Revisiting immunity vs. exposure in schistosomiasis: A mathematical modeling study of delayed concomitant immunity

Abstract

The relative contributions of exposure vs. acquired immunity to the epidemiology of human schistosomiasis has been long debated. While there is considerable evidence that humans acquire partial immunity to infection, age- and sex-related contact patterns with water bodies contaminated with infectious cercarial schistosome larvae also contribute to typical epidemiological profiles of infection. Here, we develop a novel schistosome transmission model that incorporates both partially protective “delayed concomitant” acquired immunity—stimulated by dying worms—and host age- and sex-dependent patterns of exposure. We use a contemporary Bayesian approach to fit the model to historical individual data on exposure to infectious cercaria, eggs per gram of feces, and immunoglobulin E antibodies specific to Schistosoma mansoni Tegumental-Allergen-Like protein 1 collected from a highly endemic community in Uganda, estimating the relative contributions of exposure and acquired immunity. We find that model variants incorporating or omitting delayed concomitant immunity describe equally well the age- and sex-specific immunoepidemiological patterns observed before intervention and 18 months after treatment. Over longer time horizons, we find that acquired immunity creates subtle differences in immunoepidemiological profiles during routine mass drug administration that may confer resilience against elimination. We discuss our findings in the broader context of the immunoepidemiology of schistosomiasis.

Introduction

Schistosomiasis is a parasitic neglected tropical disease of considerable public health importance. Schistosomiasis affects approximately 240 million people worldwide (1), causing chronic and acute morbidity, including anemia, wasting, and damage to intestinal or urogenital systems. The global burden of schistosomiasis was estimated in 2021 at 1.75 (95% uncertainty interval [UI] 1.04–2.99) million disability adjusted life years, with 12,900 deaths (95% UI 11,400–14,700), the majority of which occurred in sub-Saharan Africa (2). The World Health Organization (WHO) has targeted schistosomiasis for elimination as a public health problem (<1% prevalence of heavy-intensity infections) by 2030, principally using a strategy of annual or semi-annual mass drug administration (MDA) with praziquantel (1).

Schistosomes are transmitted indirectly via intermediate freshwater snail hosts; consequently, exposure is linked to contact with water bodies containing infectious cercarial larvae. Five decades ago, Warren (3) first noted the difficulty of distinguishing the contributions of “immunology or ecology” to the convex age-intensity profiles of schistosome infections (4). Repeated exposure may promote a partially protective immune response, causing peak infection intensity to occur in earlier childhood in areas of intense transmission due to the more rapid acquisition of immunity (“peak shift”) (5–8). However, childhood peaks in infection intensity are also likely driven by age-dependent contact rates with contaminated water bodies, which are often greater in children than adults (9, 10). Hence, the relative importance of acquired immunity (“immunology”) vs. exposure (“ecology”) to the epidemiology of schistosomiasis has remained elusive.

Resolving the contributions of immunity and exposure is critical for understanding the transmission dynamics of schistosomiasis and for modeling responses to interventions. Protective immunity increases the resilience of helminthiases to intervention, making it more difficult to achieve elimination targets (9, 11). Recent advances have indicated that immunoglobulin E (IgE) responses against Tegumental-Allergen-Like (TAL) proteins are likely a major component of so-called “delayed concomitant immunity” (12). By this proposed mechanism, protective immunological responses are mediated by the death of adult schistosomes which induce antibodies that are cross-reactive with TAL proteins present on invading cercariae (12, 13). Delayed concomitant immunity may also provide a strong passive vaccination effect following praziquantel treatment of heavily infected individuals, initially boosting immunity and enhancing protection against reinfection (14), but waning following subsequent treatments.

In recent years, schistosomiasis transmission models have been increasingly used for public health decision-making, including on the frequency and population coverage of MDA required to achieve the 2030 elimination goals (reviewed previously (15)). However, due to uncertainty on the relative contributions of exposure vs. immunity, these models have typically attributed age-infection patterns exclusively to exposure (e.g. (16), although see also Refs. (9–11, 17, 18)). Depending on the nature and strength of acquired immunity, this assumption may make current projections on the long-term effectiveness of MDA too optimistic. Improved understanding of acquired immunity in natural populations would also benefit the development of effective schistosomiasis vaccines (19, 20).

Here, we develop a novel mathematical transmission model to resolve the contributions of exposure and acquired immunity to the epidemiology of intestinal schistosomiasis using a unique immunoepidemiological dataset collected from 110 individuals aged between 7 and 50 years living on the shores of Lake Albert, Uganda, before the initiation of MDA (21, 22). We use data on water contact and malacological surveys (23, 24) to infer age- and sex-dependent patterns of cercarial exposure and estimate the protective effect of delayed concomitant immunity using individual-level parasitological data on infection intensity and immunological data on Schistosoma mansoni Tegumental-Allergen-Like protein 1 (SmTAL1) IgE antibodies. We compare model variants with and without protective acquired immunity to determine its importance in capturing observed immunoepidemiological profiles. We also explore how immunity affects transmission dynamics during MDA. We discuss our findings in the context of the longstanding debate on schistosome exposure vs. immunity and the broader immunoepidemiology of schistosomiasis.

Results

Cercarial exposure shows a similar convex-like pattern in males and females, peaking at 11 and 15 years, respectively, before gradually declining throughout adulthood (Fig. 1). The fit of the transmission model (Fig. 2) to preintervention immunoepidemiological data is shown in Fig. 3A and B (“Baseline”) for the model variant without protective immunity (“standard model”) and with protective immunity (“immunity model”). The immunity model shows a sharper peak in infection intensity (eggs per gram of feces [epg]) in school-aged children (SAC) and a steeper decline into older ages than the standard model (Fig. 3A, “Baseline”; see Fig. S2 for an overlaid plot). Accordingly, in older ages the rate of increase in SmTAL1-IgE optical density (OD) is lower for the immunity model than for the standard model (Fig. 3B, “Baseline”). The deviance information criterion (DIC) indicates the standard model to be a more parsimonious and adequate fit to the data than the more complex immunity model (standard model 50.94 [95% credible interval, crI, 50.67–51.44] and immunity model 51.12 [95% crI 50.73–51.59]).

Parameter posterior distributions from each model are summarized in Fig. 4 (see also Figs. S3 and S4, Figs. S5 and S6 for pairwise correlations, and Fig. S7 for posterior log likelihoods). The median SmTAL1-IgE half-life (12 ln(2)/σ, where σ is the rate of loss; Fig. 4) is 1.1 months for both models, with 95% crI of 1.0–1.6 months for the standard model and 1.0–1.5 months for the immunity model. For the immunity model, the strength of protection against cercarial infection establishment is highly uncertain (ω, posterior distribution uniform over prior range; Figs. 4 and S4) despite mild and statistically nonsignificant pairwise correlations with other fitted parameters (correlation coefficient range −0.15 to 0.11, Fig. S6). Some resolution was obtained for the human-to-snail population density parameter (N₁/N₂, greater posterior vs. prior densities for values over ∼0.3; Figs. 4, S3, and S4), but other parameter posterior estimates were largely indistinguishable from priors (Fig. 4).

Both model variants show good predictive performance against epg and SmTAL1-IgE OD values collected at 5 weeks (Fig. 3A and B, “5 weeks”) after praziquantel treatment and against epg collected additionally at 12 months (Fig. 3A, “12 months”), and 18 months (Fig. 3A, “18 months”) post-treatment (note that SmTAL1-IgE was only measured up to 5 weeks after treatment) (24). There was no difference in model performance by root mean squared error (RMSE) in capturing post-treatment epg (standard model, 255 [95% crI 47.3–617]; immunity model, 260 [95% crI 47.7–628]) or SmTAL1-IgE (standard model, 0.24 [95% crI 0.18–0.34]; immunity model, 0.23 [95% crI 0.17–0.32]) (Fig. S8).The long-term epg dynamics through 10 rounds of annual MDA—assuming 75% coverage of SAC (1) and a nominal 30% of adults (≥16 years old), yielding a mean coverage of 50.5% in line with Ugandan coverage surveys (26, 27)—are consistently lower for the immunity model compared to the standard model (Fig. 3C, “epg”), likely since infection intensities are more concentrated in SAC in the former (Fig. 3A, “Baseline”). SmTAL1-IgE OD dynamics are similar in both models until approximately 5 years after MDA initiation, after which antibodies decline faster in the immunity model (Fig. 3C, “SmTAL1-IgE (OD)”), reflecting a lower antigenic stimulus from lowered worm burdens (and, consequently, epg; Fig. 3C). For both models, SmTAL1-IgE ODs fall below preintervention levels (Fig. 3C). The immunity model projections are associated with wide uncertainty intervals, particularly for epg (Fig. 3C), reflecting uncertainty in both preintervention fits (chiefly in SAC; Fig. 3A, “Baseline”) and the strength of protection afforded by SmTAL1-IgEs (ω, Fig. 4). This translates into more uncertain prevalence dynamics for the immunity model, further compounded by uncertainty in the among-host worm overdispersion parameter (k, used to calculate prevalence; Fig. S9). Some immunity model simulations initially show stark epg decreases which thereafter remain persistently low (Fig. 3C “epg,” immunity model lower 95% crI). These are associated with low values of basic reproduction number (⁠$R0$⁠), human-to-snail population density (N₁/N₂), SmTAL1-IgE decay (σ), and immune constant (ϕ), and high values of immune protection (ω) (Fig. S10).

Discussion

The relative importance of age-dependent exposure vs. acquired immunity to typical convex age-intensity profiles is a major unresolved question in schistosomiasis epidemiology (3, 9–11, 30). Understanding the relative contributions of these two processes has important implications for transmission dynamics, responses to MDA, including the feasibility of achieving the WHO 2030 goals (1). Here, we have used a unique immunoepidemiological dataset (24)—which includes individual data on exposure to infectious cercariae (23)—combined with a mathematical transmission model to attempt to resolve the effects of delayed concomitant immunity and age-dependent exposure on the epidemiology of intestinal schistosomiasis in a highly endemic region of western Uganda. We found that the model variants incorporating or omitting acquired immunity captured equally well observed age profiles of infection intensity and SmTAL1-IgE antibodies before and up to 18 months after praziquantel treatment. It therefore appears that delayed concomitant immunity is not required to adequately describe the transmission of intestinal schistosomiasis in this highly endemic epidemiological setting and its protective effect is highly uncertain. However, we also stress that we have taken a conservative approach to parameter inference, that the protective effects of immunity may only become apparent over longer time horizons in populations subject to MDA, and that the proxy of immunity used is a necessary simplification of the proposed delayed concomitant mechanism (12). We also emphasize that these conclusions are based on findings from a single highly endemic setting, using data from 110 individuals. It is conceivable that a larger sample size would be required to detect acquired immunity if its effects are relatively weak. Therefore, acquired immunity should not be entirely discounted from future modeling efforts aimed at projecting the effectiveness of interventions.

Many previous mathematical modeling studies have considered the role of acquired immunity on the epidemiology of schistosomiasis, but these have generally focused on the theoretical implications of immunity for control and elimination efforts (9–11, 18). While some models have been calibrated against exposure and infection data (9–11), data have typically been lacking on antibody responses, limiting inference on the role of immunity (3). The key novelty of our study is the use of individual data on exposure, infection, and immunological responses which permits, in principle, the relative contributions of exposure and immunity to be disaggregated. We focused on delayed concomitant immunity (as opposed to other postulated mechanisms of acquired immunity (31)) to align with the available immunological data on SmTAL1-IgEs which comprise the primary antibody response following the death of adult worms (13, 14, 17, 32–35) and which have been previously associated with reduced rates of reinfection (e.g. (36) and references therein). Notwithstanding, SmTAL1-IgE is a proxy for a proposed immunity-generating process acting via TAL family members such as SmTAL3, SmTAL5, and SmTAL11, since SmTAL3/SmTAL5/SmTAL11 responders are known to be a subset of SmTAL1 responders (12, 13, 37). SmTAL3 and SmTAL11 are expressed predominantly in adult worms, and IgE specific to these antigens is cross-reactive with SmTAL5 (13) expressed on cercariae and early schistosomulae. Hence, while immunity is stimulated by worm death (i.e. “delayed”), the protective response is thought to be protective against invading stages (i.e. “concomitant”).

We found that most model parameters were only weakly identifiable from the data, such that many different parameter combinations could yield similar immunoepidemiological patterns that adequately captured the data. We deliberately chose a conservative approach to parameter inference by assigning vague priors to all unknown parameters and avoided harder assumptions associated with fixing values to uncertain model parameters. While this avoids potential bias, it also highlights the limitations and challenges of parameter inference using transmission models with multiple unknown or highly uncertain parameters (38). The data were most informative on the half-life of SmTAL1-IgE antibodies, with both model variants yielding a central estimate of 1.1 months and an uncertainty range of 1–1.6 months. Without replacement, IgE has a circulating half-life of 2 days, and a half-life in the skin, the proposed site of parasite killing, of 16–20 days (39). The half-life of IgE secreting plasma cells—which replace circulating IgE—is more contentious, spanning from months to years. Blockage of plasma cell production with the immunotherapeutic drug Dupilumab shows >80% loss of circulating IgE over 1 year (reviewed in Ref. (40)), a slower kinetic than found here. While no other studies have examined SmTAL1-IgE half-life, previous unpublished work found significantly lower titers of schistosome worm antigen-specific IgE 1 year after treatment of Kenyan SAC (37). Further, S. haematobium models developed to investigate antibody isotype switching (32, 33) estimated an initial antibody response from short-lived plasma cells of 5–46 days (33) and 3.2–32 days (32), while estimates of the longer-lived antibody response were 9–90 years (33) and 10 months–87 years (32). The difference in estimates between previous work and ours could reflect biological differences between schistosome species, the modeled mechanism of protective immunity, and other modeling assumptions (e.g. age-exposure patterns and worm life-span). While we do not find that delayed concomitant immunity is necessary to explain immunoepidemiological profiles at endemic equilibrium and 18 months post-treatment in this population, we cannot rule out its importance for the longer-term epidemiology and transmission dynamics of intestinal schistosomiasis.

A principle of helminth epidemiology is that acquired immunity increases the resilience of parasite populations to interventions via the release of immune-mediated density-dependent constraints (11, 41). Previous transmission dynamics modeling of schistosomiasis has demonstrated this explicitly, showing that the loss of acquired immunity during MDA increases rates of reinfection (9, 18). By contrast, our modeling of MDA shows that stronger declines in infection and decreased rates of reinfection occur when immunity is incorporated compared to when it is omitted. This occurs because of the protection against reinfection afforded by SmTAL1-IgE antibodies—which are transiently boosted following the death of adult schistosomes after treatment—but also because the accumulation of protective antibodies during childhood increases the aggregation of worms in SAC, the demographic with the highest coverage of MDA.

Past models have often assumed that cumulative live worms are the source of antigenic stimulation for acquired immunity against schistosomes (9, 11, 18) but a compelling body of evidence now suggests this role is instead fulfilled by cumulative exposure to dying worms (12–14, 17, 32–34). These different sources of immunostimulation will yield tangible differences in projected dynamics. Assuming a live-worm stimulus, reductions in worm burden caused by treatment will cause a faster decline in protective immunity leading to more rapid rates of reinfection (18). Assuming dead worms provide the stimulus, worms killed by treatment will boost the immune response, initially reducing rates of reinfection. However, in keeping with previous serological (42, 43) and modeling (17) studies, our model also highlights that the immunological boosting effect of praziquantel is transient, both immediately after treatment—due to the relatively short estimated half-life of SmTAL1-IgE—and longer-term due to reductions in infection. Thus, in the long term, the qualitative effect on transmission dynamics of different sources of immunostimulation will become similar as population immunity is diminished under both mechanisms. Indeed, premature cessation of MDA can result in a rapid rebound in infections, even “overshooting” pretreatment levels (17, 44).

Our approach to modeling delayed concomitant immunity is necessarily a simplification of the underlying immunological mechanisms and incurs limitations. First, it is possible that our strategy of leveraging the relationship between SmTAL1 and SmTAL3/5 gave insufficient resolution to detect an immune effect. Previous work from Lake Victoria, Uganda, found that of 89 individuals with measureable SmTAL1-IgE titers, only 43% (n = 38) were SmTAL3-IgE responders, and 19% (n = 17) SmTAL5-IgE responders (13, 37). If these proportions were similar for our dataset, it might be challenging to find a protective signal—particularly if it were weak. Second, we do not account for a possible counter-balancing effect of IgG4, which is thought to downregulate IgE by competitively binding to the same epitopes (45). Several studies have reported that the balance of IgE/IgG4 determines the resistance vs. susceptibility to reinfection (24, 45, 46). Finally, we use a deterministic model which gives a mean-field approximation of transmission dynamics. A stochastic individual-based model would permit individual variability in worm burdens and antibody responses to be captured; albeit this approach would incur considerable technical difficulties in fitting to data. Notwithstanding these limitations, we emphasize the significant utility of the modeling framework presented here, which could be adapted to fit other data or other proposed immune mechanisms.

Future research could seek to design alternative population sampling strategies to better distinguish the two hypotheses represented by the model variants. Since datasets on long-term TAL family IgE dynamics are lacking, it would be helpful to introduce longer-term immunological monitoring of endemic populations. Data collected from such monitoring could then be used in conjunction with longer-term longitudinally collected parasitological data—such as programmatic data collected during MDA—to better validate and compare the long-term model dynamics. Future modeling studies could also focus on other markers of acquired immunity, including the newly characterized SmTAL11, which has also been implicated in the generation of cross-reactive IgEs specific to SmTAL3 and SmTAL5 (12).

In conclusion, the contributions of exposure vs. immunity to schistosome epidemiology, first noted five decades ago (3), remain unresolved. This article illustrates that disentangling these two processes is challenging even with complete individual-level immunoepidemiological data. We found that model variants incorporating or omitting delayed concomitant immunity could describe equally well immunoepidemiological patterns of intestinal schistosomiasis before intervention and 18 months after MDA in a highly endemic community of western Uganda. Despite this unresolved uncertainty, our modeling also indicates that the effects of acquired immunity may become apparent after repeated years of MDA and thus immunity should not be discounted from future modeling projections on the effectiveness of schistosomiasis interventions and the feasibility of reaching the WHO 2030 elimination goals.

Materials and methods

Dataset and study population

Details of the data used in this study have been described previously (21–24). Briefly, a longitudinal reinfection study was conducted in Booma, a fishing village on the eastern shore of Lake Albert, Uganda, highly endemic for S. mansoni with an estimated community prevalence in 1998 of 94% (47). Parasitology, serology, and intensive water-contact observations for 12 h per day over a 10-month period were conducted, in conjunction with systematic snail surveys, between 1998 and 2000 (23). Baseline data were collected before the introduction of MDA and can, therefore, be assumed to reflect the epidemiology at endemic equilibrium. This is an advantage for the current analysis as it avoids the need to model explicitly the effect of interventions or other secular changes (e.g. environmental or ecological change) which would introduce additional complexity and uncertainty. Following the collection of baseline (pretreatment) data in May 1998, from July 1998 a randomly sampled subset of the community was treated twice, 2 weeks apart, with praziquantel at 40 mg/kg of body weight. S. mansoni eggs per gram of feces (epg) and SmTAL1-IgE OD were longitudinally measured at follow-up at 5 weeks, 12 months, and 18 months after the second treatment (24). The sample used in the current study comprises 110 individuals of the Alur ethnic group (the now dominant ethnic group in the Lake Albert region (48)) aged 7–50 years old from whom data were collected on cercarial exposure and a subset of 85 individuals (aged 7–47) with paired data on epg and SmTAL1-IgE OD (Table S1). At each sampling time point, individual epg data were calculated as the average of two Kato-Katz egg counts, from three stool samples, collected on consecutive days. Data on SmTAL1-IgE ODs were calculated from triplicate repeats using control serum to account for between-plate variation.

Ethics

The original study obtained informed consent from all participants and ethical clearance was granted by the Uganda National Council for Science and Technology (UNCST) (23, 24). The current analysis is in line with the approved study aim of determining the roles of exposure and immunity on levels of infection and associated morbidity.

Transmission dynamics model

We developed a mathematical model in R (49) (using packages Rcpp (50), RcppNumerial (51), and RcppEigen (52)) to describe age- and sex-dependent epg and SmTAL1-IgE antibody OD (Fig. 2; www.github.com/gcmilne/SchistoTransmissionModel). The model is based on the population-based deterministic structure originally presented by Anderson and May (4) and more recently in Neves et al. (53) but with the miracidial and cercarial schistosome life-stages compressed into two components of the basic reproduction number, $R0$⁠, and additional complexity introduced to model the production and loss of SmTAL1-IgE. Here, we give a brief overview of the model (Fig. 2) with a complete derivation given in the Supporting Information.

We model the mean number of schistosomes in i = 1…n worm compartments within human hosts of age a, sex s, and time t, W_i(a, s, t), the cumulative number of dead schistosomes D(a, s, t), and the average degree of acquired immunity, I(a, s, t), using a set of integro-partial differential equations,

Here, $N1/N2$ is the human-to-snail population density, $μ1$ and $μ2$ are the per capita mortality rates of schistosomes and infectious snails, respectively, ρ is the schistosome sex ratio (proportion female), and $ψ1(a,s)$ is a normalized exposure function, with shape matching observed age- and sex-dependent ceracial exposure scores (23) (Fig. 1). The function $Φ[W(a,s,t),k(t)]$ denotes the (monogamous) mating probability (54), where $k(t)$ is the overdispersion parameter of a negative binomial distribution which is assumed to adequately capture the distribution of schistosomes among human hosts (4, 55), and which is dynamically affected by MDA (aggregation is transiently increased following rounds of treatment) but is assumed constant for age and sex (56). The function $Θ[I(a,s,t),ω]∈(0,1)$ is the acquired immunity-dependent probability of cercarial establishment,

where parameter ω describes the strength of protection afforded by acquired immunity. Parameter σ is the rate of decay of acquired immunity, such that $12ln(2)/σ$ is the antibody half-life in months. Our approach to modeling acquired immunity aligns with previous methodologies (9, 18) but uses cumulative dead, not live, worms as the source of antigenic stimuli (13, 57).

Model calibration

We calibrated the model with data on epg and SmTAL1-IgE OD. The epg is given by,

where λ is the per worm fecundity of mated female schistosomes and $W(a,s,t)=∑inWi(a,s,t)$⁠, i.e. the per host total mean worm burden. Note that in the absence of strong evidence for density-dependent fecundity in S. mansoni (58), we assume, in line with previous modeling (59), that λ is independent of $W(a,s,t).$ The OD of SmTAL1-IgE antibodies is given by,

where ϕ is a proportionality constant (analogous to λ) that maps acquired immunity to observed OD.

Parameter estimation

We fitted two variants of the model: the “standard model” in which SmTAL1-IgE is a passive marker of past exposure and has no impact on the probability of cercarial establishment (i.e. $ω=0$⁠; Eq. 2, Fig. 2), and the “immunity model,” in which SmTAL1-IgE can reduce the probability of cercarial establishment (i.e. $ω∈(0,1)$⁠; Eq. 2, Fig. 2). We employed a Bayesian approach, sampling 100,000 parameter sets from a 6D (“standard model”) or 7D (“immunity model”) Latin hypercube, assuming uniform priors for each parameter (Fig. 4). For each parameter set, the model was run to endemic stability and the simulated age- and sex-specific epg (Eq. 3) and SmTAL1-IgE OD (Eq. 4) profiles extracted and matched to corresponding data (24). A likelihood function was derived that accounts for the dependence among pairs of epg and SmTAL1-IgE OD observations made from the 85 individuals sampled at preintervention endemic equilibrium (Supporting Information, Likelihood function section). Parameter posterior distributions were approximated by selecting parameter sets obtaining the top 1% of log likelihoods (25) (i.e. 1,000 parameter sets), with each set of unique parameter values giving rise to a distinct age- and sex-specific epg and SmTAL1-IgE profile.

Model comparison

We calculated the DIC for each of the two model variants to compare their goodness-of-fits (60). We calculated the median DIC and 95% crI for each model using 1,000 bootstrap replicates from the parameter posterior distribution, sampling without replacement. The model with the lowest DIC score was assumed to be the most parsimonious and adequate description of the data (Supporting Information, Comparing model performance section).

Model prediction

We compared the predictive performance of the two model variants against epg and SmTAL1-IgE OD data collected at 5 weeks, 12 months, and 18 months after MDA for each of the 1,000 posterior parameter sets estimated using the preintervention data. We modeled MDA as an instantaneous reduction in worm burden in the treated proportion of the population (59) with the magnitude defined by the product of the age-specific treatment coverage and praziquantel efficacy (Supporting Information, Simulating mass drug administration section). Worms killed by treatment transitioned to the dead worm state, causing treatment to transiently increase SmTAL1-IgE antibodies (Eqs. 1, 2, 4; in line with observational data (14, 35)). We fixed the efficacy of praziquantel to 95% (29) and, since the whole study cohort was treated (24), set the coverage to 100% across the range of minimum and maximum ages of individuals in the study (7 and 50 years, respectively) (24) and to 0% outside these ages. We quantified the predictive performance of the model variants against the data for each post-treatment time point (24) by calculating an age- and sex-matched RMSE statistic (61) (Supporting Information, Predictive performance section).

Dynamics through MDA

We compared the longer-term treatment dynamics of the model variants (via median posterior dynamics and 95% crI) by simulating 10 rounds of annual treatment. Praziquantel coverage was set to 75% in SAC (5–15 years old), reflecting the WHO 2020 target coverage for this demographic group (1), and 30% in adults (≥16 years old), providing a mean (population structure-weighted) coverage of 50.5% in individuals ≥5 years. These assumptions concorded with community-wide coverage surveys in Mayuge District, which estimated population coverage at 52.6% in 2013 (26) and 46.5% in 2016 (27), and demographic group-specific coverage at 70.7% in SAC and approximately 25% in individuals ≥20 years (27).

In a sub-analysis, we classified the immunity model simulations into iterations reaching the lowest 10% of epg values after ≥5 years of MDA and iterations in the remainder of the epg distribution. This allowed for the investigation of areas of parameter space associated with effective treatment and—where stubbornly low transmission was maintained—immunity-mediated resilience against elimination.

Acknowledgments

This work could not be realized without the friendly co-operation of the people of Buliisa District. We would also like to acknowledge VCD technicians, especially the late J. Kemijumbi, A. Wamboko, and D. Nionsaba, and A. Babyesiza and N. Okumu for their dedication in leading the water contact observation team at Lake Albert. We deeply appreciate the field team leadership by Drs F. Kazibwe and E.M. Tukahebwa. The IgE data were generated under the leadership of Prof. D.W. Dunne; SW thanks him for the wealth of immunity data that allowed this work. Dr A. Pinot de Moira is thanked for her tireless work in generating the cercarial exposure scores. Finally, we are most grateful for the late Prof. A.E. Butterworth, Dr J.H. Ouma, Prof. B.J. Vennervald, and A.J.C. Fulford, whose inputs were extremely valuable in shaping the original study design.

Supplementary Material

Supplementary material is available at PNAS Nexus online.

Funding

G.C.M., R.C.O., J.P.W., M.W., and S.W. were supported through the FibroScHot project which is part of the EDCTP2 programme supported by the European Union (RIA2017NIM-1842-FibroScHot). R.C.O. was initially funded by the UK Medical Research Council UK Doctoral Training Partnership (grant MR/K50127X/1). C.W. was supported by a Sir Henry Wellcome Postdoctoral Fellowship award from Wellcome (224190/Z/21/Z). C.W. and M.-G.B. acknowledge funding from the MRC Centre for Global Infectious Disease Analysis (MR/X020258/1), funded by the UK Medical Research Council (MRC). This UK-funded award is carried out in the frame of the Global Health EDCTP3 Joint Undertaking. M.-G.B. and M.W. also acknowledge funding from the Bill & Melinda Gates Foundation through the NTD Modelling Consortium (grant INV-030046).

Author Contributions

Conceptualization: R.C.O., M.-G.B., M.W., and S.W.; Investigation (data collection): R.C.O., N.B.K., and S.W.; Methodology: G.C.M., R.C.O., C.W., and M.W.; Formal analysis: G.C.M. and R.C.O.; Supervision: J.P.W., M.W., and S.W.; Funding Acquisition: J.P.W., M.W., and S.W.; Writing – Original Draft Preparation: G.C.M. and M.W.; Writing – Review and Editing: all authors.

Data Availability

The data used to fit the model in this publication have been deposited in a UK Data Service ReShare Repository. Age has been collapsed into 5-year age groups to improve participant anonymity. [dataset]* Oettle, Rebecca C and Pinot de Moira, Angela and Vennervald, Birgitte J and Dunne, David W and Kabatereine, Narcis B and Wilson, Shona (2024). Revisiting Immunity Versus Exposure in Schistosomiasis: A Mathematical Modelling Study of Delayed Concomitant Immunity, 1998–2000. [Data Collection]. Colchester, Essex: UK Data Service. 10.5255/UKDA-SN-857172. The model code is publicly available as an R package at www.github.com/gcmilne/SchistoTransmissionModel.

References

Author notes