Apparent absence of Batrachochytrium salamandrivorans in wild urodeles in the United Kingdom

Whether an infectious disease threat to wildlife arises from pathogen introduction or the increased incidence of an already-present agent informs mitigation policy and actions. The prior absence of a pathogen can be difficult to establish, particularly in free-living wildlife. Subsequent to the epidemic emergence of the fungus, Batrachochytrium salamandrivorans (Bsal), in mainland Europe in 2010 and prior to its detection in captive amphibians in the United Kingdom (UK), we tested archived skin swabs using a Bsal-specific qPCR. These samples had been collected in 2011 from 2409 wild newts from ponds across the UK. All swabs were negative for Bsal. Bayesian hierarchical modelling suggests that Bsal was absent from, or present at very low levels in, these ponds at the time of sampling. Additionally, surveillance of newt mortality incidents, 2013–2017, failed to detect Bsal. As this pathogen has been shown to be widespread in British captive amphibian collections, there is an urgent need to raise awareness of the importance of effective biosecurity measures, especially amongst people with captive amphibians, to help minimise the risk of Bsal spreading to the wild. Continued and heightened wild amphibian disease surveillance is a priority to provide an early warning system for potential incursion events.


Introduction
This document describes a Bayesian model for estimating prevalences of Batrachochytrium salamandrivorans (Bsal) in newts in the United Kingdom, based on swab data obtained in 2011. This document (along with the corresponding data file), contain the code necessary to repeat the analysis in R 1 (using RStudio 2 ) and the freely available WinBUGS package 3 .
The analysis requires the R2WinBUGS 4 , coda 5 , magrittr 6 and tidyverse 7 suite of packages. This document was written using the bookdown 8 package, which is based on the core rmarkdown 9 and knitr 10 packages.
Firstly, we can load the necessary libraries. We assume throughout this document that WinBUGS has been installed in its default location and that all model and data files are in the same working directory in R. If not, then some amendments to the code will have to be made to ensure correct linkage to the necessary files.

## load libraries library(R2WinBUGS) library(coda) library(tidyverse) library(magrittr)
Data are provided as a Supplementary Data File ("counts.csv") online. Now we read in and summarise the data. ## read in newt data newtcounts <-read_csv("Supplementary Worksheet S1.csv") %>% mutate(Y = 0L) %>% select(SiteID, Y, X = Total number of newts sampled ) head(newtcounts) SiteID Y X <chr> <int> <int> This has 103 ponds, with SiteID corresponding to an identifier for each pond, Y being the number of positive swabs per site, and X the number of swabs taken per site. We can summarise the data as follows: ## produce a summary of the data summary(newtcounts) Hence there are a range of sample sizes, due to the fact that the data were collected as part of an earlier study not focussed on Bsal. However, most ponds have at least 29 samples, with only a small number having very low numbers of samples. Of all the 2409 samples tested, none were positive for Bsal.

Bayesian model
Let Y i be the number of positive swabs in pond i (i = 1, . . . , P ), where P = 103 is the number of ponds. Similarly, let X i be the number of newts sampled in pond i. To capture whether ponds are infected or not, we introduce a latent variable Z i , such that The number of positive swabs is then modelled as: i is the prevalence of the disease in pond i, given that pond i is infected, and p sens is the sensitivity of the diagnostic test (we assume 100% specificity here).

Posterior probabilities of infection for individual ponds
We can calculate the posterior probabilities of infection for each pond by simply take the average of the corresponding Z i samples for each pond as below. The larger estimated values are for those ponds with low numbers of samples, corresponding to a higher probability of missing infection if it was present in that pond. ## extract posterior samples for Z and amalgamate ## into correct form for plotting using ggplot Z_results <-model_mcmc %>% map("sims.matrix") %>% map(~{.