Network construction
The OTUs with relative abundance<0.01% were removed to reduce the rare OTUs in dataset. Only the OTUs occurred in more than half of the samples were screened for the co-occurrence networks. The co-occurrence networks were constructed based on the Spearman correlation matrix with “Hmisc” package (Harrell, 2008). The nodes of the networks indicated the species and the edges between them implied the correlation between the OTUs. The P values were controlling symmetrically for the further multiple test via the false discovery rate (FDR) procedure (Benjamini, Krieger, & Yekutieli, 2006). A cutoff wall was set up based on the Spearman correlation values and P values. All the cutoff of correlation coefficient were 0.9 via random matrix theory-based method. The cutoff of the P values for fungi and bacteria were determined as 0.001 (Luo, Zhong, Yang, Scheuermann, & Zhou, 2006). To simplify the network for better visualization, only fungal and bacterial OTUs over than 30 and archaeal OTUs over than 5 were employed. The matrix containing the information of nodes and edges were visulized into image by Gephi platform using the algorithm of Fruchterman Reingold (Bastian, Heymann, & Jacomy, 2009).