|Home | About | Journals | Submit | Contact Us | Français|
Ecological changes are recognized as an important driver behind the emergence of infectious diseases. The prevalence of infection in ticks depends upon ecological factors that are rarely taken into account simultaneously. Our objective was to investigate the influences of forest fragmentation, vegetation, adult tick hosts, and habitat on the infection prevalence of three tick-borne bacteria, Borrelia burgdorferi sensu lato, Anaplasma phagocytophilum, and Rickettsia sp. of the spotted fever group, in questing Ixodes ricinus ticks, taking into account tick characteristics. Samples of questing nymphs and adults were taken from 61 pastures and neighboring woodlands in central France. The ticks were tested by PCR of pools of nymphs and individual adults. The individual infection prevalence was modeled using multivariate regression. The highest infection prevalences were found in adult females collected in woodland sites for B. burgdorferi sensu lato and A. phagocytophilum (16.1% and 10.7%, respectively) and in pasture sites for Rickettsia sp. (8.7%). The infection prevalence in nymphs was lower than 6%. B. burgdorferi sensu lato was more prevalent in woodlands than in pastures. Forest fragmentation favored B. burgdorferi sensu lato and A. phagocytophilum prevalence in woodlands, and in pastures, the B. burgdorferi sensu lato prevalence was favored by shrubby vegetation. Both results are probably because large amounts of edges or shrubs increase the abundance of small vertebrates as reservoir hosts. The Rickettsia sp. prevalence was maximal on pasture with medium forest fragmentation. Female ticks were more infected by B. burgdorferi sensu lato than males and nymphs in woodland sites, which suggests an interaction between the ticks and the bacteria. This study confirms the complexity of the tick-borne pathogen ecology. The findings support the importance of small vertebrates as reservoir hosts and make a case for further studies in Europe on the link between the composition of the reservoir host community and the infection prevalence in ticks.
Ecological modifications are recognized as one of the main forces behind the emergence of infectious diseases (37). As vectors and wildlife are very sensitive to environmental conditions, ecological changes are expected to have a particular impact on the epidemiology of vector-borne diseases and those with a wildlife origin (29, 48). Several studies have highlighted the influence of factors such as climate change and habitat fragmentation on the risk of tick-borne diseases (20, 67). The risk of a tick-borne disease being transmitted to humans or to animals is closely linked to the prevalence of pathogens in ticks questing for hosts (38, 58). In turn, infection prevalence directly depends on the probability of ticks feeding on an infected reservoir host. This probability results from a combination of the intrinsic characteristics of the species involved (e.g., the host species feeding preference of the tick and the ability of the pathogen to infect different host species) and the characteristics of the host community (e.g., the likelihood of contact between ticks and infected reservoir hosts) that vary in time and space. Due to the difficulty of directly assessing the host community, it may be characterized indirectly by studying landscape and habitat features (3, 9). The increased fragmentation of deciduous forests, for example, favors infection prevalence in ticks that are the agents of Lyme disease in the eastern United States because this fragmentation pattern favors the abundance of a highly competent host reservoir, the white-footed mouse (Peromyscus leucopus) (1, 12). However, studies of the effect of habitat fragmentation on different tick-borne pathogens are scarce (25, 40, 67). Most only report on the infection prevalence of pathogens in ticks according to sampling locations, the stage of tick development, and their sex (18, 56); few studies take into account the interplay or simultaneous effects of explanatory environmental factors (2).
In Europe, the Ixodes ricinus tick is one of the most important vectors for animal and human pathogens, especially bacteria (21). These include pathogenic species of the complex Borrelia burgdorferi sensu lato, the agent of Lyme borreliosis, the most prevalent vector-borne human disease in Europe (57); Anaplasma phagocytophilum, the agent of human and animal granulocytic anaplasmosis, considered to be an emerging disease both in human and in animals (8, 61); and Rickettsia helvetica of the spotted fever group, known to be responsible for nonspecific fevers in humans (28).
Although they share the same tick vector, B. burgdorferi sensu lato, A. phagocytophilum, and R. helvetica have different ecological cycles and transmission patterns which influence the infection prevalence at different stages of a tick's life. For B. burgdorferi sensu lato, the maintenance cycle of the bacteria depends on immature stages of I. ricinus ticks feeding on infected reservoir hosts, mainly small rodents and birds that feed on the ground (36, 62). For A. phagocytophilum, small mammals and ruminants are reservoir hosts (16, 22, 69). In contrast to the other two pathogens, R. helvetica is transovarially and sexually transmitted in ticks (13, 33). Ticks are thus considered to be a reservoir host for the bacteria. Small rodents are also suspected to be reservoir hosts in Europe (45), while the role of ungulates remains unknown (60).
It is increasingly recognized that a better understanding of the variation of the prevalence of pathogens in ticks within regions of endemicity is critical to the rational design and monitoring of control programs (47). Our objective was to run an exploratory analysis to test the influence of a range of factors on variations in the prevalence of B. burgdorferi sensu lato, A. phagocytophilum, and Rickettsia sp. of the spotted fever group in questing I. ricinus ticks. The factors considered were two habitats (pasture and woodland), forest fragmentation metrics, the vegetation around and near the pasture, and adult tick hosts (deer and cattle); and the analysis took into account factors linked to tick characteristics (tick sex, tick stage, and the density of questing nymphs). Consequently, we analyzed ticks collected in the field for the presence of DNA from the three bacteria and ran an exploratory statistical model using multivariate regression.
Questing adult and nymph ticks were collected in the spring of 2003 from a representative set of pastures and woodlands located in the Combrailles region of central France (latitude, 45.8° to 46.2°N; longitude, 2.5° to 3.1°E; altitude, 329 to 830 m) during the study of Boyard et al. (10). Briefly, Combrailles is characterized by a bocage landscape, i.e., woodlands and pastures enclosed by hedgerows, earth banks, or rows of trees. The woodland surface represents 33% of the total surface. Ticks were collected by dragging the vegetation on 61 permanently grazed pastures and the woodland area nearest each pasture (Fig. (Fig.1).1). The mean surface size of the pastures was 1.97 ha (standard deviation, 1.05 ha). In the pastures, the ticks were collected on 10-m2 subtransects along the inner perimeter. In the woodlands, 10 subtransects of 10 m2 were randomly selected. In all cases, two consecutive subtransects were separated by a gap of 20 m. Once they were collected, the ticks were immediately stored in 70% ethanol. The total number of I. ricinus ticks collected included 4,465 nymphs, 123 adult males, and 102 adult females.
Prior to DNA extraction, individual ticks were identified under a microscope (19). The ticks then were washed in three sterile water baths and then in one absolute ethanol bath to remove the microorganisms present on the surface of the ticks, before they were air dried and collected in sterile microtubes. DNA was extracted using a Qiamp DNA extraction kit (Qiagen, Hilden, Germany) for the tissue protocol described by Halos et al. (31). DNA was extracted from adult ticks individually. For nymphs, the extraction was run on pools of 5, 10, or 50 nymphs from the same sampling site, according to the number of nymphs sampled at the site. We limited the total number of pools per site to 1 pool of 5, 5 pools of 10, and 5 pools of 50 nymphs. Thus, the nymphs that were not analyzed came from sites where less than 5 nymphs remained after the pooling or from three pasture sites that had more than 305 nymphs. We obtained DNA extracts from the pasture site and the associated woodland in 37 locations.
All PCRs were performed in a Genamp thermocycler (Applied Biosystems, Courtaboeuf, France). Each reaction was carried out in 25 μl containing 0.5 μmol/μl of each primer, 2.5 mM each deoxynucleoside triphosphate, 2.5 μl of 10× PCR buffer, 1 U of Taq DNA polymerase (Takara Biomedical Group, Shiga, Japan), and 5 μl of the DNA extract. Negative (sterile water) and positive controls were included in each experiment. The efficiency of tick DNA extraction was evaluated by amplification of a fragment of the tick mitochondrial 16S rRNA gene using tick-specific primers (7). The samples were tested for the presence of B. burgdorferi sensu lato, A. phagocytophilum, and Rickettsia sp. DNA by PCR using primers specific for each of the pathogens (41, 42, 53). DNA electrophoresis was carried out in 2% agarose gels containing ethidium bromide, and the DNA fragments were visualized under UV light. Ten to 20 amplified fragments for each PCR were sent to Qiagen for sequencing. The sequences were compared with known sequences listed in the GenBank nucleotide sequence databases. The BLAST search option of the National Center for Biotechnology Information (NCBI) was used to confirm the origin of the sequence.
Factors related to tick characteristics were the stage-sex (adult females, adult males, and nymphs) and the density of questing nymphs per site.
Forest fragmentation was estimated through five landscape metrics. The metrics were (i) the number of forest patches, (ii) the sum of the surface areas of the forest patches, (iii) the total edge length, measured by the sum of the perimeters of all forest patches within the buffers, (iv) the perimeter length/surface area ratio, and (v) the mean patch surface area (surface area/number of patches) (26). The forest fragmentation metrics were extracted from the CORINE Land Cover database (CLC2000; 100-m spatial resolution; European Environment Agency) using the sum of all forest types (deciduous, mixed, coniferous). On each woodland and pasture site, the five metrics were calculated with circle buffers that were centered on each site and had radius sizes of 1 km, 2 km, and 5 km, which yielded a total of 15 metrics-buffers for each site. The smallest radius defined the buffer that was just large enough to contain the entire surface of an individual site. The largest radius was the buffer that was just small enough to avoid having buffers from different sites overlap. The metrics were calculated for each buffer size using the spatial module PostGIS (Refractions Research; http://postgis.refractions.net/) of the PostgresSQL database (The PostgreSQL Global Development Group, GNU license; http://www.postgresql.org/).
For the pasture sites, the factors related to vegetation were the percentage of the pasture perimeter with tree or bush layers, the major landscape cover lying outside the pasture (i.e., shrub-tree, pasture under cultivation, or mixed composition), and the percentage of the pasture perimeter that was less than 20 m away from woodland. For the woodland sites, the factor related to vegetation was forest type (deciduous, coniferous, mixed).
Roe deer (Capreolus capreolus) and cattle are hosts for adult ticks and a reservoir host for A. phagocytophilum but not for B. burgdorferi sensu lato. For all of the sites, the factor that we considered was the number of deer killed by hunters per county (tableau de chasse) provided by the Fédération des Chasseurs. For pasture sites alone, we considered the duration of cattle presence in the pasture (not present, present for less than 1 month at the date of sampling, present for more than 1 month).
The probability of individual tick infection was estimated on the basis of individual tick PCR data (for the adults) or pooled data (for the nymphs). For the pools, three hypotheses were considered: (i) a pool was found to be infected if at least one nymph within the pool was infected, (ii) the probability of a nymph being infected was the same for all nymphs of the same pool, and (iii) within each pool, the probability of a nymph being infected was independent of the probability of infection of the other nymphs of the same pool. We set p to be the probability that a tick was infected, and then (1 − p) was the probability that a tick was not infected. The probability for pool i of size ni (i.e., 1 adult or 5, 10, or 50 nymphs) not to be infected was . We then derived qi, the probability that pool i would be positive, by 1 − . To ensure that the probability p was between 0 and 1, we used the logit(p): θ = ln]p/(1 − p)]. θ was estimated using maximum likelihood.
Screening for an association between the presence of bacterial DNA in ticks and the explanatory factors was performed using generalized linear models (GLMs) (43) through R statistical software, version 2.4.0 (52). The prevalence of infection by each bacterium was analyzed separately. The null model (i.e., the intercept-only model) was used to compute the overall infection prevalence and confidence intervals in adult and nymph ticks collected in woodland and pasture sites. Quantitative explanatory factors were categorized into tertiles. Because the woodland sites chosen were those that were closest to the pastures, the two types were not independent. We therefore first tested the significance of the explanatory factors on the pasture and woodland sites separately. We then tested whether the infection prevalence varied between the 37 associated pasture and woodland sites.
To avoid multicollinearity between identical fragmentation metrics at different buffer sizes, univariate analyses were performed between each metric-buffer and the response variable. A maximum of one buffer size per metric was selected by choosing from among the buffers that had P values of <0.05 the buffer that had the most significant P value. The potential explanatory factors were then tested in a multivariate model using a backward-selection procedure. For a given significant factor, two levels were significantly different when the likelihood ratio test between the final multivariate model and the model in which the two levels had been merged was significant (P < 0.05).
To test whether the individual infection probabilities varied between the 37 associated pasture and woodland sites in the locations where ticks were analyzed in both associated pasture and woodland sites, we took the estimated infection probability per site, computed the difference between the paired sites, and tested whether the distribution of the differences was symmetrical around zero with a Wilcoxon signed-rank test (P < 0.05).
We obtained a total of 521 DNA extracts that came from 123 adult males, 102 adult females, and 296 pools of nymphs that summed up to 3,480 nymphs. Among the 521 DNA samples, 255 were found to be positive for at least one of the three pathogens (Table (Table1).1). Sequence analysis of 17 fragments related to B. burgdorferi sensu lato DNA showed that the sequences corresponded 100% to a part of the 16S rRNA gene of either B. garinii or B. afzelii. The sequences were not discriminant enough to allow distinction at the genospecies level. Sequence analysis of 10 fragments related to A. phagocytophilum DNA showed that the sequences had 95% to 100% identity with the 16S rRNA genes of pathogenic strains of A. phagocytophilum. Sequence analysis of 12 fragments related to Rickettsia sp. DNA showed that they had 100% identity with the citrate synthase gene of R. helvetica.
Three adult females were coinfected, one with B. burgdorferi sensu lato and Rickettsia sp. and two with A. phagocytophilum and Rickettsia sp. In addition, 42 pools of nymphs were coinfected. The estimated individual prevalence rates of infection with B. burgdorferi sensu lato were 1.4%, 16.1%, and 5.3% for adult males, adult females, and nymphs in woodlands, respectively, and 1.9%, 4.3%, and 1.4% for adult males, adult females, and nymphs in pastures, respectively (Table (Table1).1). The individual prevalence rates of infection with A. phagocytophilum were 4.3%, 10.7%, and 1.7% for males, females, and nymphs in woodlands, respectively, and 9.4%, 2.2%, and 2.6% for males, females, and nymphs in pastures, respectively. Concerning Rickettsia sp., the respective prevalence rates were 5.7%, 7.1%, and 2.4% for males, females, and nymphs in woodlands and 3.8%, 8.7%, and 2.2% for males, females, and nymphs in pastures.
In the multivariate analysis, the bacterial infection status was influenced by tick characteristics and forest fragmentation, vegetation, and habitat variables. Deer and cattle had no observed influence on infection prevalence of ticks. The B. burgdorferi sensu lato infection prevalence in woodland ticks was influenced by stage-sex (P = 0.001) and the forest patch surface area within a 2-km buffer (P = 0.009). Adult females were more infected than males and nymphs (Table (Table2;2; see Fig. S1 in the supplemental material). Ticks were more likely to be infected on woodland sites with a small forest patch surface area. In pastures, the B. burgdorferi sensu lato infection prevalence decreased with an increase in the perimeter length/surface area ratio of the forest patch within a 2-km buffer (P < 0.001) and increased with medium and high percentages of shrub layer around the pasture (P < 0.001). The infection prevalence was lower on sites with a high perimeter length/surface area ratio and with a low percentage of shrub layer (Table (Table2;2; see Fig. S1 in the supplemental material). At the 37 locations where the ticks were analyzed in both pastures and woodlands, the infection prevalence was higher in woodlands than in pastures (P < 0.001; probability of infection in pastures = 0.044; probability of infection in woodlands = 0.099).
The A. phagocytophilum infection prevalence in ticks sampled on woodlands varied with stage-sex (P < 0.001), nymph density (P = 0.013), and the perimeter of forest patches within a 1-km buffer (P = 0.023). Adult females were more infected than nymphs, but the difference with males was not significant (Table (Table2;2; see Fig. S1 in the supplemental material). The infection prevalence was higher on sites with a high nymph density than on those with a medium one, but the difference from the prevalence on low-density sites was not significant. An increasing forest patch perimeter was correlated with an increased infection prevalence. In the multivariate model, no significant explanatory factors for infection prevalence in pasture ticks were identified; none of the fragmentation metrics-buffers was significant in the univariate tests. Moreover, the difference in infection prevalence between woodland and pasture ticks was not significant.
The Rickettsia sp. infection prevalence in pasture ticks was influenced by the forest surface area in the 1-km buffer (P = 0.002), the percentage of tree layer around the pasture (P = 0.020), and the border characteristics (P = 0.026). The most infected ticks were collected on sites with a medium forest patch surface area, a medium percentage of tree layer, and pastures surrounded mainly by shrubs-trees or by several land cover types (Table (Table2).2). We did not find any factor influencing the Rickettsia sp. infection prevalence in woodland ticks, although three of the fragmentation metrics-buffers were significant in the univariate tests. The difference in infection prevalence between woodland and pasture ticks was not significant.
We observed wide variations in the prevalence of tick-borne bacteria in I. ricinus ticks in a French bocage landscape. This prevalence was influenced by several environmental factors that varied according to the bacterial species and habitat type. Depending on the habitat and the stage and sex of the adult tick, we found that 1.4% to 16.1% of the ticks carried DNA of at least one of the three bacteria studied. Compared to the rates found in Europe (51) and France (5, 50), the infection prevalences that we found for B. burgdorferi sensu lato are low. This could be due to the inclusion in our study of pastures, where there was a lower infection prevalence, whereas most studies carried out in Europe focus on woody habitats. For A. phagocytophilum, our figures are higher than the figures previously reported in France (5, 27), but they fall within the range of figures found in other European sites (30, 34), including some that considered woodlands and pastures (39, 66). For Rickettsia sp., the sequences that we obtained were similar to the sequence of R. helvetica. Some of our infected ticks may also have been infected by other Rickettsia spp., such as symbionts (32). Regarding R. helvetica, the prevalences that we found were in the range of (49) or lower than (4) the prevalence found in previous work.
B. burgdorferi sensu lato was more prevalent in woodland ticks than pasture ticks, which is consistent with the fact that the main reservoir hosts, i.e., small rodents and passerine birds, are more abundant in woodlands than in pastures. Furthermore, ticks in pastures are less likely to become infected by B. burgdorferi sensu lato because they can feed on cattle, which are not competent as a reservoir (6). The increased infection prevalences for B. burgdorferi sensu lato and A. phagocytophilum on woodland sites were associated with increased forest fragmentation, in which greater fragmentation was considered a low forest patch surface area and a high patch perimeter. Changes in patch surface area and patch perimeter sizes are linked because when the patch size decreases, the amount of edge, measured here as the perimeter, increases relative to the patch area. A larger amount of edge increases the abundance of small vertebrate hosts—rodents, which are the reservoir for both pathogens, and birds, which are a reservoir only for B. burgdorferi sensu lato—because the vegetation complexity is greater at the edge or because several landscape elements are available (9, 70). Indeed, we found that in a bocage landscape, small mammals were more abundant in the forest-pasture ecotone than inside forests or pastures (11). Similar results were found in the United States, where patch size was found to be predictive of the B. burgdorferi sensu lato infection prevalence of I. scapularis because of the increased abundance of the reservoir host for B. burgdorferi sensu lato (1, 12). Another possible explanation for the increase in the A. phagocytophilum infection prevalence in relation to the patch perimeter is that an increased availability of forest edge favors roe deer abundance (67). Roe deer are a reservoir host for this bacterium (22) and are known to prefer forest edges (64). Since an increase in roe deer abundance leads to increased tick density (46), this could explain why we found that a high nymph density was associated with a high A. phagocytophilum infection prevalence.
The B. burgdorferi sensu lato infection prevalence was higher on pastures that had a high percentage of shrubs on the perimeter. This result is consistent with the fact that B. burgdorferi sensu lato reservoir hosts, i.e., rodents and birds, are particularly concentrated in the shrubby vegetation around pastures (11, 65). The B. burgdorferi sensu lato infection prevalence also increased in pastures surrounded by forests with low perimeter length/surface area ratios. The lowest theoretical ratio corresponds to a circle; conversely, a high ratio indicated an indented shape with more edge compared to the surface area, which, again, should favor the small vertebrate abundance. This effect was thus in contrast to what we observed on woodland sites, where the prevalence tended to be associated with fragmented forest. Infected ticks found in pastures could have become infected by feeding on infected reservoir hosts located in the pasture itself or by feeding on infected woodland hosts that then imported the ticks into the pasture (10). We therefore hypothesized that a decreased flow of tick hosts between woodlands and pastures could occur when the forest surrounding the pasture has a high perimeter length/surface area ratio, a characteristic which would consequently hinder the infection prevalence in pasture ticks. This mechanism could be mediated by factors that we have not taken into consideration in this study, such as the pasture isolation from woodlands and the lack of connectivity between woodlands and pastures.
The Rickettsia sp. infection prevalence in pastures was maximized with a medium total forest surface area within a 1-km buffer area, a medium percentage of tree layer around the pasture, and a pasture border dominated by shrub-tree vegetation. There was no difference in the Rickettsia sp. infection prevalence between woodlands and pastures, and no factor explained the prevalence variation in woodlands. We can only speculate that our findings reflect the optimum conditions for Rickettsia sp. transmission mediated by small vertebrate reservoirs and for transtadial, transovarial, and sexual tick transmission. The absence of a significant tick factor is, furthermore, consistent with the fact that transovarial transmission (13) and sexual transmission (33) should lower the differences in infection prevalence between tick stages and sexes.
In our work, we found that on woodland sites, B. burgdorferi sensu lato infected adult female ticks more than it did males or nymphs. The same tendency was observed in woodlands for A. phagocytophilum, although it was not significant. In pastures, no sex-stage effect was detected for any of the pathogens, a result which would indicate that tick sex-stage was a less important driver of infection prevalence than vegetation or fragmentation. The higher infection prevalence in adult ticks than in nymphs is commonly found for B. burgdorferi sensu lato (51) and A. phagocytophilum (17, 30) because adult ticks have had more opportunities to become infected than nymphs. The reason for the higher infection prevalence in females versus males is less clear-cut in the literature, as few studies have run statistical analyses, and even fewer have taken into account other factors. For B. burgdorferi sensu lato, although many publications report a higher infection prevalence in females (e.g., see references 59 and 68), a meta-analysis based on 11 studies in Europe did not reveal any difference in the infection prevalence between males and females (51). For A. phagocytophilum, some studies highlight a higher infection prevalence in females (e.g., see references 30 and 59), but not all have done so (17). Three hypotheses may be proposed to explain why there are fewer infected males than females. First, female nymphs consume more blood than male nymphs (24), which results in a higher probability that females will ingest bacteria. Second, adult male I. ricinus ticks do not feed on hosts; therefore, bacteria that infect adult males are not further transmitted. Processes that would decrease the infection prevalence in male populations, such as the mortality of bacteria in male ticks or the mortality of infected male ticks, should thus not be counterselected because they would not decrease bacterial fitness. However, evidence of harmful interactions between ticks and bacteria is equivocal for Ixodes spp. (14, 15, 54, 55, 63). Finally, de Meeûs et al. (23) suggest that immature males feed on hosts that have a lower reservoir competence for B. burgdorferi sensu lato than the hosts of females.
In this study, we provide an example of statistical analyses of pooled data that can replace analyses in which samples are analyzed one by one and samples are not pooled. There is, however, an associated loss of statistical power, which may explain why we did not identify any explanatory factors in multivariate models for A. phagocytophilum in pastures or Rickettsia sp. in woodlands, even though some fragmentation factors were significant in the univariate model. The failure to identify any significant factors in the multivariate models also underlines the danger of interpreting the results of univariate analyses, since the effect seen in a single-factor model may, in fact, derive from the effects of other factors that were not considered. The approach that we used may be improved by including the identification of B. burgdorferi sensu lato species, which would allow results to be interpreted in the light of the different reservoir hosts involved. The use of a higher spatial resolution for fragmentation metrics should also improve the ability to identify small patches that are important for small reservoir species. Finally, an analysis of the blood meal remnants in ticks should make it possible to identify the host species on which the tick had its last meal (35, 44). This information is crucial to disentangle the relationship between pathogens, vertebrate hosts, and habitats. It would allow the link between forest fragmentation and infection prevalence to be investigated with quantitative and precise information on the multitude of host species that are involved in the transmission cycle of tick-borne pathogens. Our results confirm the complexity of tick-borne microorganism ecology and illustrate the difficulty of defining generic epidemiological frameworks for the transmission of these pathogens.
We are thankful to Ginette Sadot, Valérie Poux, Nelly Dorr, and all our colleagues and trainees for their efficient and friendly help in field data acquisition and management. Sincere thanks go to the Etablissement Départemental de l'Elevage du Puy-de-Dôme, the Région Auvergne, the Syndicat Mixte pour l'Aménagement, le Développement des Combrailles, the Fédération des Chasseurs, and all the farmers for their collaborative spirit. We thank the Tiques et Maladies à Tiques group of the Réseau Ecologie des Ineteractions Durables for discussion and support.
We thank the Département de Santé Animal (INRA) and Merial for funding.
Published ahead of print on 7 May 2010.
§Supplemental material for this article may be found at http://aem.asm.org/.