Successional trend
A two-step approach was used to determine the processes of endophyte community assembly and to evaluate whether these fungal communities followed the deterministic successional trend observed for their woody tree hosts (Jamison-Daniels et al. , 2021).
First, we assessed whether the amount of variation in community composition of our actual endophyte assemblages, explained by a number of predictor variables, was greater than (implying deterministic processes), less than (implying deterministic processes) or not different (implying stochastic processes) to what could be expected in randomly generated communities (Dini-Andreote et al. , 2015). Initially, a canonical correspondence analysis (CCA) was conducted to test which factors, namely host identity, spatial distance between sampled trees, abiotic conditions (maximum and minimum temperature, temperature standard deviation, light intensity, soil moisture measured per BC) and biotic factors (tree basal area per BC, tree height and distance to forest edge), best explained fungal community composition. Weighted linear regression was performed on the constraining variables (Ter Braak, 1986). To incorporate the effect of spatial distance on community composition, a weighted principal coordinates of neighbourhood matrix (PCNM) analysis (created using the geographic coordinates of every sampled tree) was performed. This analysis transforms coordinates to a rectangular distance matrix that is acceptable for constrained ordination techniques like CCA (Borcard and Legendre, 2002; Legendre and Borcard, 2008). Backward and forward stepwise permutation tests, for 1000 permutations, were used to determine the best fitting model, based on the model with the lowest AIC score (Venables and Ripley, 2002). Rare species contribute heavily to the chi-squared distance used to plot site and species scores in CCA analysis (Legendre and Legendre, 2012). Therefore, to reduce the spurious effects of rare species within the CCA, an eigenvalue decomposition approach was used to determine in how many samples a particular ASV must have occurred for it to be retained when performing the final ordination (Legendre and Legendre, 2012). When a considerable drop in inertia for one of the first five eigenvalues is detected, it indicates in how many samples an ASV must have occurred to be retained for the final CCA analysis (Supporting Information Figure S2) (Legendre and Legendre, 2012). The eigenvalue decomposition detected a drop in the inertia for the third eigenvalue after dropping ASVs that occurred in less than 20 samples (Supporting Information Figure S2). Therefore, all ASVs (n= 4905) that occurred in 20 samples or less were removed before the final CCA analysis (N= 411) and the randomisations.
The amount of variation explained by the best model with the retained predictor variables (host identity, light intensity and four spatial eigenvectors (spatial eigenvector 1, spatial eigenvector 2, spatial eigenvector 13 and spatial eigenvector 17)) was calculated. Then, 10 000 randomly assembled abundance-weighted community data matrices were constructed based on the true community identity, richness and abundance (Greve et al. , 2008). The randomly assembled matrices conserved species richness per sample as observed in the true community, and set the probability of species being selected proportional to ASV read abundance (Gotelli and Graves, 1996; Gotelli, 2000). For each of the 10 000 randomly assembled communities, a new CCA was performed with the predictors from the best CCA model (see above). The 2.5% and 97.5% quantiles of the percentage variation explained by the CCAs were calculated for the 10 000 random communities, and it was assessed whether the percentage variation explained by the CCA of the true community was larger than (> 97.5% quantile), smaller than (< 2.5% quantile) or not significantly different (between 2.5% and 97.5% quantiles) to the percentage variation explained by the randomly generated communities using a z-test (Greve et al. , 2008).
Because CCA randomisation analyses provided evidence for deterministic community assembly (see Results), a second analysis was conducted to test whether changes in community composition could be explained by BC size, as BC size increases with tree succession (Jamison-Danielset al. , 2021). A significant directional change in endophyte species composition with BC size would indicate deterministic succession, while no predictable change in composition with BC size would suggest stochastic succession (Dini-Andreote et al. , 2015). We tested this for the fungal communities extracted from each of the three host species using two different pair-wise similarity indices (Morisita and Raup-Crick indices). The Morisita index is weighted towards assessing similarity in common taxa, while the Raup-Crick index gives more weight to co-occurring rare taxa (Morisita, 1962; Raup and Crick, 1979). Pair-wise similarity in fungal community composition was calculated between all possible combinations of each of the smallest ¼ of the BCs and each of the largest ¾ of the BCs for each host species (following Jamison-Daniels et al. , 2021), to establish if there was directional change in community composition as BCs increase in size. For each of the largest ¾ BCs, similarity values with the smallest ¼ of the BCs were averaged. Generalised linear mixed effects models (GLMMs) were used to model the effect of BC area (of the largest ¾ BCs) on foliar fungal endophyte community similarity for both Raup-Crick and Morista similarity indices. BC identity of the largest ¾ BCs was included as a random effect in the model (McCulloch, 1997). Since the similarity values for both the indices are scaled between 0-1, models were fitted using a binomial distribution and logit link function (Zuuret al. , 2009). If the foliar fungal community similarity between the largest and the smallest BCs decreased or increased significantly with BC area of the largest BCs, it was interpreted as an indication of deterministic succession, while no relationship was taken as an indication of communities being ecologically neutral and primarily governed by stochastic processes that structure community composition (Hubbell, 2001; Dini-Andreote et al. , 2015; Jamison-Danielset al. , 2021).