Statistical analysis
Forward raw sequences for the amplicons were filtered by QIIME to remove
the noise and chimeras. After this process, the remained sequences were
clustered as operational taxonomic units (OTUs) with the average
neighbor algorithm of 0.03 distance. For the libraries of bacteria and
archea, the taxonomy of OTUs was performed using the Silva database
(release 128) by RDP classifier (v2.2), while OTUs were classified with
UNITE database (release 6.0) for fungal library.
For bacterial and archaeal libraries, the Silva database (release 128)
was used for taxonomic assignment of OTUs by using RDP classifier (v2.2)
(Wang, Garrity, Tiedje, & Cole, 2007). For fungal libraries, OTUs were
classified against the
UNITE
database (release 6.0).
All the statistic analysis were performed in R library using packages of
“vegan”, “Hmisc”, “ggtern” and “ggplot2” except for the
otherwise annotation. To determine the α-diversity, Shannon and Simpson
index were caculated for microbial diversity while ace and chao index
for richness and shannoneven index for evenness. To identify the
microbial β-diversity among soil, phyllosphere and faeces, ANOSIM,
ADONIS and Polar Coordinate Analysis (PCoA) were performed by the
“ANOSIM”, “ADONIS” and “PCoA” function of the “vegan” package
(Dixon, 2003). The OTUs overlaps were conducted into the Venn analysis
using package of “venndiagramm”. A network-like venn plot were
generated to reflect the comparison of OTUs among three biotopes by the
software of gephi. The taxa were also clustered into
Hierarchical
clustering tree at Order level by the “stats” package built-in R
library using the algorithm of bray-curtis. To clarify the alteration of
the microbial community among biotopes, the relative abundance of the
microbial communities on the phylum level were presented via ANOVA
barplot, radar plot and coxcomb plot. The Ternary Analysis, which
visually exihbited the microbial variation among three biotopes, was
presented by the “ggtern” package (Hamilton, 2017).
To examine the environmental impact on microbial community, the
canonical correlation analysis (CCA) was visualized using the “CCA”
function in the “vegan” package and embellished by “ggplot2” package
with the environmental factors of “carbon”, “nitrogen”, “nitrate
nitrogen”, “ammonia nitrogen”, “phosphorus”, “temperature”,
“pH” and “moisture” (Dixon, 2003; Wickham, 2009). All these factors
were also calculated for the Mantel test to confirm their respective
influence on microbial communities through “vegan” package (Dixon,
2003). A structure equitation model (SEM) was constructed to examine the
effect of physical and nutrient factor effect on microbiota among
biotopes via path coefficient. SEM was commonly used as a multivariate
technique achieved by statistical methods and computer algorthms (Gupta,
Kapur, & Kumar, 2017). The appropriate variables were selected based on
the test for pre-assumed causal relationships as an initial step of SEM.
Such test was based on the statical cariteria referring to the standard
estimating results and path coefficients via Amos, 22.0 (SPSS Inc.,
Chicago, IL, USA) (Mardani et al., 2017). The Random Forest model was
employed to present the top 15 genuses of the microbial community via
random forest package in R.