Assemblage composition
The final GDM for the full dataset retained two host species (E.crispa and S. chirindensis ), tree basal area, distance to the indigenous forest edge, geographic distance between samples, light intensity and tree compositional dissimilarity as predictor variables (Figure 2). The final model for the full dataset was significant (Null Deviance = 691.996, GDM Deviance = 491.635,p -value = 0.000001), as were all the predictor variables, except light intensity (Figure 2; Table 2). The final model explained 28.95% of the deviance in the turnover of foliar fungal endophyte community composition. With host identity alone explaining ~23% of this variance. Of the remaining predictors distance to the forest and the difference in tree basal area were the most important, both having variable rates of turnover along the gradients and displaying the quickest turnover at short distances from the forest and small differences in tree basal area (Figure 2c & d). The rate of turnover in fungal endophyte composition along the tree compositional dissimilarity gradient was the steepest of any of the gradients; with the maximum magnitude of turnover reached when the difference in surrounding tree composition was only slightly dissimilar, i.e. samples not from the same BC (Figure 2f).
When GDMs were run per host species, different predictors were retained as important in explaining the composition of endophytes within different hosts. Only geographic distance between samples was retained and significant in all models (Figure 2 & Figure 3). The final GDM model for E. crispa was significant (Null Deviance = 147.69, GDM Deviance = 122.65, p­- value = 0.000001) and explained 16.96% of the deviance in turnover of foliar fungal endophyte community composition. It retained five variables (Supporting Information Table S5), of which distance to the forest edge, geographic distance between samples and the difference in tree basal area were significant (Figure 3a-c). Distance to the forest was the most important predictor, with the rate of turnover being approximately linear (Figure 3a). The rate of turnover in endophyte composition along the geographic distance and tree basal area gradients were non-linear with the highest rates of turnover occurring at larger geographic separation between samples and small differences in tree basal area (Figure 3b & 3c).
For C. inerme , the final GDM model retained five predictor variables (Supporting Information Table S6); however, only minimum temperature and geographic distance between samples were significant (Figure 3d & 3e). Minimum temperature was the most important predictor with both larger differences in minimum temperature and geographic distance between samples resulting in higher rates of turnover. The magnitude and rates of turnover in endophyte composition was almost non-existent at low differences in minimum temperatures and short geographic distances between samples but increased dramatically at the higher end of these gradients (Figure 3d & 3e). The final model was significant (Null Deviance = 52.85, GDM Deviance = 45.92, -value = 0.00503), and explained 13.12% of the deviance in turnover of foliar fungal endophyte community composition.
The final GDM model for S. chirindensis fungal communities retained five predictor variables (Supporting Information Table S7); but the only retained variable which was significant was geographic distance between samples (Figure 3e). The rate of turnover in endophyte composition increased with distance between samples; this increase was most rapid at short geographic distances (Figure 3e). The final S. chirindensis GDM model was significant (Null Deviance = 42.78, GDM Deviance = 38.622, p -value = 0.000001), and explained 9.52% of the deviance in turnover of foliar fungal endophyte community composition.