Experimental design
This experiment was set up containing two levels of soil biodiversity (high and low soil biodiversity) and seven treatments considering the number of global change factors (GCFs) (0, 1, 2, 4, 6, 8, 10) (Table 1, Supplementary Fig. 1 and Supplementary Data 1). We used the dilution-to-extinction approach to create the high and low soil biodiversity treatments (Supplementary Methods). Soil dilution can lead to a gradual loss of rare soil microbes, which can simulate a realistic loss of soil biodiversity, because rare soil microbes are more sensitive to anthropogenic pressures, e.g., warming, nitrogen addition and drought15. We note that the low soil biodiversity treatment is a subset of the high biodiversity, as many rare species have been eliminated through the dilution; this approach will likely lead to relatively more tolerant microbes in the resulting communities.
An increasing number of GCFs was created inspired by the experimental design of the studies on biodiversity-ecosystem function relationships, based on random sampling from a species pool5,6,14. The combination of multiple GCFs was replicated 15 times at each level by randomly selecting GCFs from a pool with 10 GCFs for each replicate (Table 1 and Supplementary Data 1). For each replicate of combined GCFs, there were identical GCF combinations between the high and low soil biodiversity treatments to avoid a confounding effect of GCF combination and soil biodiversity treatments. The pool of 10 GCFs included: warming, nitrogen deposition, drought, heavy metal pollution, plastic mulching film residues, salinity, agricultural fungicide, bactericide application, surfactant contaminant and soil compaction (Supplementary Methods). These GCFs frequently occur in intensively managed agroecosystems and are treated as anthropogenic pressures10,13,14,15.
Microcosms
This experiment was conducted using 50 ml conical Mini Bioreactors (Product Number 431720, Corning Inc., NY) as experimental units (Supplementary Fig. 2). Each Mini Bioreactor has four vents in the cap, where a hydrophobic membrane avoids microbial contamination but allows gas exchange. We filled each Mini Bioreactor with 40.0 g (dry weight, d.w.) of soil in total, which received the appropriate treatments.
Soil sterilization and inoculum preparation
We collected the field soil from the top 10 cm of an intensive farming system in Berlin (52.466°N, 13.303°E). Field soil was passed through a 2 mm mesh to remove large roots and stones. We sterilized 20 kg of soil for 90 min at 121 °C, and stored 2 kg of fresh soil at 4 °C. The dilution-to-extinction approach38,39,40,41,42 was used to create high and low soil biodiversity (Supplementary Fig. 1). A parent inoculum suspension was prepared by mixing 100 g of fresh soil with 200 ml of sterilized VE water. The sediment settled for 1 min. The upper 200 ml of soil suspension was treated as parent inoculum suspension. 50 ml of parent inoculum suspension was added to 500 g of sterilized soil in a plastic bag, and homogenized by turning the bag up and down 30 times to obtain the inoculum of high soil biodiversity. Another 5 ml of parent inoculum suspension was mixed with 45 ml of sterilized parent inoculum suspension to create the 10-1 dilution. This procedure was repeated five times to reach the 10-6 dilution. 50 ml of the 10-6 dilution was mixed and homogenized with 500 g of sterilized soil in a plastic bag to obtain the inoculum of low soil biodiversity. This whole dilution procedure was repeated five times to obtain 10 bags of soil inoculum (five bags for each soil biodiversity inoculum).
Sterile water was added to each plastic bag to reach the water content of the fresh soil in the field. All bags were closed with a sterilized cotton plug and a rubber band to avoid microbial contamination but allow gas exchange42. All bags were incubated in a dark room at 20 °C until similar microbial abundance was observed between the high and low soil biodiversity inoculum. Soil inoculum was homogenized by shaking and turning the bags once a week. After incubation, 2.0 g of soil in each bag was collected and stored at −80 °C for DNA extraction. Quantitative real-time PCR (qPCR) was used to determine fungal and bacterial abundance. In the present study, it took two months to recover soil microbial biomass (Supplementary Fig. 3).
The implementation of GCFs and harvest
Agroecosystems, some of the most intensively managed ecosystems, are affected by the co-occurrence of multiple GCFs13,14,15. This study focused on GCFs that frequently occur in agroecosystems, including warming, nitrogen deposition, drought, heavy metal pollution, plastic film residues, salinity, agricultural fungicide and bactericide application, surfactant contaminant and soil compaction. We present the rationale for the 10 tested GCFs in the Supplementary Methods.
Loading soils were used to achieve an effective mixing of chemical agents into 40.0 g soil in each Mini Bioreactor. We created separate ‘loading soil’ for each GCF with chemical addition by mixing an appropriate dose of a chemical agent with sterilized soil through careful homogenization. This was done to avoid exaggerated effects of more concentrated chemicals when mixed with soil. For each chemical, 1.0 g (d.w.) of loading soil contained an appropriate dose for 40.0 g soil in a Mini Bioreactor. For instance, 1 634 mg of NH4NO3 was mixed with 100 g (d.w.) of sterilized soil, to ensure that there was about 16.34 mg of NH4NO3 in 1.0 g of sterilized loading soil. We weighed 40.0 g (d.w.) of soils, including 1.0 g (d.w.) of each loading soil, 5.0 g (d.w.) of soil inoculum (high or low soil biodiversity), an appropriate amount of film (0 or 0.16 g plastic film) and sterilized soil, according to GCF combination for each experimental unit. We put 40.0 g of mixed soils into a clean and sterilized cup (200 ml) with a cap, and then homogenized it by turning the cup up and down for 5 min using a shaking machine (Heidolph Reax 2, Heidolph Instruments GmbH & CO. KG, Schwabach, Germany). After homogenization, 40.0 g of mixed soils was transferred to a Mini Bioreactor, and a mesh bag containing about 100 mg of dry Medicago lupulina leaves (65 °C for 72 h) was buried 1 cm below the soil surface. We used a stick to press soils in each Mini Bioreactor to simulate an ambient condition (1.3 g cm−3) or mechanical compaction (1.7 g cm−3) in farmland.
For the warming treatment with an increment of 5.0 °C over the ambient temperature (20 °C), we wrapped heating cables (Exo Terra PT-2012; Hagen Deutschland GmbH & Co. KG, Holm, Germany) around the outside of the bioreactors. A set temperature was maintained by temperature controllers (Voltcraft ETC-902; Conrad Electronic SE, Hirschau, Germany), which can switch off and on heating cables depending on the real-time temperature of the outside surface of Mini Bioreactors. Mini Bioreactors were placed in beakers filled with sand to reduce thermal radiation from neighboring units. At the start of the experiment, we added suitable amounts of sterilized water to each experimental unit to reach 60% of water holding capacity for the non-drought treatment and 30% water holding capacity for the drought treatment.
All Mini Bioreactors were incubated at 20.0 °C in the dark for six weeks before the final harvest. Because there was 2.0 g of weight loss on average in each Mini Bioreactor in the first three weeks, we added 2 ml of sterilized water to each Mini Bioreactor on the first day of the fourth week. During the final harvest, soil in each Mini Bioreactor was gently homogenized using a spoon, and then 2.0 g of fresh soil was collected and stored at −80 °C for DNA extraction; 5.0 g was stored at 4 °C for the determination of soil enzyme activity; the leftover was oven-dried at 40 °C for other measurements. DNA of each soil sample was extracted from 250 mg soil, using DNeasy PowerSoil Pro Kit (QIAGEN GmbH, Hilden, Germany), following manufacturer’s instructions. Soil DNA extraction was stored at −80 °C for further analysis.
The measurement of response variables
We measured the following response variables: microbial activity (soil respiration), microbial abundance (bacterial and fungal abundance), nutrient cycling (litter decomposition rate and soil enzyme activity), physical properties (water-stable soil aggregates and soil water repellency), bacterial and fungal biodiversity (richness and microbial network features) (See details in the Supplementary Methods). We measured soil respiration as CO2 concentration (ppm h−1 g−1 soil) in the third and sixth week as an indicator for soil microbial activity. Bacterial and fungal abundance was estimated by qPCR. The proportional loss of litter (Medicago lupulina leaves) dry weight during soil incubation was used as an indication of decomposition rate. We measured the activity of β-glucosidase (cellulose degradation), β-D-celluliosidase (cellulose degradation), N-acetyl-β-glucosaminidase (chitin degradation) and phosphatase (organic phosphorus mineralization) using high throughput microplate assay43,44. A modified protocol by Kemper and Rosenau was used to measure water-stable soil aggregates45. Soil water repellency was measured using the water drop penetration time method46. High throughput sequencing (Illumina MiSeq) was used to measure the taxonomic composition of soil fungal and bacterial communities with the primers fITS7 and ITS4 for fungi and 515F-Y and 806 R for bacteria47,48 (Supplementary Methods).
Statistical analyses
For diversity and community composition analysis, we excluded the samples with less than 1% of the observations of the largest sample in the ASV table. For network analysis, we then removed ASVs with low prevalence, which presented less than 20% of samples across all experimental units to reduce the high percentage of zero counts. A co-occurrence network was constructed based on both fungal and bacterial ASV tables. The PLNnetwork function in the R package PLNmodels was employed to infer the network, using a sparse multivariate Poisson log-normal (PLN) model49. According to the Extended Bayesian Information Criterion (EBIC), the best model was extracted with the function getBestModel. The network was compartmentalized into different modules using the cluster_fast_greedy function in the igraph package and visualized with partial correlations with |ρ| > 0.05. We focused on the response of the relative abundance of modules, also known as clusters, which represent the closely associated microbes, e.g., groups of coexisting or co-evolving microbes27. The relative abundance of modules was calculated by summing relative abundances for individual ASVs in modules. We used the package FUNGuildR50 to taxonomically parse fungal trait information, using the FUNGuild database51.
For each single GCF treatment, we took the average response from the 10 replicates before analysis. To confirm how the effect of soil biodiversity treatment can change along with the increasing number of GCFs, we quantified the effect size of soil biodiversity treatment for each response variable using Hedges’ g (mean and 95% CIs) at each level of the number of GCFs, using the R package effsize52. Hedges’ g is calculated as the mean difference between the high and low soil biodiversity treatments in units of the pooled standard deviation as a paired-samples because there were identical GCFs and GCF combinations for both high and low biodiversity conditions, with the exception of the zero and 10 GCF treatments.
To evaluate how each of the response variables changes along with the number of GCFs, we applied a generalized additive model (GAM)53. GAM is a penalized generalized linear model that fits a nonparametric, nonlinear smooth curve54. The degree of smoothness of model terms is estimated as part of fitting, using the generalized cross validation. We reasonably assume that the curve shapes are different between high and low soil biodiversity treatments. Therefore, we included biodiversity conditions (low/high) as the model intercept and as the “factor smooth” smoothing class, where a smooth function is created for each factor level independently55. For GAM modeling, we used the mgcv package55. The dimension of the basis used to represent the smooth term was set as k = 5 so that the model does not overfit to the data. For this, we compared some other values (from 3 to 8) and confirmed that the results are essentially the same within the tested range. The other parameters were set as default.
The relationships between soil microbial indices and other soil indices were tested using Spearman correlation in the package microbiome, and the adjustment method “fdr” was employed to control the false discovery rate for multiple testing correction56. For the further multivariate integration of soil functions/properties and composition of modules, the DIABLO (Data Integration Analysis for Biomarker discovery using a Latent component method for Omics studies) was employed to detect correlation (Pearson’s correlation |r| > 0.5) among variables using the package mixOmics57.
The Z-scores for each of the eight soil functions (as shown in Fig. 1, with the exception of soil water repellency) were evaluated, and then we computed an improved weighted multifunctionality metric to represent soil multifunctionality (Supplementary Methods)58. Structural equation models (SEMs) were used to reveal the direct and indirect effects of an increasing number of GCFs on soil multifunctionality within each soil biodiversity treatment using the package lavaan59. We assumed that an increasing number of GCFs influences soil multifunctionality by regulating the bacterial and fungal abundance and the relative abundance of modules. All response variables were standardized to the same comparison scale using the z-score transformation before constructing SEMs. Models with optimal fitting indices were reported (Supplementary Fig. 11).
The permutational multivariate analysis (ADONIS) and non-metric multidimensional scaling (NMDS) ordination based on the Bray-Curtis distance were conducted to test the effect of soil biodiversity and GCF treatments on the community composition of bacteria and fungi using the R package vegan60. For the data handling, processing, and visualization, we used the packages tidyverse61, reshaping62, cowplot63, RColorBrewer64, qgraph65, igraph66, factoextra67, phyloseq68 and itsadug69. These data manipulation and analyses were conducted using R version 4.1.370. The R script and data are available in a publicly accessible database71.
Reporting summary
Further information on research design is available in the Nature Research Reporting Summary linked to this article.
Source: Ecology - nature.com
