2.5 Statistical analyses
A one-way analysis of variance (ANOVA) followed by a Duncan’s test (p <0.05 ) was applied to test the differences in soil properties, fungal diversity and abundance across five sites. PCA and Pearson’s correlation was used to evaluate the SQI using the IBM SPSS software (SPSS Inc., Chicago, USA). For the sequence data, the α diversity indices including Observed species, Shannon-Wiener’s diversity, Chao1 and species accumulation curves were calculated using Mothur and visualized in Origin 2018 software (OriginLab, USA). The β diversity in the fungal community was visualized in CANOCO 5.0 software (Microcomputer Power, Ithaca, New York) by using non-metric multidimensional scaling (NMDS) ordinations based on Bray–Curtis distance. Network visualized by Cytoscape (Vailaya et al., 2007) was used to analyze the number of OTU which was cosmopolitan (present in all forest types), specific to forest or plantations (i.e., only detected in one forest type) or broadly dispersed (found in two or four sites). Linear discriminant analysis (LDA) effect size (LEfse) was conducted to identify the biomarkers between forest and plantations using an LDA score > 4 and p < 0.05 for the nonparametric Kruskal–Wallis test. A heatmap was conducted in R with the “pheatmap” package to compare the variation in dominant “funguilds” (top 35 in abundance). Relationships between soil parameters and the fungal community were performed with redundancy analysis (RDA) by using the CANOCO 5.0 software. Structural equation model (SEM) was further used to determine indirect or direct contributions of soil properties and fungal community with a multivariate approach using AMOS software (IBM SPSS AMOS 20.0.0). In this model, the Shannon index was selected as an informative estimate of taxonomic diversity for the fungal communities (Delgado-Baquerizo et al., 2016). The SEM fitness was examined on the basis of a non-significant chi-square test (χ2), p value (>0.05), the goodness-of-fit index (GFI, good fit when GFI>0.9) and the root mean square error of approximation (RMSEA, good fit when RMSEA<0.05) (Byrne & Erlbaums, 2009). The fit of additional indices of SEM is shown in Table S6.