Tapeworm manipulation of copepod behaviour: parasite genotype has a larger effect than host genotype

Abstract

Compared with uninfected individuals, infected animals can exhibit altered phenotypes. The changes often appear beneficial to parasites, leading to the notion that modified host phenotypes are extended parasite phenotypes, shaped by parasite genes. However, the phenotype of a parasitized individual may reflect parasitic manipulation, host responses to infection or both, and disentangling the contribution of parasite genes versus host genes to these altered phenotypes is challenging. Using a tapeworm (Schistocephalus solidus) infecting its copepod first intermediate host, I performed a full-factorial, cross-infection experiment with five host and five parasite genotypes. I found that a behavioural trait modified by infection, copepod activity, was affected by both host and parasite genotype. There was no clear evidence for host genotype by parasite genotype interactions. Several observations indicated that host behaviour was chiefly determined by parasite genes: (i) all infected copepods, regardless of host or parasite genotype, exhibited behavioural changes, (ii) parasitism reduced the differences among copepod genotypes, and (iii) within infected copepods, parasite genotype had twice as large an effect on behaviour as host genotype. I conclude that the altered behaviour of infected copepods primarily represents an extended parasite phenotype, and I discuss how genetic variation in parasitic host manipulation could be maintained.

1. Introduction

Parasitic manipulation of hosts is a classic example of an extended phenotype [1]—parasite genes have an effect beyond the parasite itself, modifying a host phenotype, like behaviour. The notion of host manipulation as an extended parasite phenotype implies that parasite genes determine host phenotypes. This seems obvious in many cases, because the changes benefit parasites [2]. Still, the phenotype of a parasitized host can be shaped by parasite genes (e.g. manipulation), host genes (e.g. a sickness behaviour) or both [3–5]. Host manipulation has been well documented for trophically transmitted parasites in their intermediate hosts [6–8]. In such systems, the parasites' ability to modify hosts can vary genetically [9–12], but it is not known whether parasite genes play a larger role than host genes in determining altered host phenotypes.

I performed a full-factorial, cross-infection experiment with five host and five parasite genotypes. I used the tapeworm Schistocephalus solidus in its copepod first intermediate host. The tapeworm has a three-host life cycle: first infecting copepods, then a fish, and finally a piscivorous bird [13]. The worm alters copepod activity, a trait linked to fish predation [14]. Infected copepods are less active while parasites are uninfective [15], which suppresses predation risk [16]. Once parasites become infective, copepod activity increases [15], resulting in increased predation by the fish next host [14]. However, the switch to elevated copepod activity is not observed in every parasite population [17]. Behavioural changes are not a by-product of pathology [18,19]. Moreover, there is genetic variation for behaviour modification in S. solidus, with heritability estimated to be approximately 0.15 [9,10]. Copepods also exhibit genetic variation in their activity [20]. Here, I disentangled the effects of host and parasite genes to ascertain whether infected copepod behaviour should be considered an extended parasite phenotype.

2. Methods

(a). Source and breeding

Tapeworms were collected from infected sticklebacks caught in Lake Skogseidvatnet, Norway (60°14′ N, 5°55′ E). Worms were size-matched to favour outcrossing [21] and bred in vitro [22] to create five full-sibling families. Tapeworm eggs were stored at 4°C. Prior to use, they were incubated at 20°C for three weeks in the dark and then exposed to light 1 day before infection to induce hatching.

Copepod cultures (Macrocyclops albidus) also originated from Lake Skogseidvatnet. I created five inbred lines. Single egg-bearing females were isolated in new tanks. After four to five weeks, enough time for their offspring to mature, a few adult males and females (less than five of each) were collected, moved to a new tank, and allowed to reproduce via sibling (sib) mating. This was continued for 10 generations. To ensure the lines' persistence, I did not control copepod mating. Thus, the inbreeding coefficient probably ranges from 0.69 (only half-sib mating) to 0.89 (only full-sib mating) [23].

I refer to genetic groups as ‘genotypes' throughout, but these groups differed for host and parasite (inbred lines for copepods, full siblings for tapeworms). Within-group genetic variance is therefore presumably larger for worms than for copepods. The main reason for the discrepancy was an inability to produce adequate numbers of copepod sibs for exposure to multiple parasite genotypes.

(b). Copepod infection

Adult male copepods were isolated in 24-well plates (approx. 1.5 ml per well) and each was exposed to one tapeworm larva (n = 1540). Controls were not exposed (n = 138). Copepods were maintained at 18°C with a 16 :8 h light : dark cycle and fed with cultured paramecia. After one week, copepods were microscopically checked for infection. At 9 days post-infection (dpi), worm development was visually recorded, specifically whether they had a cercomere, which is associated with infectivity to fish [24]. At 21 dpi, infected copepods were dissected. Worm size was recorded by tracing their area in two photographs with ImageJ [25] and averaging the measurements.

The experiment was conducted in three blocks over three months. All host and parasite genotypes were represented in each block.

(c). Behavioural recordings

Copepod behaviour was recorded every other day from 5 to 21 dpi. Worms develop infectivity to fish between 11 and 15 dpi [15], so the experiment covered the entirety of parasite development. Plates containing copepods were gently placed onto an apparatus that held them about 3 mm above the ground (see [15]). After allowing copepods to acclimatize for a minute, recording began. I recorded behaviour for a minute, then I dropped the plate and recorded an additional minute. The aim was to record both undisturbed (pre-drop) and frightened (post-drop) copepod behaviour. Videos were recorded (Panasonic WV-BP550) at 8 frames s⁻¹.

Videos were processed in two ways. First, videos were reduced to 0.5 frame s⁻¹, converted to image stacks, and then manually clicked through to track copepod position. This was time-consuming, so I also wrote a python script to automatically track copepods (https://github.com/dbenesh82/automatic_copepod_tracker). This tracker made mistakes, but the values from the two methods were well correlated (R² = 0.82). As auto-tracking allowed more recordings to be analysed, I present these data, though I also checked whether the manual dataset yielded the same results.

(d). Response variable

I defined activity as the average copepod speed over a recording; it could equivalently be the total distance moved, since recording times were constant. Copepod activity was calculated separately for the 1 min periods before and after the plate drop. Activity incorporates both movement frequency and magnitude, and infection affects both ([19]; see https://github.com/dbenesh82/GxG_analyses/blob/master/analyses/GxG_01_define_responses.md). For simplicity, I only present activity, not its components.

(e). Statistical models

I fitted mixed models and compared them with likelihood ratio tests. Copepod activity was a positive, continuous variable with an overabundance of zero values, as sometimes copepods did not move during a recording. A compound Poisson Gamma distribution is appropriate for such data, so I fitted models using the cplm R package [26]. Model building occurred in five steps: (i) establish a ‘base' model, (ii) add infection, (iii) add copepod genotype, (iv) add parasite genotype, and (v) add host by parasite genotype interactions. The last two steps were conducted with just infected copepods, as parasite genotype was nested within infection.

First, I established a ‘base' model with variables that were not of primary interest. Experiment block, dpi and recording period (pre- or post-drop) were fixed factors, while copepod individual was a random effect. I also tested whether copepod ‘drop reactions' varied with dpi, e.g. due to habituation. Second, I evaluated the impact of infection on copepod activity and drop responses. I allowed the differences between infected and uninfected copepods to vary with dpi. Third, I added copepod genotype and checked whether its effect depended on infection. Fourth, I added parasite genotype. For both copepods and parasites, I examined whether genotypes responded differently to the drop. Finally, I added a host genotype by parasite genotype interaction, and I examined whether genotype interactions were consistent over time. For all models, I estimated the variation explained by fixed and random effects [27].

I checked the robustness of the results to mortality by excluding copepods that died during the experiment. I also ran the same analyses on the dataset created with manual copepod tracking.

(f). Other fitness proxies

I measured four additional parasite fitness components: infection rate, copepod survival, worm development and worm size. For each, I fitted generalized linear models to test the effect of host genotype, parasite genotype and genotype interactions.

Analyses were conducted in R [28].

3. Results

The infection rate was 27.1%. The number of infected copepods in the 25 host–parasite genotype combinations varied from 8 to 25 (average = 16.4). Activity was measured for 1018 copepods, of which 413 were infected.

(a). Infected copepods were less active and less responsive to the drop

Unexposed controls and exposed-but-uninfected copepods behaved similarly (likelihood ratio tests of models with and without this distinction, p = 0.164; electronic supplementary material, figure S1), so they were pooled. Infected copepods were less active than uninfected copepods, though this difference varied slightly with dpi (table 1; electronic supplementary material, figure S1). In response to the drop stimulus, uninfected copepods decreased their activity more than infected copepods (table 1; electronic supplementary material, figure S2).

Table 1.

Compound Poisson Gamma mixed models of copepod activity. First, a base model was created, which included variables of secondary interest. Second, the effect of infection was tested. Third and fourth, the effects of copepod and tapeworm genotype were tested. Finally, host–parasite genotype interactions were tested. The final two steps were only assessed within infected copepods. Likelihood ratio tests (LRTs) assessed model improvement. Marginal R² (R²_m) represents the variation explained by fixed effects, while conditional R² (R²_c) represents that explained by random and fixed effects combined ([27] delta method).

	uninfected + infected (16 600 observations on 1018 copepods)				only infected (n = 6834 observations on 413 copepods)
model	d.f.	LRT	${\mathit{R}}_{\mathbf{m}}^{\mathbf{2}}$	${\mathit{R}}_{\mathbf{c}}^{\mathbf{2}}$	d.f.	LRT	${\mathit{R}}_{\mathbf{m}}^{\mathbf{2}}$	${\mathit{R}}_{\mathbf{c}}^{\mathbf{2}}$
initializing model
+ copepod random effect	—	—	0	0.425	—	—	0	0.251
+ block + dpi + drop response	11	<0.0001	0.041	0.442	11	<0.0001	0.038	0.277
+ time-dependent drop responsea	8	0.0003	0.042	0.443	8	0.0009	0.041	0.280
infection
+ infection	1	<0.0001	0.174	0.437	—	—	—	—
+ infection-dependent drop response	1	0.001	0.173	0.437	—	—	—	—
+ time-dependent effect of infectiona	8	<0.0001	0.185	0.446	—	—	—	—
+ time- and infection-dependent drop response	8	0.075	0.186	0.447	—	—	—	—
copepod genotype (inbred lines)
+ copepod genotypea	4	<0.0001	0.197	0.447	4	0.010	0.050	0.280
+ copepod genotype split by infection	4	0.197	0.198	0.447	—	—	—	—
+ genotype-dependent drop response	8	0.229	0.199	0.447	4	0.395	0.051	0.281
parasite genotype (full-sib families)
+ parasite genotypea	—	—	—	—	4	<0.0001	0.070	0.279
+ genotype-dependent drop response	—	—	—	—	4	0.533	0.070	0.279
genotype × genotype interactions
+ host by parasite genotype interaction	—	—	—	—	16	0.202	0.082	0.278
+ time-dependent genotype interactions	—	—	—	—	192	<0.0001	0.148	0.341

^aModels used for the next step.

(b). Copepod and tapeworm genetics matter

Copepod lines varied in their activity (table 1). The most active genotypes were the same in both uninfected and infected copepods (infection×genotype interaction, p = 0.197), but the differences were smaller in infected copepods (figure 1a; copepod genotype explained 2.0% of the variation in uninfecteds versus 0.9% in infecteds).

Figure 1.

(a) Mean activity (speed in pixels s⁻¹ over 1 min recording intervals) for five inbred copepod lines when they were uninfected or infected. (b) Mean activity of infected copepods split by either host or parasite genotype. (c) Mean activity of infected copepods split by both host and parasite genotype; host genotypes depicted by colours. In all panels, solid error bars depict the 95% CI as estimated by a mixed model. Dotted error bars show the interquartile range. (Online version in colour.)

Parasite genotype affected copepod activity (table 1), and its effect was stronger than host genotype (figure 1b); host and parasite genotype explained 0.9% and 2.0% of the variance in infected copepod activity, respectively. The response to the drop stimulus was not dependent on either copepod or parasite genotype (table 1), suggesting genetics did not affect fright responses.

(c). Genotype × genotype interactions?

The interaction between host and parasite genotype was not significant (table 1 and figure 1c). However, the model was improved by allowing genotype interactions to vary by dpi, despite the huge number of parameters this entailed (table 1). There was not a discernible pattern in how genotype combinations varied over time (figure 2). For instance, genotype combinations that were relatively active early in the experiment were not less active later (Spearman correlation for mean activity of 25 genotype combinations in the first and last third of the experiment, ρ = 0.34, p = 0.10), nor was activity related to parasite development (correlations between early, late and overall activity with cercomere presence, ρ = −0.18 to 0.07, all p > 0.38).

Figure 2.

Time-dependent genotype effects. Copepod activity over the experiment is plotted for each copepod line (panels) infected with each parasite family (coloured, dashed lines). (Online version in colour.)

(d). Robustness of results

Key results, such as the larger parasite genotype effect and the lack of genotype interactions, were unchanged by excluding copepods that died during the experiment or by using manually tracked movements, though there were minor discrepancies (see electronic supplementary material, tables S1 and S2).

(e). Other fitness proxies

Host genotype affected copepod infection, copepod survival and parasite growth, but not parasite development. Parasite genotype affected copepod infection, parasite development and parasite growth, but not copepod survival. None of the four traits was affected by genotype interactions (table 2).

Table 2.

Whether copepod or tapeworm genotype affected five traits. Significance was determined with likelihood ratio tests on generalized linear models that stepwise added host genotype, parasite genotype and genotype interactions. *p < 0.05, **p < 0.01, ***p < 0.001.

trait	host genotype	parasite genotype	G × G interaction
copepod activity (n = 413)	yes*	yes***	maybea
infection rate (n = 1502)	yes**	yes*	no
proportion surviving 21 dpi (n = 1502)	yes***	no	no
cercomere presence 9 dpi (n = 413)	no	yes***	no
parasite size 21 dpi (n = 315)	yes***	yes***	no

^aCombinations of host and parasite genotype did not differ in activity more than expected, given their main effects, but they did vary over time.

4. Discussion

I found that both host and parasite genotype affected copepod activity, but the effect of parasite genotype was twice as large. A priori one might have expected the opposite, because genetic similarity within groups was higher for hosts (i.e. they were inbred lines; parasites were full sibs). However, this is consistent with studies that found a stronger response to selection on S. solidus for manipulation than to selection on copepods for resistance to manipulation [10,20]. A local adaptation experiment also showed that tapeworm population has a larger effect on behaviour than copepod population [17]. Parental effects could inflate differences among worm sibships [29], but it is unlikely that this caused the larger parasite genotype effect, because heritability estimates for behaviour that include [9] or exclude [10] parental effects are similar. Together these results support the interpretation of altered copepod behaviour as an extended phenotype of the tapeworm.

Infection does not entirely override host behaviour—copepod genotype still mattered. The most active copepod lines tended to be so regardless of their infection status, suggesting different baseline activity levels. Nonetheless, infection appears to move these baselines closer together, as differences between copepod genotypes were smaller in infected than in uninfected copepods. There is likely to be little selection for resistance to tapeworm manipulation, as few copepods are naturally infected (typically less than 1%; [30–32]). Minimal resistance is further implied by the strong effect of infection (7–12 times larger than genotype effects). All infected copepods were less active than uninfected copepods—no host–parasite genotype combinations were ‘unmanipulated'.

Assuming there is an optimal level of host manipulation for parasites [33], stabilizing selection should eliminate genetic variation [34]. What maintains it in S. solidus? Genotype interactions are one possibility [35], because the relative fitness of a parasite genotype depends on host genotype, such that there is no single best-performing parasite genotype. I did not find clear evidence for genotype interactions on copepod behaviour. Rather, the activity of different genotype combinations varied over time, seemingly unrelated to parasite development. This may be a by-product of overfitting a model (25 genotype combinations × 9 dpi = 225 parameters!) to a noisy behaviour variable. Consistent with this interpretation, genotype interactions were not observed for other traits, such as infection rates or parasite growth. Thus, genotype interactions probably do not maintain genetic variation in this system. A caveat, though, is that S. solidus can infect several copepod species [36]. Manipulation can vary across host species in other trophically transmitted parasites [37,38], so it is possible that genetic variation is maintained by tapeworm genotypes faring better or worse in particular copepod species.

Even without genotype interactions, selection on manipulation can fluctuate. Factors determining manipulation's costs and benefits such as the abundance of target hosts [39], non-host predators [40] or food resources [19,41] may vary over space or time. Separate S. solidus populations modify copepods differently [17], and gene flow between populations [42] experiencing different selective regimes could maintain genetic variation. Alternatively, evolutionary fluctuations in copepod behaviour, e.g. due to changes in food or predation levels, may lead to fluctuating selection on parasites to preserve an optimal level of host alteration. Finally, it is possible that the observed genetic variation is selectively neutral. Compared with the disparity between uninfected and infected copepods, the differences among parasite genotypes were minor, and they may not be substantial enough to impact predation rates or other traits linked to parasite fitness via host activity.

Cross infections of five inbred copepod lines with five tapeworm families showed that infected copepod behaviour is determined more by parasite genes than by host genes. I did not find clear evidence for genotype interactions. I conclude that the altered behaviour of infected copepods primarily represents an extended parasite phenotype.

Ethics

Animal experiments were conducted with the permission of the ‘Ministry of Energy, Agriculture, the Environment and Rural Areas' of the state of Schleswig-Holstein, Germany (reference no. V 313-72241.123-34).

Data accessibility

Data are available from the Dryad Digital Repository: https://doi.org/10.5061/dryad.v41d698 [43].

Competing interests

I declare I have no competing interests.

Funding

I was supported by a grant from the Deutsche Forschungsgemeinschaft (BE 5336/3-1).