Results
Soil microbial diversity . The dilution-to-extinction approach was successful in creating a gradient of soil microbial (bacterial and fungal) diversity; importantly, soil microbial biomass, as indicated by respiration rates, was recovered after incubation (Fig. 2A, B and C; Table S2). Compared with the undiluted soil inocula, 57.91% of soil microbial taxa was lost through the 10-3 dilution, and 78.18% was lost during the following 10-6 dilution based on the number of soil microbial amplicon sequence variants (ASVs) (Fig. 2A). Low-abundance taxa, for instance, Chytridiomycota and Chlamydiae , were first eliminated during dilution (Fig. 2B and C). Although the microcosms were open to microbial re-colonization from the air, which increased soil bacterial diversity especially in the low soil biodiversity treatment, differences of fungal diversity were still observed among soil dilution treatments at the end of the experiment (Fig. S1 and Table S2). Besides, the diversity of fungal mutualists in the phylum of Glomeromycotawas dramatically decreased at the 10-3 dilution and was absent at the 10-6 dilution (Fig. S1A and Table S2). The diversity of plant fungal pathogens was deceased by soil dilution in the soil inocula, and was not altered at the last harvest (Table S2). Because soil biodiversity is dominated by soil microbes, soil microbial diversity was used as an indicator for soil biodiversity in the present study.
Temporal stability of plant community biomass production . Model selection with Akaike information criterion (AIC) suggests that the additive models including soil biodiversity (low, moderate, high) and plant diversity (both species and functional richness) are better than the corresponding models with the interaction term to explain all aspects of temporal stability (i.e., temporal stability, mean, standard deviation, asynchrony, and population variance) (Table S3 and S4). This indicates that soil biodiversity and plant diversity independently influenced stability-related indices (Fig. 3; Table S5 and S6). Soil biodiversity loss had consistent negative effects on the temporal stability of community biomass production and plant species asynchrony along a gradient of plant species richness, and plant species richness was positively associated with the temporal stability of community biomass production and plant species asynchrony (Fig. 3A and B). Soil biodiversity and plant species richness did not affect population variance (Fig. 3D). The temporal mean of community biomass production was positively related to plant species richness, while the standard deviation was negatively correlated with plant species richness (Fig. 3C and E). Soil biodiversity loss did not alter the temporal mean of community biomass production, while statistically significantly increasing the standard deviation (Fig. 3C and E). Asynchrony at both plant species and functional group level was positively related to the temporal stability of community biomass production, while soil biodiversity loss did not exert a significant effect on the asynchrony-stability relationship (Fig. S2). When monocultures were excluded from the analysis, plant species richness did not exert a significant effect on the temporal stability of community biomass production, although the effect of soil biodiversity was still observed (Table S5).
However, functional diversity in terms of functional richness increased the temporal stability of plant communities excluding monocultures (Fig. 3F). Plant species asynchrony, population variance and the temporal mean of community biomass production were increased by functional richness, and decreased by soil biodiversity loss (Fig. 3G, H and I). Soil biodiversity loss and functional richness did not alter the standard deviation of community biomass production (Fig. 3J). Furthermore, functional richness was closely related to plant species richness, and 78% of variance was explained by an additive model, which was much higher than that of other functional diversity indices (R 2 < 40%, Table S6). Plant species richness did not explain functional evenness and the community-weighted mean of all traits (Table S6). Functional divergence and dispersion were increased by plant species richness, but were not correlated with the temporal stability (Table S7).
Multitrophic biodiversity was positively associated with the temporal stability of community biomass production, plant species asynchrony and the temporal mean of community biomass production, and negatively correlated with the standard deviation of community biomass production (Fig. 4A, B, C and E). Multitrophic biodiversity was not related to population variance (Fig. 4D). The linear relationships indicate that multitrophic biodiversity can drive the temporal stability of community biomass production through an increase in either soil biodiversity or plant species richness.
Plant community composition . Plant species loss and community evenness were affected by either soil biodiversity or plant species richness, but were not related to the temporal stability of community biomass production (Table S5). In the monocultures, soil biodiversity did not affect the temporal stability of grasses, while increasing the temporal stability of herbs and legumes (Fig. S3A, B and C). In mixed-species communities, plant species richness increased the temporal stability of grasses, but did not affect the temporal stability of herbs and legumes (Fig. S3D, E and F). Moderate soil biodiversity increased the temporal stability of herbs, while soil biodiversity did not influence the temporal stability of grasses and legumes in mixed-species communities (Fig. S3D, E and F). Soil biodiversity loss increased the proportional abundance of grasses in mixed-species communities, but decreased the proportional abundance of herbs and legumes (Fig. S3G, H and I). The proportional abundance of grasses was positively associated with the temporal stability of community biomass production, and the proportional abundance of herbs and legumes was negatively associated with temporal stability (Fig. S3J, K and L). However, soil biodiversity loss still had negative effects on the temporal stability of community biomass production (Fig. S3J, K and L).
Soil biodiversity loss did not alter the shoot biomass of grasses and herbs during wet and dry periods in the monocultures (Table S8 and Fig. 5). Shoot biomass of legumes was increased by soil biodiversity loss during wet periods, and was not altered during dry periods in the monocultures. In mixed-species communities, the effects of soil biodiversity loss on shoot biomass of plant functional groups were similar along a gradient of plant species richness (Table S8 and Fig. 5). For instance, the loss of soil biodiversity increased the growth of grasses, while strongly decreasing the growth of herbs and legumes, independently of plant species richness. On average, there was at least twice as much shoot biomass of grasses as legumes or herbs at low soil biodiversity. Plant species richness did not affect shoot biomass of plant functional groups during wet periods, but tended to increase their shoot biomass during dry periods.