Few insecticides are available for use in malaria vector control and only pyrethroid compounds are considered safe for the treatment of ITNs [8]. Therefore, the continued success of ITNs and essentially the current vector-control paradigm depend on continued mosquito susceptibility to a single class of insecticides. Two mechanisms of insecticide resistance, which may co-occur, are commonly measured: (1) metabolic resistance and (2) target-site insensitivity [9]. The former results from the over expression or amplification of genes coding for enzymes that either biochemically alter the insecticidal compound making it less toxic to the mosquito (e.g. P450 monooxygenases), or sequester the compound preventing reactions detrimental to normal physiology (e.g. esterases) [9,10]. In contrast, target-site insensitivity involves one or more mutations that make the physiologic target of an insecticide less reactive to the chemical [11]. Target-site insensitivity in the malaria vector Anopheles gambiae s.s. (hereafter referred to as A. gambiae) includes two kdr mutations (for knock-down resistance) which result in an amino acid substitution in the S6 hydrophobic segment of domain II in the voltage-gated sodium channel of neuronal membranes [9,11]. Both mutations reduce susceptibility to DDT and to the pyrethroid class of insecticides and occur at different nucleotides of the same amino acid (residue 1014). In wild-type mosquitoes, this residue codes for leucine (TTA), while in mutant mosquitoes single nucleotide changes result in either a phenylalanine (TTT) or serine (TCA) substitution [12,13]. The former is commonly referred to as West African kdr, or L1014F; while the latter is called East African kdr, or L1014S. Molecular data suggest that each mutation has risen independently at least twice in A. gambiae [14], and, although the names of each allele suggest distinct geographic distributions, co-occurrence has been reported in multiple countries including Uganda, Gabon, Cameroon, Equatorial Guinea, and Angola [15]. L1014F is thought to confer a greater degree of insensitivity to pyrethroid insecticides than L1014S [11], although bioassay data comparing each mutation within the same genetic background do not exist. Nevertheless, it has been speculated that selection for L1014F should be stronger than for L1014S, a hypothesis supported by recent molecular data showing that signatures of selective sweeps (i.e. reduced nucleotide variability) are more extensive around L1014F than L1014S mutations in the genomes of wild-caught mosquitoes [16]. A third mechanism of resistance--behavioural--is poorly understood and difficult to measure, although the excitatory effects of DDT in IRS and permethrin in certain ITN formulations suggest that a primary mode of action of these two chemicals is on vector behaviour and not survival [17-20].

When monitoring or investigating insecticide resistance in vector populations, at least three approaches can be taken, each with advantages and disadvantages: (1) measures of phenotypic resistance provide a direct indication of how resistance mechanisms impact vector control activities but require access to testing kits, insecticide impregnated papers, rearing facilities, and large numbers of mosquitoes, any of which may be limiting; (2) frequencies of target site mutations (e.g. kdr) are easier to measure, but it is unclear how much this mechanism contributes to resistant phenotypes; (3) measures of metabolic resistance (e.g. biochemical assays or gene expression arrays) are likely to be strong indicators of phenotypic resistance but are technically challenging.

Malaria vectors in western Kenya, where the present study was conducted, include two sibling species of the A. gambiae species complex, A. gambiae and Anopheles arabiensis, as well as Anopheles funestus s.s. Historically, A. gambiae has been the primary vector in the study area, but this species has declined proportionately to A. arabiensis (a secondary vector), possibly due to greater effectiveness of ITNs against the former, more anthropophilic species [21]. Development of insecticide resistance in the local A. gambiae population could reverse this trend and might reduce effectiveness of the malaria control program in the region.

The rise in the mutation's frequency followed similar trajectories in Asembo and Seme despite substantial differences in net coverage between the areas for a decade (Figure ​(Figure2).2). Net ownership rose modestly in Seme from 2000 to 2003 and then substantially increased with subsidized distribution of ITNs to pregnant women and children <5 years of age from government health facilities beginning in 2004, followed by a mass campaign in 2006 [21,33]. The similar patterns of emergence of resistance likely reflect details of migration and selection pressure not measured in our study. While A. gambiae came into ever increasing contact with pyrethroids in Asembo over the past 13 years, selection pressures outside of Asembo due to either malaria control or agricultural pesticide use are unknown.

Other studies have also reported increases in frequency of the kdr L1014S allele in A. gambiae from Burundi [44] and Uganda [45] associated with vector control using pyrethroid insecticides; while studies in Niger [46] and Equatorial Guinea [47] have observed sharp rises in the kdr L1014F allele in response to ITNs and IRS, respectively. In Burundi and Rwanda, the rise in the kdr L1014S allele was ascribed to use of insecticides as indoor residual sprays to control adult stages of malaria vectors. However, in Burundi, the allele frequency actually increased in an unsprayed control area as well as in the sprayed area. In Uganda, there were marked regional variations in allele frequency without clear correlation with intensity of use of IRS or ITNs, and allele frequency and phenotypic resistance were noted particularly in areas with a history of cotton agriculture where insecticide use is often intense [45].

In other settings, agricultural use of insecticides has been cited as the primary cause of the emergence of insecticide resistance in A. gambiae s.l. populations in sub-Saharan Africa [48-50]. Anopheles gambiae s.l. larvae may face strong selection pressure for resistance to insecticides if exposed in breeding sites located near cultivated fields where insecticides are applied to control agricultural pests. In a cotton-growing region of northern Cameroon, investigators sampled mosquito larvae from breeding sites within cotton fields at different times during the rainy season [51]. Bioassays of emerged adults revealed that vector susceptibility to DDT and permethrin decreased over time in accordance with a spraying schedule in the cotton fields that included two applications of an organochlorine compound (endosulfan) followed by a pyrethroid/organophosphate mixture (cypermethrin/profenofos). In Burkina Faso, the spatial heterogeneity of kdr L1014F was associated with cotton agriculture; the mutation was present at 16 of 21 sites and ranged in frequency from 4.7% to 97.0% with the highest frequencies occurring in the so-called "cotton belt" [52]. Although data are lacking, current agricultural use of pesticides in western Kenya is likely low owing to the prevalence of subsistence agriculture. However, use of permethrin for cattle dips is common, and the region produces some cash crops (sugar cane and cotton) that may require insecticide use. In addition, the low level of resistance to bendiocarb implies that agricultural use of insecticides may play a role in the evolution of resistance in malaria vectors in western Kenya. However, the strong temporal association reported here between net ownership and kdr L1014S frequency (Figure ​(Figure2)2) suggests that ITNs have been the most important selection pressure in this study population.

Given the trends in net ownership and kdr L1014S frequency in Seme, it is not surprising that similarly high homozygote frequencies were found in samples of A. gambiae throughout western Kenya in 2009 and 2010. One unexpected finding is the apparently greater magnitude of phenotypic resistance near the Ugandan border, as well as the dramatic differences in species composition near Lake Victoria compared to sites further north. Due to the decline in A. gambiae along the lakeshore, inadequate numbers of this species were available for testing in phenotypic assays. It is, therefore, possible that phenotypic resistance in A. gambiae along the lakeshore is similar to that of sites further away and the few mosquitoes that were tested (<5 per insecticide) suggest that this is the case. Therefore, the question is why A. gambiae remains the predominant species in sites located further from the lakeshore. The current hypothesis is that ITNs have not been in place sufficiently long in Busia, Malaba, Bungoma, and Kakamega to drive down local abundance of A. gambiae. Qualitative comparisons of the ratio of A. arabiensis to A. gambiae amongst all these sites suggests that the former species is rising in frequency in the sites away from the lake shore as it has at sites near the lake [21]. Indeed, the presence of A. arabiensis at Bungoma and Kakamega, both sites at relatively high altitudes where A. gambiae has traditionally been the only species in the complex present, is particularly striking [53]. However, because historical data on net ownership in these areas are lacking, baseline levels prior to the national scale-up of ITN are unknown. In Asembo and Seme, the change from sub-populations dominated by A. gambiae to those dominated by A. arabiensis took about a decade and occurred in Asembo first, as would be expected if ITNs were the primary cause [21]. This hypothesis is one that could be tested by monitoring both ITN coverage and species composition over the next several years in the region.

One trend that was consistent throughout all populations examined was the high degree of susceptibility of A. arabiensis to all insecticides but moderate to high resistance to pyrethroids in A. gambiae. The persistence of a species with little to no pyrethroid resistance (A. arabiensis) compared to a species with moderate to high levels of pyrethroid resistance (A. gambiae) in an area with high ITN coverage is somewhat counterintuitive. However, it is likely explained by the behaviour of A. arabiensis which often feeds outdoors and on cattle and may avoid the insecticide on nets. Anopheles arabiensis populations are therefore able to persist, apparently with little to no selection from the pyrethroid insecticides on nets [21]. Anopheles gambiae populations, despite having some resistance to pyrethroid insecticides, are still in decline possibly due to irritancy of the insecticide on the nets or the physical barrier imposed by the nets. These observations also suggest there is a limit to the degree of resistance conferred by the molecular and biochemical mechanisms currently present in western Kenya. Similar observations were made in the early 1990s during a small scale study of permethrin-treated nets where resistance was detected within a year of implementation [24], but reached a plateau and even regressed after three years [25]. However, other resistance mechanisms, possibly coupled with secondary compensatory mutations, may lead to further increases in pyrethroid resistance which could lead to a resurgence of A. gambiae. Despite the rapid decline in A. gambiae along the lakeshore, malaria transmission-presumably maintained by A. arabiensis--remains high with parasite prevalence in children over 45% (M. Hamel, unpublished data). As ITNs and IRS are increasingly scaled up throughout Africa, behavioural avoidance of these interventions may become increasingly important and tools to address species or populations exhibiting these traits are urgently needed.