Ecology Letters, (2014) 17: 970–978

LETTER

Michaela D.J. Blyton,1,2* Sam C. Banks,2,3 Rod Peakall,1 David B. Lindenmayer2,3 and David M. Gordon1

doi: 10.1111/ele.12300

Not all types of host contacts are equal when it comes to E. coli transmission Abstract The specific processes that facilitate pathogen transmission are poorly understood, particularly for wild animal populations. A major impediment for investigating transmission pathways is the need for simultaneous information on host contacts and pathogen transfer. In this study, we used commensal Escherichia coli strains as a model system for gastrointestinal pathogens. We combined strain-sharing information with detailed host contact data to investigate transmission routes in mountain brushtail possums. Despite E. coli being transmitted via the faecal-oral route, we revealed that, strain-sharing among possums was better explained by host contacts than spatial proximity. Furthermore, and unexpectedly, strain-sharing was more strongly associated with the duration of brief nocturnal associations than day-long den-sharing. Thus, the most cryptic and difficult associations to measure were the most relevant connections for the transmission of this symbiont. We predict that future studies that employ similar approaches will reveal the importance of previously overlooked associations as key transmission pathways. Keywords E. coli, gastrointestinal pathogens, networks, social interactions, transmission, Trichosurus cunninghami. Ecology Letters (2014) 17: 970–978

Pathogen transmission has important ramifications for population viability and conservation. It is also thought to be a primary cost influencing the evolution of sociality (Alexander 1974; Freeland 1976; Altizer et al. 2003). Central to the dynamics of disease spread is how pathogens are transmitted through the host population. These transmission routes are often represented as contact networks, which provide a powerful predictive framework for modelling disease spread (Read & Keeling 2003; Keeling 2005). However, the connections between individuals in these networks must be relevant to the pathogen studied. For instance, when pathogens are transmitted by host-to-host contact, social interactions are of primary relevance, whereas, for vector-borne or environmentally transferred pathogens, spatial proximity or shared resource use may be more informative (Altizer et al. 2003). Furthermore, particular types of social interactions may be relevant for different classes of directly transmitted pathogens. For example, non-sexual associations have little impact on the spread of sexually transmitted diseases (May & Anderson 1987). The specific processes that facilitate pathogen transmission between individuals are poorly understood for most host and pathogen species. For instance, several transmission routes have been suggested for gastrointestinal pathogens, yet in few cases has the relative importance of these routes been clearly identified. These pathogens are transmitted primarily via the faecal-oral mode and have traditionally been considered to be

environmentally transferred through contaminated soil, water or food (Madigan & Martinko 2006). However, close spatial proximity has been correlated with the transmission of a gastrointestinal bacteria between humans, primates and livestock (Goldberg et al. 2008). Furthermore, direct host contact has been associated with the incidence of gastrointestinal disease in livestock and humans (Belongia et al. 1993; Besser et al. 2001; Johnson et al. 2008). This suggests that host contacts and/or greater space sharing may also be important in the transfer of these pathogens. A major impediment for disentangling transmission pathways is the need to simultaneously measure host associations, space sharing and the patterns of pathogen transfer. In this study, we took advantage of a model microbial system and animal surveillance technologies that continuously record associations between individuals, to investigate the transmission routes of gastrointestinal microbes. Specifically, we assessed the patterns of commensal Escherichia coli strainsharing in a wild population of mountain brushtail possums (Trichosurus cunninghami) to test whether transmission was facilitated by spatial proximity and/or social contacts. Commensal E. coli strains provide an ideal surrogate system for investigating the potential transmission routes of gastrointestinal pathogens because they are genetically diverse, and are likely to be transferred in a similar way to many other gastrointestinal pathogens (Gordon & Cowling 2003). Furthermore, it has be suggested that transmission pathways may be more easily identified using commensal strains than

1

3

INTRODUCTION

Evolution, Ecology and Genetics, Research School of Biology, The Australian

National Environmental Research Program Environmental Decisions Hub,

National University, Canberra, ACT, 0200, Australia

ANU Node, Canberra, ACT, 0200, Australia

2

*Correspondence: E-mail: [email protected]

The Fenner School of Environment and Society, The Australian National

University, Canberra, ACT, 0200, Australia

© 2014 John Wiley & Sons Ltd/CNRS

Letter

pathogens as they are ubiquitous and host resistance is less likely to have evolved (Bull et al. 2012). For strain-sharing patterns to inform our understanding of transmission pathways, they must be indicative of primary transmission. We expect that this will only be the case where strain diversity is high, strain turnover is frequent, and a large proportion of the local host population is sampled. The mountain brushtail possum offers an ideal host system to investigate transmission dynamics in the wild. Previous work has revealed high genetic diversity and frequent temporal turnover (in < 6 weeks) of E. coli strains in this marsupial (Blyton et al. 2013). Thus, strain-sharing between hosts is likely to represent contemporary transmission patterns. By using proximity data loggers to record all associations between collared possums, we were also able to explore whether particular classes of host contacts were more strongly associated with E. coli strain-sharing than others. Specifically, we examined whether nocturnal associations, den-sharing or social pair-bonding better explained E. coli strain-sharing among individuals. Through this work, we reveal counterintuitive patterns of strain-sharing that highlight the importance of directly assessing transfer routes, as they cannot always be predicted from the mode of transmission (e.g. faecal-oral or air-borne).

MATERIALS AND METHODS

Host study species

The mountain brushtail possum (Trichosurus cunninghami) is a medium sized, nocturnal, arboreal marsupial, with adults weighing between 2.5 and 4.5 kg. Adult individuals occupy long-term 2–6 ha home ranges after limited male biased natal dispersal (Martin 2005, 2006; Banks et al. 2008). Individuals commonly live for 8–12 years and reach sexual maturity at 2–3 years of age (Viggers & Lindenmayer 2000). Females produce a single offspring per year shortly after the 1–2 month autumn breeding season. The social system of the mountain brushtail possum can be defined as semi-social with home-range overlap common within and between the sexes (Lindenmayer et al. 1997). Individuals interact during night-time activity and regularly share tree hollows during the day (Lindenmayer et al. 1997; Banks et al. 2011). Furthermore, some male–female pairs form social pair-bonds, exhibiting highly overlapping home ranges, frequent den-sharing and genetic monogamy (Martin et al. 2007; Blyton et al. 2012). Mountain brushtail possums have hindgut fermentation with a well-developed caecum (Crowe & Hume 1997). E. coli is a ubiquitous member of the species intestinal microbiota, with each animal carrying an average of two strains (range 1–6) at a given time point (Blyton et al. 2013). Study sites

We investigated E. coli strain-sharing in mountain brushtail possums from two study sites. The sites were located in the Central Highlands of Victoria, Australia (37°65 S, 145°72 E, 600–700 m altitude). Both sites were characterised by tall wet eucalypt forest, dominated by a mountain ash (Eucalyptus

Host contacts and E. coli strain-sharing 971

regnans) overstorey. The sites were 20 ha (500 m by 400 m) in size and 160 m apart at their closest point. Despite the spatial proximity of the sites, the core home ranges of all studied animals were restricted to a single site. The adult possum populations at the sites were similar, with 31 and 28 animals trapped at each site over the study period respectively. However, Site 1 had considerably more juvenile and sub-adult individuals (between 1 and 3 years old) with 18 captured over the study, compared with eight at Site 2, leading to a higher overall population density at that site. Quantification of associations among hosts

Associations between adult (> 3 years) mountain brushtail possums at both study sites were quantified using proximity data logger collars (Sirtrack Ltd, Hawkes Bay, New Zealand), which recorded the date, starting time and duration of each association (defined by an unobstructed collar proximity of < 5 m; ultra high frequency range coefficient = 5). The collars also contained a VHF (very high frequency) transmitter that allowed an individual’s spatial location to be tracked (however, they did not have the capability to automatically record GPS locations). We fitted collars to possums during two trapping sessions using a regular grid of 40 cage traps and retrieved them 1–2 months later by tracking each possum and placing traps around the base of their denning tree. Thus, we recorded all associations between collared possums during two seasons of a year: summer (December 2010/January 2011–February 2011) and winter-spring (July 2011–September 2011). Thirty four animals were collared across the two study sites (Site 1 : 16, Site 2 : 18), which equated to 33 and 50% of the populations at Sites 1 and 2 respectively. Twenty animals were surveyed across both seasons (Site 1 : 11, Site 2 : 9). Raw association data were processed by removing erroneous 1 s contacts and associations that occurred during the trapping sessions. We combined the records from each member of an associating dyad (and merged associations that overlapped) into a single association manifest per dyad, as the entries were not always symmetrical in number or duration between dyad members (following Hamede et al. 2009). Furthermore, records that were separated by ≤ 5 min were then merged into single entries, as proximity collars are known to record extended associations as a series of short contacts (Drewe et al. 2012). The processed data were then separated into nocturnal associations and diurnal den-sharing. In general, associations that began and/or ended between civil twilight rise (the limit at which illumination is sufficient, under good weather conditions, for terrestrial objects to be clearly distinguished by humans) and 30 min after civil twilight set were considered den-sharing, and all other records were classified as nocturnal associations (see online supplement 1 for exceptions). Pair-bonded possums were identified by their strong patterns of association and overlapping home ranges as defined by Martin et al. (2007). Determining spatial proximity between possums

The spatial proximity between members of each dyad was estimated by calculating the distance between their spatial © 2014 John Wiley & Sons Ltd/CNRS

972 M. D. J. Blyton et al.

locations using the PBS mapping package (Schnute et al. 2012) in R v.2.15.2 (R Core Team 2012). First, we calculated a minimum convex polygon from all the observed locations (trapped or radio-tracked) of each possum. We then determined the centroid of each polygon and used those coordinates as the possum’s spatial location. Radio-tracking fixes of each possum consisted of between 5 and 41 (mean = 20.4) diurnal denning locations recorded over summer and winterspring by VHF telemetry. Trapping fixes consisted of between 2 and 23 (mean = 8) nocturnal trapping locations. Sampling of E. coli strains and assessment of strain-sharing

The E. coli strain communities of all collared possums were sampled when their collars were fitted and removed. This provided us with up to four samples per individual and two per season. Samples were taken from each possum under sedation (Viggers & Lindenmayer 1995) by inserting a sterile swab into the colon of each animal, via the cloaca, to a depth of approximately 10 cm. This sampling technique produces highly repeatable results between samples taken from a host at the same time point (unpublished data). E. coli strains were then isolated from each swab following Blyton et al. (2013). To identify the strains carried by each animal, we randomly selected 23 E. coli isolates from each sample for genetic analysis. This provided a 95% chance of detecting a strain at 13% relative abundance within an animal and a 50% chance of detecting a strain present at a relative abundance of 3%. The E. coli isolates were genotyped following Blyton et al. (2013) using a combination of CGG primer-based PCR (AdamusBialek et al. 2009), ERIC primer-based PCR (Versalovic et al. 1991) and phylogenetic group membership as identified by quadruplex-PCR (Clermont et al. 2013). Isolates were considered to be the same strain only if they showed the same banding pattern for both rep-PCR primers and belonged to the same phylogenetic group. Strain-sharing between individuals was assessed separately for summer and winter-spring. However, sequential sharing between the two sampling periods within each season was considered, as strain turnover between sampling periods was common. Assessment of factors affecting E. coli strain-sharing

On a fine spatial scale (similar to that of the host species’ home range), E. coli strain-sharing is likely to represent primary transmission either through a common source or by host-to-host contact, particularly when temporal turnover of strains is high. However, on a broader scale, sharing is more likely to reflect landscape features such as habitat effects on strain survival (Bergholz et al. 2011) as well as secondary transmission though intermediate individuals. Therefore, we restricted our analysis to possum dyads where the members were separated by less than the distance that mountain brushtail possums move and interact (see Results: Associations among hosts). To investigate whether host associations and/or spatial proximity influenced E. coli strain transmission, we analysed pairwise strain-sharing in a Bayesian regression framework. Regression analyses generally assume that all observations are © 2014 John Wiley & Sons Ltd/CNRS

Letter

independent. However, this is not the case for pairwise data where the same individual may be involved in multiple observations. In the case of social contacts there are two nonmutually exclusive processes that can contribute to this dyadic non-independence. First, particular individuals may be more gregarious than others. To account for this type of dependence between observations, we applied a multiple membership random effects structure of possum identity in our analyses (Browne et al. 2001; Clarke et al. 2002). Second, individuals may have a social interaction ‘time budget’ such that associations with one con-specific reduce associations with another. However, as possums are semi-social and spend considerable periods of time alone, this form of dyadic nonindependence is unlikely to apply in this study system. Thus, we did not specifically account for this form of non-independence in our models. We performed separate analyses for each study site due to potential differences between the sites in how well strain-sharing patterns reflected primary transmission, as the sites varied in host population density and the frequency of strain-sharing (probability of dyad members sharing a strain Site 1 : 24%, Site 2 : 41%; Blyton et al. 2013). For each site, we fitted the generalised linear mixed models using the MCMCglmm package (Hadfield 2010) in R v.2.15.2 (R Core Team 2012). The response variable in the analysis was whether or not members of a possum dyad had at least one strain in common during a season (Binary response variable: yes = 1, no = 0; with logit link function). The potential explanatory variables were as follows: (1) total association duration (divided by days collared), (2) nocturnal association duration (per night), (3) diurnal den-sharing (binary response: yes =1, no = 0), (4) diurnal association duration (per day), (5) pair-bonding (yes = 1, no = 0), (6) spatial proximity and (7) season. All measures of association based on duration were log-transformed because we expected that increases in E. coli sharing were likely to diminish as contact duration increased. It should be noted that, in contrast to the raw measures of association, as a result of this transformation to the log scale, den-sharing associations did not substantially increase the log total association duration above the log nocturnal association duration. Both raw and log-transformed spatial proximity were included as potential explanatory variables in the analyses, as we did not have a prior expectation regarding the relationship between strain-sharing and the distance between members of a dyad. The deviance information criterion (DIC) was used to rank models containing different combinations of explanatory variables. In addition, we determined which single variable best explained strain-sharing patterns and assessed the significance of individual terms within the best models, as DIC can overfit the model (Ando 2011). The effect sizes of the explanatory variables are reported as both posterior mean coefficients and the odds ratios. While, the odds ratios are an intuitive measure of effect size they are not directly comparable between different explanatory variables that are measured in different units. Our list of candidate models included all possible combinations of the explanatory variables. However, highly correlated terms (Pearson’s correlation ≥ │0.7│) were not fitted in the same model (see Table 1).

Host contacts and E. coli strain-sharing 973

Letter

Table 1 Pearson’s correlations between the continuous explanatory variables* included in the regression analysis of E. coli strain sharing (combined across

sites) Total association duration Total association duration Nocturnal association duration Diurnal association duration Spatial proximity Log spatial proximity

1 0.98 0.72 0.74 0.74

Nocturnal association duration

1 0.57 0.72 0.70

Diurnal association duration

1 0.54 0.73

Spatial proximity

Log spatial proximity

1 0.86

1

*All measures of association were log transformed.

We used uninformative default priors in all analyses as we did not have prior information on the parameter estimates. Each analysis was run for 2 000 000 iterations following an initial burn in phase of 2000, with a thinning interval of 1000. This returned minimal autocorrelation between recorded iterations and a sample size of 2000 for estimating the posterior distribution.

RESULTS

Associations among hosts

Across the two study sites, we recorded 2145 nocturnal associations (after processing) over the two seasons within 83 possum dyads (Site 1 = 30, Site 2 = 53; see Fig. 1a). Within each of these dyads, the members were within 300 m of each other (see Fig. 2). The proportion of surveyed nights on which members of a dyad associated varied greatly among dyads and followed a highly positively skewed distribution. Associations within 50% of dyads occurred on < 6% of nights, whereas members of one dyad associated on 89% of nights. The nocturnal association duration per night also followed a highly positively skewed distribution (mean = 232 s, SD = 640 s). We recorded a total of 463 instances of den-sharing within 13 different dyads (Site 1 = 135, Site 2 = 328). Ten of those dyads were classified as pair-bonded as they also had high levels of association during night-time activity (see Fig. 1a) and had overlapping home ranges. More pair-bonded dyads were identified at Site 2 than at Site 1 (Site 1 = 3, Site 2 = 7). E. coli diversity and strain-sharing

A total of 86 E. coli strains were identified from the 34 collared mountain brushtail possums over the two seasons. Of these strains, 74 (86%) were assigned to phylogenetic group B2, whereas the remainder belonged to phylogenetic groups D, B1, E or Cryptic Clade V (Blyton et al. 2013). In 73 mountain brushtail possum dyads, the members shared at least one strain within a season (Site 1 = 35, Site 2 = 38; see Fig. 1b), with up to two strains shared within a dyad in the same season. Only within six dyads were strains shared in both seasons. Factors affecting E. coli strain-sharing

As expected, given the host and strain differences between the sites, our ability to explain strain-sharing patterns varied between Sites 1 and 2. At Site 2, strain-sharing could be

explained in the regression models by associations among hosts and spatial proximity. In contrast, at Site 1 the patterns could not be explained and the only significant explanatory variable was season, in either the multivariate or univariate models (posterior mean = 1.474, 95%, credible interval = 2.867 to 0.270, P = 0.017). As such, consistent with our expectations given the lower proportion of individuals collared and higher strain-sharing rate, Site 1 provided little information on E. coli transmission pathways. Thus, we limit our detailed account of the observed patterns and model results to the findings from Site 2. At Site 2, the probability that two individuals shared an E. coli strain was strongly influenced by the time they spent associating. For instance, 16 of the 26 (61%) dyads whose members associated on five or more nights (over the two seasons) shared E. coli strains in at least one season. In contrast, only 17 of the 86 (20%) dyads whose members associated on less than five nights shared strains (see Fig 1). In addition, five of the seven (71%) male–female dyads classified as pair-bonded shared E. coli strains (see Fig 1). The best overall regression model of strain-sharing at Site 2 included both study season and total association duration as explanatory variables (see Tables 2 and 3). Total association duration and nocturnal association duration were explanatory variables in four of the top 10 models, and both were significant predictors (see Table 3). Log-transformed spatial proximity was a significant predictor in the 9th and 10th model. Season appeared in eight of the top 10 models but was not a significant predictor in any case. Pair-bonding, diurnal association duration and diurnal den-sharing (yes/no) were also found in the top 10 models but were not significant predictors. In the top 10 models, all measures of host association had a positive effect on the likelihood of E. coli strain-sharing, suggesting that increases in host associations are associated with an increased rate of E. coli strain transmission. Logtransformed distance between members of a dyad (spatial proximity) had a negative effect on sharing, indicating that more proximate individuals were more likely to share E. coli strains. The best sole explanatory variable of E. coli strain-sharing (when fitted in univariate models) at Site 2 was the total association duration within dyads, with a one unit increase in association [log(seconds)] improving the odds of sharing by 1.63 times (see Table 4). Nocturnal association duration was the next best sole explanatory variable and had an odds ratio of 1.76. Nocturnal association duration performed similarly well to log-transformed spatial proximity and better than raw © 2014 John Wiley & Sons Ltd/CNRS

974 M. D. J. Blyton et al.

Letter AHG014

AHG021

(a)

AHG019

AHG035 AHG020 AHG006

AHG004 AHG013 AHG030

AHG012 AHG001

AHG005

AHG007 AHG011

AHG017

AHG002

AHG018

50 m

AHG015 AHG014

AHG021

(b) AHG019

AHG035

AHG020

AHG006

AHG013

AHG004 AHG030

AHG012 AHG001

AHG005

AHG007

AHG002 AHG011 AHG018

AHG017

50 m

AHG015

Figure 1 Association network (a) and E. coli strain-sharing network (b) among mountain brushtail possums at Site 2, derived from data collected over both the summer and winter sampling periods. Squares represent males and circles represent females. White nodes represent individuals that were only collared during one season, whereas grey and black nodes represent individuals collared in both seasons. The position of the individuals indicates their geographical location (scale is approximate). In (a) lines represent nocturnal associations with the thickness of the line representing the number of nights during which pair members associated. In (b) lines indicate individuals that shared E. coli strains in either season and were within 300 m of each other.

© 2014 John Wiley & Sons Ltd/CNRS

Host contacts and E. coli strain-sharing 975

10

20

30

40

DISCUSSION

0

Number of associating dyads

50

Letter

0

50

100

150

200

250

300

350

Distance between dyad members (m) Figure 2 Distribution of the number of dyads whose members associated, plotted against the distance (m) between members of those dyads.

Table 2 The best 10 models of E. coli strain-sharing amongst mountain brushtail possums at Site 2 and the significance of each term in those models

Δ DIC

Explanatory variables in model (MCMCp)

0.00 1.37

Season (0.274), Total association duration (0.001) Pair-bonding (0.881), Season (0.271), Total association duration (0.011) Nocturnal association duration (0.001), Season (0.233) Total association duration (< 0.0005) Nocturnal association duration (0.071), Season (0.319), Diurnal association duration (0.199) Nocturnal association duration (0.088), Season (0.316), Den-shared (0.187) Pair-bonding (0.957), Total association duration (0.023) Nocturnal association duration (0.037), Season (0.261), Pair-bonding (0.416) Log spatial proximity (0.001), Season (0.397) Log spatial proximity (0.011), Season (0.396), Pair-bonding (0.578)

2.46 2.70 3.71 3.71 4.16 4.21 4.44 5.44

Table 3 The fixed explanatory variables in the top ranking model of E. coli strain-sharing amongst mountain brushtail possums at Site 2*

Variable Constant Season Total association duration

Posterior mean coefficient 2.190 0.684 0.549

95% credible interval 3.612, 0.792 0.527, 1.986 0.232, 0.874

Odds ratio

1.98 1.73

*The posterior distributions for this model were modified to obtain estimates based on the assumption that the actual residual variance equalled zero, following Hadfield (2012).

spatial proximity, den-sharing, and whether or not the dyad was classified as bonded. Nocturnal association duration also performed better than diurnal association duration and had a larger effect size (see Table 4). All explanatory variables, except season, were significant in the univariate models.

Traditionally, gastrointestinal pathogens have been considered to be environmentally transferred via faecal contamination (Madigan & Martinko 2006). Thus, research has largely focused on environmental sources of disease such as food and water (e.g. in E. coli: Besser et al. 1993; Solomon et al. 2002). However, our findings suggest that direct associations among hosts may also be an important transmission route for gastrointestinal symbionts. We found that on a fine spatial scale (< 300 m) at one of our study sites, E. coli strain-sharing was better explained by associations among mountain brushtail possums than by their spatial proximity. Furthermore, surprisingly, we discovered that the duration of brief nocturnal associations (mean per night  4 min) was more strongly associated with E. coli strain-sharing than other types of host contacts, such as day-long den-sharing. Thus, not all social contacts pose an equal transmission risk and, in this case, the most cryptic contacts were critically important. Such a discovery would not have been possible without the use of proximity collars that allowed us to detect these brief and cryptic contacts. Several previous studies indicate that host-to-host contact may be an important transmission route for gastrointestinal symbionts. For instance, reports by Belongia et al. (1993) and Besser et al. (2001) suggested that host-to-host contact in humans and cattle can facilitate the transmission of the pathogenic E. coli serotype O157:H7. Similarly, Johnson et al. (2008) found that human and animal household members, who directly interacted and were spatially proximate, were more likely to share E. coli strains than individuals from different households. However, these studies did not explicitly distinguish between host contacts and spatial proximity. In contrast, in our study, we were able to explicitly distinguish between the direct effect of associations among hosts and spatial proximity by comparing the performance of these variables in combined and separate models. Consistent with our findings that associations among hosts best predict E. coli strain-sharing, a few recent studies have specifically linked the transmission of gastrointestinal symbionts with host associations. A study by Bull et al. (2012) found that the sharing of commensal Salmonella enterica strains among wild sleepy lizards (Tiliqua rugosa) was influenced by the host’s social network and not spatial proximity. Furthermore, in a social bumble bee (Bombus impatiens), contact with nest mates infected with the intestinal pathogen Crithidia bombi was the only predictor of infection risk (Otterstatter & Thomson 2007). Johnson et al. (2008) also found that human sexual partners were more likely to share E. coli strains than other household members, suggesting an effect of direct social and physical contact. Transmission pathways can be difficult to elucidate, particularly in wild populations, as transmission events can rarely be observed. The assessment of strain- sharing as a proxy for transmission overcomes this challenge and provides a tangible opportunity to study the factors that facilitate transmission. However, for such studies to be informative, strain-sharing patterns must reflect contemporary primary transmission. To improve concordance between strain-sharing and transmission © 2014 John Wiley & Sons Ltd/CNRS

976 M. D. J. Blyton et al.

Letter

Table 4 Single explanatory variables of E. coli strain-sharing at Site 2

Explanatory variable*

Δ DIC

Total association duration Nocturnal association duration Log spatial proximity Spatial proximity Diurnal association duration Den-shared: Yes Pair-bonding: Yes Intercept Season: Winter-Spring

0.00 2.80 3.02 4.59 6.88 6.94 9.62 16.07 16.08

Posterior mean coefficient 0.516 0.563 2.602 0.012 0.426 2.248 2.042 1.211 0.426

95% credible interval 0.206, 0.203, 4.218, 0.017, 0.141, 0.696, 0.471, 1.835, 0.514,

0.827 0.958 1.018 0.004 0.701 3.695 3.502 0.522 1.571

MCMCp

Odds ratio†

Unit

Not all types of host contacts are equal when it comes to E. coli transmission.

The specific processes that facilitate pathogen transmission are poorly understood, particularly for wild animal populations. A major impediment for i...
256KB Sizes 0 Downloads 4 Views