INTRODUCTION
Alveolar echinococcosis (AE) of humans is an often fatal zoonosis, distributed in arctic and temperate regions of the northern hemisphere. It is caused by the ingestion of eggs of the cestode Echinococcus multilocularis which are disseminated into the environment by definitive host feces. Definitive hosts are mainly canids and particularly red foxes (Vulpes vulpes) in the paleartic Eurasian region (Eckert and Deplazes, Reference Eckert and Deplazes2004). Canids acquire infection by ingestion of small mammal intermediate hosts infected with the metacestode. Once released in the environment, eggs can survive up to 1 year in optimal microclimate conditions, being sensitive to dry and hot conditions (Veit et al. Reference Veit, Bilger, Schad, Schäfer, Frank and Lucius1995). Humans can become infected after accidental ingestion of eggs (e.g. via eating contaminated vegetables or after a close contact with infected canids).
The incidence of human AE in endemic areas of Europe can reach 0·74 per 105 per year (Vuitton et al. Reference Vuitton, Zhou, Bresson-hadni, Wang, Piarroux, Raoul and Giraudoux2003). However, some foci of very high prevalences in humans have been recorded in western China, where a prevalence of 14·3% was observed in villages of the eastern Tibetan Plateau (Sichuan province) (Tiaoying et al. Reference Tiaoying, Qiu, Wen, Craig, Xingwang, Xiao, Ito, Giraudoux, Wulamu, Wen and Schantz2005). The Tibetan Plateau, situated between 3200 and 5000 m above sea level, is inhabited by a large diversity and abundance of host species and is regarded as a méta-stable reservoir for Echinococcus multilocularis (Giraudoux et al. Reference Giraudoux, Pleydell, Raoul, Quéré, Wang, Yang, Vuitton, Qiu, Yang and Craig2006). The landscape is dominated by grassland and the main resource for people is provided by yak breeding (Bos grunniens). Livestock, especially grazing pressure, has been linked to the provision of optimal habitats for small mammal intermediate hosts, i.e. pikas Ochotona curzoniae and O. cansus, voles Microtus limnophilus and M. leucurus, and hamster Cricetulus kamensis, whose spatial distributions are clustered in colonies (Raoul et al. Reference Raoul, Quéré, Rieffel, Bernard, Takahashi, Scheifler, Ito, Wang, Qiu, Yang, Craig and Giraudoux2006). Four wild canid species (Vulpes vulpes, Vulpes ferrilata, Vulpes corsac and Canis lupus) as well as the domestic dog (Canis familiaris) are potential definitive hosts in this area. High prevalences of 59·1% and 57·1% for E. multilocularis have been recorded in Tibetan foxes and red foxes respectively (Qiu et al. Reference Qiu, Chen, Ren, Luo, Liu and Liu1995), though what proportion was actually E. shiquicus in Tibetan foxes is not clear (Xiao et al. Reference Xiao, Qiu, Nakao, Li, Yang, Chen, Schantz, Craig and Ito2005). Lower prevalences (12–15%) have been found in owned dogs following purgation studies in Shiqu's county (Budke et al. Reference Budke, Campos-Ponce, Qian and Torgerson2005a; Wang et al. Reference Wang, Raoul, Budke, Craig, Xiao, Vuitton, Campos-Ponce, Qiu, Pleydell and Giraudoux2010). However, dogs are numerous in Tibetan villages where they are kept as guard dogs, and stray dogs are protected by the local Buddhism religion. It is a rarity to see a house without dogs and stray dogs are frequently observed roaming in and around villages. In summer, some of the dogs go with the herdsmen in the summer pastures in high elevation grasslands.
Epidemiological studies have identified dog ownership, number of dogs owned and relationships with dogs (i.e. contact) as risk factors for human AE in Tibetan communities in northwest Sichuan (Tiaoying et al. Reference Tiaoying, Qiu, Wen, Craig, Xingwang, Xiao, Ito, Giraudoux, Wulamu, Wen and Schantz2005; Budke et al. Reference Budke, Campos-Ponce, Qian and Torgerson2005a), where small mammal population cycles have been documented (Giraudoux et al. Reference Giraudoux, Pleydell, Raoul, Quéré, Wang, Yang, Vuitton, Qiu, Yang and Craig2006). Preliminary analysis of dog feces collected in Shiqu's county showed that small mammal remains occurred in 28·8% of the samples (Wang et al. Reference Wang, Raoul, Budke, Craig, Xiao, Vuitton, Campos-Ponce, Qiu, Pleydell and Giraudoux2010). Furthermore, evidence for dog/small mammal interactions have been outlined by a positive correlation found between the proportion of fenced pastures, optimal habitats for small mammals, and both E. multilocularis prevalences and worm burden in owned dogs (Wang et al. Reference Wang, Vuitton, Qiu, Giraudoux, Xiao, Schantz, Raoul, Li, Wen and Craig2004, Reference Wang, Raoul, Budke, Craig, Xiao, Vuitton, Campos-Ponce, Qiu, Pleydell and Giraudoux2010). In such context, dogs are currently assumed to be the main definitive host in a peri-domestic cycle of E. multilocularis (parallel to the wildlife cycle). However, to date there have been no studies of the behavioural ecology of the dog population in Tibetan communities in this highly human AE endemic area.
Generally, studies on the role of behavioural ecology of dogs in parasite transmission cycles are rare despite the fact that dogs are known to be hosts for up to 60 zoonoses (Macpherson et al. Reference Macpherson, Meslin and Wandeler2000) and are often considered as the main source of contamination of the human environment (Craig et al. Reference Craig, Giraudoux, Shi, Bartholomot, Barnish, Delattre, Quéré, Harraga, Bao, Wang, Lu, Ito and Vuitton2000; Macpherson, Reference Macpherson2005; Thompson et al. Reference Thompson, Kutz and Smith2009). Behavioural ecology provides a means to study the role of dogs in transmission by identifying and quantifying defecation and predation behaviours. Most studies of dog populations in an epidemiological context provide measurements of population size, sex and age structure that can be used to guide control programmes (Anvik et al. Reference Anvik, Hague and Rahaman1974; WHO and WSPA, 1990). The spatial distribution of zoonotic pathogens in human environments can be assessed by analysing soil contamination arising from definitive host feces such as for Toxoplasma gondii (Afonso et al. Reference Afonso, Lemoine, Poulle, Ravat, Romand, Thulliez, Villena, Aubert, Riche, Rabilloud and Gilot-Fromont2008), Toxocara canis (Habluetzel et al. Reference Habluetzel, Traldi, Ruggieri, Attili, Scuppa, Marchetti, Menghini and Esposito2003) or Echinococcus spp. (Craig et al. Reference Craig, MacPherson, Watson-Jones and Nelson1988; Shaikenov et al. Reference Shaikenov, Rysmukhambetova, Massenov, Deplazes, Mathis and Togerson2004), or by studying host defecation behaviour (Anvik et al. Reference Anvik, Hague and Rahaman1974; Robardet et al. Reference Robardet, Giraudoux, Caillot, Boue, Cliquet, Augot and Barrat2008). Modern non-invasive methods have been developed to facilitate the assessment of definitive host infection status from fecal material via DNA analysis, e.g polymerase chain reaction (PCR), or immunological analysis, e.g. copro-antigen enzyme-linked immunosorbent assays (ELISA) (Dinkel et al. Reference Dinkel, Von Nickisch-Rosenegk, Bilger, Merli, Lucius and Romig1998; Raoul et al. Reference Raoul, Deplazes, Nonakac, Piarroux, Vuitton and Giraudoux2001; Eckert and Deplazes, Reference Eckert and Deplazes2004). Analysis of movements of the definitive host has proven useful in epidemiological studies: for example, telemetry technology provided relevant data to understand the spatial behaviour of foxes in cities and the risk they induce for the transmission of Echinococcus multilocularis (Robardet et al. Reference Robardet, Giraudoux, Caillot, Boue, Cliquet, Augot and Barrat2008). Concerning dogs, Watson-Jones and Macpherson (1998) have studied free-roaming dog movements by visual observations in Turkanian villages and found that the time dogs spent in the community settlement was correlated with other risk factors of Echinococcus granulosus transmission and with the incidence of the hydatid disease in the community.
Our general objective was to quantify and spatialize the role of domestic dogs in an E. multilocularis peri-domestic transmission cycle within Tibetan villages situated on the eastern Tibetan Plateau. Spatial patterns of dogs and fox's defecation and dog's movements were studied to address the following questions. (i) Which of dog or fox feces constitute the main source of E. multilocularis contamination in the human environment? (ii) How are E. multilocularis contaminated canid feces distributed within and at the periphery of the human environment (i.e. village areas)? (iii) What are the distributions (extent and frequency) of dog movements within the human and wild intermediate host environment and how and why spatial behaviours vary between individual dogs? (iv) Are there specific areas in the human environment where dogs could potentially be infected due to overlap of their movement range with small mammal colonies?
A new coprodiagnostic test (real-time PCR) combining parasite detection with discrimination of the canid host species from which the fecal sample derived, provided us with a tool to quantify the role of dogs versus foxes in E. multilocularis environmental contamination. Spatial distribution of canid feces (characterized by their specific identification and infection status) was investigated in relation to human and wild intermediate host population habitats (in and around village areas), together with the distribution of individual dog nocturnal movements in relation to dog owners' houses. The influences of dog individual traits (e.g. sex, relationships with owners) on inter-individual variability of movement were also analysed. Finally, we analysed the spatial overlap between small mammal population distributions and dog acitivity areas.
MATERIALS AND METHODS
Study sites
Shiqu township (Sichuan province, China) is situated between 3900 and 4200 metres above sea level on the Tibetan Plateau. Four villages in the vicinity (20–85 km) of Shiqu township were investigated (Fig. 1). The study started in May 2006 in Yiniu and Qiwu villages which were revisited in May 2007 along with 2 further sites: Mengsha and Jiefan. Villages were characterized by hundreds of scattered houses spread out along rivers and crossed by tracks. Within villages, areas of bare ground were common in contrast to the surrounding landscape which was dominated by grassland (fenced or grazed). Bushes were sparsely distributed on sloped grassland whilst wet grassland with micro-topography could be found near rivers.
Fig. 1. Localization of the study sites in Shiqu county (China).
The 4 villages were selected according to their epidemiological situation regarding the human AE prevalence (Tiaoying et al. 2005). Qiwu and Mengsha were identified as relatively low AE prevalence areas (2±0·9% and 3±5·8%, respectively) whilst AE prevalences of 11±13·3% and 12±8·8% were observed in Jiefan and Yiniu respectively. An epidemiological survey done twice in 2006 provided an estimate of 31–64 owned dogs per village (Jasmin Moss, unpublished data, Table 2). Dog population size was underestimated in Jiefan since only 70% of the households have been sampled.
Canid feces and environmental contamination
Fecal sampling
Canid feces were sampled in 1 hectare quadrats located inside and outside villages. A total of 33 quadrats have been sampled during 6 sampling surveys: 17 inside and 16 outside villages (Fig. 2). Inside villages, quadrats were in close vicinity to houses, while outside villages, quadrats were situated in grassland surrounding village cores. All the feces found within each quadrat were counted and geo-referred systematically. Fecal densities were estimated as the number of counted feces per quadrat.
Fig. 2. Feces sampling design per sampling survey. Village areas (90% MCP) and village houses are represented by grey areas and small green squares respectively. Large black squares represent quadrats and small black and red circles represent collected dog and fox feces respectively.
Fox and dog feces were discriminated visually (based on their shape, size and smell) by field experts who previously experienced fox feces collection in the field, from captive foxes or from necropsies. The expert accuracy to discriminate dog to fox feces was then tested by comparing the expert identification to the identifications of host species from molecular analysis (see following).
Copro-assays for host species DNA and parasite DNA
In 27 out of the 33 quadrats we collected a minimum of 25 (in 2006) and 15 (in 2007) fecal samples for coprodiagnostic tests in order to identify DNA of E. multilocularis and to discriminate canid host species directly from feces.
DNA from fecal samples was extracted as described previously by Dinkel et al. (Reference Dinkel, Von Nickisch-Rosenegk, Bilger, Merli, Lucius and Romig1998). For the specific detection of E. multilocularis and canid host species DNA (Vulpes vulpes, red fox; V. ferrilata, Tibetan fox; V. corsac, sand fox; and Canis lupus/familiaris, wolf/dog) we used 2 newly-developed hybridization probe-based real-time LightCycler (Roche, Mannheim, Germany) PCR assays (Dinkel et al. Reference Dinkel, Kern, Brinker, Vaniscotte, Giraudoux, Mackenstedt and Romig2011) as follows. (1) A real-time multiplex nested PCR (rtm-nested PCR) with fluorescence resonance energy transfer (FRET) probes for specific DNA detection and discrimination of E. multilocularis and canid host species in a single reaction, enabling specific species identification by analysis of the melting curves. The target sequence for E. multilocularis was part of the mitochondrial 12S rRNA gene, whereas a part of the mitochondrial cytochrome b gene was used as target for the canid host species identification. (2) A real-time nested PCR (rt-nested PCR) with FRET probes only for the detection and discrimination of the host species described above by melting curve analysis.
Specificity of the assays was tested with DNA isolated from Echinococcus granulosus (G1), E. ortleppi, E. canadensis (G6, G7), Taenia crassiceps, T. hydatigena, T. mustelae, T. pisiformis, T. serialis, T. taeniaeformis, Mesocestoides leptothylacus, Microtus arvalis, Arvicola terrestris, Felis catus and Martes foina. No cross-reactions were detected. The detection sensitivity of the rtm-nested PCR obtained with serial dilutions (100 ng – 1 fg) of genomic E. multilocularis DNA added to 10 μl total DNA solution from E. multilocularis-negative canid feces was 10 fg (Dinkel et al. Reference Dinkel, Kern, Brinker, Vaniscotte, Giraudoux, Mackenstedt and Romig2011). Based on a comparison of 47 fecal samples from China, the proportion of positives tested by the real-time mutliplex nested PCR was in the same range as when tested by the previously evaluated nested PCR (38% vs 30%) which showed a sensitvity of 70% to 100% depending on the number and gravidity status of worms present in the intestine (Dinkel et al. Reference Dinkel, Von Nickisch-Rosenegk, Bilger, Merli, Lucius and Romig1998).
Both, the rtm-nested and rt-nested PCR were conducted in 2 steps. For the first pre-amplification PCR, which allows the parallel amplification of DNA from various cestodes (355 bp fragment) and different carnivores (401 bp fragment) the primer mix contained a set of 4 primers (cf. Appendix 1; Dinkel et al. Reference Dinkel, Romig and Mackenstedt2006, Reference Dinkel, Kern, Brinker, Vaniscotte, Giraudoux, Mackenstedt and Romig2011). A total of 10μl of DNA from fecal samples was added to a 90 μl reaction mixture (cf. Appendix 1) and hot-start PCR was done for 40 cycles in a thermal block cycler (2720 Thermal cycler, Applied Biosystems, cf. Appendix 1). In a second step the LightCycler instrument (Version 1.5, Roche, Mannheim, Germany) was used for amplification of the target DNA and to monitor the development of the PCR product after each cycle. The rtm-nested PCR was performed with a final volume of 20μl in sealed LightCycler glass capillaries (cf. Appendix 1; Dinkel et al. Reference Dinkel, Kern, Brinker, Vaniscotte, Giraudoux, Mackenstedt and Romig2011). Differentiation was obtained by detection of mismatches with FRET probes using subsequent melting curve analysis (cf. Appendix 1).
To increase the sensitivity in samples which gave no result for the identification of canid host species we additionally used the rt-nested PCR. The detection of E. multilocularis was confirmed by using a published nested PCR assay (Dinkel et al. Reference Dinkel, Von Nickisch-Rosenegk, Bilger, Merli, Lucius and Romig1998). All PCRs included a negative control to monitor possible contamination and a PCR positive control. The isolated DNA from each fecal sample underwent an inhibition control to exclude false-negative results due to inhibition factors present in fecal samples.
Prevalences and their corresponding 95% confidence intervals were estimated. The confidence intervals were computed on the score-test statistic for binary data (Pearson chi-square test) using the prop.test function of the stats R package. Then, the difference in E. multilocularis fecal prevalence between quadrats inside and outside of village areas was tested using a Chi-squared test.
Feces distribution in the environment
Random versus aggregated distributions of feces were assessed in each sampling survey over all quadrats. The ratio of feces density variance over its mean was used as an index of the aggregation. Null models, Generalized Linear Models (GLM) assuming Gaussian, Poisson (random) or Negative Binomial (aggregated) density distributions, fitted on all quadrat densities, were then compared. Density distribution providing the lowest second-order bias corrected Akaike Information Criterion (AICc) was then used in further modelling.
The effect of the type of quadrat (inside vs outside village) was then tested on feces density. Because houses were scattered and village boundaries hard to assess, analysis was completed by investigating how feces density varied according to quadrat distances from houses. This provided additional information concerning the range of feces distribution and potential environmental contamination around houses. Generalized Linear models incorporating the effect of the type of quadrat (inside vs outside village) and the effect of the minimum distance between the quadrat and the nearest village house, these effects being considered alone and associated, were evaluated using AICc and corresponding Akaike weights (wi) that provide the probability of each model to be the best given the data and the set of candidate models (Burnham and Anderson, Reference Burnham and Anderson1998). Before running the models, we tested whether it would be beneficial to add random effects on the year and village (sampling survey). Finally, aggregation in feces distributions within each quadrat was tested by estimating the Clark and Evans index (R index) (Clark and Evans, Reference Clark and Evans1954) using the spatstat package under R statistical software (R Development Core Team, 2008). The hypothesis that the proportion of quadrats with aggregative distribution differed inside and outside village areas was tested using a Chi-squared test.
Dog movement survey
Dog class and data collection
In Tibetan communities dogs could be free roaming, partially free or always tethered. Two categories of dogs were sampled: dogs restrained all day and released at night (owned tethered=OT dogs) and dogs always free (owned free=OF dogs). Dog owners supplied information regarding the sex of the dog and whether it usually visits higher altitude summer pastures. The latter variable was named ‘pastoral activity’ in the following analysis. Relocations were recorded by GPS collars (WildTrax, Bluesky Telemetry) that were set up on dogs during the day and removed the following day, 1 location being recorded every 10 min.
Since OT dogs were only free from their tether at night, we focused on studying only nocturnal movements for the whole dog population. The time-period of relocation records was homogenized between dogs: the beginning of the night (8:00 pm) was set as the earliest hour at which owned and tethered dogs were known to be released, while the end of the night was considered as the latest hour at which the same dogs were known to be re-tethered (8:00 am), providing trajectories of 12 h for all dogs. However, the lack of accurate information on each dog's hours of freedom and when owners actually re-tethered them may mean that some OT dogs could have been free for less than 12 h and their trajectories may include an unspecified period when they were tethered.
Analysis of distributions of single-night dog movements
We characterized 3 areas of dog population space use: the core areas, the mean dog population activity areas and areas covered by excursive trajectories (extended dog movements).
First, relocation distributions were analysed in relation to the distances between the dog and its release point (RP, i.e. its owner's house). The core area, classically used in the context of home range estimation (Seaman and Powell, Reference Seaman and Powell1990; Hodder et al. Reference Hodder, Kenward, Walls and Clarke1998), was used here as a means to define areas around dog owners' houses where dog relocations were aggregated, i.e. where dogs spent the majority of their time. Minimum Convex Polygon (MCP) areas were estimated for different percentages of included relocations on the basis of their distances to their barycentre, i.e. to the dog RP. Friedman test followed by multiple comparisons were done, using the function friedmanmc of the pgirmess R package, to identify the percentage of relocations included in MCP estimation that lead to a significant increase in MCP area. The nocturnal core area was estimated per dog as the MCP area that included the relocation below such threshold. Dog trajectories were described by estimating the maximum distance travelled from the dog RP inside and oustide nocturnal core areas. Additionally, the mean distance travelled per hour provided a measure of their nocturnal activity (Ciucci et al. Reference Ciucci, Boitani, Francisci and Andreoli1997).
The only night sampling period used to characterize dog movements prevented estimation of the dogs' home ranges. Instead, we defined the ‘one night activity ranges’ of the dog population of one village – or their home range estimated on a delimited time-period (Okubo and Levin, Reference Okubo, Levin, Antman, Marsden, Sirovich and Wiggins2001) – by subsampling each dog trajectory in 1 location/h/dog. Utilization Distribution functions were then estimated by the Kernel method (Worton, Reference Worton1989). The smoothing parameter href estimated by the ad-hoc method available in the function kernelUD of the adehabitat package (Calenge, Reference Calenge2006; R Development Core Team, 2008) is known to suffer from over-estimation of the area when multiple clustering is observed in the relocation data set (Seaman and Powell, Reference Seaman and Powell1996). As dog relocations were aggregated around their RP, we estimated an optimal h for each sampling survey as a re-scaling of href which provided an area that did not connect clusters between which no movements (successive relocations) have been recorded. The dog population nocturnal activity range was defined as the area in which dogs spent the majority of their time - we defined this on 90% of the relocations.
Finally, the delineation of dog nocturnal activity range allowed us to define excursive trajectories using dog relocations falling outside the dog population activity range. The number of such excursive relocations was assessed per dog and main patterns of excursive trajectories were identified. Then, excursive visiting paths were identified as at least 3 successive excursive relocations. We finally explored the existence of aggregated relocations for each dog, as a potential surrogate for foraging behaviours, by estimating the Clark and Evans index on excursive relocations.
Correlations of inter-individual dog movement parameters
Correlations between movement parameters, i.e the maximum distance from the release point, the speed, the core area and the number of excursive relocations per dog, were assessed by Spearman correlation tests. Then, the influences of dog sex, freedom from tether and pastoral activity on normalized (log transformed) movement parameters were investigated among the dog population via linear models, except for the effects on the number of excursive relocations which were assessed by a GLM assuming a Negative Binomial distribution. We compared AICc of the null models with all possible combinations of covariates. For each movement parameter, we carried out a preliminary test for the addition of random effects on the sampling survey and individual dogs.
Spatial interactions between dogs and small mammal populations
In the 4 villages, we walked radial transects along which indices of activity of small mammals (mainly feces and holes of Ochotona spp., Microtus spp. and Cricetulus kamensis) were recorded within every interval of 10 paces (Raoul et al. Reference Raoul, Quéré, Rieffel, Bernard, Takahashi, Scheifler, Ito, Wang, Qiu, Yang, Craig and Giraudoux2006). Transects started from the centre of the village and ended in the surrounding grassland. We first estimated the minimum distance separating positive intervals for small mammal presence to all dog RP's at each village. Differences in frequency distribution of positive intervals were assessed between the dog population activity areas and the excursive areas only. This was due to low or even null numbers of transect intervals being situated within individual dog one night core areas. The presence/absence of small mammals were modelled with a Generalized Linear Mixed Effects (GLME) model assuming a logistic binomial link function, against the different dog areas and including a random effect on the sampling survey. Differences in small mammal presence indices between dog areas were assessed by comparing model AICc with the null models and by estimating their risk ratio (Bailey and Alimadhi, Reference Bailey, Alimadhi, Kosuke, King and Lau2008).
All statistical analyses were achieved using the R statistical software (version 2.10) (R Development Core Team, 2008).
RESULTS
Canid feces and environmental contamination
Fecal species identification and prevalences
A total of 980 feces were counted and geo-referred in the 4 villages, among which 284 feces were analysed for host species identification and presence of E. multilocularis contamination (Table 1).
Table 1. Fecal samples in all quadrats of each sampling survey: total number, aggregation index, number of PCR analyses, number of non-inhibited PCR analyses and prevalence of E. multilocularis DNA for all samples and dog and fox feces separately
(a) Number of feces analysed by PCR;
(b) number of identified samples (non failed PCR);
(c) 95th confidence intervals for prevalences (in%).
PCR inhibition problems were observed for 74 (46·8%) of the 158 feces collected in 2007 and 30 (23·8%) of the 126 feces collected in 2006 (Table 1). These feces could not be used for further analysis of fecal prevalences.
Among the remaining 180 feces collected, based on expert knowledge, 142 were attributed to dog, 13 to fox (Vulpes ferrilata) and 25 could not be identified. Actually, PCR analysis showed that 84·6% (11/13) of the feces attributed to fox and that 100% (142/142) of feces attributed to dog were correctly classified. Thus, a very low proportion of foxes feces (2 of 155) were confused with dog feces in the field. Among the 284 feces collected in quadrats and analysed by PCR for E. multilocularis DNA only, experts identified a total of 14% of fox feces.
The overall coproPCR prevalence for E. multilocularis DNA was 20·5%, with 22·5% and 15·4% respectively for dogs and foxes (Table 1). However, there was no significant difference between prevalences of E. multilocularis infection in dog and fox feces samples tested (Pearson's Chi-squared test with Yates' continuity correction, Chi-squared=0·06, D.F.=1, P value=0·81).
We also failed to detect a significant difference in overall fecal prevalence of E. multilocularis between feces samples located inside villages (22·1%, 32/145) versus those located outside villages (14·3%, 5/35, Pearson's Chi-squared test with Yates' continuity correction, Chi-squared=0·62, D.F.=1, P value=0·43).
Feces distributions
Aggregation indices were greater than 1 in each sampling survey as well as for the total (Table 1). Clearly, feces were over-dispersed in villages and fecal density distributions were better captured on the whole data set by a model incorporating a Negative Binomial distribution (AICc= 281·62 versus AICc=372·32 and AICc=2343·72 for Gaussian and Poisson distribution respectively). Consequently we used a Negative Binomial fecal distribution for further modelling.
We found no effects of the year and the village on feces densities (GLM, Δ AICc=0·03, w i=0·5 and Δ AICc=2·49, w i=0·2, respectively in comparaison to the null model AICc). Those variables were thus not considered as random effects in the further models. We also failed to detect an effect of inside village versus outside village quadrat on feces densities (GLM, AICc=284·3, w i=0). However, the model including the minimum RP distance provided the approximate best fit (GLM, AICc=138·1, w i=1). Feces density reached a maximum (354 feces/ha) in 1 quadrat situated at a minimum distance of 48 m from village houses while all quadrats situated farther than 200 m from village houses contained less than 14 feces (Fig. 3). Low feces density was observed for 4 quadrats situated at less than 50 m from houses suggesting that the model particularly over-fitted feces densities for those quadrats.
Fig. 3. Distribution of feces densities (number of feces per hectare) according to the minimum distance between quadrats and village houses.
Finally, feces were aggregated in 84% of the quadrats. No significant difference was observed between the proportion of quadrats showing aggregation situated inside (88%, 15/17) and outside (81%, 13/16) villages (Chi-squared=0·01, D.F.=1, P value=0·9).
Dog movement survey
Sampling effort
Seventy-eight dogs were fitted with GPS collars for 1 night, 7 were fitted for 2 nights, 2 for 3 nights and 1 for 4 nights. A total number of 96 one-night trajectories were recorded in the 4 villages (Table 2). Collected GPS collar connections to satellites induced a mean of 19·5% (S.D.=22·9%) missing values per dog trajectories that were excluded for further analysis. The mean positioning error provided by GPS collars and estimated over all dog relocations varied according to the sampling survey from: 4·7 m (S.D.=2·9) to 10·5 m (S.D.=4·6) horizontally and from 6·8 m (S.D.=5) to 15·20 m (S.D.=8·3) vertically.
Table 2. Number of GPS collared dogs (sampling effort) per sampling survey and dog category (freedom level, pastoral activity and sex)
(a) Summer pasture dog;
(b) non summer pasture dog;
(c) percentage of sampled dogs over the total number of dogs estimated (into brackets);
(d) undetermined.
Distribution of single-night dog movements
MCP areas differed according to the percentages of relocations included in their estimation (Friedman test, Friedman chi-squared=837, D.F.=9, P value <2·2e-16) (Fig. 4). Most of the relocations were aggregated in areas smaller than the total activity area which relied on few exceptional relocations. For example, when including 80% of the relocations, we observed that a median value of about 10% and an outlier below 70% of the total activity area could be captured among dogs. A very low proportion of relocations (the last 10%) provided more than 70% of the total area for half of the dogs. Because a significant difference was found between the inclusion of 80 and 100% of relocations (P value=0·05), the one night core areas were estimated for each dog as the MCP area including 80% of its relocations.
Fig. 4. Boxplots of the proportion of the maximum MCP area according to the percentage of the total number of relocations included in the estimation of MCP area.
Dogs were located at 10 to 1500 m from their release points but were generally under 250 m from this point, moving slowly (162 m/h on average) within a less than 1 ha core area for the large majority of them. Among individual dogs, one night core areas and maximum RP distances showed dispersed (standard deviations larger than the means) distributions aggregated around low values. Fifty percent of the dogs had night core areas smaller than 0·16 ha, stayed within a 36 m range around their RP and travelled a maximum distance of 152 m from their RP. However, 5% of them had night core area larger than 1·19 ha, moved at 115 m maximum from their RP and travelled a maximum distance of 921 m. Optimal h (smoothing parameter) estimates for each sampling survey ranged from 35·7 (scaling factor=0·2) to 133·9 (scaling factor=0·5) (Table 3). The main activity areas of dogs ranged from 32·5 to 174·5 ha according to the village and sampling survey; 39% of the dogs had at least 1 relocation falling outside the population activity area (Table 3). The mean number of excursive relocations per dog varied from 2·7 to 5·1 depending on the sampling survey. The mean distance of such excursive relocations from the dog RP varied between 309 and 583 m according to the village, with a minimum distance of 58 m.
Table 3. Parameters of the dog population activity areas and excursive (‘exc’) relocations for each sampling survey
(a) Distance of excursive relocations from dog RPs;
(b) dogs having at least 3 successive excursive relocations (visiting path);
(c) number.
Among those trajectories, 2 patterns of dog excursive relocations were observed: visiting paths consisting of a minimum of 2 successive relocations (Fig. 5, a–d), and ‘away and back’ movements relying on isolated excursive relocation from the dog one-night core areas (Fig. 5, e)). Eleven dog trajectories were considered as visiting paths among which 5 of them showed aggregative excursive relocations (mean(R)=0·85; S.D. (R)=0·11) such as the trajectory illustrated on Fig. 5, f).
Fig. 5. Example of trajectories (black lines) and excursive relocations (dots) for 6 dogs, from (a) to (f), in reference to the dog population main activity area (grey polygons). Dog trajectories shown from (a) to (d) correspond to visiting paths while dog trajectories shown in (e) and (d) correspond to 'way and back’ paths.
A maximum of 9 successive relocations (or 90 min excursive trajectories) were recorded for a male dog fitted in Qiwu village (2007) and which regularly visit summer pastures.
Correlations of inter-individual movement parameters
As expected, all movement parameters were significantly correlated to each other (P values <0·01). Spearman correlation coefficients ranged from 0·53 between the number of excursions and the core area, to 0·84 between the maximum distance from RP and the speed. No effects of the individual, the year and the village have been identified using AICc comparisons. Therefore those variables were not incorporated as random effects in the models. Dog sex slightly influenced dog maximum RP distances (LME, AICc=236·4, ΔAICc=2·4, wi=0·3), median area being larger for males (median=165) than for females (median=117). No effects of the other factors on other movement parameters have been detected.
Spatial interactions of dog and small mammal populations
A total of 42 standard radial transects were done. Evidence for small mammal presence was found close to dog owners' houses at a minimum distance varying between 32 and 122 meters according to the sampling survey (Table 4). Positive transect intervals (92%) were identified in 2006 inside dog individual one night core areas in Qiwu village (Table 4). The proportion of transect intervals with the presence of small mammals was higher in the dog excursive area than in the dog population activity area (15·5–56·5% versus 3–37%, GLME, AIC=5275, ΔAICc=138; Risk ratio=2·36 (1·86–2·93)).
Table 4. Small mammal spatial distributions inside the three type of dog areas of activity. Confidence intervals for the percentages and risk ratios provided in parentheses were estimated on the inverted score test statistic (Pearson chi-square) using the prop.test function of the stats R package
(a) Small mammal sampling effort: numbers of transect. Number of transect intervals in parentheses;
(b) minimum distance from dog RP;
(c) total number of transect intervals;
(d) percentage of positive transect intervals.
DISCUSSION
In western China, including Tibetan regions, domestic dogs appear to play an important role in the risk of human alveolar echinococcosis caused by exposure to eggs of the fox tapeworm E. multilocularis (Macpherson and Craig, Reference Macpherson, Craig, Macpherson, Meslin and Wandeler2000; Craig et al. Reference Craig2006). High prevalences (>10%) of E. multilocularis have been recorded in owned dogs in Tibetan communities of northwest Sichuan province located in the eastern edge of the Tibetan Plateau (Budke et al. Reference Budke, Campos-Ponce, Qian and Torgerson2005a; Wang et al. Reference Wang, Raoul, Budke, Craig, Xiao, Vuitton, Campos-Ponce, Qiu, Pleydell and Giraudoux2010). However, in contrast to studies of fox definitive hosts in European and Japanese E. multilocularis endemic areas, there is virtually no information on dog behavioural ecology and its potential role in the peri-domestic transmission of E. multilocularis. To address this issue we used a multi-disciplinary approach that combined PCR analysis (of both host and parasite targets) of spatially distributed fecal samples and dog behavioural ecology study.
In the 4 Tibetan villages situated in the area of highest endemicity for E. multilocularis on the Tibetan Plateau, 7% of feces analysed were identified by host DNA amplification as Tibetan fox (Vulpes ferrilata) origin. This endemic fox is known to inhabit grassland areas far (>1000 m) from human influences (Gong and Hu, Reference Gong and Hu2003; Wang et al. Reference Wang, Wang and Liu2007) but our research demonstrates that Tibetan foxes also forage and defecate in and around villages. Considering that 15% of fox feces were positive for E. multilocularis DNA then foxes could be considered as a source of human environmental contamination. However, fox feces were found in much smaller proportions than dog feces in and around villages (8% of feces analysed by multiplex PCR). Prevalence of E. multilocularis did not differ significantly between dog and fox feces but the larger density and number of dog feces found in human environments suggests an important role of dogs in fecal contamination of the environment in Tibetan villages in this area (Shiqu county, Ganze prefecture, Sichuan).
An important drawback of our parasitological analysis remains the large proportion of inhibited PCRs that constrains the feces sample size and consequently the interpretation of observed prevalences on their own. Since both species exhibit quite similar foraging behaviours in village areas (eating remains of human food or carrion), apart from predation behaviour, no credible argument towards a high sensitivity of fox or dog DNA to inhibitors can be assumed. Feces identification by experts in the field provides additional information on the contamination distribution and can help to compensate the current limitations of the molecular methods used here. This emphasizes the importance of using multiple methods for feces collected from the analysis of environment contamination.
All dogs that were GPS collared shared a common pattern of single-night relocation distributions: i.e. they were aggregated around owners' houses. The area over which owned dogs spent the majority of their time at night ranged from 0·004 to 10 ha, and 95% of dogs stayed for 80% of their time at a maximum distance of 115 m from their owners' houses. That area probably corresponds to the small and easily defendable territory where aggressive dog behaviours are mostly expressed (Boitani et al. Reference Boitani, Francisci, Ciucci, Andreoli and Serpell1995; Macpherson et al. Reference Macpherson, Meslin and Wandeler2000). In contrast, 40% of dogs were recorded outside their main population activity area, at distances varying from 58 to 1519 m from their release points and during a maximum time-period of 90 min. Even if recorded for only 12 h and at the dog individual level, the patterns of relocation distributions identified here are in agreement with those observed in feral or free-roaming dogs for which core areas (50% of their home range) corresponded to 5·7% of the total home range area, including dens, resting and retreat sites (Boitani et al. Reference Boitani, Francisci, Ciucci, Andreoli and Serpell1995). The large dispersion we observed in the dog maximum release point distance, core area and core release point distance emphasized the existence of 2 types of dog behaviour: fast and long distance movements versus a large proportion of relatively slow movements concentrated around the owner's home. The fact that all movement variables were correlated to each others suggests that collared dogs which moved farthest also had a tendency to have larger core areas, speeds and excursive relocations.
Our observations on dog nocturnal movements agree with Meek (Reference Meek1999) who found inter-individual variability in free-roaming dog home ranges and discriminated sedentary (mean home range=2·6 ha) from wandering dogs (mean home range=927 ha). We found that only dog sex could partly explain this observed variability (maximum distance travelled by males being more active than females) which confirms previous studies (Boitani et al. Reference Boitani, Francisci, Ciucci, Andreoli and Serpell1995). The inter-individual variation in nocturnal behaviour might interact and hide the effects of the nature of dog/human relationships (pastoral activity and free roaming) on nocturnal spatial behaviour. It is well known that dog-owner relationships influence social organization of the dog and consequently their territory definition and spatial utilization - owned dogs having smaller home range size than owner-less dogs (Macpherson et al. Reference Macpherson, Meslin and Wandeler2000). Repeated trajectory recordings for each individual dog and stratified sampling of the dog population would help to identify those effects. Additionally, one weakness of our movement survey is that dogs were followed during a limited time-span in May. However, dog movements might vary between months and this variation should be investigated to fully assess dog space uses in relation to the transmission risk of the parasite. For example, in winter, individual movements might be even more concentrated in village houses and their close surroundings due to the climatic conditions. Also, at the population level, the number of dogs and their frequentation of the village area should be higher because of the lower mobility of herdsmen.
Dog feces densities were found to be higher in quadrats situated inside villages than outside villages and were over-dispersed regarding their minimum distance from houses. This aggregative distribution of feces matched with dog nocturnal relocation distributions. The distance range of feces contamination (0 to 200 m) and the maximum density of feces observed at 50 m fell within the maximum distance from the owner's house recorded within the core areas (345 m). Therefore, dogs mainly defecated within their core areas. In urban areas, dog feces contamination has been observed in places close to houses, in parks and residential areas or in public playgrounds and were associated with a high rate of soil contamination by eggs of the zoonotic nematode Toxocara (Anvik et al. Reference Anvik, Hague and Rahaman1974; O'Lorcain, Reference O'Lorcain1994). Concerning pastoral areas of the Tibetan Plateau, the close relationship and spatial proximity between dogs and humans may cause a considerable increase in risk of Echinococcus spp. transmission. Given the average E. multilocularis fecal prevalence found (21%) and the dog spatial defecation behaviour outlined above, we suggest that the village area, and particularly within about 200 meters around dog owners' houses, might be the place for probable higher contact between dog feces and the human population, and thus increased risk of exposure to the parasite. On the other hand, the transmission risk is likely to decrease with distance from a dog owner's house. Since the Tibetan villages we studied did not vary in feces spatial distribution patterns (and prevalences), we could cautiously generalize that a similar pattern probably occurs in other villages of the study region. Furthermore, at a higher resolution, feces were found to be aggregated within quadrats inside the village area. Scent-marking behaviour and social interactions among dogs, arising within the population area of activity, might explain such aggregation (Boitani and Ciucci, Reference Boitani and Ciucci1995). Feces of different dogs are likely to be found within such micro-aggregation zones inside villages where E. multilocularis environment contamination might be particularly high.
In a control programme perspective, it would be necessary to relate contamination patterns with dog individual traits, for instance to quantify the relative role of owned vs stray dogs in environmental contamination, and to estimate whether village contamination by feces is due to a large or low numbers of dogs. The use of DNA micro-satellites to relate individual dogs to feces sample is promising for such a purpose. Concerning the predicted transmission risk to humans, interpretations of our findings are drawn assuming that risk factors related to human behaviours are constant along a gradient from house to outside village. However, human behaviours are known to influence the transmission risk and such parameters are often missing in transmission risk estimation (Macpherson, Reference Macpherson2005). Hygienic conditions and behaviours that induce direct contact with the dogs such as feeding it, playing with it and allowing it to stay inside the house, or direct contact with feces have been found to be risk factors (Craig et al. Reference Craig, Giraudoux, Shi, Bartholomot, Barnish, Delattre, Quéré, Harraga, Bao, Wang, Lu, Ito and Vuitton2000; Wang et al. 2001; Schantz et al. 2003; Tiaoying et al. 2005). Investigating the spatial distribution of ‘at risk’ human behaviours within and around the villages would probably provide a more complete assessment of the transmission risk.
We found that dogs also defecated sporadically during their excursive trajectories, further than 200 m from houses and outside village areas up to a distance of 1519 m and therefore could contribute to the contamination of the wild reservoir (small mammals) mainly inhabiting adjacent grassland areas. Small mammals infected with E. multilocularis metacestodes can contain thousands of protoscoleces and dog infection pressure appears high when they predate on small mammals (Budke et al. Reference Budke, Jiamin, Craig and Torgerson2005b). The same authors ran an epidemiological model to estimate the contact rate between dogs and small mammals and found a value of 0·52 infective insult per year to explain actual dog prevalences in the same study area (i.e. villages surrounding Shiqu township). This contact rate in the case of E. multilocularis transmission is known to rely on prey densities and prevalences (Giraudoux et al. Reference Giraudoux, Delattre, Takahashi, Raoul, Quéré, Craig, Vuitton, Craig and Pawlowski2002; Raoul et al. Reference Raoul, Deplazes, Rieffel, Lambert and Giraudoux2010) as well as on predator behaviour (Hegglin et al. Reference Hegglin, Bontadina, Contesse, Gloor and Deplazes2007 ; Raoul et al. Reference Raoul, Deplazes, Rieffel, Lambert and Giraudoux2010). On the Tibetan Plateau, relatively high prevalences of E. multilocularis have been recorded in Microtus irene (25%) and Ochotona curzoniae (up to 7·7%) (Qiu et al. Reference Qiu, Liu, Schantz, Ito, Carol and He1999). In parallel, such species are susceptible to undergo multi-annual population fluctuations (Giraudoux et al. Reference Giraudoux, Pleydell, Raoul, Quéré, Wang, Yang, Vuitton, Qiu, Yang and Craig2006) with high abundance phases in particular habitats (see Raoul et al. (Reference Raoul, Quéré, Rieffel, Bernard, Takahashi, Scheifler, Ito, Wang, Qiu, Yang, Craig and Giraudoux2006) for Microtus limnophilus and Ochotona curzoniae in the study area).
The current study suggests that small mammal density increases while intensity of spatial use by dogs decreases with the distance from houses. Spatial overlapping of the two host populations is therefore maximal in dog activity areas, excluding core areas, where small mammal densities are intermediate. Similar spatial patterns have been found between foxes and small mammals in peri-urban areas in Switzerland (Deplazes et al. Reference Deplazes, Hegglin, Gloor and Romig2004). Indices for small mammal species presence were found with largest frequency outside dog core activity areas, i.e. in the areas where dogs would only visit on occasional excursions. Nevertheless, small mammals were also observed, even at lower densities inside dog population areas, and even in dog core areas for one of the studied villages. This suggests that contact between dogs and grassland small mammals could even occur inside the village area and does not necessarily require dog excursions into the surrounding grassland. Thus, the possibilities of a peri-domestic parasite cycle are real, and it could be maintained in close vicinity of the dog owner houses. The villages occupied during winter or by non-transhumant human and dog populations are potential areas of small mammals/dog interactions and therefore dog contamination. This situation is probably similar in summer pastures where dogs are generally attached close to the tents and released for the night (Giraudoux, personal observation).
At the population level, our study constitutes a preliminary step in the quantification of the frequency of dog and small mammal interactions. Further data are required to investigate the plasticity of dog predation behaviour in relation to small mammal prey densities by dietary analysis. Since dog predation pressure could shift according to the variation in anthropogenic resources (Macpherson et al. Reference Macpherson, Meslin and Wandeler2000), particular effort should be put on studying the effects of dog ‘domestication states’ as defined by the feralization model developed by Boitani and Ciucci (Reference Boitani and Ciucci1995), on predation behaviours. In parallel, small mammal habitats have already been defined in this region of China (Raoul et al. Reference Raoul, Quéré, Rieffel, Bernard, Takahashi, Scheifler, Ito, Wang, Qiu, Yang, Craig and Giraudoux2006; Marston, Reference Marston2008) but further work is required to build an accurate predictive mapping of small mammal occurrence within the whole area surrounding villages. Ultimately such contact rate could be of particular relevance as a parameter of epidemiological models.
ACKNOWLEDGEMENTS
We address many thanks to local people for their precious help in the field and to Rémi Helder from the the URCA-CERFE research team for his critics on the analysis of GPS data.
FINANCIAL SUPPORT
This work was supported by the NIH Fogarty International Center (Grant numbers RFATW-00-002, RO1 TW001565). The content is solely the responsibility of the authors and does not necessarily represent the official views of the Fogarty International Center or the National Institutes of Health.
APPENDIX 1. Technical details of copro-assays
Pre-amplification PCR (Dinkel et al. Reference Dinkel, Romig and Mackenstedt2006, Reference Dinkel, Kern, Brinker, Vaniscotte, Giraudoux, Mackenstedt and Romig2011)
– Primers: P60 short.for (5′-TGG TAC AGG ATT AGA TAC CC-3′)/P375 short.rev (5′-TGA CGG GCG GTG TGT ACC-3′) and CVF for (5′-TTA ATG ACC AAC ATT CGA AA-3′)/CVF rev (5′-AGG/T ACA/G TAG/C CCC ATA/G AAA/T GC-3′)
– PCR reaction mixture: 15 mm Tris–HCl (pH 8·0), 50 mm KCl, 2·5 mm MgCl2, 200 μ m of each dNTP, 30 pmol of each cestode primer (P60 short.for/P375 short.rev), 40 pmol of each carnivore primer (CVF for/CVF rev) and 2·5 units AmpliTaq Gold (Applied Biosystems).
– Cycles: denaturation for 30 s at 94°C, annealing for 1 min at 54°C and elongation for 40 s at 72 °C.
Rtm-nested PCR (Dinkel et al. Reference Dinkel, Kern, Brinker, Vaniscotte, Giraudoux, Mackenstedt and Romig2011)
– LightCycler glass capillaries: 5 pmol of primer Pnest.for (5′-ACA ATA CCA TAT TAC AAC AAT ATT CCT ATC-3′) and Pnest.rev (5′-ATA TTT TGT AAG GTT GTT CTA-3′) (Dinkel et al. Reference Dinkel, Von Nickisch-Rosenegk, Bilger, Merli, Lucius and Romig1998), 10 pmol of primer CVF light.for and CVF light.rev, 2 μ m of each hybridization probe (emulti-fl, emulti-705, CaVuFe1-fl and CaVuFe2-640), 2 μl of LightCyler DNA Master HybProbe (Roche, Mannheim, Germany), 2·5 mm MgCl2 and 1 μl of PCR product of the first PCR. gg
– Cycling conditions: 15 s at 94°C, followed by 45 cycles of 4 s at 94°C, 15 s at 50°C and 25 s at 72°C.
– Melting curve analysis: 4 s at 94°C, 15 s at 50°C and 25 s heating to 90°C with a ramping rate of 0·2°C/s for emulti-fl/emulti-705 and 10 s at 95°C, 10 s at 30°C and heating to 90°C (0·2°C/s) for CaVuFe1-fl and CaVuFe2-640 (melting temperature [Tm] 60·0°C E. multilocularis; V. vulpes 56·0°C; V. corsac 45·4°C; V. ferrilata 44·8°C; C. lupus/familiaris 38·0°C).
Rt-nested PCR (Dinkel et al. Reference Dinkel, Kern, Brinker, Vaniscotte, Giraudoux, Mackenstedt and Romig2011)
– PCR reaction mixture: 10 pmol of each primer (CVF light. for and CVF light.rev), 2 μ m of each hybridization probe (CaVuFe1-fl and CaVuFe2-640), 2 μl of LightCyler DNA Master HybProbe (Roche, Mannheim, Germany), 2·5 mm MgCl2 and 1 μl of PCR product of the first PCR.
– Thermal cycling and melting curve analysis was done as described for rtm-nested PCR.
