Drivers of assemblage composition
We tested which factors, namely host identity, spatial distance between sampled trees, abiotic conditions (maximum and minimum temperature, temperature standard deviation, light intensity and soil moisture measured per BC) and biotic factors (tree basal area per BC, tree height, distance to forest edge and the tree composition per BC), affected fungal endophyte community composition.
To do this we used a generalized dissimilarity modelling (GDM) approach. This flexible approach enabled us to simultaneously incorporate categorical (i.e. host), linear (e.g. climate), compositional (i.e. tree compositional dissimilarity), and spatial data (i.e. geographic distance) predictors of endophyte composition into one analysis (Ferrieret al. , 2007). GDM is a nonlinear extension of matrix regression, which has specifically been designed to deal with two types of nonlinearity commonly encountered in biological data: 1) the curvilinear relationship between ecological or spatial separation and the observed compositional dissimilarity, and 2) non-stationarity, i.e. differences in the rate of compositional turnover along environmental or spatial gradients (Ferrier et al. , 2007; Fitzpatrick et al. , 2013).
The default of three I-spline basis functions (knots) per predictor variable was used in all GDM analyses (Ferrier et al. , 2007); and backwards selection was used to determine how many variables to retain in each of the final models (Williams et al. , 2012). The sum of the coefficients per I-spline represents the maximum amount of variation explained by a particular variable, and can be used to determine variable importance (Ferrier et al. , 2007). Since host identity explained the majority of the endophyte compositional dissimilarity for the full dataset (see Results), we repeated the GDM analyses per individual host species using the same approach as above.