Parasite-mediated manipulation? Toxoplasma gondii infection increases risk behaviour towards culling in red deer

Abstract

Parasites can modify host behaviour to increase their chances of survival and transmission. Toxoplasma gondii is a globally distributed protozoan whose ability to modify host behaviour is well known in taxa such as rats and humans. Less well known are the effects on the behaviour of wild species, with the exception of a few studies on primates and carnivores. Taking advantage of a culling activity conducted in Stelvio National Park (Italy), the serological status of T. gondii was studied in 260 individuals of red deer Cervus elaphus with respect to the risk of being culled. A temporal culling rank index was fitted as a response variable, and T. gondii serological status as the main explanatory variable in linear models, accounting for covariates such as sex, age, jaw length, bone marrow fat and culling location. The overall seroprevalence of T. gondii was 31.5%, and the selected models suggested that seropositive deer were culled earlier than seronegative ones, but this effect was only evident in females, in individuals with medium–good body condition, and in areas with greater human presence. Our results suggest that T. gondii may be involved in risk behaviour in large herbivores, supporting its role as a facilitator of predation risk.

1. Introduction

Several pathogens have developed strategies to increase their chances of transmission within animal populations [1]. One such strategy is the manipulation of host behaviour through direct or indirect mechanisms [2]. The rabies virus, for example, acts directly on the host's nervous system, causing an increase in aggressive behaviour and thus a greater likelihood of biting, thereby facilitating its own transmission [3]. Parasites can also directly influence secondary aspects of the host, such as its metabolism or immune system, thus indirectly its behaviour [2].

A parasite known for its direct and indirect impact on host behaviour is Toxoplasma gondii. This pathogen acts at the level of the central nervous system, lowering risk perception by decreasing anxiety, fear and vigilance in several species. Attenuated fear response, for example, was observed in rats [4], possibly favouring the transmission of the parasite through predation, while in humans it may cause psychological disorders [5,6]. Most of the studies on the effects and physiological mechanisms of T. gondii involving risk-taking behaviour have been obtained in laboratory conditions [7], whereas the alteration of behavioural mechanisms in the wild is understudied. Following infection with T. gondii, an increased likelihood of predation by sharks has been suggested in sea otters Enhydra lutris nereis [8], and increased risk behaviour has been observed in central chimpanzees Pan troglodytes troglodytes, spotted hyenas Crocuta crocuta and grey wolves Canis lupus [9–11]. Overall, however, we have limited knowledge of the impact of T. gondii on behaviour in wildlife populations.

Taking advantage of a population of red deer Cervus elaphus subject to a culling programme in Stelvio National Park [12], we investigate the potential effects of T. gondii exposure on altering deer behaviour, under the assumption that individuals seropositive toward T. gondii antibodies are infected by tissue cysts [13]. We tested the hypothesis that risk behaviour of seropositive deer is altered, and we expect seropositive individuals to be culled earlier than seronegative ones.

2. Material and methods

(a) . Study area and population

Stelvio National Park is located in the central Italian Alps, between 800 m.a.s.l. and 3851 m.a.s.l.; at the time of the study, the red deer population in the study area had a density of about 30 deer km⁻² in winter, with an estimated number of deer of about 1400 individuals [14]. Given the impact on forest regeneration and biodiversity, in recent years the Park has decided to reduce deer density through culling in the wintering area, over about 2679 ha. Within the area, three culling sub-areas with different levels of human activity were identified: low (772 ha), intermediate (1200 ha) and high (707 ha) [11]. In our system, the role of predation is represented solely by culling, as natural predators such as grey wolf were absent at the time of study.

(b) . Data collection

Data were collected during the culling seasons 2016–17 and 2017–18, from the end of November to the end of January. All culled deer (n = 260) were brought to a checkpoint. The culling plan included 9% yearling (1.5 years) males, 12% adult (≥ 2.5 years) males, 7% yearling (1.5 years) females, 39% adult (≥ 2.5 years) females and 33% calves (0.5 years).

First, shooting date was recorded to generate a ‘culling rank index' based on the temporal order of culling, from 1 to n (1 if the animal was shot on the first day of each culling season, n if it was shot on the last day of culling: deer culled on the same day were assigned the same rank), for each year separately. The culling rank index is intended as a proxy to investigate the probability of individuals being culled at a given time. A similar rank index was used to assess possible effects of T. gondii on the host, the wild brown rat Rattus norvegicus [15], and the probability of rats being trapped was positively associated with T. gondii seropositive status., i.e. rats that tested positive for Toxoplasma were caught earlier than rats that tested negative.

Next, assuming that individuals testing positive for IgG antibody toward T. gondii are infected (but see [13]), we investigated its occurrence in deer through an ELISA (enzyme-linked immunosorbent assay) test (ID Screen Toxoplasmosis Indirect Multi-species ELISA, IDVET, Montpellier, France) after collecting blood from the jugular vein. During inspection at the checkpoint, the veterinary staff assessed the health condition of animals in order to exclude co-morbidities that might influence the behaviour of deer, and thus their probability of being culled. To control for other potential confounding effects in the relationship between culling rank index and T. gondii seropositivity, for each animal we collected data on sex, age, jaw length, eviscerated body mass, proportion of bone marrow fat (measured through the dehydration of the marrow of metacarpals [16]) and sub-area of culling. Jaw length is a good indicator of skeletal development [17], while eviscerated body mass and proportion of bone marrow fat are representative indices for the nutritional and body conditions of the animal. As body mass and body condition both decline over winter due to food restrictions, to avoid bias, for each year these variables were adjusted to the first day of culling by fitting quadratic linear models between body mass/bone marrow and Julian date, from the first day of culling, for different age-classes, for the two sexes separately (cf. [18] and references therein).

(c) . Statistical analyses

To investigate the relationship between temporal rank of culling and T. gondii seropositivity status, a regression modelling approach was used, where the global model (sensu [19]) included all the collected variables (and their first order interactions) that were considered to be related to T. gondii seropositive status. Culling rank index was the response variable, and T. gondii serological status (negative/positive) was the main explanatory variable in interaction with sex, age-class, jaw length, proportion of bone marrow fat, level of human activity in culling area and year of culling. Such first-order interactions are considered important to reveal subtle effects of T. gondii which might affect behaviour only in specific conditions. Body mass was excluded because of collinearity issues detected during preliminary data analyses. A deer seropositive to T. gondii does not generate obvious external signs of disease, thus culling was assumed to be unbiased.

Since the response variable was a rank variable, ordered regression would be the natural approach to modelling. However, an alternative approach is to use simple linear regression on transformed ranks, as long as the number of categories is large [20], as in our case. After preliminary analyses, a boxcox function showed improved model fit after a square-root transformation for the response. Our full model was thus of the form:

$$CR{I}_i\sim N({\mu }_i,\hspace{0.17em}{\sigma }^2)$$

$$\mathrm{E}(CR{I}_i)={\mu }_i\mathrm{and}\mathrm{var}(CR{I}_i)={\sigma }^2$$

$${\mu }_i=infection\hspace{0.17em}statu{s}_i\times (se{x}_i+ag{e}_i^2+bone\hspace{0.17em}marrow\hspace{0.17em}fa{t}_i^2+jaw\hspace{0.17em}lengt{h}_i^2+level\hspace{0.17em}of\hspace{0.17em}human\hspace{0.17em}activit{y}_i+yea{r}_i)$$

where $CR{I}_i$ was the square-root value of the culling rank index for observation i, assumed to have conditional normal distribution with expected value ${\mu }_i$ and variance ${\sigma }^2$. The adequacy of the full model was inspected through residual diagnostics. To obtain a more conservative model structure, model selection was conducted through best subsets regression, based on the minimization of the AICc values. Unnested models within delta AICc ≤ 2 were averaged to obtain final parameter estimates. In the presence of interactions, post-hoc pairwise comparisons were conducted using Bonferroni adjustment for multiple contrasts. To further assess the consistency of our results, a similar analysis using an ordinal regression modelling approach on untransformed data, as well as a path analysis on transformed ranks were also performed (electronic supplementary material 1). All analyses were conducted with R [21] in RStudio [22].

3. Results

A total of 260 individuals (162 females and 98 males) were culled, 82 of which (59 females and 23 males) tested positive to T. gondii antibodies, leading to a mean prevalence of 31.5% (95% CI: 26.1–37.3).

The global model did not show major violations of assumptions. After model selection, four unnested models were considered plausible to explain variation in culling rank index. These included the interactions of T. gondii serological status with sex, level of human activity and proportion of bone marrow fat (plus their lower terms), as well as the effect of culling year (table 1). The results suggested that seropositive deer, in general, were not culled earlier than seronegative ones (the main effect of T. gondii was non-significant), but the effect was evident in females (table 2 and figure 1a), in individuals with medium–good body condition (table 2 and figure 1b), and in areas with greater human presence (table 2 and figure 1c). Furthermore, in 2017–18 deer were culled earlier than in 2016–17 (table 2 and figure 1d). Specifically, with respect to the interaction between T. gondii serological status and sex, culling rank index was significantly different between T. gondii-negative (n = 103) and T. gondii-positive (n = 59) female deer (difference = 0.805, p-value = 0.003), but not between T. gondii-negative (n = 75) and T. gondii-positive (n = 23) males (difference = 0.114, p-value = 0.695). Culling rank index was significantly different between negative females and negative males (difference = 0.505, p-value = 0.003), but not between positive females and positive males (difference = −0.187, p-value = 0.527) (figure 1a). With respect to the interaction between T. gondii serological status and level of human activity, culling rank index was significantly different between T. gondii-negative (n = 74) and T. gondii-positive (n = 44) deer in the highly used area (difference = 0.813, p-value = 0.002) but not between T. gondii-negative (n = 53) and T. gondii-positive (n = 25) deer in the areas with a low level of activity (difference = −0.003, p-value = 0.990), nor between T. gondii-negative (n = 51) and T. gondii-positive (n = 13) deer in the areas with a medium level of activity (difference = 0.570, p-value = 0.133). No difference occurred across different levels of human activity both in negative (low–medium = 0.111, p-value = 0.862; low–high = −0.338, p-value = 0.203; medium–high = −0.450, p-value = 0.069) and in positive deer (low–medium = 0.684, p-value = 0.161; low–high = 0.478, p-value = 0.196; medium–high = −0.206, p-value = 0.824) (figure 1c).

Table 1.

Model selection for the effects of human activity level (Hal), proportion of bone marrow fat (Bm), sex (sex), toxoplasma serological status (Ts), year (year) and their interactions on culling rank index. The table reports degrees of freedom (df) differences in Akaike's information criterion corrected for small sample size (ΔAICc) between each model and the model with the lowest AICc and Akaike's weights (weight). The ‘ + ’ indicates variables included in the models. Only models with delta ΔAICc value ≤ 2 are reported.

Hal	Bm	sex	Ts	year	Hal × Ts	Bm × Ts	sex × Ts	df	AICc	ΔAICc	weight
+	+	+	+	+	+	+	+	14	797.8	0.00	0.428
	+	+	+	+		+	+	10	799.4	1.53	0.199
+	+	+	+	+		+	+	12	799.4	1.61	0.191
+	+	+	+	+	+			11	799.6	1.72	0.181

Table 2.

Averaged parameter estimates of the models selected to explain the variation in culling rank index of red deer in the winter of 2016–17 and 2017–18 in Stelvio National Park. The table reports parameter estimates with associated lower and upper bounds of 95% confidence interval (95LCL and 95UCL, respectively).

parameter	coefficient	s.e.	95LCL	95UCL
(intercept)	4.086	0.184	3.724	4.447
T. gondii serological status [positive]	−0.266	0.403	−1.056	0.525
sex [male]	−0.477	0.175	−0.819	−0.135
proportion of bone marrow fat	−2.896	1.437	−5.712	−0.079
proportion of bone marrow fat2	3.181	1.321	0.592	5.770
level of human activity [medium]	−0.154	0.222	−0.589	0.281
level of human activity [high]	0.275	0.225	−0.166	0.716
year [2017–18]	−0.995	0.141	−1.270	−0.719
T. gondii serological status [positive] × sex [male]	0.673	0.343	0.000	1.346
T. gondii serological status [positive] × proportion of bone marrow fat	6.718	2.837	1.158	12.278
T. gondii serological status [positive] × proportion of bone marrow fat2	4.697	2.765	−0.723	10.117
T. gondii serological status [positive] × level of human activity [medium]	−0.553	0.434	−1.404	0.298
T. gondii serological status [positive] × level of human activity [high]	−0.817	0.341	−1.486	−0.148

Figure 1.

Marginal effects of the variables selected to explain variation in culling rank index of red deer in the winter of 2016–17 and 2017–18 in Stelvio National Park: (a) sex; (b) proportion of bone marrow fat; (c) level of human activity by serological status; (d) year of culling.

The same results were obtained using an ordinal regression modelling approach on untransformed ranks (electronic supplementary material, 1), supporting the fact that a linear model to the square root of ranks did not create any artefacts. Furthermore, a path analysis on square-root transformed ranks confirmed the occurrence of direct effects only, thereby supporting the linear model results (electronic supplementary material, 1).

4. Discussion

In our study, culling rank index was not affected by the occurrence of T. gondii serological status alone, as seropositive deer were not culled earlier than negative ones. An effect was nonetheless evident when T. gondii serological status was in interaction with other factors such as sex, body condition and level of human activity. Furthermore, deer were culled earlier in 2017–18 than in 2016–17. Previous studies conducted in laboratory settings have shown that T. gondii infection can manipulate intermediate host behaviour through changes in neurotransmitters [7]. This, in turn might increase boldness and possibly facilitate predation by the definitive host [17,23–26]. Using a similar approach to the one used in this study, [14] showed that the action of T. gondii led to infected rats being trapped earlier than non-infected rats, thereby supporting a ‘parasite-increased susceptibility to predation'. This form of parasite manipulation may also increase predation on intermediate hosts [27] in wildlife species, as recently observed in hyenas and primates [9,10].

Wildlife populations are characterized by great behavioural heterogeneity among individuals, including intersexual differences owing to differences in the trade-offs between energetic needs and risk avoidance [28,29]. For example, it may be argued that female deer naturally tend to show cautious behaviour and occupy forested areas more than males, possibly reflecting an antipredator strategy associated with their reproductive role [30]. As observed in rats, where T. gondii infection is thought to induce behavioural change through modulation of neurotransmitters and hormones involved in stress regulation [7], in red deer T. gondii infection might have favoured a relaxation of the perception of risk in females, thereby leading to the adoption of a bolder behaviour. Such a behaviour may manifest itself through increased use of open areas, with consequent higher chance of being culled. This effect was not observed in males, but we cannot rule out the impact of limited sample size. According to figure 1b, T. gondii seropositive individuals with medium–good body condition appeared to be culled earlier than individuals with poorer body condition. Given the limited data for individuals with a low proportion of bone marrow (n = 7 with bone marrow values < 0.5), any interpretation requires caution: with this caveat in mind, it might be that the mild effects of T. gondii are less evident in animals in poor condition because these effects are overwhelmed by stronger factors such as asthenia and a lower propensity for movement, possibly induced by starvation. On the other hand, artificial selection operated by hunters, biased towards individuals in better body condition, may also explain the pattern; as T. gondii does not affect body condition, it is unlikely that hunters may directly select for serological status of deer.

Toxoplasma gondii seropositive deer were culled earlier than negative ones in areas with a high presence of humans. This may be due to the fact that uninfected deer in areas with high human activity tend to be shy for antipredator reasons [31], thus are culled later than the seropositive ones, if the latter have become risk-prone following exposure with T. gondii. This suggestion is supported by the absence of difference between serological status of deer in areas with medium and low level of activity, where deer may normally be less cautious due to low human presence. Furthermore, in areas with a high level of human activity, the greater presence of domestic cats Felis silvestris catus, the final host of T. gondii, might increase the prevalence [32] and, possibly, the population effects on deer behaviour. With respect to difference between years, a weather station in the proximity of the study area, at 2300 m.a.s.l., showed that the mean daily snow cover was 86.3 cm in December 2017 and 54.8 cm in December 2016 [33]. This may explain the observed year-effect on culling rank index, as high snow cover constrains deer movement to low elevations, increasing deer density and favouring culling opportunities.

We are aware that a number of potential unmeasured variables could explain the relationship between T. gondii serological status and culling rank index. Our models accounted for several internal and external variables. Furthermore, during veterinary inspection at the checkpoint, no animal showed pathological signs with possible effect on behaviour, e.g. movement disability or impaired organ functionality, therefore the culling rank is unlikely to be biased by clinical co-morbidities. Other infections were detected in our study population. For example, red deer are an asymptomatic reservoir of Shiga toxin-producing Escherichia coli and Staphylococcus aureus, which may have zoonotic potential [34,35], but none of them are related to the probability of being culled.

Our study relies on the assumption that all infected animals are seropositive. However, it should be pointed out that, in the case of antibody waning, animals that are infected the longest may in fact be seronegative when tested. Taking this caveat into account, our study shows some evidence for possible behavioural effects of T. gondii in large herbivores under natural conditions. To our knowledge, the only other study investigating potential behavioural effects of T. gondii in herbivores was recently conducted on domestic sheep Ovis aries, providing evidence of a relationship between T. gondii infection and problem-solving behaviour, but no alteration of fear response [36]. In our study area, there are no large felines that can prey on deer, and the only species in which the parasite can complete its cycle is the domestic cat. More generally, however, modification of prey behaviour toward higher risk of predation could have evolutionary evolved to enhance the transmission to definitive hosts such as lynx Lynx lynx in Europe, or bobcat Lynx rufus and couguar Puma concolor in North America. In this respect, given the current expansion of the wolf on Alpine territory, it may be hypothesized that T. gondii exposure could become a facilitator of predation of deer by wolf, possibly with cascading effects on other ecosystem components, including definitive hosts.

Ethics

Samples were collected after the culling of animals for management purposes, according to the official culling plan for the reduction of red deer density authorised by Istituto Superiore per la Protezione e la Ricerca Ambientale (ISPRA), the Italian Ministry of Environment (Prot. 48585/T-A25-Ispra), in the Lombardy sector of Stelvio National Park. Therefore animals were not sacrificed for research purposes specific to this study.

Data accessibility

Metadata are included in the R-Script.

The dataset and R-Script used for this analysis are provided in the electronic supplementary material [37].

Declaration of AI use

We have not used AI-assisted technologies in creating this article.

Authors' contributions

M.N.: conceptualization, data curation, formal analysis, writing—original draft; L.C.: conceptualization, data curation, formal analysis, writing—original draft; N.F.: conceptualization, methodology, writing—original draft; T.T.: conceptualization, methodology, writing—original draft; L.P.: conceptualization, methodology, writing—original draft; A.G.: conceptualization, methodology, writing—original draft; P.L.: conceptualization, methodology, writing—original draft; C.L.: conceptualization, investigation, methodology, writing—original draft; N.F.: conceptualization, data curation, investigation, methodology, writing—original draft.

All authors gave final approval for publication and agreed to be held accountable for the work performed therein.

Conflict of interest declaration

We declare we have no competing interests.

Funding

We received no funding for this study.