3 RESULTS
3.1 Resistance of ecosystem multifunctionality and major microbial taxa
After conversion from primary forest to the other three land use, we found similar multifunctional resistance among SF, AL and CL (Fig. 1). C-cycling functional resistance decreased by 77%, nitrogen and phosphorus functional resistance increased by 17% and 19% in abandoned land, compared to secondary forest. Cultivated land had the lowest N functional resistance, while SF had lowest P functional resistance (Fig. 1).
At the level of the major bacterial phylum, in addition to Verrucomicrobia, the groups with relatively high abundance had higher resistance than the groups with low relative abundance, and the proteobacteria (p<0.001) had the highest resistance (Fig. 2). Land-use intensity levels affected the resistance of the major bacterial phylum (Fig. 3a). Compared with SF, the resistance of Actinobacteria (p < 0 .001), Acidobacteria (p = 0.001) and Chloroflexi (p = 0.049) were 40%, 11% and 78% lower, respectively, in AL, and 41%, 8.0% and 53% lower in CL. In contrast, Verrucomicrobia (p<0.001) was 49% more resistant in AL than in SF.
Significant effects of land-use intensity levels were also found for bacterial multifunctionality resistance at the class level (Fig. 3b). Deltaproteobacteria, Actinobacteria and Alphaproteobacteria had the highest resistance (Fig. 2). Alphaproteobacteria, Betaproteobacteria, Gammaproteobacteria and Thermoleophilia showed a pattern of lesser resistance in AL and CL than in SF (p < 0.05, Fig. 3b). In contrast, Actinobacteria was more resistant in AL (42%, p < 0.05) as well as in CL (10%, p < 0.05) compared to SF.
3.2 Linkage of bacteria communities with resistance of ecosystem multifunctionality
NMDS ordination of the bacterial Bray-Curtis distance matrices clearly separates the PF and SF from the AL and CL (Fig. 4a). Soil physicochemical properties and soil enzyme activities were also clearly separated by land-use in the PCA plot (Fig. 4b) but, in contrast to the NMDS plot, the two forest sites (PF and SF) were not grouped close together. Structural equation modelling was constructed to explore the direct and indirect effects of land-use and bacterial communities on multifunctional resistance (Fig. 4c). The bacterial communities (NMDS1, NMDS2, p<0.05) showed the largest direct positive effect on multifunctional and N-related functional resistance (Fig. 4e, Fig. S1). The positive direct effect of soil Mg content (path coefficient = 1.029) on C-functional resistance was the greatest, followed by bacterial community structure and Fe content (Fig. 4d, Fig. S1). Besides bacterial community, land use presented as the second main predictor influencing the N functional resistance via direct (path coefficient = 0.676) and indirect effects (path coefficient = -0.237) on soil pH, minerals and bacterial community (Fig. 4d, Fig.S1). Among soil variables, the positive direct effect of bacterial community on P functional resistance was the greatest, while Mg was related to the largest negative indirecteffect (Fig. 4f, Fig. S1).
3.3 Bacterial groups predicting multifunctional resistance of ecosystem and soil single function
To further elucidate the role of bacterial community structure in driving multifunctional resistance, we used Random Forest modelling to quantify the influence of bacterial abundance at both phylum and class level on multifunctional resistance of the different ecosystems in the land-use gradient (Fig. 5). The model explained 46.4% variance on multifunctional resistance (Fig. 5). The minimum cross-validation error was obtained when using 20 important classes, And the number of classes against the cross-validation error curve stabilized (Fig. 5a). Thus, we defined these 20 classes as biomarker taxa in the model. The list of the top 20 bacterial taxa at the class level across multifunctional resistance, in order of multifunctional resistance importance, is shown in Fig.5a. Verrucomicrobia showed the highest contribution to multifunctional resistance (Fig. 5a), and was mainly present in secondary forest and abandoned land (Fig. 5b). This was followed in importance by Chloroflexia, Anaerolineae and Chloroplast (Fig. 5a), which started to accumulate in the AL and CL stages (Fig. 5b). We also found Thermoleophilia and Alphaproteobacteria showed a contribution to multifunctional resistance (Fig. 5a), and started to accumulate in PF and SF (Fig. 5b). These indicates that some biomarker taxa showed high relative abundance in the corresponding land-use gradient (Fig. 5b).
Based on 20 bacterial groups as important predictors identified by random forests(Fig. 5a), we further predicted the soil single function (e.g. TOC, TN, and TP contents) and listed the corresponding significant microbial species (Fig. 6, p<0.05).Our Random Forest models indicate that the relative abundance of Chthonomonadetes and OPB35_soil_group were the best predictors of TOC and DOC, respectively. The relative abundance of Chloroflexia, Alphaproteobacteria and OPB35_soil_group were the best predictors of TN, NH4+ and NO3-, respectively. OPB35_soil_group was also the best predictor for TP.