearest distance to another house (within the dataset) where livestock were kept, the difference in An. gambiae s.l sporozoite rates between
households with and without livestock achieved statistical significance. It was significantly higher at households without livestock than with.
The contrasting predictions obtained from different statistical models are deliberately presented here to highlight that no single unambiguous interpretation of these results is yet possible, and that further investigation to disentangle potentially confounding effects is required. At least two alternative explanations could account for the observed pattern. The first is that the reduced sporozoite rates found in
An. gambiae s.l. is an indirect consequence of livestock keepers being more likely to live in villages where malaria transmission was lower; either by chance or due to co-occurring environmental conditions such as more open grassland, nearer distances to the river, etc., which could influence risk. Another potential explanation is that ‘village’ is too large or imprecise a measure over which to assume transmission is heterogeneous. The villages in this study area were not always discrete units with clear spatial separation between them. Some villages were immediately adjacent to each other whilst others covered relatively large areas with two or more population clusters within them. Recent evidence suggests that malaria exposure risk can vary significantly over distances of a few hundred metres in response to local environmental factors [
57], thus there could have been significant heterogeneity in malaria transmission within these study villages that washed out finer-scale impacts of livestock at the household level. Finally, the tendency for lower sporozoite rates at households with livestock may be due to the higher proportion of
An. arabiensis within the
An. gambiae s.l. in these settings. Sporozoite rates were moderately lower in
An. arabiensis than in
An. gambiae s.s., thus variation in the relative proportion of these two species within the vector community could influence the total exposure risk arising from
An. gambiae s.l. This could provide an explanation for the observed variation in
An. gambiae s.l. sporozoite rates, but does not help resolve whether it is likely to have a significant epidemiological impact. Further study investigating the contribution of environmental variation over multiple spatial scales to both these entomological indicators and clinical risk factors is required to definitively resolve the impact of cattle on exposure risk.
In this study, the most pronounced impact of livestock was a reduction in the human blood index of malaria vectors. This help to support the lower sporozoite rates observed at households with livestock. The higher the human-vector contact the higher the risk of malaria transmission [
58]. However, the magnitude of the changes in HBI varied between vector species. At household with livestock, about 90% of non human blood index was from cattle, (Additional file
4). Whilst the HBI of
An. arabiensis and
An. funestus s.l was ~50% at households with livestock,
An. gambiae s.s. was relatively unaffected. The consistently high human blood index of
An. gambiae s.s. is not surprising in light of its well documented highly anthropophilic behaviour [
20]. However, the sizeable reduction in the HBI of
An. funestus was unexpected given this species is typically thought to be highly anthropophilic [
20,
59]. A possible explanation is that mosquitoes identified as
An. funestus s.l. in this study included morphological cryptic species, which have more diverse behaviours.
Anopheles funestus s.l. is a species complex consisting of both the type species (
An. funestus s.s.) and 7 morphologically indistinguishable subspecies [
60]. Of these,
An. funestus s.s. was assumed to be the only member of the species complex present within the Kilombero Valley at the time of study as resources for molecular confirmation were not available. More recently, Lwetoijera
et al. have confirmed that several members of this species complex are present in this area including
An. funestus s.s, Anopheles rivulorum,
Anopheles leesoni and
Anopheles parensis [
61]. Of these,
An. funestus s.s. predominates by 98%. The presence of
An. rivulorum which is highly zoophilic and is known to be associated with cattle [
62] may account for the observed reduction in the HBI of
An. funestus s.l. at households with cattle, or it can also mean that
An. funestus s.s. did feed on livestock as well. A further study needs to be done to clear this observation.
The abundance of mosquito vectors collected by different sampling methods also raises the possibility of human exposure to mosquito bites was overestimated in this study. Between ten and twenty times more vectors were sampled by CDC light traps than in all resting collections combined. Although clearly more efficient for sampling, the number of vectors captured in CDC light traps may not accurately reflect the proportion that would succeed in feeding. In our study, the abundance of blood fed mosquitoes found resting indoors was very low (on average <0.5 mosquito per collection), whereas 3–4 times more found in resting catches inside cattle sheds. This may indicate that few mosquitoes who attempt to feed indoors are successful due to the presence of bed nets, with most leaving the house to seek blood elsewhere (possibly in cattle sheds). Under such a scenario, CDC light traps might have overestimated actual exposure rates in the presence of bed net use. Further investigations involving detailed study of house entry and exit behaviors under varying scenarios of bed net usage and cattle presence would be useful to test this possibility.
Whilst this study yielded no clear evidence of a protective effect of cattle on exposure to malaria vectors, the possibility of a detrimental, zoopotentiative effect was refuted. Neither the abundance nor sporozoite rates of indoor biting vectors were higher at households with livestock.
It has been hypothesized that keeping cattle could increase malaria risk by attracting more mosquitoes to nearby houses, providing an additional source of blood to fuel mosquito reproduction, and create more larval habitats (through the puddles their footprints create, etc.) [
63]. This phenomenon has been observed in Ethiopia and Pakistan where the density of human-biting vectors increased in association with livestock [
16,
64]. However, these studies were conducted in communities where livestock were kept either inside human dwellings [
16], or where people slept outside close to livestock [
64]. In the Kilombero Valley, residents generally sleep indoors at night, with livestock being situated in separate cattle sheds that are an average of ≤25 m away. The separation of human and animal dwellings on this scale appears to be sufficiently large to avoid a zoopotentiation effect.
Differences in mosquito vector ecological and epidemiological factors described may be the cumulative impact not only of the presence of livestock, but of variation in socioeconomic and housing conditions that could be correlated with livestock keeping. For example, several household factors such as roof type, the presence of open eaves, window screens and ITN usage are significantly related to the abundance of malaria vectors that are find indoors at households within the Kilombero Valley [
65,
66]. Additionally, these factors are associated with wealth both in this part of Tanzania [
67], and other parts of sub-Saharan Africa [
68,
69]. Thus any systematic variation in house type, bed net usage, and socioeconomic status between households with and without livestock here could confound our ability to identify the specific impact of cattle. Unfortunately it has not possible to collect contemporary data on these associated household factors within the scope of the current study, so we cannot rule this out as a possibility. We note anecdotally however, that no systematic differences in house construction between households with and without livestock were obvious in this study. Almost all houses in this area have open eaves (>90%, Mnyone
et al., unpublished data) and households spanning the range of very low (generally thatched roof and walls, no window screens) to moderate income (bricked walls, aluminium roofs, screened windows) were evident in both livestock classes. Bed nets were observed in almost every household visited, although the insecticidal property could not be ascertained. Additionally, variation in mosquito numbers between households may also have been influenced not only by local households features, but the proximity and density of hosts (human and cattle) at neighbouring households. Time and logistic constraints meant that it was not possible to simultaneously map the distribution of people and cattle at all surrounding households, and include this as additional explanatory variables in our analysis. To fully resolve the direct impact of cattle on malaria risk, we encourage further more detailed studies in which associated demographic and socioeconomic factors from both focal and neighbouring households are taken into consideration.
Analysis of mosquito resting site use presented here was based on comparison of the abundance of vectors found inside individual houses versus individual outdoor resting boxes. Generally these abundances were similar. However, when the total number of mosquito vectors captured inside a house versus all outdoor resting boxes (four to eight per site) onsite was summed, significantly more individuals were caught inside than outside. This indicates that if resting collections were made only from inside houses, as is typical in many vector surveillance studies, at least half of the local resting vector population (those resting outdoors) would be missed. By failing to monitor what can clearly be a significantly sized outdoor resting population, conventional indoor-based surveillance methods risk misrepresenting vector ecology, and missing opportunities to identify settings in which vector control could be significantly strengthened by targeting mosquitoes outside houses.
A limitation of the present study was that it only estimated exposure rate in terms of the number of infectious bites that people would be expected to receive when they were indoors between 18.00 and 06.00 hours. Given that
An. arabiensis is exophilic [
70], it is possible that outdoor biting rates and associated exposure risk is higher at households with livestock [
15]. Further work to simultaneously quantify outdoor and indoor exposure risk at households with cattle is required to resolve this. However, a number of studies, including others from the Kilombero Valley [
47,
71] have shown that the biting activity of malaria vectors mainly occurs between 22:00 pm −06:00 am, a period when most people are asleep indoors [
72]. Consequently, assessment of mosquito biting indoors is a relevant index of the majority of human exposure.
These results add to a growing body of research that suggests the potential effectiveness of zooprophylaxis will vary with ecological context. For example, a previous study in West Africa found no evidence that cattle could provide a zooprophylactic effect in reducing exposure or disease risk [
17,
21]. The dominant vector species in this study was the highly anthropophilic
An. gambiae s.s. [
21], whose innate host preference may render it less susceptible to a zooprophylaxis approach. In contrast, other studies conducted in areas of Kenya and Zambia where
An. arabiensis is dominant found a significant reduction of malaria prevalence in areas where livestock were kept [
31,
33,
34]. This variability highlights the need for detailed study of vector ecology and behaviour to identify settings in which combining relatively simple household-level interventions such as extending insecticide coverage to cattle and their holding facilities [
63,
73,
74] with existing frontline measures (e.g., LLINs and IRS) could yield substantial improvements in malaria vector control.
Authors’ contributions
VSM coordinated and performed all mosquito sampling, laboratory analysis, data analysis, and developed the first draft of the manuscript. GN contributed to guiding the experimental design, laboratory analysis, interpretation of data. INL contributed to interpretation of data and drafting of the manuscript. JK, HM and HN contributed to mosquito collections. TLR contributed to data analysis and interpretation. HMF conceived the study, guided experimental design, data analysis, interpretation and the drafting of the manuscript. All authors read and approved the final manuscript.