Cary Institute of Ecosystem Studies, United States
Abstract: Understanding the dynamics of zoonotic and wildlife pathogens depends on accounting for how infection status and shedding of pathogens respond to processes on multiple scales. These range from landscape features affecting spatiotemporal location of hosts and transmission to within host processes affecting immunity and infectiousness. Nevertheless, most studies of infectious diseases focus on a single scale. Here we develop a multi-scale model of Hendra virus shedding to inform spillover risk. Hendra virus circulates in Australian flying fox populations and spillover occurs when virus is shed in fields and paddocks where horses can become exposed and infected. Flying foxes are colonial roosting bat species that are hypothesized to have primarily latent (inactive) Hendra virus infections. Re-activation of infections and viral shedding in urine is expected with increased stress. We develop modular mechanistic models incorporating roost to continental scales to describe ecological, epidemiological, and biological processes affecting Hendra virus shedding. We parameterize these models using existing datasets and compare performance of alternative multi-scale model structures to under-roost viral shedding data from 2011-2014 in 26 longitudinally sampled roosts across a 2300-km transect. We used boosted regression trees to identify important environmental predictors and temporal lags of these environmental features for 1) reservoir species roost locations, 2) flying fox food shortage, 3) formation of new flying fox roosts, and 4) flying fox rehabilitation admission. Our comparison of seven model structures finds that the most accurate multi-scale model includes two of three stress proxies (food shortage and rehabilitation admissions). While model predictions do not always precisely track the magnitude of observed prevalence, they successfully differentiate low and high prevalence areas that are consistent with observed shedding. Within model components, some environmental features are only informative when temporal lags (up to 18 months) are considered. This study identifies environmental variables that influence Hendra virus shedding, and highlights how incorporating multiple scales into shedding predictions improves risk forecasts.