Statistical analysis
Principal component analysis (PCA) was used to measure the overall
functional gene structure. Bray–Curtis distance was used to obtain
dissimilarity matrices in the adonis algorithm of the dissimilarity test
for comparing GeoChip data of four elevations. The similarity test,
Mantel test, Canonical correspondence analysis (CCA) and Variation
partitioning analysis (VPA) were used to evaluate the linkages between
microbial gene compositions and environmental attributes. In the
similarity test, Euclidean distance was used to calculate the distance
between samples, followed by calculation of Pearson correlation coeffi-
cient. To select attributes in CCA modeling, we used variation inflation
factors (VIF) to examine whether the variance of canonical coefficients
was inflated by the presence of correlations with other attributes. If an
attribute had a variation inflation factor value larger than 20, we
deemed it to depend on other attributes and consequently removed it
from the CCA modeling. Correlation coefficients (r) were calculated
using Pearson's correlation. The normalized total gene abundance for
each functional gene was the average of the total gene abundance
from all the replicates and all data are presented as mean ± s.e. The
least significant difference (LSD) test was used to compare the signifi-
cance of differences in relative abundance among four elevations. All
of the analyses were performed with the Vegan package (v.1.15-1)
using R, version 2.8.1 (R Foundation for Statistical Computing, Vienna,
Austria).