Prevalence of infection as a predictor of multiple genotype infection frequency in parasites with multiple-host life cycle

Authors


Summary

  1. In nature, parasites commonly share hosts with other conspecific parasite genotypes. While adult parasites typically show aggregated distribution in their final hosts, aggregation of clonal parasite genotypes in intermediate hosts, such as those of trematodes in molluscs, is not generally known. However, infection of a host by multiple parasite genotypes has significant implications for evolution of virulence and host–parasite coevolution. Aggregated distribution of the clonal stages can increase host mortality and reduce larval output of each infecting genotype through interclonal competition, and therefore have significant implications for parasite epidemiology.
  2. The aim of this study was (i) to find out how common multiple genotype infections (MGIs) are in aquatic snails serving as intermediate hosts for different trematode species; (ii) to find out whether the prevalence of infection could be used to predict MGI frequencies and (iii) to use the relationship to infer whether MGIs aggregate in molluscan hosts.
  3. We determined the prevalence of trematode (Diplostomum pseudospathaceum) infections and the frequency of MGIs in snail (Lymnaea stagnalis) host populations as well as compiled corresponding literature data from a range of snail-trematode systems. We used Bayesian simulations to explore the relationship between prevalence of infection and MGI frequency in these data, and tested whether genotypes aggregate in snails by comparing the simulated relations with null model (Poisson and demographic Poisson) expectations.
  4. Our results show that MGIs are common in aquatic snails with up to 90% of the infected snails carrying MGIs. Parasite prevalence is a good predictor of MGI frequencies at a range of prevailing natural prevalences of infection (0–50%). The frequency of MGIs was higher than expected by both null models, indicating parasite aggregation at genotype level.
  5. These findings are in sharp contrast with the absence of multiple infections in snails at level of trematode species, suggesting that co-infections by multiple species and multiple genotypes of one species are controlled by different biological processes. Aggregation of MGIs in snail hosts appears to be widespread across different snail-trematode systems.

Introduction

Host individuals are frequently infected by several strains or genotypes of the same parasite species (Read & Taylor 2001). It is important to understand the ecological and evolutionary processes that determine the parasite distribution in the host populations for several reasons. For example, the coexisting parasite groups in the host effectively form interactive units against which the host defences are directed (Karvonen et al. 2012). Similarly, the population of reproductive parasite stages in a single host often is the reproducing unit that determines the genetic properties of the next generation (Criscione & Blouin 2005). Furthermore, in trophically transmitted parasites, the population of parasites in a single host is often transmitted together to the next host (Criscione & Blouin 2006) and thus is coupled epidemiologically. All of these processes may have significant group-level effects on parasite fitness and host defences that affect epidemiology, host–parasite coevolution and general health of the host populations. Before we can formulate and test specific hypotheses of these group-level effects, we need to estimate how likely it is for two, or several, parasite individuals (genotypes) to occupy the same host individual.

Under random expectations, parasite distribution in the host population should follow a Poisson distribution (Shaw, Grenfell & Dobson 1998). Risk of infection (density of parasites) should affect the frequency of multiple genotype infections (hereafter MGIs), assuming that already infected hosts can be re-infected by another parasite genotype and that all parasite genotypes are present in similar proportions. Hence, under Poisson expectations, assuming parasite individuals do not directly interact and hosts do not differ in susceptibility, the more there are infective stages in the environment, the higher should be the frequency of hosts that carry MGIs. In nature, practically all observations of adult macroparasite burdens in host populations show aggregated [higher frequency of co-infections than expected by the Poisson null model (PNM)] parasite distributions (Shaw & Dobson 1995). Aggregated distributions follow when host individuals differ in the risk of acquiring an infection because of ecological or demographic reasons, or when they differ in susceptibility (see Wilson et al. 2002). Significant variation in aggregation patterns has been reported among host–parasite systems, and even among host–parasite populations within a system (Shaw, Grenfell & Dobson 1998). Aggregation may shape direct interactions (antagonistic or synergistic) between parasites, which can further influence host–parasite dynamics.

Clonal stages of parasites reproducing in intermediate hosts represent a challenge for the study of parasite distributions, as genetic identification of individual parasites is needed to distinguish MGIs. Consequently, studies screening MGIs across replicated intermediate host populations are still rare. Understanding the distribution of clonal parasite stages among the hosts is especially interesting as the genotypes are likely to compete directly for limited host resources and because such competition should have negative effect on parasite fitness.

In this study, we focus on trematodes, which are parasites of vertebrates and have multiple-host life cycles involving also invertebrate intermediate hosts. Recent studies suggest that MGIs of trematodes parasitizing snail intermediate hosts are common (e.g. Minchella, Sollenberger & Desouza 1995; Dabo et al. 1997; Sire et al. 1999; Eppert et al. 2002; Rauch, Kalbe & Reusch 2005; Lagrue et al. 2007; Keeney et al. 2008), making these systems ideal for studies on the relationship between overall prevalence of infection (surrogate of parasite density in the environment) and frequency of MGIs. Recently, Keeney et al. (2008) suggested a positive relationship between parasite prevalence and frequency of MGIs in marine trematodes. It is, however, not known whether this is a common trend in trematodes or other macroparasites with a multiple-host life cycle. If the frequency of MGIs could be readily predicted from parasite prevalence, it would help in designing and performing MGI surveys, and facilitate the study of aggregation patterns and their ecological, epidemiological and evolutionary consequences.

Here, we test whether prevalence of infection can be used to estimate the frequency of MGIs in aquatic snail-trematode systems. To achieve this, we first tested whether the correlation between parasite prevalence and the frequency of MGIs is positive in our model system, natural populations of great pond snails (Lymnaea stagnalis Linnaeus, 1758) infected with the trematode Diplostomum pseudospathaceum Niewiadomska, 1984. We further investigated whether a similar positive correlation was generally observed in the published studies of MGI frequencies in other aquatic snail-trematode systems. Finally, we developed a Bayesian simulation-based approach to describe the general relationship between prevalence of infection and the frequency of MGIs in these two data sets. We then compared the simulation results to expected frequencies given by two null models (a standard Poisson model and a novel demographic Poisson model). The former model is commonly used in testing, whether parasites are aggregated within their hosts and the latter provides an extension to the Poisson model by taking into account the demographic structure of the host population. These null models use the Poisson process to calculate the expected frequency of MGIs at different parasite prevalences. Comparison of the data to these models allowed the estimation of genotype aggregation patterns. Finally, we discuss what these results imply to evolutionary ecology of host–parasite interactions.

Materials and methods

Parasite Life Cycle

Diplostomum pseudospathaceum reproduces sexually in the intestine of fish-eating birds, such as gulls and terns. Parasite eggs, each carrying a different parasite genotype, are released into water by bird faeces and hatch to free-swimming miracidia larvae that infect the first intermediate host, aquatic snails (mainly L. stagnalis in Finland; Louhi et al. 2010). Within the snail, the parasite multiplies asexually in sporocysts, producing many thousands of free-swimming clonal cercariae larvae that leave the snail and infect fish, which serve as the second intermediate hosts (several species of freshwater fish can be infected, Rellstab et al. 2011). In the fish, parasites migrate to the eye lenses and develop to metacercariae, but do not multiply. The complex life cycle of the parasite is completed when the bird host preys upon an infected fish.

Sampling

We collected L. stagnalis snails from seven locations in central Finland in July–August 2007 and 2009. We sampled three of these locations in both years. As the life span of L. stagnalis is c. 1 year (Noland & Carriker 1946) and each infection in a snail is genetically unique, the populations re-sampled in 2009 represented a new cohort of snails and a new set of parasite genotypes. Two of the sampling sites (Pynnölänniemi and Ämmänlahti) were located in the same drainage (Lake Konnevesi and Lake Liesvesi), whereas the other locations were situated in separate watersheds. Distance between the sampling sites varied from 1·5 to 300 km. During each sampling campaign, we collected all adult snails (shell length > 2 cm) that we encountered (Table 1).We sampled the snails at the same time of the year in all populations. Therefore, the average age of the snails was roughly the same across the populations. In the laboratory, we placed the snails individually in cups containing 2 dL of water. After 24 h, we determined the percentage of D. pseudospathaceum infected snails (prevalence of infection) by observing released cercariae under a light microscope. We did not consider undeveloped infections (sporocysts not releasing cercariae) and therefore the prevalence values represent minimum estimates. The proportion of undeveloped infections was most likely similar for all populations as all samples were taken within a short time interval during the peak period of cercarial release (see Karvonen et al. 2004). We picked 15 cercariae from 20 haphazardly chosen infected snails from each population (note that we obtained fewer infected snails from some of the populations) and stored the cercariae individually in 1·5-mL Eppendorf tubes at −20 °C.

Table 1. Overview of the sampling sites in Finland in 2007 and 2009
PopulationSampling dateCoordinatesnysmxzALLzINF
  1. Coordinates = WGS84 geographic coordinates; n = total number of collected snails; y = number of snails infected with D. pseudospathaceum; s = number of snails studied for multiple genotype infections; m = number of snails with multiple genotype infections; x = prevalence of D. pseudospathaceum infections (y/n); zALL= frequency of multiple genotype infections among all snails (x * zINF); zINF = frequency of multiple genotype infections among infected snails (m/s).

Huumojärvi 200714/08/200765°06′N, 26°08′E28310420100·370·1840·500
Kuivasjärvi 200714/08/200765°04′N, 25°29′E906620·070·0220·333
Peurunka 200706/08/200762°27′N, 25°51′E93111150·120·0540·455
Peurunka 200931/07/200962°27′N, 25°51′E141212040·150·0300·200
Pirtti-Herttu 200914/08/2009, 21/08/200962°19′N, 26°29′E168151530·090·0180·200
Konnevesi (Pynnölänniemi) 200723/07/2007, 25/07/200762°37′N, 26°22′E1083020110·280·1530·550
Vuojärvi 200706/08/2007, 08/08/200762°25′N, 25°56′E264432040·160·0330·200
Vuojärvi 200914/07/200962°25′N, 25°56′E195242040·120·0250·200
Liesvesi (Ämmänlahti) 200731/07/2007, 01/08/2007, 02/08/200762°36′N, 26°20′E68262090·380·1720·450
Liesvesi (Ämmänlahti) 200931/07/200962°36′N, 26°20′E71121260·170·0850·500

DNA Extraction and Microsatellite Analysis

We used microsatellite markers to distinguish between different multilocus genotypes of parasites (see Louhi et al. 2010 for details of the microsatellite analyses). We determined the genotypes of the parasites at the loci Diplo06, Diplo09 and Diplo23 (Reusch, Rauch & Kalbe 2004) that are highly polymorphic in Finnish D. pseudospathaceum parasites (Louhi et al. 2010). We classified the snails as being infected with multiple parasite genotypes if they carried more than one parasite genotype. We determined the frequency of MGIs among all snails (zi) in a snail population by zi = (mi/si)*(yi/ni), where mi is the number of snails with MGIs, si is the number of infected snails that were examined for MGIs (the term takes into account the fact that parasites of <20 snails were analysed in some populations; Table 1), yi is the total number of infected snails and ni the number of sampled snails (see Table 1). We estimated population genetic parameters using fstat 2.9.3.2 (Goudet 2001) and Genepop 4.0.10 (Raymond & Rousset 1995).

Compilation of Literature Data

To address the generality of our findings, we searched the literature for MGI frequencies in a range of snail-trematode systems. We included studies that analysed trematode genotypes from at least five infected snail individuals. However, we excluded studies that pooled snail data from two or more consecutive years, studies that did not provide information on parasite prevalence and studies in which parasite prevalence was > 50% (this was performed because the relationship between prevalence of infection and frequency of MGIs in such populations differed markedly from populations with a prevalence < 50%, see Bayesian simulations of the infection prevalence–MGI frequency relation). We calculated the zi as described in the section ‘DNA extraction and microsatellite analysis’.

Data Analysis

We compared the mean number of D. pseudospathaceum genotypes per snail among the sampling sites using a Kruskal–Wallis test. We examined the relationship between parasite prevalence and the frequency of MGIs in the field data and in the literature data by calculating Pearson correlation coefficients (r) between the logit-transformed variables. We used spss 16.0 for Windows (SPSS, Chicago, IL, USA) for all calculations.

Bayesian Simulation of the Infection Prevalence–MGI Frequency Relation

We applied Bayesian analysis to describe the relationship between parasite prevalence and frequency of MGIs. The approach was inspired by the analysis of toxicity data in Racine et al. (1986) and the model design was aided by the book of Gelman et al. (2004).

The preliminary correlation analyses (see 'Results') of the data indicated that the frequency of MGIs (z) increased as a function of observed prevalence of infection (x). The simplest model for the xizi relation in the whole host population (θi) was linear: θi α + βφi, where φi denotes the actual population prevalence, which is estimated by xi. As the probability θi varies between 0 and 1, we used a logistic transformation of the θ's as given by eqn 1 The same transformation was applied to φ's to achieve a better fit.

display math(eqn 1)

The basic idea of the Bayesian model was to consider all the φi–θi relations of this type, that is, all the possible combinations of (α, β), and to generate a probability distribution for them. The obtained posterior distribution for the linear coefficients (α, β) summarizes the available information about the relations that could have yielded the observed data. The posterior distribution allowed us to make statistical inference about the nature of the φi–θi relation. We sampled 10 000 random draws from the posterior distribution to examine the central posterior 95% interval for any given prevalence φ.

We simulated random draws from the joint posterior distribution for (φ, α, β) by a combination of direct sampling from the posterior distribution (in case of φ) and Monte Carlo Markov Chain sampling from the joint conditional posterior distribution for (α, β). We used the Metropolis-Hastings algorithm in the Markov Chain simulations and tested chain convergence with the Gelman–Rubin test (Gelman et al. 2004). We ran the simulations either using the field data or the combination of the field and literature data.

We restricted the analysis to snail populations where the prevalence of infection was <50%, because the three high-prevalence samples in the data set were deviant from the majority of the data and tended to increase the uncertainty of the analysis. The applied linear model may therefore not be an adequate approximation of the relationship when the parasite prevalence is high (>50%). The low number of such populations (three), however, did not allow us to estimate the exact shape of the relationship. We computed all Bayesian analyses with the open source statistical package R version 2.10.1 (R Development Core Team 2009). We give a detailed description of the method in Appendix S1, and the R code for the simulations in Appendix S2.

Poisson Null Model

In parasitology, the Poisson model is commonly used as a null model for parasite aggregation (e.g. Shaw, Grenfell & Dobson 1998) and the distribution of within-host parasite richness (Dobson 1990; e.g. Goater, Esch & Bush 1987). The PNM can be applied to calculate the expected frequency of MGIs in relation to parasite prevalence. Applied to our data, the model assumes that (i) frequency of the different parasite genotypes is equal; (ii) the genotypes are randomly distributed within a study site; (iii) all hosts are equally susceptible; (iv) there are no interactions among the genotypes that would influence their establishment in the snails; and (v) parasites do not generate new genotypes within individual host. For graphical comparison of the model with our data, we computed the expected frequency of MGIs as a function of parasite prevalence (the R code of the model is embedded in the mgi_V101 file in Appendix S2).

Demographic Poisson Model (Infection State Null Model)

We constructed an alternative null model to examine how inclusion of host infection state in the PNM would change the null expectations of parasite aggregation (the R code is available in Appendix S2). Essentially, the model represents a simple extension to the traditional PNM, which does not, for example, account for aggregation of parasite genotypes into some host individuals over time. Similar to the PNM, the demographic PNM also assumes equal exposure and susceptibility of the hosts.

In the model, host survival parameters that affect the infection state transitions consist of the survival rate of newly infected susceptible hosts after a simulation cycle (σs) and of the survival rate of previously infected hosts after a simulation cycle (σi). Symbol b denotes the birth–death rate of susceptible hosts and was held dynamic in the sense that it kept the number of susceptible hosts stable during each simulation cycle. The population vector for each simulation cycle n classifies hosts into three different states: susceptible I0(n), singly infected I1(n) and multiple infected I2(n), and hosts are allowed to move progressively from the ‘susceptible’ state towards the ‘multiple-infected’ state. For each cycle n, we denote the exposure risk of a host being infected by at least one new parasite genotype with p1 and the exposure risk of a host being infected by at least two new parasite genotypes with p2. Transitions between the infection states can be described using p1 and p2 as given by the eqns 1–2.

display math(eqn 2)
display math(eqn 3)
display math(eqn 4)

The exposure risk during a cycle is assumed to be Poisson distributed and is given by eqns 5 and 6.

display math(eqn 5)
display math(eqn 6)

Here the parameter λ represents host's mean exposure rate to new parasite genotypes.

By allowing the model to run until the population becomes stable, we obtain different prevalence values xi = (I1 + I2)/(I0 + I1 + I2) and a corresponding frequency of MGIs z1 I2/(I1 + I2 + I3) for each chosen λ. By varying λ, we can examine the expected relation between prevalence of infection and MGI frequency.

We report the results of one of the simulation trials (with parameter values σs = 0·95; σi = 0·9; minimum λ = 0·001 and maximum λ = 0·13) to compare the demographic PNM to the standard PNM. We estimated weekly survival rates of uninfected and infected snails based on the data from Potamopyrgus antipodarum snails infected with Microphallus sp. trematodes (Jokela et al. 1999), and thus they represent rough estimates. We chose minimum and maximum mean infection rates so that the prevalence of infection did not exceed 50%.

Comparison of the Bayesian Simulations to the Null Models

We compared the results from the Bayesian simulations with the expectations of the null models by plotting the observed data, the chosen null model and 250 random draws from the simulations into the same figure. The Bayesian approach allowed us to test whether the observed frequencies of MGIs in the snail populations were significantly different from the null expectations. The comparisons and visualizations were performed with the statistical package R.

Results

Prevalence of D. pseudospathaceum infections in our field surveys varied between 6·7% and 38·2% (Table 1). Microsatellite genotyping was highly reliable; the success rate (all loci scored per individual) varied between 94 and 99% depending on the population. In total, multilocus genotypes of 2394 individual cercariae were determined. All loci were highly variable with 16–36 alleles per locus, the total number of alleles being 86. The number of alleles in the populations ranged from 29 to 57. Allele numbers per locus were similar in all host populations (except for Kuivasjärvi, which had fewer parasite samples because of the low prevalence). Three snails (all from Ämmänlahti 2009 population) from a total of 164 analysed carried a parasite with the same multilocus genotype. As the probability of finding two different genotypes exhibiting the same multi-locus genotype was <10−5 owing to the highly polymorphic loci (Karvonen et al. 2012), these cases most likely represented contamination during laboratory procedures. We excluded these genotypes from the data leaving 248 genotypes for further analysis.

The genetic data revealed that 20–55% of the infected snails carried MGIs depending on the population (Table 1). Frequency of MGIs among all collected snails varied between 1·8% and 18·4% (Fig. 1a). Three snails had five parasite genotypes, which was the maximum observed. The mean number of parasite genotypes per infected snail varied between 1·2 and 2·1 (Table 2) and was significantly different among the sampling sites (Kruskal–Wallis test, = 17·38, d.f. = 9, = 0·043). In the literature, we found 13 snail populations that matched our search criteria. The majority of the data came from schistosome populations (Table 3). The other studies included Maritrema novaezealandensis, Coitoceacum parvum and D. pseudopathaceum. In the compiled data set, 5–90% of the infected snails harboured MGIs and frequency of MGIs among all collected snails varied between 0·04% and 36% (Fig. 1b).

Figure 1.

Frequency of multiple genotype infections in different snail populations (dots) at different levels of parasite prevalence. (a) Frequency of multiple genotype infections in Finnish Lymnaea stagnalis populations (Table 1). (b) Frequency of multiple genotype infections in literature data (Table 3).

Table 2. Distribution of the number of Diplostomum pseudospathaceum genotypes among infected freshwater snails (Lymnaea stagnalis) in 10 Finnish snail populations
 Number of parasite genotypes/infected snail
Population12345Mean ± SD
Huumojärvi 20071072011·75 ± 1·02
Kuivasjärvi 2007420001·33 ± 0·52
Peurunka 2007631101·73 ± 1·01
Peurunka 20091631001·25 ± 0·55
Pirtti-Herttu 20091230001·20 ± 0·41
Konnevesi (Pynnölänniemi) 2007992001·65 ± 0·67
Vuojärvi 20071631001·25 ± 0·55
Vuojärvi 20091640001·20 ± 0·41
Liesvesi (Ämmänlahti) 20071134111·90 ± 1·21
Liesvesi (Ämmänlahti) 2009630212·08 ± 1·44
Table 3. A synopsis of the multiple infection data collected from published studies
Parasite speciesSnail hostPopulationnysmxzALLzINFReference
  1. n = total number of collected snails; y = number of infected snails; s = number of snails analysed for the number of parasite genotypes; m = number of snails with multiple infections; x = prevalence of infected snails (y/n); zALL = frequency of multiple genotype infection among all snails (x * zINF); zINF = frequency of multiple genotype infection among infected snails (m/s).

Coitocaecum parvumPotamopyrgus antipodarumLake Waihola, New Zealand99117046120·170·0450·261Lagrue et al. (2007)
Diplostomum pseudospathaceumLymnaea stagnalisKleiner Plöner See, Germany50201090·400·3600·900Rauch, Kalbe & Reusch (2005)
Maritrema novaezealandensisZeacumantus subcarinatusTurnbull Bay, New Zealand407322120·087·5 × 10−30·095Keeney et al. (2008)
M. novaezealandensisZ. subcarinatusCompany Bay, New Zealand208392140·190·0360·190Keeney et al. (2008)
Schistosoma haematobiumBulinus truncatesBankoni 1, Mali289121210·043·5 × 10−30·083Dabo et al. (1997)
S. haematobiumB. truncatesBankoni 2, Mali1568840·050·0260·500Dabo et al. (1997)
S. haematobiumB. truncatesFarako 2, Mali3366620·026·0 × 10−30·333Dabo et al. (1997)
Schistosoma mansoniBiomphalaria glabrataCabana, Brazil236640·260·1740·667Minchella, Sollenberger & Desouza (1995)
S. mansoniB. glabrataBarreiro de Baixo, Brazil708840·110·0570·500Minchella, Sollenberger & Desouza (1995)
S. mansoniB. glabrataGuadeloupe, 1995, France5231111120·0023·8 × 10−40·182Sire et al. (1999)
S. mansoniB. glabrataGuadeloupe, 1996, France1321121220·0091·5 × 10−30·167Sire et al. (1999)
S. mansoniB. glabrataGuadeloupe, 1997, France420202010·052·4 × 10−30·050Sire et al. (1999)
S. mansoniB. glabrataDionisio, Brazil31151590·480·2900·600Eppert et al. (2002)

The correlation between the logit-transformed variables x and z indicated a strong positive relationship between both in the field (= 0·917; d.f. = 8; < 0·01) and in the literature data (= 0·937, d.f. = 11; < 0·01). In other words, the proportion of MGIs increased with the overall parasite prevalence. As the correlation between the logit-transformed variables x and z was significant, the model logit (θi) α + β logit (φi) was considered as an adequate description of the relationship. The visual comparison of the Bayesian simulations with the null models revealed that the φi–θi relation was significantly different from the neutral expectation of the PNM, as 99·9% of the simulated relations for the field data, and 100% of the simulated relations for the combined data were above the neutral expectations (Fig. 2). Thus, all snail populations harboured a significantly higher proportion of MGIs than expected by the PNM (Figs 2a and b). Combining field and literature data increased the certainty in estimation of the relation (see Fig. 2b). The inclusion of host infection state transitions and demographic parameters substantially increased the expected frequency of MGIs compared to the traditional PNM. As a result, the proportion of relations that were above the null expectations decreased to 75% for the field data and to 90% for the combined data (Figs 3a and b). However, all relations were above the expectations at < 30% prevalence.

Figure 2.

Simulated parasite prevalence and multiple genotype infection frequency relations compared to the Poisson null model (PNM). Light green lines show 250 random draws from the 10 000 simulations. Dashed red lines show the median and the 95% central interval of the posterior distribution, and dotted black line shows the point estimate given by the fitted linear model. The PNM relation is shown with a thick continuous black line and data from the snail populations with dots. (a) Data from the Finnish Lymnaea stagnalis populations (white dots). (b) Data from Finnish L. stagnalis populations (white dots) combined with literature data (yellow dots).

Figure 3.

Simulated parasite prevalence and multiple genotype infection frequency relations compared to the infection state null model. Light green lines show 250 random draws from the 10 000 simulations. Dashed red lines show the median and the 95% central interval of the posterior distribution, and dotted black line shows the point estimate given by the fitted linear model. The infection state null model relation is shown with a thick continuous black line and data from the snail populations with dots. (a) Data from the Finnish Lymnaea stagnalis populations (white dots). (b) Data from Finnish L stagnalis populations (white dots) combined with literature data (yellow dots).

Discussion

Co-infections of parasites in their host populations have significant implications for various biological processes such as evolution of virulence (e.g. van Baalen & Sabelis 1995; May & Nowak 1995; Frank 1996; de Roode et al. 2005; Bell et al. 2006; Choisy & de Roode 2010), parasite transmission dynamics (e.g. Taylor, Walliker & Read 1997) and epidemiology (e.g. Cox 2001). The distribution of parasites in a host population also determine the likelihood for the interactions between conspecific parasites (Poulin 2007), which may have consequences for parasite fitness, population dynamics (Jaenike 1996) and ecology. For example, while sharing host resources with another individual might be a constraint for growth and fecundity (Keymer, Crompton & Singhvi 1983; Jones, Breeze & Kusel 1989; Shostak & Scott 1993), it may increase mating opportunities at the sexual stage (Shaw & Dobson 1995) and in some cases, induce host immunosuppression (Hudson et al. 1976; Read & Taylor 2001; Lello et al. 2004) or increase infection success (Karvonen et al. 2012).

In the present study, we examined the relationship between MGI frequency and prevalence of infection to investigate how accurately parasite prevalence can be used to predict MGI frequencies in natural snail-trematode populations. For this purpose, we developed a Bayesian model to estimate the expected number of snails carrying MGIs in any given snail population where the overall prevalence of infection is known. We then used the results of the Bayesian analysis to compare the observed frequencies of MGIs to predictions given by two null models. We suggest that our approach provides powerful opportunities to test the alternative processes that may affect the distribution of parasites in their host populations.

The strength of the Bayesian analysis is that it flexibly incorporates variation in sample size and gives more weight to those estimates that have higher statistical confidence (i.e. larger sample). It also allows the combination of information from separate data points (populations) to estimate the shape of the relationship between variables. This is useful in the study of natural infections patterns as the number of sampled hosts and the number of analysed parasites often differ. Furthermore, observed MGI frequencies can be compared to any desired null model or to an experimental dose–response curve, which allows flexible hypothesis testing.

In our study system, the frequency of MGIs in the snails (L. stagnalis) was positively associated with parasite (D. pseudospathaceum) prevalence. The pattern was consistent also across different published studies on aquatic snail-trematode systems, despite of the differences in temporal and spatial sampling designs and host–parasite systems that the studies covered. We found that the observed MGI frequencies were significantly higher than predicted by the PNM at prevalence values < 50% and also significantly higher than predicted by the demographic PNM, when prevalence values were < 30%. These results indicate aggregation of parasite genotypes to few host individuals, a pattern that was similar for the field and for the literature data. The demographic PNM predicted higher frequencies of MGIs than the PNM as it incorporated a heterogeneous age distribution of the host population, which contributes to the likelihood of MGIs. However, the demographic PNM does not account for the effects of unequal exposure or unequal susceptibility of the host individuals, which would be logical alternative factors that should be examined. Unfortunately, the empirical data required for the estimation of such parameters are scarce, emphasizing the need for detailed population studies of host–parasite interactions. Deviance from the null expectations could be caused by several different biological processes and we discuss some of them later.

The frequency of MGIs is likely to be affected by ecological factors that generate heterogeneity in exposure risk of host individuals. Exposure risk, on the other hand, depends on the abundance of infected definitive vertebrate hosts (Fredensborg, Mouritsen & Poulin 2006), which determines the frequency at which the snails become exposed and infected with new parasite genotypes (Keeney et al. 2008). Prevalence of trematode infections in natural snail populations has been shown to be spatially variable even within the host populations (e.g. Sousa 1993), which suggests that there is variation in the local risk of infection and differences in local parasite transmission dynamics (e.g. Kuris & Lafferty 1994; Lively & Jokela 1996). The relative size of the sampling site, water movements and patterns of final host mobility are all important factors determining whether infective stages are heterogeneously or evenly distributed within a study site (Eppert et al. 2002). For example, local concentration of the definitive hosts can create infection ‘hot spots’, resulting in high infection rate within the spot, but low infection rate elsewhere (Jokela & Lively 1995). Because Diplostomum eggs are disseminated by avian definitive hosts, a local bird colony could in theory create a hot spot structure even within one lake. However, this was unlikely in our study as we did not collect snail samples close to bird colonies. Also, the larval stages (miracidia) that infect the snails can disperse by active swimming or with water currents (Steinauer et al. 2009), despite of their short lifetime. In addition, L. stagnalis snails can disperse short distances by floating in the water surface. These factors were likely to reduce heterogeneity in the exposure risk within the snail populations, although the exact estimation of the exposure risk would require controlled field experiments.

Another explanation for the aggregated parasite distributions are genetic factors that generate heterogeneity in host susceptibility (see Shaw, Grenfell & Dobson 1998; Wilson et al. 2002). If host individuals differ in the array of parasite genotypes that they succeed to deflect after the contact then highly susceptible hosts will carry MGIs more frequently than expected by chance. Recently variation in susceptibility of individual hosts was shown to enhance the frequency of MGIs in microparasites (van der Werf et al. 2011). Similar processes could operate also in our study system where populations of L. stagnalis show significant among-family variation in immune defense traits, although the specific causes for such a variation remain unknown (Seppälä & Jokela 2010). It is also possible that specific parasite strategies in host finding could play a role in generating aggregation, for example, if some host individuals attract more infections through chemical perception by the infective stages. This has not been studied in detail in the present system, but such behaviours of larvae infecting their snail hosts are well known among other trematode species (reviewed by Sukhdeo & Sukhdeo 2004).

A large body of evidence suggests that the interactions between co-infecting trematode species in snails are primarily antagonistic (e.g. Sousa 1992; Kuris & Lafferty 1994; Hechinger, Wood & Kuris 2011). However, interactions among co-infecting genotypes of one parasite species have received less attention. Adult parasites can clearly benefit from sharing the final host with potential conspecific mates as sexual reproduction enables genetic recombination and production of new parasite genotypes. Moreover, the competition among parasite genotypes is likely to be less severe in parasite stages that use the host mainly as a transmission vehicle (Karvonen et al. 2012). However, within the snails competition is likely to be strong among parasites genotypes as they multiply rapidly and compete directly for host resources (Karvonen et al. 2012). Thus, the aggregation of MGIs across the studied snail populations seems to strike against the fair assumption that parasite genotypes compete for host resources and therefore the interactions should be antagonistic. Our recent experimental evidence actually supports antagonistic interactions between D. pseudospathaceum genotypes within snails. Cercariae originating from co-infected snails have a higher infection success in the next host (fish), but the net result of co-infection is negative because of decreased rate of cercarial production in the multiple-infected snails (Karvonen et al. 2012). This should select against MGIs in the snail. However, it is possible that variation in individual host susceptibility and ecological factors that lead to small-scale differences in the infection risk are enough to maintain the tendency of the genotypes to accumulate to the same snail individuals despite of the costs of MGIs.

In conclusion, we present a statistical model to predict frequency of MGIs in aquatic snails by measuring parasite prevalence. We found that aquatic snails carry unexpectedly high frequencies of trematode MGIs. This finding is in sharp contrast with the absence of multiple species infections in snails (reviewed by Kuris & Lafferty 1994), suggesting that co-infections by multiple species and multiple genotypes of one species are controlled by different processes. High prevalence of MGIs should be taken into consideration when conducting experiments with naturally infected snails as MGIs may have significant consequences for the experimental outcome. For example, MGIs at one host stage can influence infection success of the parasite in the consecutive host of the life cycle (Karvonen et al. 2012). MGIs can be identified by genotyping a subsample of the cercariae released by each snail or excluded by using experimentally infected snails. The commonness of MGIs in snails is nevertheless a bit of a paradox because it is detrimental in terms of reduced larval output for each co-infecting genotype (Karvonen et al. 2012) and/or impaired survival of the co-infected host (Davies, Fairbrother & Webster 2002). More studies on the susceptibility profiles of the host populations are needed to explore if the parasites entering an already occupied snail are merely doing the best of a bad job in terms of host availability, or if the co-infection is enhanced by suppressed host resistance. Regardless of the underlying mechanisms, frequent between genotype interactions emerging from the co-infections may have significant evolutionary consequences for both parasite and host populations. The statistical method we developed allows flexible testing of alternative null models against data, which should prove useful in various host–parasite systems where different alternative hypotheses need to be contrasted to understand parasite aggregation and its consequences.

Acknowledgements

We greatly appreciate the help from staff at Konnevesi Research Station and Oulu Experimental Zoo. We are grateful to Elina Virtanen and Sari Viinikainen for help in the laboratory and to Kirsten Klappert for handling the samples. We thank the reviewers for useful comments that improved this paper. This project was funded by the Center of Excellence in Evolutionary Research at the University of Jyväskylä (K-RL, AK and CR) and the Swiss National Science Foundation (JJ).

Ancillary