Different agroecosystems have similar Anopheles – Microsporidia MB infection dynamics in the Ahero region of Western Kenya: implications for symbiont control strategies

Microsporidia MB is a maternally inherited symbiont of several Anopheles species. In An. arabiensis, it is characterized by strong Plasmodium falciparum transmission blocking, stable vertical transmission, and no detectable virulence, making it a promising candidate for a symbiont-based malaria control intervention. Previous studies have shown that Microsporidia MB prevalence varies across regions of Kenya; however, the scale at which Microsporidia MB prevalence varies is unknown. Longitudinal sampling was conducted simultaneously in 10 sentinel houses per area within an irrigated rice scheme and an adjacent non-irrigated sugarcane-growing zone located 8 km away. Collected mosquitoes were screened for Microsporidia MB and Plasmodium via PCR, and their abundance and infection rates were correlated with remote sensing-derived microhabitat variables. Both areas harbored similar Anopheles species pools, dominated by An. arabiensis, An. gambiae, and An. funestus. Larval surveys mirrored this composition, though An. funestus was absent. The irrigated rice scheme had a seven-fold greater Anopheles mosquito abundance, indicating strong spatial heterogeneity in abundance driven by location. Microsporidia MB was detected across both areas in all three major species and in other uncharacterized Anopheles species (prevalence 2–15.6%). Prevalence did not vary significantly in adult mosquitoes between areas (Chi−square test, χ ² = 1.716, df = 1, p = 0.1902) or across sentinel houses (Ahero: Kruskal−Wallis H−test, H = 10.02, df = 9, p = 0.35; Miwani: H = 8.89, df = 9, p = 0.45), nor between larvae in Ahero and Miwani (Chi−square test, χ² = 0.01612, df = 1, p = 0.8990). The vegetation index (NDVI) positively influenced Microsporidia MB prevalence, while wind speed had a negative influence. Furthermore, the higher vector-producing rice scheme also had greater spatio-temporal detection of Plasmodium. These findings provide vital context on Microsporidia MB and its hosts; consistent prevalence across contrasting agroecosystems, despite a seven-fold variance in host population density, indicates that its transmission is resilient to local ecological shifts. While specific to this study’s spatial scale, this provides a crucial proof of concept that the efficiency of deploying the symbiont as a malaria control tool is therefore unlikely to be hindered by local-scale differences in landscape heterogeneity.

Malaria remains a major global public health challenge across Africa (World Health Organization, 2024). Anopheles mosquito vectors are highly efficient and adaptable, enabling sustained transmission across diverse ecological settings that are likely increasingly influenced by climate change (Caminade et al., 2014; Ferguson and Govella, 2023; Chapoterera et al., 2025; Megersa and Luo, 2025; van der Deure et al., 2025). The effectiveness of existing vector control tools is increasingly constrained; residual transmission persists, as mosquitoes adopt outdoor and daytime biting behaviors that reduce the impact of long-lasting insecticidal nets and indoor residual spraying, while resistance to all major classes of insecticides is now widespread (Ranson, 2017; Kleinschmidt et al., 2018; Sherrard-Smith et al., 2019). Combined with difficulties in achieving and maintaining high intervention coverage (Hawley et al., 2023; Oxborough et al., 2024), these limitations render current strategies only partially effective, highlighting the urgent need for complementary, biologically grounded vector control approaches that can suppress malaria transmission without reliance on chemical insecticides.

One such promising innovation is Microsporidia MB, a vertically transmitted endosymbiont that prevents host mosquitoes from transmitting the malaria parasite, Plasmodium (Herren et al., 2020; Bukhari et al., 2022). Although initially discovered in An. arabiensis in Kenya (Herren et al., 2020), the endosymbiont has been found to occur naturally in several other important malaria vectors across geographical regions (Akorli et al., 2021, 2024; Nattoh et al., 2023; Ahouandjinou et al., 2024; Moustapha et al., 2024, 2025; Parry et al., 2025; Tchigossou et al., 2025), indicating the potential for broad utility if developed into a malaria control strategy. Since Microsporidia MB is naturally occurring, this type of strategy is likely to enjoy high community acceptability, as demonstrated in pilot studies conducted in western Kenya (Bukhari et al., 2024). One of the major knowledge gaps relates to how different environments impact Microsporidia MB infection dynamics (Mfangnia et al., 2025). We have previously found that across regions in Kenya that are over 100 km apart, Microsporidia MB prevalence varies between 1% and 9% in An. arabiensis (Herren et al., 2020), whereas a lower variation of 0.9-2.2% has been observed for An. gambiae s.s. in sites within approximately 15 km of each other (Nattoh et al., 2023). However, the scale at which Microsporidia MB prevalence varies is not known. While differences in Microsporidia MB dynamics at the regional level can be incorporated into large-scale deployment strategies, a very high level of variation at a smaller scale within areas could hinder the efficacy of a Microsporidia MB- based malaria control approach. In addition, understanding the drivers of any variation in prevalence is likely to contribute important knowledge that could help maximize the efficacy of Microsporidia MB- based malaria control.

Land use and agricultural practices play a pivotal role in shaping mosquito ecology (Stryker and Bomblies, 2012; Abigail Nana Afia, 2023). Agricultural expansion alters local hydrology, vegetation, and microclimate, creating breeding habitats and resting sites for disease vectors (Kweka et al., 2016; Abigail Nana Afia, 2023). The use of pesticides (Kibuthu et al., 2016), the crop type, and irrigation can influence mosquito species composition, abundance, and behavior, often enhancing conditions for malaria vectors. For instance, rice paddies are well-documented hotspots for Anopheles breeding, often supporting high vector densities and prolonged transmission seasons (Muturi et al., 2006, 2007; Kibret et al., 2019; Orondo et al., 2023). Similarly, sugarcane cultivation has been associated with microhabitats that support Anopheles mosquitoes (Wondwosen et al., 2018; Demissew et al., 2020), with irrigation of these areas resulting in increased mosquito abundance (Demissew et al., 2020). Such contrasting crop and irrigation practices may present unique scenarios in vector control, especially in regions where rice and sugarcane dominate the landscape, such as western Kenya.

Given these intersecting challenges, this study investigates the spatial and temporal dynamics of Anopheles species, Plasmodium, and Microsporidia MB infection patterns in two distinct yet adjacent agroecological zones in Western Kenya: an irrigated rice scheme and a non-irrigated sugarcane-growing region located 8 km away. The study integrates remote sensing and meteorological data to identify the critical environmental drivers shaping vector ecology and the dynamics of associated Microsporidia MB.

Adult mosquito sampling

A purposive longitudinal sampling of early morning indoor-resting mosquitoes was conducted from sentinel houses using a CDC backpack aspirator. A total of 10 sentinel houses were selected per region, in two distinct ecological areas: Ahero, a rice-irrigated scheme, and Miwani, a sugarcane-growing, non-irrigated area. Selection was based on accessibility, informed participant consent, and a deliberate spatial distribution to minimize overlap of at least 500 m between surrounding microhabitats around each house (Figure 1).

A total of 15 sampling rounds were completed for each household between May 2023 and February 2024. Sampling followed a target biweekly schedule, achieving a median interval of 14 days between visits (mean: 17 ± 8 days). To ensure data comparability despite minor logistical variations in timing, each round was executed as a synchronized cohort, with all households within a site visited within a consistent sampling window. These 15 rounds sufficiently captured the three main ecological seasons observed during the study period: the Long Rains transition, covered in Rounds 1–2 (May-June), Round 3 (July), and Rounds 4–5 (August 2023); the Short Rains, covered in Rounds 6–7 (September), Round 8 (October), Rounds 9–10 (October–November), Round 11 (November), and Round 12 (November–December 2023); and the Dry Season, represented by Round 13 (December–January), Round 14 (January), and Round 15 (January–February 2024).

Mosquitoes from each sentinel house per collection round were transported in 15 x 15 x 15 cm netted cages to a sorting facility at the International Centre of Insect Physiology and Ecology, Thomas Odhiambo Campus in Mbita Point, western Kenya. The mosquitoes were separated based on morphological characteristics to distinguish Anopheles from other genera. Female Anopheles were sectioned and separated into two parts: head and thorax (HT) and abdomen. These sections, in addition to individual male Anopheles, were stored as individual specimens in 1.5 mL microfuge tubes containing absolute ethanol at -80 °C until further processing.

Mosquito larval sampling

Larval sampling rounds were conducted in parallel with adult collections, with 13 successful rounds completed when suitable rainfed roadside pools were available within 1 km of sentinel households. Larvae from different pools in each region were pooled into a single regional sample per collection. A 350 mL standard dipper (BioQuip Products, Inc., California, USA) was used to retrieve mosquito larvae from the edges of sampled pools. Larvae were transported in pool water in 10 L clear plastic PET bottles to the sorting facility, where Anopheles larvae were identified and separated from Culicine larvae based on morphological keys (Pratt et al., 1969) and preserved as individual specimens under the same storage conditions as adult mosquitoes until further processing and molecular identification.

Molecular identification

DNA was extracted from each adult and larval specimen using a protein precipitation protocol (Puregene, Qiagen, Netherlands). Species identification within the Anopheles gambiae complex was performed using a diagnostic PCR assay targeting the intergenic spacer (IGS) region of rDNA, following the protocol set out by Scott et al. (1993). This assay employed the universal reverse primer UN (5′-GTGTGCCCCTTCCTCGATGT-3′), alongside the species-specific forward primers: GA for An. gambiae s.s. (5′-CTGGTTTGGTCGGCACGTTT-3′) and AR for An. arabiensis (5′-AAGTGTCCTTCTCCATCCTA-3′). Members of the An. funestus group were identified using an ITS2-based PCR assay described by Koekemoer et al. (2002), using primers FUNF (5′-TGAACTGCAGGACACATGAA-3′) and FUNR (5′-GTTGTGATAGCTGGTGTTC-3′). Specimens that did not amplify with either the An. gambiae complex or the An. funestus assays were designated as other Anopheles species.

All reactions were run in 10 μL final volumes consisting of 2 μL HOT FIREPol ® EvaGreen ® qPCR mix plus (ROX) (Solis Biodyne, Estonia; mix composition: HOT FIREPol ® DNA Polymerase, ROX dye, 12.5 mM magnesium chloride, dNTPs, 5× EvaGreen ® HRM buffer, and EvaGreen ® dye), 0.25 μL forward primer, 0.25 μL reverse primer, 6 μL nuclease-free PCR water, and 1.5 μL of template DNA. Reactions were run on a MIC qPCR cycler (Bio Molecular Systems, Australia) using the following thermal profile: initial denaturation at 95 °C for 15 min, followed by 35 cycles of denaturation at 95 °C for 30 s, annealing at 50 °C for 30 s, and extension at 72 °C for 40 s, with a final extension at 72 °C for 5 min. For the An. gambiae complex assay, amplicons were visualized by electrophoresis on a 1% agarose gel stained with SafeView™ Red Nucleic Acid Stain (NBS Biologicals Ltd., UK) and run in 1× TAE buffer. Species-specific bands were identified by size: An. gambiae s.s. (390 bp), An. arabiensis (315 bp).

Microsporidia MB and Plasmodium screening

The DNA from all specimens except the female Anopheles HT sections was screened for Microsporidia MB infection using the 18S rRNA gene primer rRNA gene primer MB18SF (5′-CGCCGGCCGTGAAAAATTTA-3′) and MB18SR (5′-CCTTGGACGTGGGAGCTATC-3′) (Herren et al., 2020). Relative quantification of Microsporidia MB was done through normalization against the host’s ribosomal S7 gene using the S7 gene using the S7F (5′-TCCTGGAGCTGGAGATGAAC-3′) and S7R (5′-GACGGGTCTGTACCTTCTGG-3′) (Liu et al., 2011) primer pair. In addition, the female Anopheles HT sections were used for the screening of infective stages of the malaria parasite Plasmodium, targeting the parasite’s cytochrome b gene using the primer pair F (5′-TGGTAGCACAAATCCTTTAGGG-3′) and R (5′-TGGTAAATTGACATCCAATCC-3′) (Xu et al., 2015). Similar reaction volumes and cycling conditions were used, with the exception of the annealing temperature, which was set at 62 °C for the three primers.

Environmental data acquisition and processing

Environmental covariates were extracted for each house location using Google Earth Engine (GEE) (Gorelick et al., 2017). The Normalized Difference Vegetation Index (NDVI) was calculated from the Sentinel-2 Level-2A Surface Reflectance product (COPERNICUS/S2_SR) at 10 m resolution within a 500 m radial buffer around each house. A custom cloud-masking routine prioritized QA60, SCL, and MSK_CLDPRB layers to exclude contaminated pixels. Daily meteorological data (June 2023 to February 2024) were extracted from ERA5-Land (Muñoz-Sabater et al., 2021) and CHIRPS (Funk et al., 2015) datasets, including temperature, precipitation, and humidity. Additional derived variables were calculated using standard formulas, including wind speed, computed as sqrt (u² + v²), where u and v represent the zonal (east-west) and meridional (north-south) wind components, respectively. Hargreaves potential evapotranspiration was estimated as ET 0 = 0.0023 × (Tmax − Tmin)^0.5 × (Tmean + 17.8) × Ra, where Tmax, Tmin, and Tmean are daily temperatures in °C, and Ra is extraterrestrial radiation in MJ/m²/day.

To align environmental signals with mosquito ecology, raw data were transformed into antecedent variables. NDVI values were retrieved as averages over a ±7-day window centered on each house’s acquisition date. All other meteorological predictors were derived as 14-day antecedent values preceding each collection. Specifically, average temperature, temperature range, relative humidity, and wind speed were averaged; rainfall was summarized as a median; and ET 0 was included as a cumulative total.

Data analysis

Spatial heterogeneity in Anopheles populations was assessed using standard ecological diversity metrics. Alpha diversity (Simpson and Shannon indices) quantified within-site species variation, while beta diversity (Bray-Curtis dissimilarity and PERMANOVA) compared species composition across and within areas.

Microsporidia MB prevalence and intensity across species and sexes were analyzed using non-parametric tests (Kruskal-Wallis for inter-house comparisons with Dunn’s post-hoc tests). Household-level stratification of Microsporidia MB infection dynamics was evaluated to account for fine-scale micro-environmental heterogeneity, in light of evidence that environmental factors such as temperature and humidity are implicated in shaping infection dynamics of this symbiont (Boanyah et al., 2025; Otieno et al., 2025). Overall site-level differences in Microsporidia MB prevalence were compared using Chi-squared and Fisher’s exact tests, with odds ratios used to quantify the effect sizes of location on infection status.

Temporal variations in Microsporidia MB prevalence and infection intensities were analyzed using Generalized Linear Mixed-Effects Models (GLMMs) in R via the glmmTMB package (Brooks et al., 2017). Prevalence was modeled using a binomial distribution (logit link), while infection intensity among MB-positive individuals was modeled using a Gamma distribution (log link). ‘Household ID’ was included as a random intercept in all models to account for repeated sampling within households. For each site, models incorporated collection round, host abundance, and environmental covariates (median rainfall, relative humidity, average temperature, NDVI, and wind speed) as fixed effects. The included covariates met criteria for statistical stability, with all variance inflation factors (VIF)< 10 and variance decomposition proportions (VDP)< 0.80; maximum VDP values were 0.49 (relative humidity) in Ahero and 0.38 (wind speed) in Miwani (Craney and Surles, 2002; Kim, 2019). In Ahero, a fully integrated model was fitted. In Miwani, where rounds with zero prevalence were more frequent, a partitioned approach was adopted: (i) an environmental model to identify ecological drivers, and (ii) a round-only model to evaluate overall temporal significance.

Statistical significance for infection prevalence was determined using Wald Type II chi-square tests (car package), and post-hoc comparisons were performed using marginal contrasts (emmeans) to compare round-specific adjusted predictions to the site-wide mean, with p -values adjusted for multiple comparisons using the False Discovery Rate (FDR) method. Biological effect sizes were expressed as Odds Ratios (OR) derived from the model-predicted prevalence relative to the study average. Model suitability was rigorously validated using the DHARMa diagnostic suite (Hartig, 2024); the Ahero prevalence model showed no significant zero inflation (p = 0.986) or overdispersion (p = 0.974). For the partitioned Miwani analysis, both the temporal and environmental driver models demonstrated excellent fit with the binomial distribution, showing no evidence of zero inflation (both p = 1.0) or overdispersion (temporal: p = 0.980; environmental: p = 0.966). Infection intensity (density) models were similarly validated using the Gamma distribution, with no significant overdispersion detected in Ahero (p = 0.194) or Miwani (p = 0.932).

Plasmodium infection rates were evaluated using normalized indices per house and weighted scores per collection period:

Species-specific Plasmodium infection proportions were also compared between areas using Fisher’s exact tests. Anopheles larval species composition across collections was evaluated using NMDS ordination of Bray-Curtis dissimilarity and PERMANOVA.

To identify the key environmental drivers of Anopheles abundance and Microsporidia MB prevalence, Bayesian hierarchical models were fitted using the combined dataset from both study areas. Temperature (average and range) was excluded from the final models due to high multicollinearity with Hargreaves potential evapotranspiration (ET 0; VIF > 14). ET 0 was retained as a composite thermal-hydrological indicator that integrates temperature and solar radiation. By capturing both thermal energy and atmospheric evaporative demand, and reflecting the influence of regional crop canopy architecture, ET 0 potentially provides a more ecologically meaningful measure of the conditions shaping mosquito habitats and crop-associated microclimates than raw temperature alone (Brenner et al., 1995; Tesfuhuney et al., 2013; Barnard and Bauerle, 2016; Banerjee and Linn, 2018). The modeled environmental covariates (NDVI, median rainfall, relative humidity, wind speed, and ET 0) exhibited low multicollinearity (VIF< 3) and were standardized (mean-centered and scaled by standard deviation) to facilitate the comparison of effect sizes. Location was included as a fixed predictor, with Ahero as the reference level, and Household ID as a random intercept to account for hierarchical clustering within sampling units.

Models were fitted using the brms package (Bürkner, 2017) in R with the cmdstanr backend (Gabry et al., 2025), utilizing weakly informative priors (the expected effect centered at 0 with a standard deviation of 0.5) (Gelman et al., 2008). Anopheles abundance was modeled via a negative binomial distribution, while Microsporidia MB prevalence was modeled as a binomial outcome. Effect magnitudes and relevance were interpreted using Rate Ratios (RR), Odds Ratios (OR), and 95% credible intervals (CrIs). Statistical significance and practical relevance were further assessed using the probability of direction (pd), the posterior probability of an effect being greater than zero (Pr > 0), and the Region of Practical Equivalence (ROPE), computed via the bayestestR (Makowski et al., 2019). All analyses were done in R v4.5.1 (R Core Team, 2025).

Anopheles species composition

A similar adult Anopheles species profile was recorded in both study areas, dominated by An. arabiensis (Ahero: 89%, n = 2258; Miwani: 76%, n = 266). This was followed by An. gambiae s.s. and An. funestus, with minor proportions of other Anopheles species comprising the remainder of the collections (Figure 2A).

Alpha diversity metrics revealed a more evenly distributed adult Anopheles assemblage in Miwani compared to Ahero, with lower single-species dominance (Simpson index, p < 0.05; Figure 2B, panel 1). Correspondingly, species richness and evenness were significantly higher in Miwani (Shannon index, p < 0.05; Figure 2B, panel 2).

Beta diversity analysis revealed pronounced spatial heterogeneity between regions, with households clustering strictly by site (Bray-Curtis R 2 = 0.70, p = 0.001; Figure 2C). Within regions, ordination patterns suggested moderate homogeneity in Ahero (eight houses clustered together) and higher dispersion in Miwani. However, these intra-regional differences were not statistically significant (Ahero: PERMANOVA F = 1.65, R 2 = 0.17, p = 0.212; Miwani: F = 0.96, R 2 = 0.11, p = 0.426; Supplementary Figure 1), and therefore the observed clustering may reflect subtle ecological variation rather than strong compositional divergence.

Microsporidia MB prevalence

Household collections in Ahero yielded a seven-fold higher number of Anopheles mosquitoes compared to Miwani (N = 2525 in Ahero vs. N = 349 in Miwani), with a correspondingly higher absolute number of Microsporidia MB- infected mosquitoes, predominantly An. arabiensis. In addition, at least one Microsporidia MB- infected Anopheles mosquito was found in all sentinel houses during the study except house 4 in Miwani, which had none (Figures 3A, B).

In females (all Anopheles species combined), Microsporidia MB prevalence was 11.9% (n = 189/1592) in Ahero compared to 12.3% (n = 25/204) in Miwani, whereas in males, prevalence was 14.3% (n = 133/933) and 7.6% (n = 36/145) for the two areas, respectively (Figure 3C, panel i).

At the species level, Microsporidia MB prevalence was highest in An. arabiensis females (12.5%, n = 204/1626) and males (15.8%, n = 142/898), notwithstanding a raw prevalence of 21.4% (n = 6/28) in other Anopheles species females, with significantly lower rates observed in An. gambiae s.s. and An. funestus (Figure 3C, panel ii).

Overall (sexes combined), Microsporidia MB prevalence was comparable between Ahero and Miwani across all species (Figure 3C, panel iii). Statistical analysis confirmed that there was no significant difference between the two sites (Chi-square test, χ ² = 1.716, df = 1, p = 0.1902). Similarly, statistical comparisons of Microsporidia MB prevalence by species and sex yielded no significant differences between areas (Supplementary Table 1).

In addition, there was no significant house-level variation in the prevalence of Microsporidia MB in Anopheles mosquitoes among sentinel houses in Ahero (Kruskal-Wallis H-test, H (9) = 8.89, p = 0.45; Figure 3D), nor among sentinel houses in Miwani (Kruskal-Wallis H-test, H (9) = 10.02, p = 0.35; Figure 3E).

In Ahero, significant temporal variation in Microsporidia MB prevalence was observed across collection rounds (Wald type II chi-square test, χ ² = 77.50, df = 14, p < 0.001; Figure 3F). Significant peaks in prevalence were identified in Round 5 (38.8%, Odds Ratio (OR) = 1.03, p < 0.001) and Round 13 (26.4%, OR = 4.30, p < 0.001) relative to the average prevalence for Ahero (12.8%), while a notable decline occurred in Round 11 (0.6%, OR = 0.16, p < 0.001) and Round 14 (3.9%, OR = 0.55, p = 0.014) (Figure 3F). These variations were positively associated with mosquito host abundance (Wald type II chi-square test, χ ² = 5.80, df = 1, p = 0.016), and seasonal variations in the relative humidity (Wald type II chi-square test, χ ²= 3.90, df = 1, p = 0.048). However, prevalence did not vary in response to seasonal fluctuations in rainfall, average temperature, wind speed, or vegetation greenness (NDVI) (Supplementary Table 2).

Similarly, in Miwani, significant temporal variation in Microsporidia MB prevalence was observed across collection rounds (Wald type II chi-square test, χ ² = 24.84, df = 14, p = 0.036; Figure 3G). GLMM analysis revealed a continuous predicted prevalence trend; for rounds where no infections were recorded (Rounds 2, 3, 9, 10, 11, and 14), the model predicted a low but non-zero background prevalence (ranging from 1.2% to 4.8%) by accounting for underlying environmental drivers. Significant peaks were noted in Round 13 (47.6%, OR = 7.9, p < 0.001) and Round 15 (24.1%, OR = 2.8, p = 0.016) relative to the overall Miwani prevalence (10.3%) (Figure 3G). These temporal fluctuations were significantly associated with mosquito host abundance (Estimate = 0.74, Wald type II chi-square test, χ ² = 8.32, df = 1, p = 0.004) and seasonal changes in NDVI (Estimate = 0.41, Wald type II chi-square test, χ ² = 4.75, df = 1, p = 0.029). However, prevalence remained stable across seasonal fluctuations in relative humidity, rainfall, average temperature, and wind speed (Supplementary Table 2).

Microsporidia MB infection intensities

Regardless of the area from which Anopheles mosquitoes were collected, Microsporidia MB infections exhibited similar ranges of infection densities (Mann-Whitney U-test, U = 5449, p > 0.05; Figure 4A). However, Microsporidia MB densities varied significantly between mosquito species (Kruskal-Wallis H-test, H (3) = 8.89, p = 0.03), with An. arabiensis exhibiting the highest densities and An. funestus the lowest. Both An. gambiae s.s. and other Anopheles species showed intermediate density levels that were not significantly different from either extreme (Figure 4B). In addition, no significant inter-house variation in Microsporidia MB densities was observed among sentinel houses in Ahero (Kruskal-Wallis H-test, H (9) = 10.8, p > 0.05; Figure 4C) or in Miwani (Kruskal-Wallis H-test, H (8) = 6.62, p > 0.05; Figure 4D).

In Ahero, Microsporidia MB infection densities varied significantly across the 15 collection rounds (Wald type II chi-square, χ 2 = 165.31, df = 14, p < 0.001). Relative to the site’s mean Microsporidia MB infection density (899.6), significant fold-change increases in infection density were recorded in Round 4 (7.56-fold, p < 0.001), Round 8 (4.92-fold, p<0.001), and Round 12 (3.57-fold, p < 0.001). No significant associations were detected between these variations and host abundance or any of the modeled environmental covariates (p > 0.05) (Supplementary Table 3).

Similarly, infection intensities also varied significantly in Miwani across collection rounds (Wald type II chi-square, χ 2 = 62.36, df = 8, p < 0.001), with a recorded grand mean density of 252.9. Relative to this mean, a significant fold increase in density was observed in Round 15 (3.41-fold, p = 0.009). This variation was not significantly associated with host abundance or any of the modeled environmental covariates. (Supplementary Table 3).

Plasmodium infection rates in Anopheles mosquitoes

Plasmodium infections were more frequently detected in Anopheles mosquitoes from Ahero than from Miwani, with 15 of the 17 total infections being recorded in Ahero. Positives were found in eight sentinel houses in Ahero (Figure 5A), compared to only one in Miwani (Figure 5B). Within Ahero, houses 2 and 4 exhibited multiple infections (more than one mosquito infected with Plasmodium) within single collection rounds, while houses 3, 4, and 8 recorded infections across multiple rounds, suggesting persistent transmission foci. These spatial patterns were supported by normalized detection indices, with houses 3, 4, and 8 in Ahero having the highest composite scores, whereas only house 1 in Miwani showed any infection signal (Supplementary Table 4).

Temporally, infections in Ahero were concentrated in the early sampling rounds, particularly Collection 1, which yielded the highest weighted score, followed by moderate detection in Collections 2–4 and 14 (Figure 5A). Miwani had minimal temporal variation, with only two positives recorded in Collection 2 and none thereafter (Figure 5B). These patterns were consistent with normalized temporal indices, which revealed concentrated infection signals in the initial sampling rounds in Ahero and minimal persistence in Miwani (Supplementary Table 5).

Across both areas, the majority of the Plasmodium infections (16/17) occurred in Anopheles mosquitoes that were not co-infected with Microsporidia MB, with a single coinfection observed in an An. arabiensis individual from Ahero. The majority of the Plasmodium infections were found in An. arabiensis (77%, n = 13) (Figure 5C). Notably, in An. arabiensis, the observed Plasmodium infection rate was lower in Microsporidia MB -infected females (0.49 %, n = 1/204) compared to uninfected females (0.84 %, n = 12/1422). There were no significant differences in infection between species from both locations (Fisher’s exact test, p > 0.05; Supplementary Table 6).

Anopheles larvae and Microsporidia MB infections

Among Anopheles larvae collected from both Ahero and Miwani, An. arabiensis was predominant in both areas, occurring at near equal proportions (Ahero: 71%, n = 198/282; Miwani: 70%, n = 157/226; Figures 6A, B). This was followed by An. gambiae s.s. (Ahero: 21%, n = 60; Miwani: 11%, n = 25). No An. funestus larvae were retrieved from any collections in either region. Larvae of other (unknown) Anopheline species comprised 9% (n = 23) in Ahero and 19% (n = 43) in Miwani (Figures 6A, B).

Fewer larvae were retrieved during the first five collections in Ahero (Figure 6C) and the first four collections in Miwani (Figure 6D), with a seasonal decline observed at Collection 9 in both areas. Species distribution varied by collection round, with the first two collections in Ahero dominated by other Anopheline species, while subsequent collections showed increasing dominance of An. arabiensis. Notably, An. gambiae s.s. was substantially present in the last four collections (Figure 6C). Similarly, in Miwani, other Anopheline species dominated the first four collections, followed by consistent dominance of An. arabiensis in the remaining rounds, except for Collection 12, where An. gambiae s.s. was most prevalent and also notably present in the two preceding collections (Figure 6D). These distinct collection-based species compositions were found to be significant in both Ahero (PERMANOVA, F = 7.2, R ² = 0.4, p < 0.05; Figure 6E) and Miwani (PERMANOVA, F = 3.86, R ² = 0.26, p = 0.04; Figure 6F).

The prevalence of Microsporidia MB did not differ significantly between larval species within either study region (Ahero: Fisher’s exact test, p = 0.134; Miwani: Fisher’s exact test, p = 0.110) (Figure 6G). Specific infection rates across taxa ranged from 1.7% to 12.5% in Ahero and 0% to 7.0% in Miwani (Figure 6H). Furthermore, with all species combined, no significant difference in infection rates was observed between the two study areas (Chi-square test, χ² = 0.01612, df = 1, p = 0.8990).

Habitat-specific drivers of Anopheles mosquito abundance and Microsporidia MB prevalence

Spatial patterns in the abundance of Anopheles mosquitoes and Microsporidia MB infections were evaluated using Bayesian hierarchical models. Compared to Ahero, households in Miwani experienced a highly significant 81% reduction in expected Anopheles counts (Rate Ratio [RR] = 0.19, 95% CrI: 0.16 – 0.24, ROPE = 0%). In contrast, the effect of location on Microsporidia MB prevalence was inconclusive (OR = 0.78, 95% CrI: 0.54 – 1.11, ROPE = 36.1%) (Figure 7). Random intercepts revealed robust house-to-house variation in mosquito abundance (SD = 0.21, 95% CrI: 0.08 – 0.43, ROPE = 2.8%), but variation in infection probability between houses was practically negligible (SD = 0.12, 95% CrI: 0.01 – 0.37, ROPE = 73.5%) (Supplementary Table 7).

Environmental drivers exerted a more robust influence on symbiont prevalence than on vector abundance. Microsporidia MB prevalence was most strongly associated with vegetation greenness and wind speed. A one standard deviation (SD) increase in NDVI (0.067 units) was linked to a 30% increase in infection odds (OR = 1.30, 95% CrI: 1.15 – 1.46, ROPE = 7.1%). Conversely, a one SD increase in wind speed (0.12 m/s) reduced infection odds by 24% (OR = 0.76, 95% CrI: 0.64 – 0.89, ROPE = 9.0%). While relative humidity and ET 0 showed high probabilities of a positive direction (pd > 99%), their practical relevance was more moderate, with 13.4% and 26.6% of their posterior distributions contained within the ROPE, respectively (Figure 7). Similarly, although median rainfall exhibited high probabilities of a negative effect (pd > 97.6%) for both Anopheles abundance and Microsporidia MB infection, the effect was practically negligible (ROPE = 48.7% and >75.6%, respectively). All other environmental effects on mosquito abundance were negligible, with their 95% credible intervals overlapping the null effect (Supplementary Table 7).

A relatively consistent structure of Anopheles mosquito species composition was observed in both Ahero and Miwani, with a predominance of Anopheles arabiensis and notable detections of Anopheles gambiae s.s. and Anopheles funestus. Together, these three species have been identified as the principal malaria vectors in East Africa (Okara et al., 2010; Msugupakulya et al., 2023; Kamau et al., 2024; Kweyamba et al., 2025). Despite this shared overall structure, significant differences emerged in the local ecological context: Miwani exhibited higher alpha diversity (richness and evenness), indicating a more balanced community, while Ahero was strongly dominated by An. arabiensis. Since there were no statistically significant intra-area (house-to-house) compositional differences in Anopheles species within either Ahero or Miwani, the observed strong inter-regional beta diversity (distinct clustering) is likely attributable to the environmental differences between the two study areas. This spatial heterogeneity is frequently documented in East Africa, where different land use patterns and proximity to water sources drive distinct vector species compositions over short distances (Imbahale et al., 2011; Platt et al., 2018; Debrah et al., 2021; Ondeto et al., 2022; Orondo et al., 2023; Zhou et al., 2024). These contrasting ecological patterns are highly suggestive of location-specific aspects, likely reflecting variations in microhabitat availability or anthropogenic influences, such as irrigation.

Unlike Anopheles species compositions, there were no systematic spatial variations in the prevalence of Microsporidia MB between the two areas or between sentinel houses in each region. In An. arabiensis, the overall prevalences of 13.8% and 12.8% in Ahero and Miwani, respectively, were within the range of commonly reported variability in wild Anopheles populations (Herren et al., 2020; Akorli et al., 2024, 2021; Moustapha et al., 2025; 2024; Nattoh et al., 2023; Tchigossou et al., 2025). Nonetheless, the prevalence estimates for the less abundant species, An. gambiae s.s., An. funestus, and other (unknown) Anopheline species, may be limited by low statistical power due to the relatively small sample sizes in this study. The detection of the endosymbiont in both sexes from either or both sites is a likely indicator of shared transmission mechanisms across infected Anopheles species, as both vertical and horizontal routes have been shown to result in transmission to both sexes (Herren et al., 2020; Nattoh et al., 2021, 2023; Maina et al., 2025).

However, significant temporal variation in infection prevalence collectively across all Anopheles mosquitoes was noted, suggesting an influence of transient ecological factors in Microsporidia MB transmission. Such seasonal fluctuations have been observed, for instance, with Microsporidia MB often peaking several weeks after heavy rainfall in wild Anopheles populations (Herren et al., 2020). In this study, seasonal variation in area-specific environmental parameters, such as relative humidity in Ahero and NDVI in Miwani, was identified as the primary driver of temporal variations in Microsporidia MB prevalence.

Interestingly, temporal fluctuations in Microsporidia MB prevalence were also positively associated with Anopheles host abundance in both study areas. This suggests a host density-dependent component to endosymbiont transmission, in which larger host populations likely facilitate increased horizontal or vertical transmission events through greater environmental contamination or higher encounter rates. For instance, larger lekking swarms, which are more likely to form during high host abundance, have been found to be more attractive to female An. gambiae and correlate with mating success (Diabaté et al., 2011). However, this relationship appears to be primarily temporal rather than spatial; despite Miwani exhibiting significantly lower overall mosquito abundance than Ahero, the baseline prevalence remained comparable between the two regions. This implies that while seasonal surges in Anopheles abundance drive Microsporidia MB transmission ‘peaks,’ both landscapes maintain sufficient host populations to support stable endosymbiont persistence.

Similarly, Microsporidia MB infection densities did not vary overall between the two locations or between sentinel houses within each location, indicating that once established, Microsporidia MB maintains relatively stable within-host densities across geographic space. However, a significant variation in infection densities was recorded between Anopheles species, with the highest density levels observed in An. arabiensis. This pattern may reflect species-specific physiological compatibility or ongoing physiological adaptation if the association with these species occurred more recently than with An. arabiensis. It was notable that both prevalence levels and infection intensities of Microsporidia MB were lower in An. funestus. This suggests that the infection dynamics in An. funestus differ from those in An. gambiae s.l., which warrants further investigation to determine potential significant differences with implications for malaria control.

In addition, there was significant temporal variation in Microsporidia MB infection densities collectively among infected Anopheles mosquitoes from Ahero, but not from Miwani, indicating that the influence of seasonal or environmental factors on both prevalence and infection densities is complex and context-dependent. Specifically, the environmental conditions that promote the transmission of the endosymbiont (affecting prevalence) may not identically support its optimal replication within the host (affecting density). Such divergent responses have been demonstrated for Microsporidia MB (Boanyah et al., 2025) and for Wolbachia (Hague et al., 2020). Notably, in Ahero, the significant peaks and troughs in prevalence did not align with those in infection density, indicating that the two metrics varied independently rather than in synchrony. Nevertheless, the observed variations in infection densities could not be linked to any of the modeled environmental covariates. This likely reflects infection densities being more closely tied to host–symbiont interactions than to external environmental factors, or governed by complex environmental processes not captured within the scope of this study.

The spatial and temporal distribution of Plasmodium -infected Anopheles mosquitoes indicated a significantly higher and sustained malaria risk in Ahero households compared with Miwani. Focal transmission clusters were inferred in Ahero, with specific sentinel houses showing recurrent Plasmodium detections in more than one collection round, indicating possible localized environmental or anthropogenic drivers (Zhou et al., 2007, 2024; Baidjoe et al., 2016; Platt et al., 2018; Mategula and Gichuki, 2023). Indeed, irrigation, which distinguishes the two areas, has been implicated in increased Anopheles exposure and malaria transmission (Beke et al., 2023; Zhou et al., 2024). In contrast, there were lower Plasmodium levels and minimal evidence of sustained transmission in Miwani. This spatial concentration and the temporal clustering observed during the early rounds of sampling align with known seasonal peaks in vector abundance in western Kenya (Nyawanda et al., 2024).

Importantly, our longitudinal data provide a rare field-based perspective on the interaction between Microsporidia MB and Plasmodium. Despite the prevalence of both organisms in high-risk zones, we observed a near absence of co-infection (n = 1). Specifically, in An. arabiensis, the Plasmodium infection rate was lower in Microsporidia MB -infected females (0.49%) than in uninfected individuals (0.84%). While the low frequency of natural infections precludes a definitive statistical correlation, this 1/17 co-infection ratio is consistent with the antagonistic interference phenotypes reported in laboratory settings (Herren et al., 2020). These findings suggest that the symbiont’s blocking potential is maintained across varying field conditions and vector densities, further supporting its feasibility as a control tool. Future large-scale longitudinal studies with increased sampling intensity will be essential to provide the statistical power necessary to formally quantify the strength of this antagonistic correlation under field conditions.

Beyond house-based adult collections, larval sampling from temporary rainfall-fed pools also revealed a consistent dominance of Anopheles arabiensis in both Ahero and Miwani. The absence of An. funestus larvae may reflect habitat-specific breeding preferences, as this species is known to favor vegetated aquatic habitats with specific physicochemical characteristics (Nambunga et al., 2020; Omondi et al., 2023). The larval community structure exhibited non-random temporal variation, with an early dominance of the other (unknown) Anopheline species, followed by the later emergence of An. gambiae s.s., a pattern consistent with seasonal shifts in habitat availability and species abundance (Imbahale et al., 2011; Hinne et al., 2021; Tsegaye et al., 2023). Importantly, Microsporidia MB prevalence was highest in An. arabiensis larvae, with infection rates slightly higher in Miwani than in Ahero. The consistency between larval and adult prevalence in Miwani and Ahero, together with the relatively lower prevalence rates in larvae, is indicative of transovarial and sexual transmission (Onchuru et al., 2024; Parry et al., 2025).

Bayesian hierarchical modeling provided robust support for spatial structuring in Anopheles abundance, with households in Miwani exhibiting a markedly lower expected mosquito count compared with Ahero. Proximity to larval habitats, as typified by the more permanent irrigation water bodies in Ahero, likely drives this abundance, as previously observed in western Kenya (McCann et al., 2017). In contrast, the effect of location on Microsporidia MB infection probability was inconclusive, suggesting that infection dynamics may be decoupled from regional mosquito abundance. Random intercepts for sentinel house ID captured residual house-level variation that was meaningful for Anopheles abundance but negligible for Microsporidia MB infection probability, suggesting the influence of unmeasured microenvironmental or anthropogenic factors on mosquito ecology, whereas the likelihood of an infected mosquito being sampled at a particular house is largely stochastic.

Among the environmental covariates tested, green vegetation (NDVI) within a 500 m radius around sentinel houses emerged as a strong positive predictor of Microsporidia MB infection. Conversely, increased wind speed was significantly associated with reduced infection odds. The two factors might influence infection prevalence through opposing modulation of behaviors essential for survival and transmission, with vegetation promoting conditions suitable for resting, foraging, swarming, and mating, thereby enhancing transmission of the symbiont, while increased wind speeds could interfere with these processes or even cause overdispersion of infected mosquitoes from nearby aquatic habitats. These findings contribute insights to the current understanding of factors influencing Microsporidia MB infection dynamics, including diet (Boanyah et al., 2024; Bilgo et al., 2025), mating characteristics (Maina et al., 2025), chemical properties of breeding sites (Tchigossou et al., 2025), and climatic factors (Herren et al., 2020; Bilgo et al., 2025; Boanyah et al., 2025; Otieno et al., 2025).

Given the close proximity (approximately 8 km) of the two study sites, movement of adult mosquitoes, potentially favoring dispersal from the high-abundance irrigated zone (Ahero) to Miwani, is likely. Such movement may indicate a direct transmission route of Microsporidia MB from Ahero, where it was first discovered (Herren et al., 2020), to the Miwani population, where the symbiont achieved comparable prevalence in both adults and larvae, indicating similar levels of establishment across both sites. Importantly, larval prevalence may provide a more accurate measure of environmental effects on Microsporidia MB dynamics, as larvae are unable to disperse between sites and therefore reflect local ecological conditions. Such dispersal in adults, combined with localized larval prevalence, highlights the immense potential of deploying Microsporidia MB as a malaria control strategy, where the expected impact can naturally surpass the target geographical sites and spread to neighboring sites, regardless of agroecological differences.

The study highlights agroecology as a key factor affecting the abundance of malaria vectors, which in turn impacts the relative risk of malaria transmission, as seen in the irrigated rice scheme, where Anopheles abundance was seven-fold higher, and Plasmodium detection was more frequent compared to the sugarcane-growing area. In addition, infection dynamics with Microsporidia MB were mainly decoupled from agroecological context, but were shaped by temporal seasonality and environmental variables, such as vegetation (NDVI) and wind speed, highlighting the importance of local environmental context over simple regional confines. Longitudinal sampling from sentinel houses revealed fine-scale, house-level variations in vector species that suggest that microhabitat structure, host availability or host density (human/animal), and specific household practices play a critical, independent role affecting vector distribution. While the extent of Microsporidia MB variation at small spatial scales remained largely unexplored, our results demonstrate that, at a local scale (~8 km), prevalence remains consistent across distinct agroecosystems. This finding addresses a key concern for symbiont-based malaria control, as the lack of extreme fine-scale variation suggests that deployment efficacy may not be significantly hindered by local ecological shifts. While these results are localized, they provide the first evidence that symbiont dynamics can be decoupled from agroecological land use, offering a promising baseline for wider-scale assessments.