Soil Microcosm Experiment

Standard chemicals for three herbicides (glyphosate, glufosinate, and dicamba) and antibiotics were acquired from Sigma-Aldrich (Germany) with a purity higher than 99.0%. Soil samples were collected from a representative vegetable cultivation base in Fujian Agriculture and Forestry University (N26°05′, E119°14′). All samples (from five fields) were collected at a depth of 0 to 15 cm, sieved (2 mm) to remove stones and debris and placed in an artificial climate room (25°C) for two weeks for stabilization before the experiments. The detailed physicochemical soil properties were measured as described previously (Yang et al. 2018) and are summarized in Supplementary table S7.

The prestabilized soil samples were homogenized and mixed, after 150 g of mixed dry soil was transferred into individual glass bottles (150 ml volume). Soil microcosms were then exposed to three herbicide treatments (a final concentration of 10 mg/kg of glyphosate, glufosinate, and dicamba dissolved in Milli-Q water [18 MΩ·cm; Millipore, Billerica, MA, USA]), while control treatment received only the solvent (Milli-Q water). Herbicide concentrations were chosen based on residual quantity of these herbicides that typically vary between 10.19 and 33.03 mg/kg of soil(Voos et al. 1994; Dennis et al. 2018). The soil water content was adjusted to 60% of the maximum water holding capacity using sterile deionized water after all the microcosms were covered with caps equipped with breathable film. The microcosms were then transferred into a climatic chamber, arranged randomly and incubated at 30°C in the dark. Soil water contents were maintained by adding deionized water at every 2 days. Each treatment was replicated 12 times and one replicate per treatment was destructively sampled at the beginning (day 0) and 15, 30, and 60 days after the start of the experiment (chosen randomly). Destructive sampling was chosen to avoid disrupting the soil and microbiome structure in contrast to repeated sampling, which would have required throughout homogenization of the soil. All samples were stored at −20°C to later quantify residual herbicides and to extract total DNA for bacterial community analyses as described below.

We used previously developed methods to analyze glyphosate, glufosinate, and dicamba residues at 15, 30, and 60 d during the soil microcosm experiment (Druart et al. 2011; Valle et al. 2019). After thorough homogenization of microcosm contents, independent soil samples (1.0 g) were mixed with Millipore water (10 ml), sonicated for 1 h, shaken at 25°C for 1 h and finally centrifuged at 5,000×g for 15 min. The supernatants were filtered (0.45 µm) to remove solid particles and evaporated in the freeze dryer. The residues of herbicides were dissolved in 0.5 ml of methanol (1%) for further following analysis. The concentrations of glyphosate and glufosinate were detected as described previously using precolumn derivatization followed by high-performance liquid chromatography (HPLC) on C18 column (Druart et al. 2011). The concentrations of dicamba was measured directly using HPLC with UV detector. The average recoveries (n = 3) for all the herbicides ranged between 80.5% and 94.3% and a more detailed description of the analytical methods is included in Supplementary materials (Supplementary Text 1).

Total genomic DNA of soil samples was extracted using a Fast DNA spin kit (MP Biomedicals, Cleveland, OH, USA) following manufacturer’s instructions. The changes in bacterial community composition and diversity were determined using 16S rRNA gene amplicon sequencing using Illumina NovaSeq 6000 PE250 platform (Guangdong Magigene Biotechnology Co. Ltd, Guangzhou, China). The universalprimers 515F (5′-GTGCCAGCMGCCGCGGTAA-3′) and 907R (5′-CCGTCAATTCMTTTRAGTTT-3′) targeting the V4–V5 region of 16S rRNA gene were used. The raw 16S rRNA gene sequences were processed using QIIME 2 as follows (version 2019.7) (Bolyen et al. 2019). Raw reads were first quality-filtered (i.e., filtered, dereplicated, denoised, merged, and assessed for chimaeras) to produce amplicon sequence variants (ASVs) using DADA2 pipeline available in QIIME2 (Callahan et al. 2016). The ASVs observed at low frequencies (less than two) were removed and remaining ASVs were classified using the QIIME2 naive Bayes classifier trained on 99% operational taxonomic units from the SILVA rRNA database (v 132) (Quast et al. 2012).

To analyze changes in ARG (resistome) and MGE (mobilome) abundances during the soil microcosm experiment, soil samples collected at 0 d (n = 3), 30 d (n = 12), and 60 d (n = 12) time points were subjected to metagenomic sequencing. Total DNA samples were sent to Guangdong Magigene Biotechnology Co. Ltd for library construction and shotgun metagenomic sequencing on HiSeq platform (Illumina) with 150 bp paired-end sequencing. The raw sequencing reads for each sample were independently processed for quality control using the Trimmomatic (version 0.39) (Bolger et al. 2014). Changes in resistome and mobilome were analyzed using local ARGs-OAP (v2.2) using clean reads (Yin et al. 2018). To compare the relative ARG abundances in different samples, normalization against the total cell numbers in each sample (copies/cell) was performed. The details and data analysis are shown in Supplementary materials (Supplementary Text 2). Briefly, extracted DNA from 27 samples resulted in 15 billion paired-end reads, comprising 235 Gb of DNA. The sequencing yielded an average of 58 million paired-end reads per pooled sample (Supplementary table S8).

To verify changes in the ARG and MGE abundances based on metagenomic analysis, quantitative PCR was used to measure the abundance of a total of 27 ARG subtypes conferring resistance to four common antibiotics (tetracycline, sulfonamide, aminoglycoside, and macrolide) and five MGEs including two integrases (intI1, intI2), two plasmids (ISCR1, IncQ), and one transposon (Tn916/1545, abbreviated Tn916). The total bacterial abundances in soil samples were determined using 16S rRNA gene (341F/515R). The quantification was carried out on a Light Cycler 96 system (Roche, Mannheim, Germany) as described previously (Liao et al. 2018) and all the details of the qPCR assay for all target genes (primers, annealing temperatures, reaction conditions, and amplification cycles) and results are listed in Supplementarymaterials (Supplementary Text 3).

Isolation of culturable antibiotic-resistant bacteria was performed at the mid-point (30 d) of soil microcosm experiment as described previously (Liao et al. 2019). Briefly, 1 g of soil sample was suspended in 9 ml of phosphate-buffered solution by shaking at 200 rpm for 30 min. Samples were then serially diluted in phosphate-buffered solution and plated on LB (10 g tryptone, 5 g yeast extract, 10 g sodium chloride, and 15 g of agar in 1 l Milli-Q water; pH: 7.4) agar supplemented with one of the four antibiotics at the following concentrations: amoxicillin (32 mg/l), chloroamphenicol (16 mg/l), erythromycin (10 mg/l), and tetracycline (16 mg/l). The numbers of colony forming units (CFU) were counted after 24–48 h of incubation at 37°C using dilutions that resulted in 20–200 CFUs per plate. A total of 115 culturable antibiotic-resistant colonies (20–30 strains resistant to each antibiotic) were randomly isolated and preserved as clonal monocultures at −80°C in 50% glycerol. To identify the taxa of isolated bacterial colonies, we extracted genomic DNA from all isolates using a Bacteria DNA Kit (Tiangen, Beijing, China) following manufacturer’s instructions. The 16S ribosomal RNA (rRNA) gene was then amplified by PCR using universal primers 27F (5-AGAGTTTGATCCTGGCTCAG-3′) and 1492R (5-GGCTACCTTGTTACGACTT-3′) and PCR products sequenced by Shanghai Shengong Biotechnology Co. (Shanghai, China).