Emergence of knock-down resistance in the Anopheles gambiae complex in the Upper River Region, The Gambia, and its relationship with malaria infection in children

Between 2000 and 2015, the prevalence of Plasmodium falciparum infection in sub-Saharan Africa (SSA) has halved due to the mass deployment of long-lasting insecticidal nets (LLINs), and to a lesser extent, indoor residual spraying (IRS) [1]. This has, however, increased selection pressure for insecticide resistance in malaria vectors, particularly resistance to pyrethroids, the only insecticide class used currently for treating bed nets. The strength and distribution of insecticide resistance has increased over time and there is growing concern that this will lead to control failure, which has the potential to reverse many of the gains seen in malaria control [2]. One of several mechanisms through which mosquitoes become resistant to insecticides is through mutations in the insecticide target site. Three different point mutations in the voltage-gated sodium channel gene confer knockdown resistance (kdr) to pyrethroids and organochlorines such as dichlorodiphenyltrichloroethane (DDT) in Anopheles gambiae s.l. [3‐5].

There has been little longitudinal insecticide resistance monitoring in The Gambia, but the general impression is that levels of insecticide resistance are low, but rising. Shortly after the introduction of permethrin-treated bed nets in The Gambia in the early 1990s, there was little or no resistance to DDT or permethrin [6, 7]. Prior to the nationwide DDT IRS campaign in 2009, DDT resistance was found in one site bordering Senegal, but mosquitoes from the site in the Upper River Region (URR) were fully susceptible to permethrin, deltamethrin and DDT [8]. Tests performed during a cluster-randomised controlled trial in the URR which compared the efficacy of LLINs versus LLINs and IRS with DDT against malaria in children found complete susceptibility of An. gambiae s.l. to DDT and permethrin in 2010 and some loss of susceptibility the following year [9]. Another study, conducted in the same year but outside the study area indicates that there may be pockets of high resistance in the URR [10]. Research also suggests that insecticide resistance may be partly responsible for the heterogeneities in malaria transmission across the country [11]. Although malaria has been declining in The Gambia since 2000 [12, 13], there is continued moderately high seasonal transmission in the URR despite high vector control coverage [14].

The study aimed to examine risk factors for kdr resistance in An. gambiae s.l. and its relationship with malaria infection in children in The Gambia.

A total of 6853 An. gambiae s.l. were caught in the 32 sampling sites over the two transmission seasons. Of these, 6828 (99.6%) were identified to species: 71.3% were An. arabiensis, 15.0% An. gambiae s.s., 12.3% An. coluzzii, and 0.1% hybrid (Fig. 2). Higher numbers were caught during 2010 when there was unusually high rainfall and extensive flooding, compared to 2011 when flooding was limited to areas beside the river. During the 2010 transmission season, 76.1% of An. gambiae s.l. were An. arabiensis, 12.0% An. gambiae s.s., 10.1% An. coluzzii and 0.2% hybrid (Additional file 1). During the 2011 transmission season, 57.8% of An. gambiae s.l. caught were An. arabiensis, 23.0% An. gambiae s.s., 19.0% An. coluzzii and 0.1% hybrid. Twenty-nine mosquitoes were caught during the dry season (of these 25 were An. arabiensis, 2 An. gambiae s.s. and 2 An. coluzzii). Spatial autocorrelation was found in species distributions with peak autocorrelation operating between 9 and 14 km depending on the species and year.

Analysis of species distributions over the two transmission seasons, showed that An. gambiae s.s. was more common further away from the river (Odds ratio, OR for every km away from the river = 1.29, 95% CI 1.21–1.38, p < 0.001) (Figs. 3, 4). Conversely, both An. arabiensis and An. coluzzii were more common closer to the river (An. arabiensis OR = 0.88, 95% CI 0.83–0.94, p < 0.001; An. coluzzii OR = 0.91, 95% CI 0.85–0.98, p = 0.01). Similar patterns were found when each year was analysed separately.

In 2010, An. arabiensis comprised 81.8% of mosquitoes caught in the LLIN only arm (63.3% in 2011) and 67.8% in the IRS-LLIN arm (50.3% in 2011). As a result, there was a significantly lower odds of collecting An. arabiensis in the double intervention arm compared to the LLIN arm of the study in both years (2010: odds ratio, OR = 0.51, 95% CI 0.31–0.84, p = 0.008; 2011: OR = 0.44, 95% CI 0.25–0.78, p = 0.005). In 2010, An. gambiae s.s. comprised 6.5% of mosquitoes in the LLIN only arm (17.9% in 2011) and 19.8% in the IRS-LLIN arm (29.9% in 2011). There was a significantly higher odds of finding An. gambiae s.s. in the IRS-LLIN arm compared to the LLIN arm of the study in both years (2010: OR = 3.02, 95% CI 1.38–6.57, p = 0.005; 2011: OR = 2.71, 95% CI 1.18–6.19, p = 0.02). There was no difference in the odds of catching An. coluzzii between the two study arms in both 2010 and 2011 (2010: OR = 1.12, 95% CI 0.71–1.79, p = 0.63; 2011: OR = 0.81, 95% CI 0.41–1.60, p = 0.54), nor was there a difference in the odds of catching hybrids between the two study arms in 2010 (OR = 1.93, 95% CI 0.43–8.62, p = 0.39).

Vgsc-1014 mutations were found in all species sampled but at differing levels. An. arabiensis were predominantly wild-type (73.1% during 2010 and 58.1% during 2011), although the proportion with heterozygous Vgsc-1014S mutations increased from 20.5% in 2010 to 28.3% in 2011 (OR = 1.58, 95% CI 1.32–1.89, p < 0.001) (Table 1). Anopheles gambiae s.s. were predominantly homozygous Vgsc-1014F and this proportion increased almost to saturation from 64.8% in 2010 to 90.9% in 2011 (OR = 8.24, 95% CI 4.99–13.63, p < 0.001). An. coluzzii were predominantly wild-type (73.1% during 2010 and 70.9% during 2011).

In both years, the odds of having any type of Vgsc-1014 mutation was significantly higher in the IRS-LLIN arm compared to the LLIN only arm (2010: OR = 1.54, 95% CI 1.07–2.22, p = 0.02; 2011: OR = 2.26, 95% CI 1.24–4.11, p = 0.01) (Table 2). This was primarily due to the higher proportion of Vgsc-1014F mutations, particularly homozygous Vgsc-1014F mutations, in the double intervention compared to the single intervention arm. In 2010, IRS-LLIN villages had 2.24 times the odds of mosquitoes carrying homozygous Vgsc-1014F mutations compared to LLIN only villages (95% CI 1.12–4.49, p = 0.02), while in 2011, double intervention villages had 2.52 times the odds (95% CI 1.20–5.29, p = 0.01). There was an increased odds of a mosquito carrying the heterozygous Vgsc-1014F mutation in 2010 (OR = 2.17, 95% CI 1.28–3.68, p = 0.004), but not in 2011 (OR = 1.32, 95% CI 0.75–2.33, p = 0.34). No significant difference in the odds of heterozygous or homozygous Vgsc-1014S mutations was found between IRS-LLIN villages and LLIN villages in 2010 or 2011.

Distribution maps of Vgsc-1014 mutations show an increase in the proportion of mosquitoes carrying homozygous Vgsc-1014F mutations in villages on the south bank and in the northern part of the study area bordering Senegal between 2010 and 2011 (Fig. 5), which mirrors the increase in the proportion of An. gambiae s.s. in these areas (Fig. 3, 4).

In a multivariable model, species, study arm and year of survey were associated with odds of any Vgsc-1014 mutation (Table 3). Adjusting for year and study arm, An. gambiae s.s. mosquitoes had 18.49 times the odds of having any Vgsc-1014 mutation compared to An. arabiensis (95% CI 14.48–23.61, p < 0.001), while An. coluzzii had 0.76 times the odds of having any Vgsc-1014 mutation (95% CI 0.63–0.92, p = 0.004) compared to An. arabiensis. Adjusting for species and year, mosquitoes caught in the LLIN-IRS arm had 1.27 times the odds of having any Vgsc-1014 mutation compared to mosquitoes in the LLIN only arm (95% CI 1.03–1.55, p = 0.02). Adjusting for species and study arm, mosquitoes caught in 2011 had 1.80 times the odds of having any Vgsc-1014 mutation compared to those caught in 2010 (95% CI 1.56–2.09, p < 0.001).

P. falciparum infection status was ascertained for 1543 children at the end of the 2010 transmission season and 1564 children in 2011. Multivariable analysis showed that girls were more likely to be infected with P. falciparum in 2010 (OR = 1.47, 95% CI 1.08–1.98, p = 0.01), but were less likely to be infected in 2011 (OR = 0.73, 95% CI 0.56–0.96, p = 0.02) (Table 4). Older children were more likely to be infected with P. falciparum at the end of the transmission season in both years (OR for 1 year increase in age: 2010 = 1.06, 95% CI 1.02–1.10, p = 0.007, 2011 = 1.12, 95% CI 1.08–1.17, p < 0.001). In 2011, children sleeping under an LLIN the previous night were less likely to be infected than children who had not slept under an LLIN (OR = 0.49, 95% CI 0.28–0.83, p = 0.009), although there was no significant association between LLIN use and infection in 2010. There was no significant association between cluster level prevalence of any Vgsc-1014F mutation and malaria infection in children in either year. In univariable analysis there was a tendency towards an increased odds of P. falciparum infection among children living in clusters with a high proportion of mosquitoes specifically carrying homozygous Vgsc-1014F mutations (OR for a 10% increase in the proportion of An. gambiae s.l. mosquitoes with any Vgsc-1014 mutation 2010: OR = 1.25, 95% CI 0.99–1.58, p = 0.07, 2011: OR = 1.13, 95% CI 1.01–1.27, p = 0.04). There was also a similar magnitude increase in the odds of P. falciparum infection among children living in clusters with a high proportion of mosquitoes identified as An. gambiae s.s. in univariable analysis (OR for a 10% increase in the proportion of An. gambiae s.l. mosquitoes identified as An. gambiae s.s. 2010: OR = 1.27, 95% CI 1.03–1.55, p = 0.02, 2011: OR = 1.12, 95% CI 1.01–1.25, p = 0.03). The proportion of homozygous Vgsc-1014F mutations and proportion of An. gambiae s.s. at each cluster were colinear (2010: R² = 0.94, VIF = 17.3, 2011: R² = 0.91, VIF = 10.8). As a result, these variables could not be combined in the multivariable model. Comparison of the AIC for models including either cluster level proportion of An. gambiae s.s. or mosquitoes carrying homozygous Vgsc-1014F mutations was not able to distinguish which model provided better goodness of fit (2010: AIC for model including An. gambiae s.s. = 1195.38, AIC for model including homozygous Vgsc-1014F mutations = 1196.90, 2011: AIC for model including An. gambiae s.s. = 1413.25, AIC for model including homozygous Vgsc-1014F mutations = 1413.71). Looking at total number of mosquitoes collected in each cluster, rather than percentage composition, there was no significant association between either the absolute number of An. gambiae s.s. mosquitoes, mosquitoes with Vgsc-1014 mutations or mosquitoes with homozygous Vgsc-101F mutations and malaria infection in children at the end of the transmission season in both years.

These findings illustrate the temporal and spatial pattern of the An. gambiae complex and Vgsc-1014 mutations in the URR of The Gambia and their association with malaria infection in children from 2010 to 2011. To the authors knowledge, this is the first study to adopt a landscape approach with intensive entomological sampling to understand factors related to the distribution of kdr in malaria vectors.

As in previous work in the URR [21‐24], An. arabiensis was the most abundant member of the An. gambiae complex and persisted longer into the dry season than the other species. The frequency of hybrids was 0.1–0.2%, a slightly lower proportion than that shown by others in The Gambia [24, 25]. Anopheles gambiae s.s. were more common in villages away from the River Gambia which corresponds with previous studies that reported An. gambiae s.s. prefers small rain-dependent larval habitats on free-draining soil covered with open woodland savannah or farmland [26, 27]. Anopheles arabiensis was more common near the river suggesting that their aquatic habitats are common in wetlands, such as rain-fed ricefields, adjacent to the river [28]. Indeed, previous studies found An. arabiensis in water bodies along the edge of the alluvial soils, particularly in areas of rice cultivation in the Central River Region [29]. Anopheles coluzzii was also more common closer to the river which supports research showing this species exploits semi-permanent aquatic habitats that are also frequented by An. arabiensis [26, 30, 31]. Although the literature suggests the distribution of species is likely to be due to differential larval habitats, this result may be because An. gambiae s.s. has less flexible host choice behaviours than An. arabiensis [32‐34] and so contributes a larger proportion of the mosquito catch further away from the river.

Interestingly, fewer An. arabiensis were caught in villages in the double intervention compared to the single intervention arm, while the opposite pattern was seen with An. gambiae s.s. If the Vgsc-1014F mutation translates into phenotypic resistance, this may have given An. gambiae s.s. a competitive advantage over An. arabiensis in the double intervention arm. This is a different pattern from that in East Africa where, with the scale-up of interventions, An. arabiensis is starting to dominate over An. gambiae s.s. since the latter is more endophagic and endophillic and so is thought to be preferentially killed by LLINs [35‐37].

Species, year of survey and study arm were associated with odds of any Vgsc-1014 mutation. There was a 1.27 increase in the odds of any Vgsc-1014 mutation in the double intervention compared to the single intervention arm and a 1.80 increase in the odds of any Vgsc-1014 mutation between 2010 and 2011. The odds of any Vgsc-1014 mutation was 18.49 times higher in An. gambiae s.s. compared to An. arabiensis. Taken together this suggests that (i) LLINs and DDT used together provide more selection pressure than LLINs alone and (ii) there was an increase in selection pressure over the two years most likely due to the second IRS round, and (iii) selection pressure favours An. gambiae s.s. since it has a higher frequency of kdr. Several studies have shown an increase in the frequency of kdr mutations following implementation of vector control interventions [38‐41]. High levels of kdr in An. gambiae s.s. compared to An. arabiensis may be explained by a greater propensity for indoor resting and feeding of An. gambiae s.s. and, therefore, potential for increased contact with insecticides on walls or LLINs [17, 34, 42]. The study used DDT for IRS and pyrethroid-treated LLINs, and as such is it unsurprising that selection pressure for development of Vgsc-1014 mutations was high in the double intervention arm. The IRS in the study was performed by government teams using the insecticide selected by the National Malaria Control Programme but an alternative insecticide class, such as an organophosphate or carbamate would have been a better option to reduce selection pressure [10]. In fact, based on insecticide resistance monitoring, the control programme recently started to implement rotation of IRS insecticides beginning with bendiocarb in 2015 and 2016 and pirimiphos-methyl in 2017.

Older children had a higher odds of P. falciparum infection in both of the end of season surveys. It is unclear why females were at increased risk of infection in the first survey but at lower risk in the second survey and this may be an anomalous result. In 2011, children sleeping under an LLIN had half the odds of being infected compared to children not sleeping under a net, but no significant difference was observed in 2010. This result is probably due to chance because of low numbers of children not using an LLIN in this study. There was no significant association between the cluster level proportion of mosquitoes with any Vgsc-1014F mutations and malaria infection in children. However, univariable analysis did show an association between P. falciparum infection in children and the cluster level proportion of An. gambiae s.s. and homozygous Vgsc-1014F mutations specifically, especially in the second year of the study. However, due to colinearity between An. gambiae s.s. and homozygous Vgsc-1014F mutations these variables could not be combined in the multivariable model. Calculation of the AIC was not able to distinguish between models including these two variables and so it was not possible to say whether high proportions of An. gambiae s.s. or homozygous Vgsc-1014F mutations increased P. falciparum infection in children.

Increased malaria infection in the study children may be explained by differences in the species distribution in the villages, specifically possible higher efficiency transmission by An. gambiae s.s. Anopheles gambiae s.s. is a more efficient vector than An. arabiensis, although it is less clear whether there is a difference in transmission efficiency between An. gambiae s.s. and An. coluzzii [43‐45]. Alternatively, heterogeneity in malaria infection could also be due to the impact of kdr. Indeed, previous studies have highlighted a lack of decline in malaria in the URR [14, 46] and a study of paired high and low malaria prevalence villages in The Gambia suggested that heterogeneous transmission may be partly due to insecticide resistance [11]. Opondo et al. showed that DDT mortality for An. gambiae s.s. was significantly lower in high prevalence compared to low prevalence villages and that there was a significant association between the Vgsc-1014F mutation in An. gambiae s.s. and resistance to DDT and deltamethrin [11]. This mutation was a strong predictor of insecticide resistance and effectively masked the effect of other mutations in this study such as those associated with metabolic resistance. However, the role of kdr is not clear cut [47, 48] and several studies show that pyrethroid LLINs were still able to kill An. gambiae despite high kdr frequencies [49‐52]. Analysis did not show any significant relationship between childhood malaria infection and the absolute number of An. gambiae s.s. mosquitoes, mosquitoes with any Vgsc-1014 mutation or mosquitoes with the homozygous Vgsc-1014F mutation per cluster. This is most likely because of low vector numbers in some of the village clusters.

In conclusion, the homozygous Vgsc-1014F mutation occurred predominantly in An. gambiae s.s. and increased almost to saturation during the course of the study. It also occurred at higher frequencies where IRS was used in addition to LLINs, probably because the kdr mutation confers a selective advantage in the presence of insecticides. There was a 13% increase in the odds of malaria infection in children associated with a 10% increase in the proportion of An. gambiae s.l. carrying the Vgsc-1014F mutation in 2011. Moreover there was a 27% increase in the odds of malaria infection with a 10% increase in the proportion of An. gambiae s.s. mosquitoes in 2010 and a 12% increase in 2011. It was, however, impossible to determine whether resistance or species increased the odds of childhood malaria infection since the homozygous Vgsc-1014F mutation was colinear with An. gambiae s.s.