Impact of seasonality and malaria control interventions on Anopheles density and species composition from three areas of Uganda with differing malaria endemicity

Background Long-lasting insecticidal nets (LLINs) and indoor residual spraying (IRS) are the malaria control interventions primarily responsible for reductions in transmission intensity across sub-Saharan Africa. These interventions, however, may have differential impact on Anopheles species composition and density. This study examined the changing pattern of Anopheles species in three areas of Uganda with markedly different transmission intensities and different levels of vector control. Methods From October 2011 to June 2016 mosquitoes were collected monthly using CDC light traps from 100 randomly selected households in three areas: Walukuba (low transmission), Kihihi (moderate transmission) and Nagongera (high transmission). LLINs were distributed in November 2013 in Walukuba and Nagongera and in June 2014 in Kihihi. IRS was implemented only in Nagongera, with three rounds of bendiocarb delivered between December 2014 and June 2015. Mosquito species were identified morphologically and by PCR (Polymerase Chain Reaction). Results In Walukuba, LLIN distribution was associated with a decline in Anopheles funestus vector density (0.07 vs 0.02 mosquitoes per house per night, density ratio [DR] 0.34, 95% CI: 0.18–0.65, p = 0.001), but not Anopheles gambiae sensu stricto (s.s.) nor Anopheles arabiensis. In Kihihi, over 98% of mosquitoes were An. gambiae s.s. and LLIN distribution was associated with a decline in An. gambiae s.s. vector density (4.00 vs 2.46, DR 0.68, 95% CI: 0.49–0.94, p = 0.02). In Nagongera, the combination of LLINs and multiple rounds of IRS was associated with almost complete elimination of An. gambiae s.s. (28.0 vs 0.17, DR 0.004, 95% CI: 0.002–0.009, p < 0.001), and An. funestus sensu lato (s.l.) (3.90 vs 0.006, DR 0.001, 95% CI: 0.0005–0.004, p < 0.001), with a less pronounced decline in An. arabiensis (9.18 vs 2.00, DR 0.15 95% CI: 0.07–0.33, p < 0.001). Conclusions LLIN distribution was associated with reductions in An. funestus s.l. in the lowest transmission site and An. gambiae s.s. in the moderate transmission site. In the highest transmission site, a combination of LLINs and multiple rounds of IRS was associated with the near collapse of An. gambiae s.s. and An. funestus s.l. Following IRS, An. arabiensis, a behaviourally resilient vector, became the predominant species, which may have implications for malaria vector control activities. Development of interventions targeted at outdoor biting remains a priority.

biting [37]. Anopheles funestus s.l. remains an important vector in dry seasons as a result of its breeding habits [38,39].
With the expansion of vector control, changes in Anopheles species composition and mosquito density have been observed in Uganda [10,15], and elsewhere in sub-Saharan Africa [26,40,41]. Changes in malaria vector species composition in response to vector control interventions are not a new phenomenon and have been described previously [42]. Recent studies have demonstrated an increase in the relative abundance of An. arabiensis, when compared to sympatric An. gambiae s.s. following deployment of LLINs and/or IRS [10,11,14]. Similarly, the apparent replacement of highly anthropophilic An. funestus s.s. by less anthropophilic (zoophilic) and more exophilic Anopheles rivulorum in response to IRS in neighbouring Tanzania, was observed in the An. funestus s.l. complex in the 1960s [42]. Due to their more zoophilic and exophilic behaviour, vector control interventions have been less effective in controlling certain malaria vector species, such as An. arabiensis [41,43], and An. rivolurum [42]. To further explore the species-specific impact of vector control interventions, the impact of LLINs and IRS on sympatric An. gambiae s.s., An. arabiensis and An. funestus s.l. was examined on mosquito density in areas with differing malaria endemicity in Uganda.

Study sites
This study was conducted from October 2011 to June 2016 in three sites with differing malaria endemicity, within Walukuba, Kihihi and Nagongera sub-counties ( Fig. 1), as part of the PRISM1 (Programme for Resistance, Immunology, Surveillance and Modelling of Malaria) project [10,44,45] [46]. Walukuba sub-county (00°26′33.2″N, 33°13′32.3″E), located on the fringes of Lake Victoria in Jinja District, eastern Uganda is a periurban area at an elevation of 1,215 m with low malaria transmission [baseline annual human biting rate of 537 and P. falciparum entomological inoculation rate (EIR) of 3.2 infective bites per person per year] [46,47]. Anopheles arabiensis has been the predominant malaria vector species in this area [46,48]. Kihihi sub-county (00°45′03.1″S, 29°42′03.6″E), located in Kanungu District, southwestern Uganda, is a rural and hilly area 1,310 m above sea level, with moderate malaria transmission (baseline annual human biting rate of 1,337 and P. falciparum EIR of 14.2 infective bites per person per year) [46]. Anopheles gambiae s.s. has been the main malaria vector species in Kihihi [46,48]. Nagongera sub-county (00°46′10.6″ N, 34°01′34.1″ E), located in Tororo District, eastern Uganda, is a rural area bordering Kenya with an elevation of 1,185 m with high malaria transmission (baseline annual human biting rate of 16,606 reported in 2014 and P. falciparum EIR of 310 infective bites per person per year) [46]. Anopheles gambiae s.s. has been described as the main malaria vector in Tororo [48], however, in 2014 increasing proportions of An. arabiensis were documented [46]. Seasonality in Uganda is characterized by alternating rainy and dry seasons and a bimodal rainfall pattern. The longer rainy season occurs between July and November and the shorter rainy season between February and May [33].
During 2011-2016, the primary malaria control interventions deployed in Uganda included artemisinin-based combination therapy for treatment of uncomplicated malaria, distribution of LLINs through mass campaigns, and IRS in select districts [5]. LLINs were delivered to  [45] Walukuba and Nagongera in November 2013, and to Kihihi in June 2014. In Nagongera, three rounds of IRS with a carbamate insecticide (bendiocarb) were implemented between December 2014 and June 2015 (1st: December 2014 to Feb 2015, 2nd: June-July 2015, and 3 rd : November-December 2015).

Household selection
During the initial enrollment period in 2011, 100 households per site were randomly selected from a list of enumerated of households, as previously described [44]. In 2013, additional households were enrolled to replace households that had dropped out of the study to increase the number of enrolled households back to 100 per site (Fig. 2).

Mosquito collection
Mosquitoes were collected monthly from cohort study households using miniature CDC light traps (Model 512; John W. Hock Company, Gainesville, FL, USA) set at 19:00 h and collected the following morning at 07:00 h. One trap was set per household each month from October 2011 to June 2016. Light traps were positioned indoors, 1 m above the ground at the foot end of the bed, next to a study participant, sleeping under a LLIN [46]. Data were excluded from analysis if the target occupant did not sleep in the selected room or if the light trap was faulty.

Mosquito species identification
All anophelines collected were scored morphologically under dissecting microscopes at the study sites using taxonomic keys [21,49]. A subset of 30-50 mosquitoes was randomly selected per month per site for the entire study period for purposes of identifying members of the An. gambiae species complex using PCR [50]. The An. funestus species complex was not processed beyond morphological identification due to resource limitations (henceforth referred to as An. funestus). Results from the species identification were extrapolated to the total dataset to establish the species composition of all Anopheles collected at each site every month. Approximately, 10% of the Anopheles collected were non-malaria transmitting Anopheles christyi, classified as 'other Anopheles species' and were not processed further.

Data management and analysis
Field entomologists recorded CDC light trap data on standardized forms. The data collection forms were double-entered into a Microsoft Access database and checked for discrepancies. Any subsequent inconsistencies were resolved using original data entry forms. The primary independent variables investigated were; seasonality (dry versus wet season) and the combined vector control interventions (pre-intervention versus post-intervention). The outcomes of interest were vector density and species composition. Seasonality, denoted by rainy and dry seasons was generated for each site independently. For each site, the same consecutive months were divided into 2 rainy seasons and 2 dry seasons over 1 calendar year. Months with rainfall above and below the median value for the entire observation period were classified as rainy or dry season, respectively, after including a 1-month lag period. Vector density was determined by the number of mosquitoes collected per household per month per site and stratified by seasonality and the period before intervention implementation versus the period after intervention implementation. Simple proportions were compared using a log-binomial regression model with generalized estimating equations to adjust for repeated measures from the same house.
Here, we expand on the PRISM1 results previously reported by Kilama et al. [46] from observations carried out over 12 months (October 2011 to September 2012), by describing species-specific changes in response to vector control interventions carried out over 57 months (October 2011 to June 2016). Musiime et al. also used PRISM1 data to examine the impact of vector control interventions on Anopheles mosquito composition in Nagongera only, as measured using indoor and outdoor human landing catches [10]. This study analyses mosquitoes collected indoors using CDC light traps using longitudinal sampling in the three study sites. The PRISM1 dataset can be accessed at https ://cline pidb. org/ce/app/recor d/datas et/DS_0ad50 9829e .

Ethical approval and consent
In each study site, the head of household or adult representative was approached for consenting before household recruitment. A written informed consent was obtained as permission to conduct CDC light trap collections within the household. The study was approved by the Uganda National Council for Science and Technology   Fig. 2). The number of Anopheles mosquitoes collected per household per night (vector density) varied across the sites from 0.89 in Walukuba to 25.11 in Nagongera (Table 1). Overall, An. arabiensis (n = 2,391) was the predominant malaria vector species in Walukuba accounting for 52% of all collections. In Kihihi, nearly all Anopheles collected (98%) were An. gambiae s.s. (n = 18,135), while in Nagongera, 65% were An. gambiae s.s. (n = 87,936) ( Table 1). Of the 1,413 'other' Anopheles species collected in the sites, 1,385 (98%) were identified morphologically as An. christyi, which is classified as a non-malaria vector [51]. There is historical evidence that An. christyi has the ability to transmit malaria parasites [52], however, subsequent reports argue that this ability was either lost or suppressed independently [51] and is thus now considered to be a non-malaria vector. As expected, more Anopheles mosquitoes were collected during rainy seasons, compared to the dry seasons (Table 2).

Trends in Anopheles mosquitoes in Walukuba
In Walukuba, the rainy season was associated with approximately a three-fold increase  Table 2). This corresponded with the pattern of distribution observed in the graphical plots examining the absolute numbers of Anopheles collected in Walukuba (Fig. 3a) and the relative proportions (Fig. 4a) of mosquito species.

Trends in Anopheles mosquitoes in Kihihi
In Kihihi, the rainy season was associated with over a fivefold increase in  (Fig. 3b). When focusing only on trends in relative proportions of Anopheles over time, however, this finding is not obvious (Fig. 4b).

Trends in Anopheles mosquitoes in Nagongera
In Nagongera, there were substantially more An. gambiae s.s. ( (Fig. 3c). The absolute number of An. arabiensis changed less after the introduction of the mass vector control measures, and, as a result, the relative proportion of An. arabiensis increased markedly as the populations of An. gambiae s.s. and An. funestus collapsed, with An. arabiensis left as the predominant species after IRS (Fig. 4c).

Discussion
Over the past 13 years (2007-2020), vector control interventions have been scaled-up substantially across Uganda. Whilst the impact of LLINs and IRS on epidemiological outcomes has been assessed routinely [4,5,7,32,53,54], the effect of these interventions on malaria vector species is less commonly investigated. This study characterized vector species composition and density in three epidemiologically diverse settings from 2011 to 2016, while vector control interventions were implemented across the country by the Uganda Ministry of Health (National Malaria Control Division). As expected, Anopheles densities were higher during the rainy season in all study sites, consistent with other studies [48,55]. Prior to the widespread implementation of vector control interventions, Anopheles species were sympatric but composition varied between the sites, with An. arabiensis predominant in Walukuba (the lowest transmission site) and An. gambiae s.s. predominant in both Kihihi and Nagongera (the moderate and high transmission sites respectively). Delivery of LLINs was associated with significant declines in vector density for An. funestus in Walukuba, An. gambiae s.s. in Kihihi and in both An. gambiae s.s. and An. arabiensis in Nagongera. Addition of IRS to LLINs in Nagongera was associated with a decline in all vector species, albeit with a greater impact on An. gambiae s.s. and An. funestus, as reported elsewhere [56,57]. Consequently, An. arabiensis became the predominant species in this area. Understanding the impact of vector control interventions on local malaria vector species is paramount for assessing gaps in current vector control tools.
Malaria vector control interventions, mainly LLINs and IRS have been associated with changes in sympatric Anopheles species composition in Uganda [10], and elsewhere in East Africa [11,39,43]. However, a shift in vector species composition and a decline in vector numbers has also been reported in absence of systematic vector   control in north-east Tanzania [58,59], which underscores the possibility of other causes for these changes, such as epidemics of mosquito pathogens, improvements in housing, and changes in climate and land use. Inherent differences in malaria vector ecological characteristics [25], host preference [17], and exophagic and exophilic behavior [29,60,61], could be a threat to vector control especially for An. arabiensis [41]. Anopheles arabiensis is considered to have a lower vectorial capacity than An. gambiae s.s. and An. funestus in parts of East Africa [38]. In other settings, however, where An. arabiensis is the principal vector, evidence of strong anthropophagic behaviour and outdoor malaria transmission have been described [60]. The opportunistic feeding behavior of An. arabiensis, enables this species to avoid contact with LLINs and walls sprayed with insecticides which are applied indoors [27,60,62,63]. Empirical evidence shows that highly anthropophilic malaria vectors, such as An. gambiae s.s. and An. funestus s.s., are more responsive to vector control, particularly IRS programmes [ 39,42]. A shift in biting patterns of An. funestus, however, including early morning biting [37,64], and broad daytime biting [65], following introduction of LLINs has been documented. Current vector control tools target highly anthropophagic and endophilic behaviour [63]. However, there is growing evidence of outdoor biting especially in An. arabiensis [62,66], which poses a threat to vector control. A similar study, within the study area in Nagongera found a high proportion of An. arabiensis biting outdoors [10]. In this study, the combination of LLIN and IRS had a lower impact on An. arabiensis vector density compared to An. gambiae s.s. and An. funestus, making it the predominant malaria vector post-intervention. The impact of this apparent increase in An. arabiensis vector density on malaria transmission remains unclear, however. A similar study in Nagongera showed limited malaria transmission despite relatively abundant An. arabiensis [10]. In Kenya, there was a decline in malaria transmission following increased LLIN coverage, coincident with the replacement of primary malaria vectors, An. gambiae s.s. and An. funestus by An. arabiensis [39]. It is plausible that An. arabiensis may maintain residual transmission until the primary malaria vectors An. gambiae s.s. or An. funestus 'bounce back' . This occurred in western Kenya, where previously dominant An. funestus was suppressed following long term use of LLINs, but then recovered, becoming the predominant vector again within a period of almost 20 years, possibly due to high levels of pyrethroid resistance in this species [67]. In a key example of vector control failure in Kwazulu Natal, previously 'eliminated' An. funestus was replaced by less endophilic An. arabiensis, but returned after almost 40 years, highly resistant to pyrethroids, and associated with a malaria resurgence in this area [68].

Table 2 Stratified analysis of vector density (density ratio) by seasonality and intervention period
Outdoor biting behaviour of An. arabiensis poses a challenge to malaria vector control. Larval source management with microbial larvicides combined with LLINs has been shown to be protective against malaria infections in rural Kenya [69], and there are several measures including treating cattle with insecticide [60], use of odour-baited traps dispensing spatial repellents [70], and transfluthrin-treated chairs and ribbons [71], which could be deployed as control interventions in the future. In Uganda, there is still an information gap regarding the zoophilic behaviour of An. arabiensis and host choice in the presence of animals and humans. There is need for further research to assess the efficacy of interventions for controlling An. arabiensis.
This study had several limitations. First, the findings presented are from three sub-counties from only three districts. Thus, the study has limited geographical scope and the results may not be generalizable to other settings. Notably, however, the selected sites represented markedly different transmission settings, and all mosquito collections were made from randomly selected households after enumeration. Second, only indoor mosquito collections were done using light traps. Therefore, these results are subject to inherent biases presented by the mosquito trapping method used. Third, species-specific sporozoite data were not collected, therefore, implications to malaria control regarding residual transmission are implied. Within the study area, pyrethroid resistance was documented in both An. gambiae s.s and An. arabiensis [32], with evidence of carbamate resistance observed in An. gambiae s.s. from Nagongera and Kihihi [32]. However, the extent to which insecticide resistance affected mosquito survival under field conditions was not assessed. Study sites were not randomized to receive particular interventions; longitudinal measurements of mosquito density were made alongside vector control interventions delivered by the Uganda Ministry of Health. Whilst monthly rainfall measurements were used in the analysis and interpretation of the results, temperature and humidity data were unavailable for the study period.
Anopheles species composition may change from highly anthropophagic to less anthropophagic malaria vectors in response to vector control. However, the implications of these shifts in species composition on malaria transmission and control programmes are not well understood and require an in-depth examination of Anopheles species specific contribution to local malaria transmission. This study found that LLINs and IRS affected vector densities and species composition differently in different settings. Measuring absolute numbers of mosquitoes to quantify the impact of interventions instead of relying on relative proportions is important in order to understand the full picture.

Conclusions
In areas of low-and moderate-malaria transmission large-scale deployment of LLINs resulted in substantial reductions in An. gambiae s.s. and An. funestus s.l. In the area of intense malaria transmission, the introduction of LLINs and IRS resulted in the near collapse of these main vectors, with An. arabiensis becoming the principal vector, but at lower densities than prior to wide-scale vector control. Measuring the impact of vector control interventions using absolute numbers of mosquitoes collected increased precision. These findings suggest that the impact of LLINs and IRS on the primary malaria vectors (An. gambiae s.s., An. arabiensis and An. funestus) may be affected by the behaviour of these mosquito populations. Current vector control interventions are effective against malaria, but will not lead to elimination of the disease unless additional tools are included as supplementary interventions. Larval source management using chemical or microbial larvicides, combined with environmental management, could be used to improve control, especially in areas of high transmission.