This study was conducted on Rusinga Island (0°21′ and 0°26 south, 34°13′ and 34°07′ east) along the shore of Lake Victoria in Homa Bay County, western Kenya ²¹ ( Figure 1). The area is endemic for malaria and the estimated prevalence of malaria in the population of Rusinga Island is around 10% ^{18, 22, 23} . Rusinga Island is only around 100 metres away from the mainland and connected via a bridge. The island has an area of 44 km ² with altitude ranging from 1100 m to 1300 m above sea level and a population size of about 25,000 ²⁴ . The daily average temperatures of Rusinga Island range from 16°C to 34°C and peak in dry seasons ²⁴ . The area experiences bimodal rainy seasons with long rains from March to June and short rains from November to December. Malaria transmission peaks following the end of the long rainy season in June/July ²⁵ . The field survey was implemented between May and June 2018, towards the end of the long rainy season.

Habitat surveys were done along stretches of 700 m long and 300 m wide (clusters of approx. 0.2 km ²). A total of 13 sampling clusters were selected around the lake shores of Rusinga Island ( Figure 1B). The areas were selected with the help of Google Earth, aiming at a homogeneous distribution around the island. Inaccessible areas with steep rocks at the shoreline were excluded. Within each sampling cluster all aquatic habitats’ locations were recorded using a smartphone, a unique identifier allocated and sampled as outlined below.

The aquatic habitat types were categorized as either swamp, puddle, fishpond, drainage/trench or artificial pit. The perimeter of every habitat was estimated by walking in large steps around the habitat. Water turbidity was measured using a turbidity meter (TRB 355IR, WTW Germany) and water pH and temperature were measured using a portable multi-parameter probe (Multi, WTW Germany).

Every aquatic habitat was inspected for the presence of larvae using the sweep-net method as described by Ndenga and others ²⁶ . The sweep-net (40 cm × 15 cm × 30 cm) was made from fine cotton cloth with a 150 cm long handle. It was chosen for sampling due to its better efficiency in sampling the diverse aquatic fauna including freshly hatched mosquito larvae and mosquito pupae than the standard dipper ^{26, 27} . A dipper was used for sampling when the habitat was too small to be sampled by a sweep-net. Sampling of mosquito larvae using either sampling tools was randomly done at different points of the habitats since different species of malaria vectors prefer different conditions and vegetations. The duration of sweeping was dependent on the perimeter of the habitat. About 10 minutes were taken to sweep habitats with perimeters exceeding 10 metres, while 5 minutes were taken to sweep habitats <10 m in perimeter. All sweeps were emptied into white trays and mosquito immature stages were counted separately for the two encountered genera, Anopheles and Culex. Culex and Anopheles larvae were identified morphologically. Culex larvae possess siphon on the posterior part of their abdomen for breathing through at the interface of the water surface during resting whereas Anopheles larvae have no siphon and rest horizontal to the water body ²⁸ . The larvae were grouped as early (1 ^st and 2 ^nd instar) and late (3 ^rd and 4 ^th instar) instars based on their body size. In addition, macroinvertebrates sampled from a habitat were grouped as Odonata (dragonfly and damselfly larvae), Coleoptera (water beetle larvae and adults), Heteroptera (Notonectidae, Naucoridae and Nepidae), fish and tadpoles. All late instar Anopheles larvae and mosquito pupae were transferred to water bottles (1 L) containing habitat water and taken to the International Centre of Insect Physiology and Ecology-Thomas Odhiambo Campus ( icipe-TOC) for rearing to adults. Rearing of the field collected larvae was done in 1 L plastic rectangle food mate (H67 × W126 × L184 mm, Kenpoly manufacturer, Nairobi, Kenya). Larvae were fed with a pinch of ground dry cat food (Nestlé Purina PetCare Company, Nairobi, Kenya) once daily. The emerged adults were killed in a -20°C refrigerator, sorted by genera and all Anopheles adults stored in Eppendorf tubes (1.5 ml) at -71°C until they were identified morphologically using printed keys ²⁹ and molecularly using polymerase chain reaction (PCR) followed by gel-electrophoresis ³⁰ . Randomly selected mosquito samples were used for molecular identification. Polymerase chain reaction was implemented for the amplification of the ribosomal internal transcribed spacer (ITS2) gene using primers ³¹ . Positive controls of An. gambiae s.s. and An. arabiensis (from Mbita insectary colony) were analyzed with the samples from field. Extraction of genomic DNA was done for each mosquito separately using Tissue Kit (Quagen, GmbH Hilden, Germany). The PCR in a 10 µl (per sample) was prepared by mixing PCR mix of 2 µl of 5XHot Firepol Blended Master Mix (Ready to Load), primers (0.5 µM each), DNA template (2 µl) and nuclease-free water (5 µl). The thermal recycling conditions involved initial denaturation for 5 min at 95°C, after which 30 cycles of denaturation followed for 30 s at 94°C, annealing for 30 s at 50°C, extension for 30 s at 72°C and final extension for 5 min at 72°C. We used Kyratec Thermal Cycler (SC300T-R2, Australia) for the thermal reactions. Agarose gel-electrophoresis (2.0%) stained with 2 µl ethidium bromide against a 100 bp DNA ladder (Bioline, A Maridian Life Science ^@ Company, UK) and a positive control was conducted to identify the species.

Vegetation coverage, vegetation types and the dominant vegetation type were recorded separately for habitat edge and water surface. Habitat edge was defined as the area along the waterline, approximately 10 cm inside and/or outside the water. Vegetation coverage was estimated visually as the proportion of the habitats covered with vegetations and categorized as (1) 1–25% (2) 25–50% (3) 50–75% (4) 75–100%. Graminoid plants across the edge and inside water were recorded as Poaceae, Cyperaceae, Juncaceae, Typhaceae. The graminoid plants were identified to family using morphology of their leaves (two or three-ranked; open or closed sheaths), and their stem type (three-sided or round; hollow or solid) using Revuelta ³² . Furthermore, herbaceous (not woody and non-graminoid plants) were collectively recorded as forbs. The presence of water plants and algae in the aquatic habitats was also recorded. The percent coverage of water plants and algae on the water surface was visually determined as above. For each habitat, the dominant type of vegetation was identified and recorded. Full specimens of all dominant graminoid plants found in the aquatic habitats were collected and planted at icipe-TOC for further identification ^{33, 34} .

Generalised estimating equations (GEE) with Poisson distribution fitted to a log function and exchangeable correlation matrix were used to test for associations between biological and environmental factors and the abundance of early instar Anopheles larvae. The cluster ID in which habitats were located was included in the model as repeated measurement. A GEE model was also used to analyse associations between factors and the presence of early instar Anopheles larvae. Here we included the presence of early instar Anopheles larvae as dependent variable in the model with binomial distribution fitted to a logit function and exchangeable correlation matrix to analyse its association with biotic and abiotic factors of the habitats (independent variables). The presence and abundance of early instar Anopheles larvae (rather than eggs which are difficult to identify from field samples) were used as dependent variable as a proxy for oviposition. This is based on recent work confirming that early instar density correlates with the abundance of females selecting a habitat for oviposition ³⁵ . The statistical outputs were reported as incidence rate ratios (RR) for the abundance of the first instar larvae and odds ratios (OR) for the presence with their 95% confidence intervals (95% CI). R statistical software version 3.5.1 ³⁶ was used for the analyses.

This field survey was largely descriptive and observational and had no human study participants. Habitat surveys on privately owned lands were made after seeking consent from the landowners and were implemented in their presence.

A total of 110 aquatic habitats were identified during the survey ³⁷ . As expected, given the targeted areas within 300 metres of the lake shore, the most prevalent aquatic habitat types were swamps (65.5%, n=72) defined as permanent or semi-permanent water-logged sections of land with tall graminoid vegetation and/or floating plants ( Figure 2A). The water sources of these were largely groundwater supplemented by rainwater. Other habitats (see Figure 2B, Figure 3C, 3D and 3E) included ponds formerly used for breeding fish but abandoned at the survey time (11%, n=12), rainfed puddles (9%, n=10), drainages (9%, n=10) and artificial pits (5.5%, n=6). Given that all non-swamp habitats were few in number, they were pooled for statistical analysis and the swamp habitats used as the reference group ( Figure 3). Early instar Anopheles larvae were found frequently during the survey in the habitat types: artificial holes (n=6, 100%), drainages (n=9, 90%), ponds (n=5, 42%), puddles (n=7, 70%) and swamps (n=61, 85%). The majority of these habitat types were characterized by possessing graminoid plants: graminoids dominated the vegetation in 100% of the swamps, in 83% of the ponds, in 80% of the puddles, in 70% of the drainages, and in 50% of the artificial pits.

All the swamp habitats were boarded by graminoid plants along the water edges and had a high surface coverage. Similarly, 84% (32/38) of non-swamp habitats had graminoids along their edges and 76% (29/38) had graminoids at their surfaces. Unexpectedly, swamp grasses were not the most frequently found graminoid plants in the survey. Representatives of the Cyperaceae family were found only in 39% of the aquatic habitats sampled. Among the Poaceae family, torpedo grass ( Panicum repens) and Bermuda grass ( Cynodon dactylon) were the dominant species ( Figure 4).

Of the surveyed habitats, 42 (38%) were found covered by P. repens along their edges and 47 (43%) of the habitats at their surfaces. Cynodon dactylon was found covering the habitats both along the edges in 35 (32%) habitats and surfaces in 25 (23%) habitats ( Table 1). Overall, graminoid plants dominated in 96 habitats whilst forbs dominated only in five habitats during the survey. Nine habitats had no vegetations at their surface and five of them were colonized by early instar Anopheles larvae. We found water plants in 26 (24%) out of the 110 habitats and most (n=20, 77%) of them in swamp habitats. Filamentous algae were recorded in 21 habitats. Contrary to our hypothesis, there was no significant association between the presence or abundance of early instar Anopheles larvae and the dominant graminoid plant present in a habitat ( Table 1).

A total of 14,145 early and late instar Anopheles larvae and 402 pupae were collected. Out of those, 4,650 emerged into adults and were morphologically identified ( Table 2). Anopheles gambaie s.l. represented 96% of all Anopheles specimen collected. Molecular identification was done for a random sample of 10% of the An. gambiae s.l. (n=480) and revealed 100% An. arabiensis.

Anopheles coustani, An. rufipes and An. maculipalpis were found only in aquatic habitats covered with graminoid plants, whereas An. arabiensis, An. ziemanni and An. pharoensis were found in both habitats with and without graminoid plants. These six species of Anopheles mosquitoes were recorded in swamp habitats. All of these Anopheles species were found in aquatic habitats with dense graminoid vegetation (50–100%) ( Table 3). However, only three species of Anopheles mosquitoes ( An. arabiensis, An. ziemanni, and An. pharoensis) were collected in habitats sparsely (1–25%) covered by graminoids.

Aquatic habitats populated with Panicum repens and forbs at their edges had all the six Anopheles species identified ( Table 4a). Megaloprotachne albescens was found dominant in six out of 110 habitats surveyed but was found to have all the six different species of Anopheles ( Table 4b). Anopheles arabiensis, An. coustani, and An. pharoensis were coexisting with all the graminoid types and forbs found dominating along the surfaces of the habitats.

The presence of early instar Anopheles larvae in habitats was significantly and positively associated with swamp-type habitats (OR=22, 95%CI=6-86, P<0.001), presence of late instar Anopheles larvae (OR=359, CI=33-3941, P<0.001), and presence of Culex larvae (OR=17, 95%CI=3-107, P=0.002) ( Table 5). In habitats containing pupae the odds of finding early instar Anopheles larvae was lower (OR=0.08, CI=0.01-0.42, P=0.003) than habitats without pupae. Notably, the majority of habitats with Anopheles larvae were also well colonised by other invertebrates, many of which are considered predators of mosquitoes, such as Odonatan, Notonecta, and Coleoptera larvae. However, the presence of early instar Anopheles was only significantly and negatively associated with presence of tadpoles (OR=0.09, 0.01-0.53, P=0.003). Correspondingly, the abundance of early instar Anopheles larvae significantly decreased with the presence of tadpoles (RR=0.5, CI=0.2-0.9, P=0.03). Similarly, larval abundance was negatively associated with presence of Odonata (RR=0.5, CI=0.3-0.9, P=0.019) and presence of Coleoptera (RR=0.4, CI=0.2-0.8, P=0.004). There was no significant association between the presence and abundance of early instar Anopheles larvae and habitat size, habitat depth, distance to the nearest house, water pH, water turbidity, biofilm, debris, algae, and water plants.

Harvard Dataverse: Association between graminoids and the prevalence of immature malaria. https://doi.org/10.7910/DVN/NAT0YY ⁵⁷ .

This project contains the following underlying data:

Data are available under the terms of the Creative Commons Zero "No rights reserved" data waiver (CC0 1.0 Public domain dedication).