Prevalence and distribution of schistosomiasis in human, livestock, and snail populations in northern Senegal: a One Health epidemiological study of a multi-host system

Summary Background Schistosomiasis is a neglected tropical disease of global medical and veterinary importance. As efforts to eliminate schistosomiasis as a public health problem and interrupt transmission gather momentum, the potential zoonotic risk posed by livestock Schistosoma species via viable hybridisation in sub-Saharan Africa have been largely overlooked. We aimed to investigate the prevalence, distribution, and multi-host, multiparasite transmission cycle of Haematobium group schistosomiasis in Senegal, West Africa. Methods In this epidemiological study, we carried out systematic surveys in definitive hosts (humans, cattle, sheep, and goats) and snail intermediate hosts, in 2016–18, in two areas of Northern Senegal: Richard Toll and Lac de Guiers, where transmission is perennial; and Barkedji and Linguère, where transmission is seasonal. The occurrence and distribution of Schistosoma species and hybrids were assessed by molecular analyses of parasitological specimens obtained from the different hosts. Children in the study villages aged 5–17 years and enrolled in school were selected from school registers. Adults (aged 18–78 years) were self-selecting volunteers. Livestock from the study villages in both areas were also randomly sampled, as were post-mortem samples from local abattoirs. Additionally, five malacological surveys of snail intermediate hosts were carried out at each site in open water sources used by the communities and their animals. Findings In May to August, 2016, we surveyed 375 children and 20 adults from Richard Toll and Lac de Guiers, and 201 children and 107 adults from Barkedji and Linguère; in October, 2017, to January, 2018, we surveyed 386 children and 88 adults from Richard Toll and Lac de Guiers, and 323 children and 85 adults from Barkedji and Linguère. In Richard Toll and Lac de Guiers the prevalence of urogenital schistosomiasis in children was estimated to be 87% (95% CI 80–95) in 2016 and 88% (82–95) in 2017–18. An estimated 63% (in 2016) and 72% (in 2017–18) of infected children were shedding Schistosoma haematobium–Schistosoma bovis hybrids. In adults in Richard Toll and Lac de Guiers, the prevalence of urogenital schistosomiasis was estimated to be 79% (52–97) in 2016 and 41% (30–54) in 2017–18, with 88% of infected samples containing S haematobium–S bovis hybrids. In Barkedji and Linguère the prevalence of urogenital schistosomiasis in children was estimated to be 30% (23–38) in 2016 and 42% (35–49) in 2017–18, with the proportion of infected children found to be shedding S haematobium–S bovis hybrid miracidia much lower than in Richard Toll and Lac de Guiers (11% in 2016 and 9% in 2017–18). In adults in Barkedji and Linguère, the prevalence of urogenital schistosomiasis was estimated to be 26% (17–36) in 2016 and 47% (34–60) in 2017–18, with 10% of infected samples containing S haematobium–S bovis hybrids. The prevalence of S bovis in the sympatric cattle population of Richard Toll and the Lac de Guiers was 92% (80–99), with S bovis also found in sheep (estimated prevalence 14% [5–31]) and goats (15% [5–33]). In Barkedji and Linguère the main schistosome species in livestock was Schistosoma curassoni, with an estimated prevalence of 73% (48–93) in sheep, 84% (61–98) in goats and 8% (2–24) in cattle. S haematobium–S bovis hybrids were not found in livestock. In Richard Toll and Lac de Guiers 35% of infected Bulinus spp snail intermediate hosts were found to be shedding S haematobium–S bovis hybrids (68% shedding S haematobium; 17% shedding S bovis); however, no snails were found to be shedding S haematobium hybrids in Barkedji and Linguère (29% shedding S haematobium; 71% shedding S curassoni). Interpretation Our findings suggest that hybrids originate in humans via zoonotic spillover from livestock populations, where schistosomiasis is co-endemic. Introgressive hybridisation, evolving host ranges, and wider ecosystem contexts could affect the transmission dynamics of schistosomiasis and other pathogens, demonstrating the need to consider control measures within a One Health framework. Funding Zoonoses and Emerging Livestock Systems programme (UK Biotechnology and Biological Sciences Research Council, UK Department for International Development, UK Economic and Social Research Council, UK Medical Research Council, UK Natural Environment Research Council, and UK Defence Science and Technology Laboratory).


S1. Miracidia hatching technique (MHT): human and livestock (adapted from Yu et al. 2007 1 ).
Faecal samples: faeces are mixed, weighed (15g for cattle; 5g for small ruminants; 30g or the whole sample for humans) and passed through a 400μm metal sieve into a plastic container (Schistosoma eggs pass through this mesh) while rinsing with de-chlorinated water. The filtrate is then passed through both parts of a Pitchford funnel, 2 while agitating and rinsing further. Any eggs present sink to the bottom of the Pitchford funnel and will then be dispensed into a specimen pot with additional de-chlorinated water and placed under light to facilitate sedimentation of any remaining organic material, and hatching of eggs into miracidia. Samples are then examined microscopically for the presence of miracidia. Tissue samples: tissue samples are first macerated and then processed in the same way as faeces.

S2. DNA extraction and PCR protocols.
Miracidia, cercariae and snail voucher collections were archived with their contextual data in the Schistosomiasis Collection at the Natural History Museum (SCAN). 3 DNA was extracted from adult worms using the QIAGEN DNeasy Blood & Tissue Kit (QIAGEN, Hilden, Germany), following manufacturer's instructions. A maximum of 81 (median 11; IQR 22.5) adult worms and eight to 16 miracidia or cercariae per individual host were analysed (or all if <eight available). This strategy was adopted as more robust estimates of genetic diversity within Schistosoma populations are found when the number of hosts sampled is increased, rather than the number of samples per host. 4 DNA amplifications were performed in an Applied Biosystems SimpliAmp ™ Thermal Cycler. The PCRs were carried out in a total volume of 25 μl including PuReTaq Ready-To-Go ™ PCR Beads (GE Healthcare UK Limited, UK), 0·4 μM of each primer and 3 µL of DNA template. Four µL of each PCR product were run on 2% GelRed ® agarose gel for 45 min at 120 V and visualized on a Synoptics U:Genius GelDoc system. Selected PCR products were sent to Eurofins Genomics (Cologne, Germany) for purification and sequencing using dilutions of the original PCR primers. Table S1. Categorisation of genotypes for the Schistosoma haematobium group using partial fragment of the mitochondrial cytochrome c oxidase subunit 1 (cox1 mtDNA) and the complete nuclear ribosomal DNA internal transcribed spacer (ITS rDNA). Definitive hosts from which miracidia of each category were isolated in this study (and other studies in West Africa) are indicated.

S3. Prevalence estimations and estimation of diagnostic test performance in livestock species: Bayesian approaches.
Relatively poor sensitivity of diagnostic tests for human schistosomiasis that rely on egg detection methods are well documented and can lead to biases in estimation of prevalence. 7,8 Bayesian approaches which allow incorporation of uncertainty around diagnostic test performance into adjusted estimation of prevalence have been described for schistosomiasis in human and non-human populations and are advantageous in enabling incorporation, as priors, of additional information regarding model parameters including test sensitivities and specificities. 9,10 Where one diagnostic test is used, the probability that an individual in a population tests positive for that diagnostic test, p(Test+), is given as: Where  is the true prevalence of infection that population, SeTest is the test sensitivity and SpTest is the test specificity. (Note: This framework was also used for estimating true prevalence in the human and snail surveys).
The Bayesian approach also allows incorporation of results from two or more diagnostic tests such as in the animal survey where the MHT and Kato-Katz (KK) tests have been used in parallel on a randomly selected subset of each animal population. However when diagnostic tests are based on measuring similar biological phenomena, such as is the case for the MHT and KK methods (which are both based on the presence of schistosome eggs in a single faeces sample), the outcomes of the tests for a given individual are likely to be correlated and failing to take this into account may lead to bias in estimates. 11,12 In this framework the probabilities of an individual testing positive (Test+) or negative (Test-) for two diagnostic tests (Test1, Test2), can be described by the model: This enables estimation of schistosomiasis prevalence whilst adjusting for conditional dependence between the two tests, where covDp and covDn are the covariances between two tests for disease positive and disease negative animals respectively. 11,13 S4. Estimation of sensitivity for diagnostic tests used in livestock: abattoir data. Due to lack of data from literature on the performance of diagnostic tests for schistosomiasis in livestock species, abattoir data was used to estimate the sensitivity of KK and MHT as diagnostic tests, and these estimates used to generate priors to be applied to estimation of prevalence using the live animal data. The sensitivity of a diagnostic test is the proportion of truly positive individuals that are identified as positive by the test. In the absence of a gold-standard diagnostic test for animals for use on abattoir specimens, a pseudo-gold standard was used, whereby any animal positive for worms or miracidia in any tissue or faeces was defined as positive (excluding any animals that did not have a minimum database of results from faeces sample, liver sample and inspection of the mesentery). Sensitivity for each test was then based on the proportion of animals positive on this pseudo-gold standard that were detected by the diagnostic test.
Allowing for the possibility of the post-mortem pseudo-gold standard not detecting all schistosome positive animals, the following adapted form of Model 2 was used to generate all sensitivity estimates for each diagnostic test: With PM+/-, SePM and SpPM representing post-mortem pseudo-gold standard positive/negative and sensitivity/specificity respectively. In this framework π can be considered the probability an animal is truly positive, although it is not used to estimate the prevalence due to biases in the abattoir populations. We identified evidence for a significant difference between the performance of the hatching test in cattle depending on infecting schistosome species, with the proportion of post-mortem positive animals that were MHT positive being higher for those infected with S. bovis than those infected with S. curassoni +/-S. bovis:S. curassoni hybrids (Fisher's Exact: p=0·01). Separate sensitivity estimates are therefore presented for cattle. Within sheep and goats, no significant difference in the proportion testing positive was found between those infected with Sc and those infected with S. bovis (Fisher's Exact: p=0·56 within sheep; p=0·43 within goats), therefore estimates of test sensitivity for the two small ruminant species are presented which can be considered valid for both genotypes. Furthermore, no significant difference was identified between the proportions of sheep or goats testing positive for each test (infected with S. bovis or S. curassoni), (Fisher's Exact: p=0·81 MHT; p=0·56 KK), so overall estimates of test performance for small ruminants (sheep and goats) are also presented, which can be considered valid for both species.

S5. Estimation of genotype prevalence.
Given that molecular data was not available for all positive individuals, of those where molecular data were available for each survey (Nm), the number testing positive for schistosome genotype i (ni) was used to estimate the probability (pi) of a positive individual being infected with that schistosome genotype: This was then combined with the adjusted estimation of true prevalence π (using Model 1 or Model 2 depending on dataset) within an adapted Bayesian framework the to give the estimation of prevalence for each genotype (πi) presented in Figure 2:

= * (Equation 2)
Results plotted in Figure 2 are given in tables S4 and S5.

S6. Priors used in Bayesian frameworks: prevalence estimation.
Prior distributions for test sensitivities and specificities are commonly described using a Beta distribution defined by two coefficients, α and β (~Beta(α,β)), as this provides a flexible means of modelling parameters ranging from zero to one. For all estimations of prevalence in the human population, snail population and living livestock populations, uninformative priors were used for the true prevalence (~Beta(1,1)). For the estimation of test sensitivity from the abattoir data, uninformative priors were also used (SeTest~Beta(1,1)). Given that the MHT, KK, cercarial shedding technique and the abattoir pseudo-gold standard are based on microscopic visualisation of clearly identifiable schistosome eggs, miracidia, cercariae or worms, specificity for all tests was set at 100%. The mean ( ) and standard deviation ( ) of test sensitivity estimates from the abattoir data (Table S1) were used to generate α and β coefficients of the β distribution for use as priors in estimation of the live animal prevalence, where: And: Given the differing profiles of schistosome genotypes found in the live cattle populations of the two sites (Table  3) priors for the diagnostic tests used in the live animal survey estimates of overall prevalence (Table S2) were based on estimates for S. bovis for cattle in Richard Toll/Lac de Guiers (RT) and S. curassoni for Barkedji (BK) respectively. For sheep and goats, the estimates of test sensitivities for small ruminants was used (Table S2).
Prior distributions for sensitivity of cercarial shedding in snails and of KK and urine filtration methods applied to the human population, were based on literature and expert opinion, with α and β parameters derived using the betaExpert function from the Prevalence package in R based on point estimate and 95% confidence intervals (Table S2). 14 Following Dendukuri and Joseph (2001), 10 in a Bayesian framework for estimating prevalence using two conditionally dependent tests (such as Model 2 using KK and MHT for livestock), ranges for covariances of disease positive and disease negative animals are specified in the following way: The feasibility range for covariance of disease positive animals was specified in this way as priors in the estimation of livestock prevalence using a uniform distribution. However given that specificity of tests was set at 100%, the framework used did not allow for estimation of the covariance for disease negative animals (covDn).    Table S5. Bayesian estimation of prevalence by schistosome genotype: livestock surveys (as plotted in Figure 2 main text).