Statistical analyses
For each biogeographic region, we normalized the OTU table for subsequent statistical analyses by rarefying the number of high-quality fungal sequences to the smallest library size in that region (14 241 reads for the Argentinian Yungas, 24 812 for Borneo, and 2000 for Panama). For the characterization of putative core communities of pantropical montane and lowland fungi, we used a dataset rarefied to 2000 reads to estimate species overlap among geographic regions and to evaluate and distinguish the possible effects of biogeography and environmental variables on community composition.
In each geographic region, total fungal richness as well as OTU richness of taxonomic classes and functional groups were compared among the three major elevational forest types via ANOVA with Tukey’s HSD test, implemented in R (R Development Core Team 2013). We calculated values of proportional richness and proportional abundance of functional groups on a per-sample basis and compared them among forest types as above. We used quadratic regression analyses, also in R, to examine relationships between elevation and environmental variables such as MAT, MAP, soil pH, organic matter, total nitrogen content, and carbon/nitrogen (C/N) ratio, as well as total P content in Argentina and Panama.
To compare community composition across elevational forest types, we used the vegan R package (Oksanen et al . 2015) to run Generalized Nonmetric Multidimensional Scaling (GNMDS) ordinations on the Hellinger-transformed OTU table and a secondary matrix containing environmental variables mentioned above. Ordinations were run separately for functional groups as well as for all fungi in each geographic region with the metaMDS function, which uses several random starts to find a stable solution. Data were subjected to 999 iterations per run with Bray-Curtis distance measure. Pearson correlation coefficient (r ) values and statistical significance between environmental variables and fungal community composition were calculated with theenvfit function, and vectors of variables with statistically significant correlations were plotted in ordinations. We plotted isolines of elevation on the GNMDS ordinations with the ordisurffunction.
Statistical tests of the equality variances via the betadisperfunction indicated no significant difference in multivariate homogeneity of group dispersions across elevational forest types in any region. To estimate the relative importance of forest type (categorical) and environmental (continuous) variables as sources of variation in fungal community composition, permutational multivariate analysis of variance (PerMANOVA) was carried out for all fungi and each functional group with the adonis function in vegan . To account for correlations among environmental variables, we performed a forward selection of parameters, including only significant environmental variables in the final model. For each geographic region as well as for the cross-regional comparison, we used partial Mantel test in veganto differentiate the effects of spatial distance and abiotic environmental variables, standardized with the scale function in R, on community structure. Finally, we carried out indicator species analysis (Dufrêne and Legendre 1997) to quantify associations of individual OTUs with specific elevational forest types at a pantropical scale with the multipatt function in the indicspeciespackage in R (De Cáceres et al . 2012).
Results