Recent progress in reducing malaria morbidity and mortality in Africa is founded upon expanded coverage of insecticide-treated bed nets (hereafter, bed nets), indoor residual spraying, and combination drug therapy 1 . For this progress to translate into the ambitious goal of malaria elimination, most agree that vector control has a central role 1 2 3 . Yet, there is an incomplete understanding of how these insecticide-based interventions affect vector populations during long-term implementation, even though a long-term perspective (10+ years) is required to comprehend well the relationship between effectiveness of anti-vector measures and prevalence of malaria infection in humans 4 .

Vector populations can respond behaviourally, numerically, or evolutionarily to insecticides implemented against them in malaria control programmes. With regard to behaviour, females of some Anopheles species show elevated activity due to the excitation effects of the active ingredients in some insecticide formulations of indoor residual sprays or insecticide-treated bed nets, resulting in their movement away from the sprayed walls or treated nets, with or without having obtained a human blood meal 5 6 7 8 . With regard to numeric responses to these interventions, malaria vector populations typically diminish in density and have reduced longevity 9 10 11 . For example, Anopheles gambiae s.l. and Anopheles funestus population density declined markedly in a randomized evaluation trial of permethrin-treated bed nets in treatment compared to control villages in western Kenya 12 , an effect which persisted for three years after the trial ended and after all villagers were given treated nets that were retreated at 6-9 month intervals 13 . Evolutionary responses typically involve changes in phenotypic sensitivity to the insecticides being used, when alleles associated with reduced target site sensitivity or enhanced metabolic detoxification increase in frequency 14 .

In the present study, research was focused on the population numeric responses of Anopheles gambiae s.l. mosquitoes to long-term implementation of insecticide-treated bed nets in western Nyanza Province, Kenya. This species complex contains six species whose members are indistinguishable morphologically but which differ in certain behavioural and ecological attributes that are important to their vectorial capacity for malaria and for sampling 15 16 17 18 . Anopheles gambiae s.s. and Anopheles arabiensis are the two most common members of this complex and the only two found in western Kenya; A. gambiae s.s. feeds mostly on humans, whereas A. arabiensis feeds mostly on cattle and other animals, less so on humans, making it a less efficient but still capable malaria vector 17 18 .

The region where the research reported here was conducted, in the Asembo Bay area of Nyanza province in western Kenya, has been an area of active research on effectiveness of insecticide treated bed nets in reducing malaria transmission, and malaria-related morbidity and mortality in people 13 19 20 21 . In a randomized trial of the effectiveness of permethrin-treated bed nets on malaria infection and transmission commencing in late 1996, all houses in selected villages in Asembo received bed nets, whereas another set received no nets and served as controls; in 1999, houses in all villages received them, leading to high coverage of bed nets there that has been maintained to 2007 through the provision of free retreatment services and periodic net replacement 13 20 . The original trial showed that indoor density of Anopheles vectors of malaria diminished substantially, villagers' health improved, and child mortality declined 19 21 . These trends were sustained for four years after the trial ended and as net coverage was sustained 13 20 . Populations of A. funestus diminished to negligible levels, when bed nets were used at high coverage in the trial in western Kenya, whilst mosquitoes of the A. gambiae s.l. complex persisted as transmission declined 19 . There was no bed net distribution programme in a nearby and identical community called Seme, bordering Asembo to the east 13 . In samples of adult female mosquitoes taken from inside houses between 1999 and 2002, Lindblade et al 13 observed that the proportion of A. gambiae s.s. was significantly less in Asembo (51.2%) compared to Seme (77.4%), suggesting that the greater number of permethrin-treated bed nets in Asembo was disproportionately affecting populations of the former species. Building upon this observation, we postulated that populations of A. gambiae s.s. would decline when bed nets were owned and used at high rates, compared to the local sibling species, A. arabiensis. The increasing and well-documented patterns of bed net coverage in Asembo and Seme allowed a test of this hypothesis by measuring changes in numbers of adult and larval mosquitoes of both species over several years. Further, historical data were obtained to examine multi-decadal trends in changes in the proportions of these two species as the national malaria campaign in Kenya resulted in increases in bed net ownership regionally.

Historical review of data on the relative proportions of A. gambiae s.s. and A. arabiensis females sampled indoors from 1970 to 2002, as well as more contemporary data from sampling efforts of larvae and adult females of these two species reported here, showed a decline in the predominance of the former species with a comparative proportionate increase in the latter species (Figures 3, 4, and 7). Any sampling bias would likely be against A. arabiensis females in indoor collections due to their relatively reduced likelihood of entering and resting in houses, compared to A. gambiae s.s. females 17 18 30 thus differential sampling in favor of A. arabiensis is highly unlikely to be an explanation for the trend. As larvae of these species show no habitat segregation in this study area 34 35 , changes in larval numbers should accurately reflect population densities and true proportions of the two species, particularly because sampling was done during wet periods, when both species were abundant. The quantitative sampling data from 2003 (Figure 3) and qualitative sampling data to 2008-09 (Figure 4, Figure 7) clearly illustrate a process of gradual extirpation of A. gambiae s.s. in the study area, but persistence of A. arabiensis. Larval sampling facilitated delineation of this process and should prove useful to others who wish to compare relative changes of the two species under similar conditions.

In Asembo, Seme and regionally, the decline in A. gambiae s.s. coincided geographically and temporally with scale-up of national programmes leading to high rates of household ownership (and presumably, use) of bed nets, suggesting that presence of the bed nets in most houses caused the observed population decline. Alternative explanations seem less likely. First, biased sampling, if having an effect, would have worked against the trend. Second, there was no evidence of an environmental or climate change that could have affected the species distributions locally; indeed, temperature and rainfall were consistently within a normal range across a two decade period (Figure 6). More importantly, a recent prediction from ecological niche models based on climate change scenarios was that A. gambiae s.s. should increase while A. arabiensis should remain stable or decline regionally 36 , opposite of what was observed. Third, the changes might be related to differential abundance of cattle and human hosts. However, in a survey conducted before the current study commenced, both hosts were common in both areas, with about 2.5 cattle and 3.6 people per compound in Asembo and 3.8 cattle and 3.2 people per compound in Seme (J. Gimnig, unpublished data). Cattle are commonly husbanded throughout western Kenya, however, this observation suggests that a greater abundance of A. arabiensis in Asembo could not be explained by more cattle there compared to Seme. Fourth, a decline in A. gambiae s.s. populations in Seme subsequent to programmatic scale-up was predicted, based upon the observations from Lindblade et al 13 on proportionate differences of indoor resting, female A. gambiae s.s. relative to A. arabiensis in Asembo and Seme in 2002, but before the national programme commenced. Results from the present study confirmed this prediction for the prolonged period from 2003-2009, when A. arabiensis adults and larvae profoundly outnumbered A. gambiae s.s. in Asembo and became proportionately dominant in Seme, in sharp contrast to the historic trend prior to arrival of bed nets in that community and in nearby ones. Finally, the correlation through time between increase in bed net ownership (Figure 2) and decrease in A. gambiae s.s. (Figure 4, 7) could be a mere coincidence, an interpretation which seems highly unlikely given the historical dominance of this species in the region west of Kisumu; and given the results of the intensive transect sampling in 2003 over a relatively short distance (12 km) (Figure 3).

The most plausible biological mechanism for our primary result is straightforward: bed nets acted as lethal, human-baited traps or as strong repellency devices for the highly anthropophilic, female A. gambiae s.s., causing their population to crash. Aside from direct mortality, blood feeding inhibition, partially due to the excite-repellency effect of bed nets, could induce mortality through deprivation of blood. Fewer and shorter-lived adult A. gambiae s.s. laid fewer eggs in larval habitats, resulting in fewer larvae, reducing larval habitat occupancy and larval density. The decline in A. gambiae s.s. populations is provocative on several levels. First, any malaria control programme is imperfect, with some families not receiving bed nets or, if possessing them, not using them nightly or not retreating them regularly. In Asembo, recent observations indicate that of those families owning bed nets, only 77% use them regularly when sleeping (M. Hamel, unpublished data). Thus, ownership does not equate to use. Nets in Asembo were retreated at regular intervals by house to house campaigns through 2003. Thereafter, retreatment was available at central locations at regular intervals and the service was free through 2007, but retreatment rates (not quantified) were certainly never 100% (M.N. Bayoh, M. Hamel, unpublished observations). In Seme, household ownership of nets increased through efforts by the Kenya Ministry of Health yet remained incomplete after the second roll-out (Figure 2). Nonetheless, a massive population decline in a major, anthropophilic vector occurred despite these imperfections. Second, while A. gambiae s.s. historically showed considerable flexibility in resting behaviour when confronted with widespread indoor residual spraying 8 10 11 14 , results from the present study suggest less flexibility in host choice (Figure 5B). There was no strong blood host shift to non-humans, nor a shift to predominantly outdoor resting 28 , in the face of strong pressure from bed nets. Bogh et al 37 found that A. gambiae s.l. females shifted slightly in host selection away from humans towards cattle when permethrin-treated bed nets were distributed in villages on the Kenya coast. However, the mosquitoes were not identified to sibling species in the complex by PCR, thus any species-specific changes in host selection were not revealed in that study. Third, results reported here are consistent with negligible density-dependent effects influencing A. gambiae s.s. population dynamics. This is in contrast to strong density-dependent controls operating in Aedes mosquito populations 38 , but in agreement with results of field studies of A. gambiae s.s., which demonstrate only moderate density-dependence 39 . Cumulative adult female mortality due to exposure to pyrethroid toxins in bed nets appears not to be buffered by density-dependent modulation in immature stages, where density-independent processes such as disturbance dominate 26 , thus the killing effect of bed nets remains strong even as vector densities are driven low. The increase in bed net coverage described here (Figure 2) likely resulted in reductions in survival, total lifetime fecundity, and basic reproductive number of A. gambiae s.s. females in the study area cumulatively over many generations. The relatively lower parity rate observed in A. arabiensis compared to A. gambiae s.s. in Asembo compared to Seme in 2005 might be interpreted as a greater effect of bed nets on the former species, potentially confounding the interpretation of the mechanism of decline of the latter species. However, sampling bias against A. arabiensis resting and feeding outside of the peridomestic setting would result in over-sampling of those female A. arabiensis affected by bed nets indoors, therefore explaining the apparent discrepancy 17 18 .

The decline of an anthropophilic, anopheline mosquito species; and corresponding proportionate rise of a zoophilic one; during malaria vector control has rarely been reported in Africa. In the Pare-Taveta region of northern Tanzania and southeastern Kenya, indoor residual spraying with dieldrin resulted in the near elimination of A. funestus, whilst absolute numbers of the closely related but zoophilic species, Anopheles rivulorum, increased dramatically 40 . Even though spraying ceased in 1959, A. funestus populations and malaria transmission remained suppressed into 1966 41 , demonstrating long-term and vigorous effects of the original programme. The increase in not just proportion, but density of A. rivulorum, was unexpected and difficult to explain. Anopheles funestus has not been replaced by A. rivulorum in the Asembo area, A. funestus populations remain very low 27 30 , nor did A. arabiensis increase in absolute numbers as A. gambiae s.s. declined (see Figure 3). In western Kenya, near the Asembo study site, an indoor residual spray programme using fenitrothion resulted in a moderate increase in the proportion of adult A. arabiensis compared to A. gambiae s.s. 42 , but species structure of larval populations did not shift in tandem, and both the intervention and evaluation periods were short-term, not allowing for analysis of long-term effects as done here. In the Garki project in northern Nigeria, there was no observed shift in proportions of the two species after a period of indoor residual spraying, although entomological surveillance was a minor component of that evaluation 43 . In South Africa, where indoor residual spraying was implemented effectively to reduce malaria burden, A. gambiae s.s. apparently disappeared whereas the zoophilic species Anopheles quadriannulatus (also a member of the A. gambiae s.l. complex, but not a malaria vector) persisted, and residual malaria transmission was attributed to A. arabiensis 4 . However, these changes were qualitatively documented and no larval data were available for unbiased comparisons of changes in relative species abundance. During the malaria eradication campaign in British Guiana from 1945 to 1949, involving application of DDT on the inner walls of houses as a residual insecticide, larvae and adults of the primary vector (Anopheles darlingi) were originally numerous but disappeared, whereas larvae and adults of a zoophilic species, Anopheles aquasalis, persisted 44 .

The marked decline in A. gambiae s.s. in western Kenya has been associated with a simultaneous decline in malaria prevalence from 70% between 1997-1999 45 to ca. 25% in 2008 in children < 5 years old (M. Hamel, unpublished data). In eastern Kenya, malaria cases declined in children over the time period of 1997 to 2007, with a steep drop after 2004 46 , when the national bed net distribution programme began in earnest; however, O'Meara et al 46 could not conclude definitively that the decline in malaria cases was due solely to increased bed net use. By contrast, a similar marked decline in malaria cases in The Gambia over the same time period appeared to be related mainly to increased use of bed nets 47 . However, neither the eastern Kenya nor The Gambia study provided mosquito community composition data to correlate with the declines in malaria in humans. The implication of the research reported here is that sustained, high coverage of bed nets should dramatically reduce malaria transmission by A. gambiae s.s., leaving residual transmission by A. arabiensis (see Additional File 2). Indeed, the ratio of A. gambiae s.s. to A. arabiensis under conditions where both species occur and are transmitting malaria may be a useful relative index of programme effectiveness in places where the former species has been historically the dominant vector, as was the case in parts of southern Africa 4 . With wide coverage of expanded interventions like the one described here, malaria transmission should suffer a precipitous decline mediated through profound effects on vector populations, driving transmission downward and significantly closer to the goal of elimination. Killeen et al 48 proposed the need for higher coverage of bed nets when A. arabiensis becomes the dominant vector, if elimination is to be achieved; the opportunity to test this hypothesis is now available. The need for alternative control methods for A. arabiensis is also apparent.

Recent perspectives on the process of elimination propose a shift from population-based coverage of interventions to a clinical surveillance-based system with expanded drug treatment 3 . Results provided here, by contrast, illustrate the crucial importance of long-term maintenance of high coverage interventions against transmission (such as insecticide-treated bed nets) to ensure continual suppression of key vector species, coupled with long-term vector surveillance as a means of continually assessing programme effectiveness, such as by quantifying species ratios, host selection patterns, and parity rates.

The authors declare that they have no competing interests.

MNB, JEG, WAH, JMV, and EDW designed the study and wrote the manuscript. MNB, DKM, FM, MRO, JEG, and EDW sampled and processed mosquitoes; DKM, LK, MRO, and EDW identified mosquitoes of the A. gambiae s.l. complex with PCR. MJH, JEG, DM, WAH, and JMV obtained data of bed net ownership. All authors read and approved the final manuscript.