Main

The wild rabbit population in the United Kingdom is dominated by five gut helminths: the strongylid nematodes Graphidium strigosum (in the stomach) and Trichostrongylus retortaeformis (small intestine); the anoplocephaloid cestodes Mosgovoyia pectinata (small intestine) and Cittotaenia denticulata (small intestine); and the oxyurid nematode Passalurus ambiguus (large intestine and colon). Our objectives were to identify (1) whether interspecific interactions could be detected within this parasite community; (2) quantify the strengths of these interactions; (3) identify the putative mechanisms by which interactions may be mediated; and (4) propose the possible consequences of such interactions for future parasite control.

We used a combination of generalized linear modelling (GLM) and residual maximum likelihood (REML) linear mixed model analyses of parasite count data to test the null hypothesis that parasite interactions do not influence parasite numbers. We included aspects of the host biology and the external environment that if excluded would be most likely to create the appearance of interspecific interactions where none exist (for further details see Supplementary Information). All predictions are presented as the percentage difference between the predicted level of the dependent variable (after back-transformation) when the interacting species has a zero count, compared with the predicted level of the dependent variable (after back-transformation) when the interacting species is at its geometric mean (in concomitant infection).

According to the models, same-locale, downstream (anterior to posterior) and upstream (posterior to anterior) interactions occurred between the helminths. One significant (P < 0.001) interaction was seen between two species in the same location in the gut, with the model predicting a positive effect of C. denticulata on the intensity of T. retortaeformis infection (Fig. 1a and Table 1). The geometric mean of C. denticulata when in a concomitant infection with T. retortaeformis was around two individuals, which was predicted to increase the number of T. retortaeformis by 51% relative to a rabbit harbouring no C. denticulata.

Figure 1: Predictions from generalized linear and restricted maximum likelihood models of the relationships between the dependent variable (affected species log(x) helminth intensity with zeros removed) and the independent variable a–e, effector species log(x + 1) helminth intensity, or f, effector species presence-absence data (in rabbits of average mass, 1,590 g).
figure 1

All analyses were conducted on adult rabbit data. Lines of predicted fit and residual fit points (or 95% confidence intervals) are shown in each case. Trichostrongylus retortaeformis with C. denticulata (a); T. retortaeformis with G. strigosum (b); P. ambiguus with M. pectinata (c); M. pectinata with P. ambiguus (d); G. strigosum with M. pectinata (e); and G. strigosum with T. retortaeformis (f).

Full size image
Table 1 Significance of factors affecting the log intensity of gut helminths of O. cuniculus
Full size table

Trichostrongylus retortaeformis was subject to a predicted, positive downstream influence from G. strigosum (P < 0.001; Fig. 1b and Table 1). In concomitant infections, the geometric mean of 18 G. strigosum was predicted to increase the number of T. retortaeformis by 36%. A more complex downstream interaction was seen with M. pectinata in the small intestine, which was predicted to have a positive effect in female (P = 0.003), but not male, hosts on the intensity of P. ambiguus in the large intestine/colon (Fig. 1c and Table 1). The geometric mean of three M. pectinata was predicted to increase the intensity of P. ambiguus by 163% relative to a host containing no M. pectinata.

The model predicted that the association between these two species also operated in the opposite, upstream direction, with P. ambiguus having a positive effect on M. pectinata. Again, this association was complicated by host sex, such that P. ambiguus increased the intensity of M. pectinata in male but not female rabbits (P = 0.008; Fig. 1d and Table 1). The geometric mean of P. ambiguus (in concomitant infection) in male rabbits was 127, which was predicted to increase M. pectinata numbers by 67% relative to hosts not infected with P. ambiguus. Mosgovoyia pectinata also had an upstream effect, reducing the intensity of G. strigosum in the stomach (P = 0.011; Fig. 1e and Table 1) by 19% on average. Finally, the presence of T. retortaeformis was predicted to affect negatively the numbers of G. strigosum in an upstream position (P = 0.018; Fig. 1f and Table 1). This effect was complicated by host mass (a proxy for age20), such that the effect decreased as rabbit mass increased; however, in a rabbit of average mass (1,590 g in concomitantly infected rabbits), T. retortaeformis was predicted to reduce G. strigosum numbers by 29%.

Analyses indicate the existence of a network of interactions between these helminths (Fig. 2). In most cases, based on the biology of the parasites, we can postulate the likely mechanisms underlying these interactions. The observed same-locale or downstream interactions may be due to direct influences of one parasite on another through by-products, manipulation of gut physiology, competition or physical crowding. However, for upstream interactions, the most likely route would be interaction mediated through the host's immune system, although changes in gut physiology, for example pH change, cannot be entirely ruled out21.

Figure 2
figure 2

Relative positions of the rabbit gut helminths in the host gut, with interaction directions and strengths, based on observed mean burdens.

Full size image

The putative explanation for the observed interactions between G. strigosum and T. retortaeformis is cross-immunity between these species. Age–intensity curves for T. retortaeformis and G. strigosum (with mass used as a proxy for age) support previous findings22,23,24 that T. retortaeformis stimulates an acquired immune response whereas the G. strigosum numbers show no decline, suggesting no response to host immunity (Supplementary Note and Fig. S1). It is likely that an ability of G. strigosum, as seen in other helminths25, to modulate host immunity explains the positive effect this species appears to have on T. retortaeformis. Conversely, stimulation of acquired immunity by T. retortaeformis may negatively affect the related G. strigosum.

The interaction of P. ambiguus and M. pectinata is affected in both directions by host sex, once again implying mediation through host biology. Similarly, the negative effect of M. pectinata on upstream abundance of G. strigosum must be host mediated but the mechanism for these interactions is not discernible from the current data. The mechanism for the relationship between the cestode C. denticulata and the nematode T. retortaeformis cannot be determined, as the current data do not indicate whether the interaction is direct or host mediated.

These analyses provide credible evidence for the presence of interspecific parasite interactions throughout a community of gut parasites, suggesting that host immunity may have a role in shaping that community. Although we wish to stress that the predictions are based on analysis of correlated data and must be validated experimentally, the biology of the helminths provides support that the interactions are genuine.

Notably, there is no reason to believe that the rabbit system is unusual and therefore such analyses applied to other systems, particularly those of economically important livestock (where laboratory experiments have provided evidence of interaction21,26), are likely to reveal similar networks of interactions. The presence and detection of interspecific parasite interactions, particularly where they are immune mediated, could have important implications for parasite control strategies. This is particularly pertinent to the future of helminth control, where rapidly developing resistance to chemotherapeutics is leading to increased efforts in the field of vaccine development27. The work described here suggests that within-host interspecific parasite interactions may have a role in determining the overall impact of such vaccines on gut helminth communities. To illustrate this we used a simple generic model, which may be adapted to represent any system. The model consisted of three parasite species within a single host (for details see Supplementary Methods). Each parasite stimulated a specific immune response at a rate proportional to the abundance of that parasite, and this immune response acted on the establishment rate of the parasite. Interspecific interactions were incorporated by allowing the immune response against one parasite to affect one of the other species, either increasing or suppressing the action of the immune response against that species. A simple network of interactions between the parasites was constructed such that parasite 1 (P1) had a negative effect on parasite 2 (P2), P2 a positive effect on parasite 3 (P3), and the effect of P3 on P1 (described by the parameter γ3) was varied from highly negative to highly positive. A species-specific vaccine was then applied, targeted solely against P1, and the impact of the strength of interaction between P3 and P1 on the efficacy of this vaccine was assessed.

Vaccination against the target parasite always suppressed its abundance below pre-treatment levels (Δ1 < 1; Fig. 3b). However, vaccination against P1 tended to have a positive effect on P2 (Δ2 > 1) (Fig. 3c), but little impact on P3 (Δ3 ∼ 1; Fig. 3d), presumably because P3 was two steps away from the target of the vaccine and was to some extent buffered from its effects.

Figure 3: Three species parasite interaction model.
figure 3

a, Schematic of three parasite species interaction model (for details see Supplementary Methods) where P1, P2 and P3 are the parasite species, I1, I2 and I3 are the species-specific immune responses, V is the vaccine applied against species 1 and γ1, γ2 and γ3 are the parameters describing the strength of the effect of one species on another. b–d, Proportional change (Δi) in equilibrium abundances of parasite species i where i = 1 (b), i = 2 (c) and i = 3 (d), following vaccination (that is, post-vaccination equilibrium level/pre-vaccination equilibrium) at varying levels of γ3, the parameter describing the strength of the effect of P3 on P1. The dotted horizontal line represents where the post-vaccination level = pre-vaccination level.

Full size image

When γ3 was 0 (that is, P3 did not interact with P1), vaccination reduced P1 to approximately 15% of its pre-vaccination level (Fig. 3b). Furthermore, as P1 had a negative interaction with P2, its suppression meant that vaccination increased the abundance of P2 (Fig. 3c). When P3 had a strong negative interaction with P1 (γ3 = -1) the vaccine, combined with this negative interaction, reduced P1 abundance to approximately 1% of its pre-vaccination level. As P1 was at very low abundance, its impact on P2 was negligible and so P2 approximated its pre-vaccination level (Δ2 approached 1). However, when γ3 was strongly positive (+1) P1 abundance was only reduced to about 50% of its pre-vaccination level; the positive impact of P3 on P1 diminished the efficacy of the vaccine. Notably, given a positive interaction between P3 and P1, vaccinating to control P1 resulted in a massive relative increase in the abundance of P2, up to an order of magnitude greater than its pre-vaccination level. This is surprising because P1 abundance is increased by the presence of P3 and so P2 abundance should be further suppressed due to the negative interaction between P1 and P2. The explanation for this counterintuitive result lies in the subtle nonlinear effects of the interactions. Before vaccination, the positive impact of P3 on P1 meant that P1 was at a very high equilibrium level. Hence, P2 was suppressed to a very low level. By applying the vaccine, P1 abundance was reduced to approximately half of its pre-vaccination level, thereby decreasing the negative impact on P2. P2 abundance therefore increased substantially, resulting in a massive relative increase compared with its pre-vaccination level. Hence, vaccination against one parasite species produced unexpected and marked increases in the abundance of another, previously innocuous parasite species, by releasing it from the negative impact of the targeted parasite. These, and other, important effects would be missed if we ignored the possibility of such interactions.

The models used here have revealed an array of potential interactions between parasites, which would not have been elucidated by less powerful techniques. As concomitant infections are the norm rather than the exception28, application of these techniques to other mammalian-parasite data sets is likely to reveal similar interactions. Given the increase of anthelmintic resistance among gut helminths29, alternative control strategies are urgently needed. A clearer understanding of parasite community ecology may provide such an alternative. One implication is that parasites could be used to control other infections. For example, artificial infection with one relatively benign species could be used to stimulate cross-immunity and prevent subsequent infection with a deleterious species without resorting to anthelmintics. However, without the knowledge of the community relationships, treatment against one deleterious species could allow an increase in the intensity of a second harmful species. Previously unexplored interactions may be one explanation for the failure of laboratory proven vaccines to work under more natural conditions. It is clear that such interactions need to be incorporated into future dynamical studies of parasite communities where (with a few exceptions30) parasite interspecific conflicts and mutualisms are ignored in the null models.

Methods

Rabbits were collected from a 400 ha site in Perthshire, Scotland (ordnance grid reference NO 280 340) between the months of January 1977 and December 1999. Total gut helminth counts, presence of myxomatosis and presence of Eimeria stiedae were recorded along with details of the external environment and aspects of host biology. Further details of the study site, collection protocols and data may be found in ref. 17.

All analyses were conducted using the GENSTAT statistical package (VSN International Ltd) and all models were initially run as REMLs with the random terms of warren code and year of capture. Interactions between all model terms were initially included up to third order. Only adult rabbits (mass >1,249 g; n = 1,526) were used for this analysis because they were likely to have been exposed to infective stages of all parasites by this age. In addition, the distributions of the gut helminth parasites in different age classes of the rabbit have been shown to be different and must therefore be analysed separately17. The data for the gut helminths do not conform to the negative binomial distribution and in order to obtain a ‘near’ normal distribution for the dependent variable, zeros had to be removed from the data before a log transformation was applied17. We therefore considered those factors that affected the intensity of the species treated as the dependent variable (in infected animals only). Each helminth species in turn was treated as the dependent variable. The helminths not being treated as the dependent variable were included in the model (as determined from submodels) as independent factors (presence/absence) or as independent variables (log(x + 1) intensity data) as appropriate. Nonsignificant terms were removed from the models in a stepwise manner until a minimal model was produced. This model was further refined following bootstrapping of the model effects and removal of any terms that were not significantly different from zero. Once the bootstrapping refinements were completed, each model was run once again in its final form to obtain effect sizes and significance levels. As model terms were only retained after the bootstrap of their effect size, log(x + 1) transformation was considered sufficient for the species treated as independent variables and allowed comparison of concomitantly infected rabbits with those only infected with the species treated as the dependent variable.