Statistical analysis
We conducted all analyses using package “lme4”123 in R v.3.6.0 (R Core Team 2019). We constructed separate models for each disease severity metric, each of which included an interactive term of site*origin site (“origin site” was the site from which bats were collected at the beginning of the experiment, whereas “site” was that within which bats were ultimately caged) as the predictor variable and a unique cage ID as a random effect. Pathogen growth rate was the only disease severity metric with a significant effect of origin site, so we report the effect for this metric but drop the origin site term in the other models. In all analyses, we only used data from alive bats except for analyses on tissue invasion. We did not use pathogen load data from dead bats because Pseudogymnoascus destructans is a poor competitor on bat carcasses and, therefore, swab data from dead bats does not accurately convey the pathogen load at the time of mortality.
To measure pathogen growth rate for each individual bat, we subtracted its early hibernation fungal load value from its late hibernation value to quantify the change in fungal loads. Bats that had no detectable fungus at the time of swabbing were assigned a value of 4.35e -03 pg (equivalent to a Ct value of 40) for that swab sample24, or a single P. destructans conidia. We then added a constant of 10 and log10-transformed the growth rate values. We used a linear regression to assess the differences in change in fungal loads on bats at each site. Incorporating cage ID as a random effect did not add explanatory power to this model and was dropped. In addition, to understand the relationship between roosting temperature and the change in fungal loads, we constructed a separate linear mixed model with the average roosting temperature (data collected by iButtons within each cage), origin site, and their interaction as fixed effects and cage ID as a random effect.
To test for differences in the severity of tissue invasion across the sites, we used a logistic regression with orange pixels indicating infection as successes and non-orange pixels as failures (generalized linear mixed model with binomial error distribution and logit link function), site as a fixed effect, and cage ID as a random effect. Additionally, to test for differences in tissue invasion between caged bats and free-flying bats opportunistically sampled at the end of hibernation in each of the persisting sites, we used a logistic regression with the same response variable (generalized linear mixed model with binomial error distribution and logit link function) and site, caging status (caged vs. free-flying), and their interaction as fixed effects. To explore the differences in weight loss across the three sites, we used a generalized linear mixed model (gamma error distribution, log link function) with weight loss as the response variable, translocation site as a fixed effect, and cage ID as a random effect. To test for differences in late hibernation body mass between caged and free-flying bats in each persisting site, we used a multiple linear regression with body mass as the response variable and an interaction term of site and caging status as the explanatory variable. We used a generalized linear mixed model (binomial error distribution, logit link function) to explore the relationship between early hibernation body mass and survival. Finally, we used a generalized linear mixed model (binomial error distribution, logit link function) to investigate how survival varied across the sites.