tion of taxa at the tips of the community phylogenetic tree 1,000 times. This is equivalent to randomly sampling the taxa while constraining the number of taxa in each community, the number of taxa shared by any two communities, and taxa occurrence across all communities. The observed phylogenetic slope was assumed significantly different from the null if it was greater than or less than 975 of the slopes of the randomizations (two-tailed test, P < 0.05). To address the unequal sampling of microbial communities, we repeatedly calculated and tested the significance of distance–decay slope on subsampled communities, where communities were subsampled by the number of clones in the smallest library. The significance of results did not differ between repeated subsamples.
To determine the correlation between elevation and potential environmental drivers with the observed diversity patterns, we used polynomial regression analysis. For each environmental variable, we fit a linear and a quadratic regression model. The best model was determined based on Akaikes Information Criterion differences (Burnham and Anderson, 2002). Using a stepwise regression to select variables and interactions, a multivariate model was constructed for each alpha diversity measurement.
Mantel tests (999 permutations) were used to determine whether compositional and phylogenetic similarity decayed significantly with elevational distance (Legendre and Legendre, 1998). Similarity values between pairwise comparisons of microbial communities were the averages of 1,000 rarefaction samples, as described above. The best fit and the most homoscedastic residuals were found in models that used the log transformation of similarity against elevational distance, with the exception of angiosperm taxa similarity, which was best described by a linear–linear distance–decay model. We used Mantel tests to examine correlations between community similarity and environmental similarity [for a discussion of these methods, see Legendre et al. (2005) and Toumisto and Ruokolainen (2006)]. We chose the combination of environmental variables that best explained the changes in angiosperm and Acidobacteria community composition between samples with BIO-ENV (Clark and Ainsworth, 1993) and tested the importance of these variables after controlling for geographic distance and vice versa by using partial Mantel tests. For all analyses, moisture, carbon, and nitrogen were arcsin(sqrt(y))-transformed and aspect was 1/y-transformed (Legendre and Legendre, 1998).