Distribution of the disease pathogen Batrachochytrium dendrobatidis in non‐epidemic amphibian communities of western Canada

Abstract

Chytridiomycosis is an emerging infectious disease of amphibians caused by the waterborne pathogen Batrachochytrium dendrobatidis (Bd) and is responsible for the recent decline of species worldwide. Elucidating patterns in disease prevalence has proved challenging, as small‐scale studies to date have provided conflicting results. We present data on the prevalence of Bd (the proportion of individuals with Bd) collected from amphibians sampled throughout British Columbia. We had two different datasets: our original dataset includes 1129 individuals sampled in 103 sites representing 14 species. A second dataset has 839 individuals sampled in 19 sites representing 10 species. We use a Bayesian state‐space occupancy modelling approach to estimate the probability that an individual is Bd+ as a function of individual and site characteristics. Cross‐validation techniques and the original dataset were used to find the best model: this model includes species, life stage, and geographic location. Our results suggest that Bd prevalence is not strongly related to seasonality, latitude or site type. Within a species, Bd prevalence depended on life stage; the watershed in which a site occurs may also usefully predict prevalence. Overall observed infection prevalence was ∼16%. Our best model accurately assigns Bd status to an individual ∼ 42% of the time. Taking advantage of the Bayesian framework, we ran an analysis with the second dataset using estimates from the original model as prior values. We present posterior density distributions for those sites and species with narrow credible intervals, and show that sites tend to be either highly likely or highly unlikely to have individuals with Bd, while individuals of some stages of some species have an intermediate likelihood of being Bd‐positive. The Bayesian model using informed priors had increased accuracy rates in assigning Bd status to both individuals and groups of individuals.

Keeping the Heat On: Weighted Surveillance for Chytrid Fungus (Batrachochytrium Dendrobatidis) in Dixie Valley Toads (Anaxyrus [= Bufo] Williamsi)

“…The likelihood of detecting Bd at a site where it occurred varied among species and life stages. Surveillance weights calculated from the data presented in Richardson et al (2014) indicated that detection of Bd in larval western toads was only 0.04 (0.01-0.11) times as likely as in adult western toads for the same number of individuals sampled (Table 2). In contrast, detection of Bd was 1.69 (0.96-2.93) times more likely in adult American bullfrogs than adult western toads and 10.2 (4.62-22.9) times more likely in larval American bullfrogs than adult western toads (Table 2).…”

“…In contrast to the relatively small difference in surveillance weights between adult toads and bullfrogs, larval bullfrogs had much higher surveillance weights. Indeed, if relative prevalence (i.e., risk of infection) of Bd among taxa in our study area is similar to that of Richardson et al (2014), then sampling a single larval American bullfrog would be equivalent to sampling about 10 adult Dixie Valley toads. Sampling bullfrog larvae for Bd could have the benefit of removing individuals of an invasive species while increasing the probability of detecting Bd in Dixie Meadows.…”

“…where µ ref was an intercept term that represented the reference class (adult western toads) against which the other surveillance classes were compared, x I was an indicator variable for the surveillance class that each sampled amphibian i belonged to, β was a vector of coefficients that estimated the surveillance class log infection rate ratios, and a was a random effect for spatial unit j . We applied the watershed designation from Richardson et al (2014) for each sample as the sample unit to account for spatial variation in detection rate. We used a Bayesian approach with Gibbs sampling implemented in JAGS (Plummer 2017) to estimate the relative surveillance weights using uninformative priors, norm (0, 1000), on the cloglog scale for the parameters for reference (µ ref ) and surveillance class coefficients (β).…”

“…We estimated surveillance weights and minimum detectable prevalence of Bd in two steps. First, we estimated surveillance weights focusing on four classes of Bd host (adult American bullfrogs, larval American bullfrogs, adult western toads, and larval western toads) using data from Richardson et al (2014) on prevalence of endemic Bd in these and other species of amphibians. We used western toads as a surrogate for Dixie Valley toads because Dixie Valley toads were formerly considered western toads and western toads are the most closely related species for which data were available.…”

“…We used western toads as a surrogate for Dixie Valley toads because Dixie Valley toads were formerly considered western toads and western toads are the most closely related species for which data were available. We chose to use these data to estimate surveillance weights because Richardson et al (2014) sampled Bd from multiple amphibian host classes (species and life stages) across the same time and space where Bd occurrence was widespread. We applied a discrete proportional hazards model (Heisey et al 2014;Jennelle et al 2018) to estimate the baseline Bd detection rate (i.e., apparent prevalence) in adult western toads and the relative Bd detection rate among I additional amphibian classes defined by species and life stage using a Bernoulli likelihood, Lðp I; j jy i;I; j Þ ¼ p I; j à ð1 À p I; j Þ; where y i,I,j was the observed detection or nondetection of the ith individual, belonging to surveillance class I, sampled from spatial unit j, with probability of detection p I,j .…”