Sampling sites

We surveyed 109 ponds monthly in six Tanzanian districts of the Lake Victoria watershed from 23 August 2021–27 July 2022 (Figure 1). These ponds are created or modified by village communities to increase year-round water availability for the purpose of human household use (“Kisima”), for cattle use (“Lambo”), or for longer term water storage dams with unspecified use (“Bwawa”). A small number of ephemeral rivers and streams (“Mto” or “Kijito”) were also included in the study. Many of these ponds dry completely for several months of the year (Figure 2A) or dramatically decrease in size (Figure 2B) in the dry season. All ponds were chosen with approval of, and surveys were conducted in collaboration with, local village leaders. A larger pilot study was conducted in 2020–2021 including 467 pond sites that were identified by local leaders as potential transmission sites across the six districts, based on frequent human and cattle use (31). In some cases, this included all ponds in the village and in other villages leaders identified 6–8 ponds. For the current study, only villages where schistosome infected snails were found were retained and up to five ponds were retained per village. All sampling was conducted with permission from the Medical Research Coordination Committee of the National Institute for Medical Research (NIMR) in Mwanza (ethics approval certificate number NIMR/HQ/R.8a/Vol.IX/3462).

The Lake Victoria watershed is typically characterized by short and long rainy seasons, Vuli (October-December) and Masika (March-May), respectively. However, a delayed Vuli commencing in December 2021 resulted in a combining of the two rainy periods in our sampling period (Figure 3A). Rainfall data (in mm per day) was obtained from the Mwanza weather station between 23 August 2021 and 27 July 2022 (32).

During monthly site visits, each pond was surveyed for maximum length and width perpendicular to maximum length in meters using a tape measure, and depth at center in meters using a pole and tape measure. Dry ponds for which these dimensions were 0 meters were noted as such. All ponds that were too deep in the center to measure were assigned a depth of 2 meters.

Snail surveying and collection

We conducted snail surveys in small rainfall catchment areas in Northern Tanzania which are occupied by Bulinus nasutus snails (22). This species is morphologically differentiated from other Bulinus species in Tanzania, B. africanus and B. globosus. Two researchers conducted time-constrained net sampling using metal mesh scoop nets to collect B. nasutus snails for 15 minutes (leading to a total of 30 minutes of surveying per pond per month). This provides a representation of the contact experience of a person standing in the water for 30 minutes. In addition, this method is as effective as quadrat methods in assessing snail population dynamics across long time periods (33). Researchers searched for snails across the area of the pond, with special focus on microhabitats (submerged and floating vegetation, and other floating objects).  Snails were placed in Nalgene containers in a cooler and brought back to the lab at the NIMR Mwanza Centre for cleaning, counting, and quantifying parasites (“shedding”).

Identifying and quantifying parasite shedding

Bulinus nasutus snails were shed for patent infections in individual 30 ml beakers with 25 ml bottled water for 24 hours in natural light conditions. Following this full day shed, beakers were examined under a dissecting microscope at 10–25x for the presence of cercariae (larval forms) of schistosome and non-schistosome trematodes. The cercariae of these two groups are distinguished by size, shape, and movement (34). Schistosoma haematobium cannot be morphologically distinguished from S. bovis or their hybrids (35), therefore we represented all these individuals as “schistosomes”. Non-schistosomes were overwhelmingly represented by xiphiodiocercariae. If the presence of trematodes was confirmed, cercarial intensity was quantified after staining with Lugol’s Iodine and homogenization by gentle pipetting. For schistosomes, if the estimated number of cercariae was below 200, all cercariae were counted. If the number was larger, a subsample of 18.5% of the beaker’s bottom area was counted and multiplied by 5.412 to extrapolate for the total area of the beaker. For non-schistosome trematodes, only the subsample approach was taken due to higher intensities being typical.

Identifying potential infected aestivators

We identified if snails carried infections through aestivation using the timing of infections relative to emergence from aestivation in desiccating ponds. The prepatent period that is necessary for infections to develop until cercariae release typically take 6–18 weeks in a laboratory setting for Schistosoma haematobium (36). Ponds need to refill for snails to revive from aestivation and be receptive to miracidia once the water has returned. This could take just a few days or several weeks from the onset of rain, depending on the size of the pond and where in the pond snails are aestivating. As a result, we conservatively infer that infections that were detected less than 60 days following the last dry survey were acquired before aestivation, with increased confidence in those detected <30 days after the last dry survey. While Bulinus snails aestivate in dry microhabitats within non-desiccating ponds when water conditions are not ideal, it was not possible to identify infected asetivators using our methodology as these ponds did not have any dry surveys as a time reference point.

Statistical analyses

In this study, we assessed transmission risk in four ways: (1) snail abundance, (2) the proportion of snails infected, (3) per capita release of parasite cercariae from sampled snails, and (4) the total parasitic cercariae observed from all sampled snails. Snail abundance is defined as the number of B. nasutus snails recovered per 30-minute pond survey. Thereafter, we assess the proportion of these snails that sheds schistosome or non-schistosome parasites within a 24-hour shedding window. In terms of cercarial production, we first looked at the average number of cercariae released by individual snails of each parasite group (per capita) and we also summed the total number of cercariae observed of each parasite group by the infected snails that we collected in each pond survey (total observed release).

We utilized Generalized Additive Mixed Models (GAMMs) in the R package mgcv to evaluate how several metrics changed as a function of time and binary pond status (desiccating or non-desiccating). Specifically, we ran individual analyses to test how the following dependent variables changed over the course of a year: water depth (Gamma distribution using depth+0.01cm and link=“log”), snail abundance (Quasipoisson distribution), and infection prevalence (binomial distributions) and per capita and total observed cercariae in pond surveys (Quasipoisson distributions) of the two parasite groups. GAMMs are effective at evaluating smoothed, non-linear relationships over time. Thus, in each analysis we represented the annual trend as a smooth term. For Gamma and binomial error distributions, we fit models with Restricted Maximum Likelihood (REML), whereas as for Quasipoisson models, we fit with Quasi-Penalized Likelihood (37). Our independent variable of time is defined as days from the first day of sampling (23 July 2021) ranging from 0 to 338. We fit models with continuous autoregressive-1 error structures to account for repeated measures, except for the prevalence and per capita cercariae models due to failed convergence. For all models, we fit the dynamics of non-desiccating waterbodies with a reference temporal smooth and tested for significantly different dynamics in desiccating waterbodies with a temporal difference smooth (37). Lastly, we included pond ID as a random effect in the GAMMs to account for nonindependence in the monthly repeated observations from these replicated sites. All GAMMs were set up with a similar structure to this schistosome prevalence model below. See script for variation per model type.

gamm(cbind(schisto_pos, schisto_neg)~ s(ContDate)+
s(ContDate, by=Aest) + s(Waterbody, bs="re"), family = binomial("logit"),
data=infected_ponds, method="REML")

We ran Generalized Linear Models (GLMs) with the R package glmmTMB with binomial error distributions on all snails collected from each pond to assess if cumulative yearly infection prevalence varied among the two ephemerality categories. We also ran similar binomial GLMs to assess if yearly transmission risk differs with intensity of ephemerality (% pond reduction in pond area in a year). We summed the total number of infected and uninfected snails per pond per year for each parasite group to evaluate cumulative yearly parasite transmission risk. We then calculated the surface area of each pond on each visit by assuming an elliptical shape (Surface area = 1/2length*1/2width*p ) and the percentage reduction in area of each pond ((max-min)/max*100) as a measure of ephemerality intensity.