Soil and vegetables were collected from three gardens (E, G, and O) located in the metro Detroit area. Gardens E and G were located near the city center while garden O was in an outer suburban area. Agricultural practices on the garden plots are provided in Supplementary Document 1. Fifteen soil samples (5 from each garden) and 45 vegetables (21 from E, 5 from G, and 19 from O) were collected. The five soil samples from each garden were spread out in a 20 m × 20 m plot and collected at a depth of 0–7.5 cm. Each sample of 1 kg was pooled by five subsamples collected approximately 10 cm from each other at the center and four corners around the same sampling spot. Vegetables were based on the availability during the collection time and included root vegetables and leafy greens. Samples were collected in clean Ziploc bags and transported using an ice-filled cooler to the lab within 2 h.

Antimicrobial-resistant bacteria were isolated and identified by following a previous protocol (Mafiz et al., 2018). Briefly, 50 g of soil or vegetable sample was mixed with 450 ml of brain heart infusion (BHI) (Difco, Sparks, MD, United States) and homogenized, followed by streaking on R2A agar (Difco) containing 20 μg/mL of ampicillin, streptomycin, or tetracycline. Agar plates were incubated at 30°C for 4 to 5 days and up to five colonies of different morphology were picked for purification. Bacteria DNA was extracted using a boiling method, followed by PCR amplification of the 16S rRNA gene. PCR reactions of 25 μl included 5 μl of template DNA, 0.5 μM of each primer, 1 × PCR buffer, 4.0 mM of MgCl₂, 200 μM of each dNTP, and 1U Taq DNA polymerase (Promega, Madison, WI, United States). PCR products were sent to Eton Bioscience Laboratories, NJ for purification and sequencing. Resulted DNA sequences were analyzed using the Ribosomal Database Project (RDP) website¹.

The Minimum Inhibitory Concentration (MIC) of the isolates was measured using the Gram-positive and Gram-negative Sensititre plates from the Sensititre Antimicrobial Susceptibility System (Trek Diagnostic Systems, Westlake, OH, United States). The resistance breakpoints of E. coli ATCC 25922 and S. aureus ATCC 29213 according to the Clinical and Laboratory Standards Institute (CLSI) were used to interpret the MICs of Gram-negative and Gram-positive bacteria, respectively. A high MIC was defined as the MIC higher than or equal to the resistance breakpoint of the reference strain.

Microbial composition in soil was determined by high-throughput 16S rRNA sequencing. Soil DNA was extracted using the DNeasy PowerSoil kit (Qiagen, Valencia, CA, United States) according to the manufacturer’s instructions. After checking the quality and quantity of the extracted DNA, a 16S rDNA sequencing library was constructed using the 16S Metagenomic Sequencing Library Preparation protocol (Illumina, San Diego, CA, United States) modified by Dr. Karen Jarvis from FDA. Briefly, Omni Klentaq PCR Kit (DNA Polymerase Technology, St. Louis, MO, United States) was used for initial PCR using locus-specific primers to amplify the V1-V3 hyper-variable region of the bacterial 16S rRNA gene (Chakravorty et al., 2007; Lusk et al., 2012; Klindworth et al., 2013; Yarza et al., 2014). The Illumina sequencing adapters and dual-index barcodes were added to the purified PCR products using Nextera XT Index Kit (Illumina) by a limited cycle PCR. Prepared libraries were sequenced on the MiSeq sequencing platform (Illumina) using paired 300-bp reads and MiSeq v3 reagents, following standard Illumina sequencing protocols. Data were analyzed using the Metagenomics workflow to perform a taxonomic classification against the Greengenes database.

The conjugation experiment was conducted to demonstrate the horizontal transfer of tetracycline resistance from soil bacteria to E. coli or Enterococcus. Bacteria of an MIC ≥ 16 μg/ml of tetracycline were selected as donors for conjugation. All donor strains were tetM positive as identified by PCR. E. coli DH5α (amp^r, kan^r) and Enterococcus faecalis JH2-2 (rif^r, fus^r) were used as recipient strains for Gram-negative and Gram-positive isolates, respectively. Conjugation was performed by the filter mating method (Agersø et al., 2006) with modifications. Briefly, overnight culture of the donor strain was grown in BHI (Difco) containing tetracycline (15 μg/ml). The recipients E. coli DH5α and Enterococcus faecalis JH2-2 were grown in BHI (Difco) containing 50 μg/ml of kanamycin and rifampicin, respectively. The mating mixture with a donor-to-recipient ratio of 1:1 was then placed on a 0.45-μm-pore-size filter and incubated on BHI agar (Difco) at 30°C for 2 days. The filter was washed and vortex-mixed in BHI broth. The mating mixture was spread onto BHI agar containing tetracycline (15 μg/mL) and kanamycin (50 μg/ml) for Gram-negative bacteria and tetracycline (15 μg/mL) and rifampicin (50 μg/ml) for Gram-positive bacteria. Transconjugants were confirmed by PCR detection of tetM.

A total of 23 isolates, seven or eight from each garden, based on the diversity of antimicrobial susceptibility profiles, were selected for whole-genome sequencing. Genomic DNA was extracted, followed by the preparation of paired-end libraries using a Nextera DNA Flex Library Prep Kit (Illumina, San Diego, CA, United States). Pooled library was diluted and denatured before being loaded into a MiSeq reagent cartridge version 3 (Illumina, San Diego, CA, United States). Sequencing by synthesis of the paired-end 300 bp reads was conducted using an on-site MiSeq platform (Illumina, San Diego, CA, United States). FastQC.0.11.6² was used to analyze the quality of the read data. Trimmomatic 0.36³ (Bolger et al., 2014) was employed to remove the adaptors and low-quality bases from the reads. The trimmed reads were assembled using the de novo assembler SPAdes 3.11.1⁴ (Bankevich et al., 2012), using default parameters with a broad range of k-mer values (from 21 to 127). Genome assembly quality was assessed by QUAST version 5.0.2 (Supplementary Table 1). Genome size was estimated by KmerGenie (Chikhi and Medvedev, 2014). All statistics were based on contigs of size ≥ 500 bp (Mikheenko et al., 2018). The assembled contigs for each genome were used as queries against the Comprehensive Antibiotic Resistance Database (CARD⁵) to identify ARGs (McArthur et al., 2013).

Chi-square tests were conducted to compare the MIC distribution of isolates between soil and vegetables using SPSS v. 21.0 (IBM SPSS, Chicago, IL, United States). P < 0.05 is statistical significance.

A total of 226 bacteria were isolated, including 54 from soil (13 Gram-positive and 41 Gram-negative) and 172 from vegetables (33 Gram-positive and 139 Gram-negative). Gram-negative bacteria (n = 180) were predominant. Bacteria from vegetables belonged to four phyla, including Proteobacteria (n = 114, 66.28%), Bacteroidetes (n = 32, 18.60%), Firmicutes (n = 18, 10.47%), and Actinobacteria (n = 8, 4.65%) (Figure 1A). Twenty-nine genera were identified, Stenotrophomonas (n = 26, 15.12%), Chryseobacterium (n = 21, 12.21%), Pseudomonas (n = 19, 11.05%), Agrobacterium (n = 10, 5.81%), and Lysinibacillus (n = 9, 5.23%) being most prevalent. The remaining genera combined accounted for 50.58% (n = 87) of total isolates (Figure 1B). Isolates recovered from soil also belonged to the same four phyla, with Proteobacteria being predominant (n = 34, 62.96%), followed by Firmicutes (n = 9, 16.67%), Bacteroidetes (n = 8, 14.81%), and Actinobacteria (n = 3, 5.56%) (Figure 1C). They consisted of 21 genera. Stenotrophomonas (n = 9, 16.67%), Lysinibacillus (n = 7, 12.96%), Pseudomonas (n = 6, 11.11%), Lysobacter (n = 6, 11.11%), and Chryseobacterium (n = 4, 7.41%) were most common. The remaining genera combined made up 40.74% of the isolates (n = 22) (Figure 1D).

The MICs of Gram-negative bacteria were skewed toward higher antimicrobial concentrations, whereas a more even distribution was observed on Gram-positive bacteria (Figures 2, 3). Of 180 Gram-negative bacteria isolated, the majority showed a high MIC to most antibiotics tested regardless of origin (p > 0.05) (Figure 2). For example, bacteria with a high MIC to amoxicillin/clavulanic acid were 63.4 and 69.8% in soil and vegetable isolates, respectively, compared to 97.6 and 90.7% for ceftriaxone, 85.4 and 81.3% for cefoxitin, 65.8 and 70.5% for chloramphenicol, 43.9 and 46.0% for gentamicin, 97.6 and 96.4% for sulfisoxazole, 46.4 and 48.2% for tetracycline, and 51.2 and 56.8% for trimethoprim/sulfamethoxazole. Significant difference was observed between soil and vegetable isolates in the profile for ampicillin, ciprofloxacin, and nalidixic acid (p < 0.05). Isolates with a high MIC to ampicillin were demonstrated by 97.5 versus 82.7% of bacteria from soil and vegetable samples, respectively. The prevalence was 80.5% (soil) and 59.7% (vegetable) for ciprofloxacin, and 58.5% (soil) and 37.4% (vegetables) for nalidixic acid.

The 46 Gram-positive bacteria showed a wide distribution of MICs across the tested concentrations of antibiotics (Figure 3). All Gram-positive bacteria, regardless of soil or vegetable origin, showed an MIC of 128 μg/ml to gentamicin and kanamycin compared to the S. aureus resistance breakpoint of 16 μg/ml. Similarly, all isolates demonstrated a high MIC for penicillin. Bacteria showing a high MIC for chloramphenicol had a prevalence of 15.4 and 6.1% in soil and vegetables, respectively, compared to 38.5 and 24.2% for ciprofloxacin, 38.5 and 30.3% for erythromycin, 15.4 and 27.3% for nitrofurantoin, and 7.7 and 18.2% for quinupristin/dalfopristin. High MIC against linezolid, tetracycline, and vancomycin was observed in 6.1, 12.1, and 6.1% of isolates, respectively, of vegetable origin, whereas no isolates from soil demonstrated an MIC greater than or equal to the S. aureus resistance breakpoint for the corresponding antibiotics.

Of more than 30 phyla identified by 16S rRNA sequencing, Proteobacteria, Actinobacteria, Firmicutes, Bacteroidetes, and Acidobacteria were most prevalent, though the order of prevalence varied by garden (Figure 4). Proteobacteria were most common in garden E and identified in 35.94% of the reads, followed by Actinobacteria (19.41%), Firmicutes (14.44%), Bacteroidetes (7.58%), and Acidobacteria (6.32%). The 16S rRNA sequencing failed to classify 9.03% of the reads to the phylum level in garden E (Figure 4A). The top five phyla identified in Garden G were comprised of 79.92% of the total reads, with Proteobacteria being most prevalent and identified in 38.11% of the reads, followed by Actinobacteria (16.28%), Bacteroidetes (10.58%), Firmicutes (8.75%), and Acidobacteria (6.20%) (Figure 4B). Top five phyla identified in Garden O were Proteobacteria (36.25%), Actinobacteria (18.23%), Bacteroidetes (10.05%), Acidobacteria (8.42%), and Firmicutes (8.24%) (Figure 4C). When comparing the cultured and 16S rRNA sequencing methods, except for Proteobacteria that predominated as revealed by both methods, the ranking of other phyla differed between the two methods.

A total of 55 Gram-negative and 2 Gram-positive bacteria with a high MIC to tetracycline were tested for conjugation. Forty of 57 bacteria (70.2%) were able to make the recipient tetracycline resistant after conjugation (Table 1). The successful transfer of tetracycline resistance was confirmed by tetM presence in all 40 transconjugants. Out of 55 Gram-negative isolates tested, 28 were Stenotrophomonas, followed by 17 Chryseobacterium, five Sphingobacterium, and one each of Dyadobacter, Lysobacter, Pseudomonas, Agrobacterium, and Variovorax. Approximately 72% (20/28) of Stenotrophomonas successfully transferred tetracycline resistance to the recipient at a conjugation frequency of 2.28 × 10^–4 to 3.36 × 10^–3/recipient cell. Conjugation was successful in 58.8% (10/17) of Chryseobacterium and 80% (four of five) of Sphingobacterium. The only two Gram-positive bacteria tested, Microbacterium and Curtobacterium, were positive for conjugation. No distinction was observed in conjugation frequency among gardens or between soil and vegetable isolates. However, more soil isolates from Garden E were positive for conjugation, which is consistent with a higher number of bacteria with high MIC to tetracycline observed in Garden E soils. This could be due to special soil factors in the garden that may select certain antibiotic resistance and will require further research.

A number of ARGs were detected in the sequenced isolates, with multidrug efflux genes being most common (Supplementary Table 2). Acinetobacter calcoaceticus carried genes related to multidrug efflux pump AdeIJK conferring resistance to β-lactams, chloramphenicol, erythromycin, fluoroquinolones, lincosamides, and tetracycline. Multidrug mex efflux genes (mexB, mexF, mexK, and mexW) were detected in Lysobacter gummosus and Pseudomonas atacamensis. SmeDEF is a multidrug efflux pump for resistance to quinolones, tetracyclines, macrolides, chloramphenicol, and novobiocin in Stenotrophomonas maltophilia. The three genes, smeD, smeE, and smeF, were detected in four Stenotrophomonas isolates that also had oqxB, an efflux pump conferring quinolone resistance. High MIC to multiple antibiotics, including ciprofloxacin, was observed in these isolates. emrB is an efflux pump gene targeting fluoroquinolone antibiotics. Both emrB and oqxB were detected in two Pantoea, one of which had high MIC to nalidixic acid whereas the other showed low MIC to all antibiotics, suggesting pan-susceptibility. cat encoding chloramphenicol resistance was detected in three Agrobacterium. Multidrug resistance gene crp was detected in Rahnella and Pantoea, and so was tetracycline resistance gene tet42 identified in two Microbacterium. Resistance genes to additional antibiotics such as β-lactam (cgb-1), landomycin (lnd-4), rifampin (rifampin resistance gene), and vancomycin (vanRO) were also detected but at a lower frequency.

Antimicrobial resistance gene identity did not always agree with antimicrobial susceptibility profiles. For example, despite the presence of crp, emrB, and oqxB, Pantoea sp. (OVA07A) appeared to be pan-susceptible with low MICs to all tested antibiotics. On the other hand, no ARG was detected in Sphingobacterium GVS05A with high MICs to ampicillin, cefoxitin, chloramphenicol, gentamicin, and nalidixic acid. The same situation applied to one Chryseobacterium (OSA05B) and two Lysinibacillus (EVS05B and OSS05C) that demonstrated high MICs to multiple antibiotics but with no ARGs identified. Microbacterium (EST19A) carried a rifampin resistance gene; however, broth microdilution failed to detect a high MIC based on the rifampin resistance breakpoint of S. aureus.