Results
The laboratory experiment revealed that geographic distance between the host and Bd collection locations, a metric for host-pathogen novelty (hereafter, distance), was negatively associated with host mortality (β = -0.06, SE = ±0.03, z = -2.10, p = 0.04; Figure 2A & Table S3), Bd prevalence (β = -0.53, SE = ±0.23, z = -2.33, p = 0.02, Table S5), and pathogen abundance on the host (β = -0.03, SE = ±0.02, z = -2.01, p < 0.05, Figure S1 & Table S4). Distance was also negatively associated with host mortality in the global-scale meta-analysis (while accounting for among-study variance, Bd strain, and host taxonomic group; β= -0.77, SE = ±0.25, t = -3.05, p < 0.01; Figures 2B & Table S4).
While local host-pathogen interactions generally result in worse outcomes for hosts, there was substantial variation among host-pathogen outcomes. We found significant variation in mortality, infection success, and pathogen load among Bd strains (main effect of strain identity on mortality: χ2 = 61.05, p< 0.001; prevalence: χ2 = 56.98, p< 0.001; pathogen load: χ2 = 351.65,p < 0.001; Figure 3 & S2). Specifically, when averaging across host species, Bd from Louisiana was most deadly, most likely to cause infection, and produced the highest infection burdens (Figure 3 & S2). We also found significant variation in mortality, prevalence, and pathogen load among host species (main effect of host on mortality: χ2 = 136.22, p < 0.001; prevalence: χ2 = 9.95, p = 0.04; and pathogen load = χ2 = 12.36, p = 0.01; Figures 3 & S2). Specifically, when averaging across Bd strains, toads from Arizona and Tennessee had the highest mortality and infection prevalence (Figure S2). Consistent with the experimental results, in the meta-analysis, we found significant variation in mortality among taxonomic groups with Bufonoidea species being especially susceptible to Bd-induced mortality (χ2 = 29.92, p < 0.001; Figure 3).
Importantly, we compared fits of all models from the experiment with and without the inclusion of a two-way interaction between host and Bd identity and determined that models without the interaction fit better and thus did not include the interaction term in any model (ΔAIC >7). This means that the main effects of host and Bd strain are more important than any potentially idiosyncratic interactions between these two factors. Finally, our log-likelihood ratio test of proportional hazard model fits from the laboratory experiment revealed that host species accounted for the most variation in the model (χ2 = 140.91, df = 4, p < 0.001), followed by Bd strain (χ2 = 60.22, df = 5, p < 0.001) and distance (χ2 = 4.19, df = 1, p = 0.0406).