3.2 Responses of different phyla to grassland degradation
To decipher the responses of bacterial communities to grassland degradation, we explored the shifts in the abundance and species richness among the top eight phyla (relative abundance > 3%).
Proteobacteria, Bacteroidetes, and Firmicutes exhibited relatively high species richness values, whereas other phyla, such as Acidobacteria, Actinobacteria, Gemmatimonadetes, Planctomycetes and Chloroflexi, exhibited relatively low richness values as degradation level increased (1–5 in G1; 1–4 in G2) (Table 1) . The decrease in species richness observed in the deserted regions (region 6 in G1; region 5 in G2) could be attributed to the substantial decrease in richness at the community level. The proportions of Proteobacteria (21.2–32.4% in G1 and 21.9–28.6% in G2) and Bacteroidetes (3.8–11.0% in G1 and 4.7–7.6% for G2) increased with an increase in grassland degradation level (Figure 2A and 2B) , indicating that Proteobacteria and Bacteroidetes exhibited positive responses based on biodiversity in the course of grassland degradation.
Conversely, with advancement in degradation, bacterial communities had more significantly altered OTUs (Figure S3) . After classifying the altered species into corresponding phyla, we observed that the numbers of the enriched OTUs in Proteobacteria and Bacteroidetes increased with an increase in the degradation level in the two grasslands (Figure 2C and 2D) . On the contrary, Actinobacteria, Acidobacteria, Chloroflexi, and Planctomycetes had greater depleted OTUs than other phyla with an increase in degradation level. The results indicate that Proteobacteria and Bacteroidetes had more positive responses with regard to survival than other phyla under grassland degradation.
To determine the dominant bacterial biomarkers along the grassland degradation gradient, we conducted RF to selected the top 30 most important OTUs and classified them into phyla. Subsequently, their correlation with environmental factors was determined using Pearson correlation analysis. The major microbial biomarkers were classified into the phyla Proteobacteria, Acidobacteria, and Actinobacteria in the process of degradation (Figure S4) . In the case of Acidobacteria, the biomarker OTUs were almost all classified into the Subgroup_6, and were significantly and positively correlated with soil nutrient contents (OM, TN, and TP) in G2 (Figure 2E and 2F,P < 0.05) . The Actinobacteria biomarker species in the two grassland types were largely classified into the Thermoleophilia and Actinobacteria classes, which were positively and negatively correlated with nutrient content, respectively (Figure 2E and 2F) . Furthermore, the biomarker OTUs in Proteobacteria were mostly classified into Gammaproteobacteria in G1, and into Gammaproteobacteria and Alphaproteobacteria in G2. The abundance of biomarkers classified into Alphaproteobacteria were positively correlated with soil nutrient (OM, TN, and TP) contents in G2 (Figure 2E and 2F) . However, Gammaproteobacteria biomarker species exhibited significant and negative correlations with nutrient contents in both grassland types(Figure 2E and 2F) .