Additional file 1 of Specialized metabolic functions of keystone taxa sustain soil microbiome stability

Additional file 1: Figure S1. (A) Schematic diagram of soil microcosm establishment. (B) Soil incubation experimental design. Figure S2. The Lg transformed 16S rRNA gene copies in all microcosms at the end of incubation period (sixteen weeks) by quantitative real-time PCR. Error bars represent standard deviations (SD, n = 12). Dil: Dilution level. No statistically significant differences were found among different dilutions and pH levels. Figure S3. The average error rate of 10-fold cross validation based on seven machine learning classification methods. Error bars represent standard deviations (SD, n = 12). Different letters above bars indicate significant differences (P-value < 0.05) between machine learning methods according to Duncan’s multiple comparison. Figure S4. Weighted Unifrac distances between bacterial communities from different dilution levels and the untreated initial soil (CK). Lg(Dil) indicates the Lg transformed dilution level. Lg(Dil) = 0 represents the untreated soil. Asterisks indicate significance: **, P-value < 0.01 based on Tukey’s HSD test. n.s.: Not significant. Boxplot: median, 25%/75% percentiles, and the highest, lowest and extremely values are shown. Figure S5. The relationship between βNTI and differences in soil pH on each dilution level (A) and across all samples (B). Figure S6. The average variation degree (AVD) of all OTUs on each dilution level. Horizontal axis represents different OTUs and vertical axis represents the absolute AVD value of each OTU. Figure S7. The average variation degree (AVD) values under different rarefaction depths of 11,020 reads and 8,000 reads per sample. Figure S8. The average variation degree (AVD) values under the normalization method of DESeq variance stabilization. Figure S9. The average variation degree (AVD) values of a community stability test (ref [19] of the main text) by detecting the dynamics of community composition with one of the species removed from the original seven-strain synthetic community. The seven strains are E. cloacae (Ecl), S. maltophilia (Sma), O. pituitosum (Opi), H. frisingense (Hfr), P. putida (Ppu), C. pusillum (Cpu), and C. indologenes (Cin). C7, all seven strains; -Cpu, remove C. pusillum from seven strains; -Cin, remove C. indologenes from seven strains; -Opi, remove O. pituitosum from seven strains; -Hfr, remove H. frisingense from seven strains; -Ecl, remove E. cloacae from seven strains; -Ppu, remove P. putida from seven strains; -Sma, remove S. maltophilia from seven strains. Figure S10. The average variation degree (AVD) of all OTUs on each pH level for 10-1 (A), 10-4 (B), 10-7 (C) and 10-10 (D) diluted samples. Horizontal axis represents different OTUs and vertical axis represents the absolute AVD value of each OTU. Figure S11. The relative abundance of the same bacterial genera in Fig. 3 detected by 16S rRNA gene amplicon sequencing. The genera with red text were not detected in amplicon sequencing. Error bars represent standard deviations (SD, n = 240). Table S1. The packages and functions of machine learning classification methods. Table S2. The nodes identified as module hubs in the functional gene co-occurrence networks on four dilution levels. Supplementary Note 1. R codes for 10-fold cross validation using seven machine learning classification methods. Supplementary Note 2. R codes for calculating spearman’s correlation between gene categories. Supplementary Note 3. R codes for calculating βNTI metrics.