Introduction

The increased use of antibiotics in both humans and livestock has promoted the spread of antibiotic resistance genes (ARGs) worldwide1,2,3. The use of antimicrobials in livestock exerts selection pressure for propagation in animals with antimicrobial resistance (AMR) microbes, which are occasionally human pathogens4. Moreover, findings suggesting the transfer of ARGs between bacteria of human and animal origins have been reported, and horizontal gene transfer plays a greater role than the clonal spread of strains between these niches4,5.

In human medicine, β-lactam and quinolone antibiotics accounted for over 50% and 10% of the prescribed antibiotics, respectively6. Therefore, major β-lactamase families mediated by mobile genetic elements, such as extended-spectrum β-lactamases (ESBLs), plasmid-mediated AmpC cephalosporinases (pAmpCs), and carbapenemases, are particularly concerning. ESBL-positive Escherichia coli has become widespread in community infections, and particularly, the intestinal tracts of animals represent essential reservoirs for ESBL-producing E. coli7. Transferable quinolone resistance genes, such as qnr, are disseminated in the gut of humans and animals and are related to various mobile genetic elements, including plasmids, transposons, genomic islands, and other specific antibiotic resistance determinants8.

Previous studies presenting shared features of resistance genes, plasmids, or clones among animal and human sources have usually compared Enterobacteriaceae isolates9,10,11. In recent years, metagenomic sequencing technology has enabled the analysis of the arrangement of specific genes and mobilomes without cultivating isolates12,13,14, providing an effective tool for investigating ARG transmission mechanisms in natural environments. In a study characterizing interconnections involving human faeces and the environment, resistomes and phylogenetic composition were tightly linked, whereas key resistance genes crossed habitat boundaries and were associated with mobile genetic elements15.

In this study, we investigated the similarity and distribution of ARGs between food animals and humans residing in an urban setting with little direct contact with living livestock. Using bacterial genomes and metagenomes sequenced from the gut microbiomes of healthy individuals, patients with Clostridioides difficile infection (CDI), chickens, swine, and cattle, the abundance, prevalence, and arrangement of ARGs and their mobilome structures were studied. Our results indicated that the distributions of ARGs against β-lactams and fluoroquinolones in the gut microbiome were most similar between CDI patients and chickens, and the blaTEM, blaCTX-M, blaOXA, and qnrS genes exhibited a strong tendency to co-occur between humans and livestock.

Results

Distribution of bacteria and ARGs in humans and livestock

The composition of bacteria and ARGs were compared between human and livestock guts. For humans, we included two groups: 61 healthy individuals and 26 CDI patients, who were expected to show an augmented ARG profile in the gut microbiome. For livestock, 36 cattle, 41 swine, and 31 chickens were included because they produce a large proportion of edible commodities.

Each group exhibited a distinct microbial composition (p = 0 and p = 0, respectively; Wilks’ lambda test; Fig. 1 and Supplementary Table S1). When the samples were clustered using the DMM model, the optimal Laplace approximation value was five, where most of the samples in each group were clustered together (at the top in Fig. 1; Supplementary Fig. S1A,B). Each group presented different dominant genera in their gut: Bifidobacterium, Ruminococcus, and Bacteroides in healthy people; Enterococcus, Lactobacillus, and Bacteroides in CDI patients; Prevotella, Lactobacillus, and Subdoligranulum in swine; Peptostreptococcus, Butyrivibrio, and Treponema in cattle; and Bacteroides, Alistipes, and Barnesiella in chickens.

Figure 1
figure 1

Characterization of the microbiome and antibiotic resistance genes (ARGs) in each group. The two bars at the top of the plot represent the result of Dirichlet multinomial mixture (DMM) clustering based on genus and ARG compositions, respectively. The relative bacterial composition abundance was plotted for a set of genera that ranked in the top three in any group. For the ARG class, the cumulative relative abundance of ARGs in each class was visualized.

Full size image

The ARG distribution also showed different patterns between the groups (Fig. 1). When the samples were clustered by the relative abundance of ARGs, a similar pattern to that based on the microbial composition was observed, and most of the samples in each group were located close together and assigned to the same cluster when the DMM algorithm was applied (at the top in Fig. 1; Supplementary Fig. S1C,D). An average of 0.07% of the reads in each sample were aligned to the ARG sequences. CDI patients exhibited fourfold higher ARG abundances than healthy individuals (3347.9 vs. 792 RPKM in Table 1), while cattle harboured the fewest ARGs, followed by chickens and swine, among the livestock (378.4 vs. 1103.5 vs. 1281.6). Humans were found to have more diverse ARGs than livestock, and chickens harboured the most diverse ARGs and cattle the least among livestock (105 vs. 81; the number of different ARGs; Table 1). However, when the prevalence of existing ARGs was compared, chickens showed the highest prevalence, followed by CDI patients (24.6% vs. 23.0%; the average prevalence of existing ARGs; Table 1).

Table 1 Statistics of antibiotic resistance genes (ARGs) in healthy individuals, patients with Clostridioides difficile infections, swine, cattle, and chickens.
Full size table

Prevalence of β-lactamase and transferable quinolone resistance genes

In humans, the abundance of most ARGs against β-lactams, quinolones, aminoglycosides, MLS, and glycopeptides was significantly higher in patients with CDI than in healthy individuals (p < 0.0005, Student’s t test; Figs. 1 and 2; Supplementary Tables S2 and S3). Among β-lactam resistance genes (Supplementary Table S4), blaCTX-M and blaTEM showed significant differences (0.73 vs. 51.76 and 1.44 vs. 59.50 RPKM for healthy individuals vs. CDI patients; p < 0.0001 for each Student’s t test; Supplementary Tables S2 and S3). More than half of the patients with CDI had these genes (Supplementary Table S3).

Figure 2
figure 2

Abundance and prevalence of β-lactamase and transferable quinolone resistance genes. (A) Relative abundance of β-lactamase and transferable quinolone resistance genes in each sample, measured by reads per kilobase of transcript per million mapped reads (RPKM). (B) Prevalence of β-lactamase and transferable quinolone resistance gene families in humans and livestock. (C) Prevalence of β-lactamase and transferable quinolone resistance gene families in patients and chickens with Clostridioides difficile infection (CDI). The names of highly prevalent ARGs (> 30%) in both groups are enlarged.

Full size image

When the prevalence of β-lactamases and transferable quinolone resistance genes was compared between humans and each group of livestock, seven genes, ampC, blaTEM, blaOXA, blaCTX-M, cfxA, cepA, and qnrS, were prevalent in both CDI patients and chickens (prevalence in both groups > 30%) but were rarely found in cattle or swine (Fig. 2A, Supplementary Table S3).

Among the plasmid-associated β-lactamase genes, blaTEM and blaOXA were prevalent in both humans and livestock (40.2% vs. 32.4% for blaTEM and 21.8% vs. 40.7% for blaOXA; Fig. 2B,C, and Supplementary Table S3). blaTEM was the most prevalent β-lactamase gene in CDI patients and chickens (84.6% and 96.8%, respectively) but not in swine or cattle (8.3% and 4.9%, respectively). blaOXA prevalence was high in CDI patients, chickens, and swine (69.2%, 45.2, and 80.6%, respectively). However, blaSHV was distributed exclusively in the human gut (9.8% in healthy adults and 57.7% in patients with CDI vs. 0% in cattle, swine, and chickens).

Among the ESBL genes, blaCTX-M was widely distributed in CDI patients and chickens (65.4% and 45.2%, respectively) but was not identified in swine or cattle (Fig. 2B,C, and Supplementary Table S3). Most pAmpC genes (blaDHA, blaACT, blaMIR, and blaACC) were exclusively observed in human samples, whereas blaCMY was detected in healthy humans, patients with CDI, and chickens (4.9%, 23.8%, and 12.9%, respectively). The carbapenemase genes blaKPC and blaNDM were not observed in these animals. One blaIMP-containing cattle sample was identified.

In terms of chromosomal β-lactamase genes, cfxA was the most common β-lactamase gene present in anaerobic organisms, found in both human and animal guts, followed by cblA and cepA (Supplementary Tables S2 and S3). These genes exist in the chromosomes of Bacteroides and Prevotella, which are the major constituents of human and livestock guts. AmpC has been frequently detected in CDI patients and chickens, suggesting enrichment of the phylum Proteobacteria, especially Enterobacteriaceae16,17.

Concerning transferable quinolone resistance genes, qnrS was prevalent in CDI patients and chickens but was rarely found in swine and cattle (46.15% and 35.48% of CDI patients and chickens, respectively; 8.33% and 0% of swine and cattle, respectively; Fig. 2 and Supplementary Table S3). Particularly, we found that 16 of the 26 patients with CDI (61.5%) had one or more qnr genes. Fourteen chicken samples (45.2%) contained qnr genes, with qnrS (35.48%) being the most prevalent family. The distribution of qnrS exhibited a statistically significant difference between the different livestock groups (p < 0.01 for swine vs. chickens and cattle vs. chickens; Student’s t test); qnrS was found in only three swine samples (8.3% prevalence) and was not identified in any of the cattle samples.

Genomic analysis of co-occurring bla and qnr genes in humans and livestock

To investigate the genomic sequences and phylogenetic relationships involving bla and qnr genes among different hosts, the complete blaCTX-M, blaOXA, blaTEM, and qnrS genes were obtained through metagenomic assembly (Fig. 3). In the network of ARGs, two sequence homology levels were investigated: identical genes were linked together as a connected component, and homologous (similarity ≥ 95%) ARG clusters were constructed (Fig. 3B,D,F,H). In the trees, ARG sequences obtained from each identical sequence group were compared to identify their phylogenetic relationships (Fig. 3A,C,E,G), revealing interesting patterns in which the qnrS and blaTEM genes belonged to common lineages in humans and livestock. Specifically, qnrS1 and blaTEM were prevalent in both the human and livestock microbiomes.

Figure 3
figure 3

Homology and phylogenetic relationships of blaTEM, qnr, blaCTX-M, and blaOXA genes. Phylogenetic tree and prevalence of (A) blaTEM, (C) qnr, (E) blaCTX-M, and (G) blaOXA genes. The name in black is the reference sequence. The names in other colours represent sequences obtained from specific samples, whose colours indicate the group. The coloured node is a novel sequence similar to the reference sequence. The gene sequence was obtained from the metagenomic assembly. Network of (B) blaTEM, (D) qnr, (F) blaCTX-M, and (H) blaOXA genes. A node represents a sequence found in the given sample. The colour represents the host group. Nodes are connected with a solid line when both sequences are identical. Nodes are clustered with a dotted line when they have a similarity above 95%. Cytoscape (v.3.8, https://cytoscape.org) was used to build network graphs.

Full size image

The blaTEM genes were clustered to construct a very large group of blaTEM-1 genes, which were identical to the known blaTEM-1 gene and mostly from humans and chickens (identical genes are connected with solid edges in Fig. 3B). As a minor group, two genes that were identical to blaTEM-135 were identified in one human and one chicken sample, and there was a gene that was one amino acid different from the blaTEM-1 gene. Homologous blaTEM genes were clustered closely in the phylogenetic tree, as indicated by the distance at the bottom of the tree (Fig. 3A). Among the qnr genes, qnrS was the most prevalent. Most of the qnrS genes were qnrS1, commonly found in humans and chickens (Fig. 3D), and showed a pattern similar to that of blaTEM. In contrast, qnrB was primarily derived from humans.

Co-occurrence of the ARG-carrying mobilome in humans and livestock

The ARG-carrying mobilomes for blaTEM, blaCTX-M, blaOXA, and qnrS were identified and compared between humans and livestock (Fig. 4 and Supplementary Table S5). Similar to the results from the ARG sequence analysis, a dominant mobilome structure for blaTEM and qnrS was commonly found in humans and chickens. In all groups, blaTEM-1 with Tn3 (i.e., Tn3- blaTEM-1 pattern) was the most common and was found in 80% of blaTEM-carrying samples. In particular, 65.7% of blaTEM-carrying humans and 100% of blaTEM-carrying chickens showed the Tn3-blaTEM-1 pattern. The second most prevalent mobilome pattern for blaTEM was IS26- blaTEM, which was found in healthy individuals (15.4%), patients with CDI (36.4%), and chickens (30.0%). For qnrS1, the dominant mobilome pattern was qnrS1-ISkpn19, which was found in 71.4% of qnrS1-carrying humans and 70.0% of qnrS1-carrying chickens (Fig. 4 and Supplementary Table S5). This observation implies that this mobilome pattern is efficient for ARG transmission in humans and chickens (see the next section for details on the qnrS1-ISkpn19 mobilome structure).

Figure 4
figure 4

Dominant mobilomes related to the blaTEM, qnr, blaCTX-M, and blaOXA genes. For each antibiotic resistance gene (ARG), the cell in the heatmap represents ARG abundance measured by (\({\mathrm{log}}\_2\, ({\mathrm{RPKM}}+1)\)). The abundance was measured by read mapping against the ARG sequences. The cell indicates whether the mobilome pattern exists in each sample for each mobilome. The pattern was named using the insertion sequence or transposase name with the ARG. The structure of each pattern is visualized next to its prevalence heatmap.

Full size image

For blaCTX-M-14 transmission, ISEc9-blaCTX-M-14 is a more common mobilome structure than IS26 transposase in humans (i.e., IS26-blaCTX-M-14). For example, 45.5% of blaCTX-M-14 was carried by ISEc9-blaCTX-M-14, and 18.2% was carried by IS26-blaCTX-M-14. In chickens, all blaCTX-M-14 genes were in the structure of IS26-blaCTX-M-14, and ISEc9-blaCTX-M-14 was not found. BlaOXA-1 was the most prevalent in patients with CDI, where 92.9% of CDI patients harbouring blaOXA-1 carried IS26-blaOXA-1-IS26, and three-quarters of the chickens carried it.

High conservation of ARG-carrying mobilomes in bacterial genomes and metagenomes

Because qnrS exhibited a high rate of co-occurrence in humans and livestock (Fig. 4), mobilome structures were further characterized for this ARG using metagenomic contigs, E. coli genomes, and plasmid genomes. From the metagenome contigs, qnrS1 genes were identified along with upstream and downstream genes in ten human and livestock samples. From the E. coli isolates sequenced in this study, qnrS1 was identified in eight samples from humans and livestock. In the bacterial and plasmid genomes obtained from the NCBI repository, qnrS1 genes were found in three E. coli chromosomes and three plasmids of E. coli, Klebsiella pneumoniae, and Shigella flexneri (Fig. 5A and Supplementary Table S6). Notably, most qnrS1-carrying mobilomes were composed of genes with the same pattern, including the qnrS1 gene, a recombinase, the ISKra4 family transposase, a recombinase, and a plasmid-related gene (Fig. 5A). These four neighbouring genes and qnrS1 were well conserved, ranging from 99 to 100% in sequence similarity. Many transposases of the insertion sequence were observed in the regions upstream and downstream of qnrS1 (Supplementary Table S7). Tn3 family transposases were frequently found, and other ARGs, such as blaCTX-M-15, blaTEM-1, lap-2, aph(6)-Id, aph(3″)-Ib, and sul2, were found close to qnrS1.

Figure 5
figure 5

Arrangement of qnrS and its neighbouring genes. (A) QnrS and its adjacent genes were identified from metagenome assemblies, E. coli genome assemblies, and reference genomes in the NCBI repository. (B) Using the E. coli plasmid containing qnrS1 as a reference, the regions mapped by using 15 E. coli isolates resistant to fluoroquinolone are coloured orange. QnrS1, its mobilome, and other ARGs in the plasmid are annotated with coloured rectangles in the centre.

Full size image

Additionally, gene arrangements upstream and downstream of the qnrS-carrying mobilome in E. coli isolates were further investigated using sequencing reads. Using a complete plasmid sequence from sample H25 as a reference, the sequencing reads of 13 E. coli isolates with fluoroquinolone resistance were aligned to confirm that the gene arrangements upstream and downstream of qnrS1-carrying mobilomes were different among the E. coli isolates, even though qnrS1-carrying mobilomes were well conserved (Fig. 5B). Moreover, it was confirmed that nine qnrS1-containing isolates showed a well-conserved mobilome pattern of qnrS1, whereas the other four, which contained qnrS2, did not show the same mobilome pattern. We analysed chromosome multi-locus sequence typing (MLST) types and plasmid incompatibility groups for these E. coli isolates and found that all 13 E. coli isolates bearing the qnrS gene had different sequence types (STs), and 12 STs were previously found (Table 2). No specific association was found between chromosomal phylogenetic groups or plasmid incompatibility groups and the type of qnrS gene.

Table 2 Characterization of QnrS-carrying E. coli isolates and plasmids.
Full size table

Discussion

In modern megacities, the food chain is considered a carrier of resistant bacteria or to enable possible horizontal gene transfer between humans and livestock5,7,8,10,11,18. The number of foodborne outbreaks associated with fresh vegetables has increased, and contamination by animal waste is mostly driven by improperly composted animal manure, water, and soil19. Retail meat, especially poultry meat, serves as a vehicle for human exposure and infection with highly resistant E. coli20. These findings highlight the relevance of antibiotic use in animal food production to antibiotic resistance in humans.

In this study, the similarities between ARGs in humans and livestock were studied, and mobile genetic elements adjacent to resistance genes were compared.

The results revealed that the overall distribution differed between humans and livestock. Interestingly, the PCoA based only on β-lactamase and qnr genes exhibited a different distribution pattern: patients with CDI and chickens were closely related. This suggests that both harbour many ARGs against β-lactams and fluoroquinolones and that their common ARGs are carried by the same types of mobile genetic elements. In the Republic of Korea, penicillin (22.7%) and enrofloxacin (18.2%) are the best-selling antibiotics for chicken breeding, followed by phenicols, ionophores, and tetracyclines21. The high prevalence of blaCTX-M in chicken gut bacteria, despite the low usage of extended-spectrum cephalosporins (ESCs) in breeding, might be because the qnrS, blaTEM, and blaCTX-M genes are frequently located in the same plasmid or in any other mobile element and co-transferred11,22.

blaTEM, blaOXA, blaCTX-M, and qnrS were the most prevalent β-lactamase or transferable quinolone resistance genes in the guts of humans and livestock and were also found in similar genetic contexts. In contrast, several ARGs, such as blaSHV, blaDHA, blaACT, blaMIR, and qnrB, have been observed in the human gut, although some were observed in livestock in other studies9,23,24. The prevalence of these genes in the gut of healthy humans was not as high as that of blaTEM or blaCTX-M, which reside in human and livestock guts. Among pAmpC family members, blaCMY was identified in both patients with CDI and livestock in this study, was prevalent in healthy humans as well, and is also known to be a predominant pAmpC in the animal sector, especially in poultry2. These findings suggest that an enlarged pool of ARGs carried by livestock may contribute to the prevalence of these genes in human pathogens.

Although we selected a cohort of humans and livestock to study the similarity of ARGs and analysed 195 metagenomes from humans and three livestock species in this study, the number of samples might not be sufficient to represent each host species, but we will continuously expand the size of the cohort and perform comparative studies. The sampling and sequencing protocols were carefully designed to maintain compatibility between different groups of samples.

In conclusion, the level and distribution of β-lactamase and qnr genes were the most similar between humans and chickens. Chickens had the highest abundance of β-lactam and fluoroquinolone ARGs, followed by pigs and cattle. blaTEM, blaOXA, blaCTX-M, and qnrS exhibited distinct features of co-occurrence in humans and non-human animals. Additionally, humans, chickens, and swine contain homologous qnrS gene cassettes in different plasmids.

Materials and methods

Study design and description of specimens

A total of 195 metagenomic samples were used: 61 from healthy individuals from Korea, 26 from patients with CDI, 36 from swine, 41 from cattle, and 31 from chickens (Table 3). Healthy individuals were defined as those scoring zero for the Charlson comorbidity index and having no hospital admission history within the past year25. Patients with CDI were selected to represent a human group with high ARG levels in their gut because of current or recent antibiotic usage26. Human samples were collected from Hanyang Medical Centers (HMC) located in Seoul and Guri, which is a satellite city connected to Seoul. Healthy individuals were all sampled from the health examination centre of HMC in Seoul, and 6 CDI patients were hospitalized in HMC in Seoul, but 20 CDI patients were enrolled from HMC in Guri. The individuals in this study had limited direct contact with food animals. The human participants were all from the Seoul area, while the livestock samples were collected from various locations across the Korean Peninsula. Sequencing data from healthy individuals, CDI patients, swine, and cattle were obtained in our previous studies26,27. Chicken faecal samples were collected and sequenced in this study. For the chicken microbiome, 155 faecal samples were collected from 31 farms (5 chickens from each farm) in the Republic of Korea, and five specimens from each farm were pooled for metagenomic sequencing. A total of 13 E. coli isolates from 5 humans, 2 swine, 1 cow, and 5 chickens were selected on MacConkey agar plates containing 0.25 or 0.5 μg/mL ciprofloxacin and were confirmed to have qnrS by qnrS-specific PCR28.

Table 3 Characteristics of gut microbiome samples from humans, cattle, swine, and chickens.
Full size table

DNA preparation

Each faecal sample was thoroughly homogenized using a spatula and divided into 250–300 mg aliquots. Total DNA was extracted using the Fast DNA SPIN Kit for Feces (MP Biomedicals, Santa Ana, CA, USA) following the manufacturer’s instructions.

Genomic DNAs of 13 E. coli isolates were extracted from overnight cultures using the QIAamp DNA Micro Kit (Qiagen, Hilden, Germany) following the manufacturer’s instructions.

Sequencing and processing of reads

All microbiome samples were sequenced on the Illumina HiSeq platform. For each sample, all reads were 151-bp paired-end sequences with an insert size of ~ 350 bp. Quality trimming was performed using Sickle (v.1.33)29 with Phred quality scores > 20 and read length options > 90 bp (pe -q 20 -t sanger -l 90). The reads that aligned to their corresponding host genome using Bowtie2-align (v.2.1.0)30 with the sensitive-local option were removed. The host reference genomes for humans, swine, cattle, and chickens were downloaded from the NCBI repository (accession numbers: GCA_000001405.1, GCF_000003025.6, GCF_000003055.6, and GCF_000002315.6, respectively). After quality trimming, an average of 91.7% of the sequence reads were conserved.

Thirteen E. coli isolates with fluoroquinolone resistance were sequenced using the Illumina HiSeqX platform. One E. coli isolate from a healthy adult, H25, was sequenced on the PacBio platform to obtain high-quality complete genome sequences.

Composition analysis of microbial communities

The microbiota composition was determined using MetaPhlAn 2.031. To minimize noise, genera with a < 0.1% abundance were not used for the analysis. The genus composition was used for the Dirichlet multinomial mixture (DMM) analysis32. This approach constructs a frequency matrix of the given taxa for each sample and builds a probabilistic model. Figure 1 shows the compositions of the top three most abundant genera in each group.

Profiling ARG abundance and prevalence

To profile ARGs, an ARG reference database was established from 2252 sequences in CARD (v.2.0.1)33. To reduce redundant mapping, homologous sequences were clustered using cd-hit (v.4.7) (-c 0.9 -n 8)34, resulting in 848 representative ARG sequences.

To measure ARG abundance, reads were mapped using Bowtie2 (v2.3.4.1) (sensitive-local). Reads per kilobase of transcript per million mapped reads (RPKM) values were used to measure the abundance of ARGs in the given samples. Only reads that met the following criteria were used for RPKM calculations: > 50% of the read length was aligned against an ARG sequence and had a similarity of > 90%. The ARG was considered to be present only when 70% of the length of an ARG was covered by one of the mapped reads. The number of samples containing a specific ARG was counted to measure the prevalence of ARGs in each group.

ARG analysis was conducted using gene, family, or class measures. The ARG family is a cluster of ARGs (e.g., qnrS1, qnrS2, and seven other genes belonging to the qnrS family), and the ARG class is a cluster of ARG families that confer resistance to the same class of antibiotics (e.g., qnrS, qnrB, and qnrD belonging to the fluoroquinolone class).

Identification of complete ARGs and sequence analysis

A three-step procedure was performed to predict the complete ARGs and their neighbouring genes. First, the reads were assembled using MEGAHIT (v.1.1.3) (default option)35. Second, gene prediction was carried out using FragGeneScan (v.1.31) using only contigs > 500 bp36, with the option of no sequencing errors (-w 1 -t complete). Finally, genes were identified as ARGs using BLASTp (v.2.6.0+)37 against the ARG database with an e-value threshold of 1 × 10−10, a similarity > 90%, and a reference coverage > 70%.

All applicable ARGs were visualized for network analysis. Pairwise similarities between the ARGs were measured using BLAST (v2.7.1+). Identical ARGs were clustered together by solid lines. A representative ARG was selected from each cluster if the cluster did not have any annotated ARGs obtained from the CARD database. Phylogenetic analysis was performed using the selected representative ARGs and annotated ARGs from the CARD database. First, multiple sequence alignments were performed on the amino acid sequences of ARGs using Muscle38. Phylogenetic distance was calculated along the branches of the tree using megacc39 with the following options (Jones–Taylor–Thornton (JTT) model; partial deletion = 95%; bootstrap value = 50). The phylogenetic tree was visualized using a heatmap from the ggtree library.

Identification of ARG-carrying mobilomes

Contigs harbouring ARGs and neighbouring genes were analysed to define the patterns of ARG-carrying mobilomes using the NCBI database. Two conditions had to be satisfied to confirm the existence of a mobilome in each microbiome: > 70% of the reference mobilome was mapped by sample reads, and one or more paired-end reads (i.e., conjunction reads) were aligned in both the ARG and the neighbouring gene. Conjunction reads were defined on the basis of the following: (i) > 20% of each read and aligned to either the ARG or its neighbouring gene, (ii) both paired reads with < 10 bp of soft clips, which are the regions at the end of reads and are not aligned to the reference sequence, and (iii) the distance between paired reads < 350 × 3 bp.

Analysis of qnrS gene arrangements and neighbouring genes

To analyse qnrS and neighbouring genes, three different types of genomic sequences were used: (i) contigs assembled from the microbiome, (ii) contigs assembled from E. coli isolates, and (iii) E. coli chromosomes and plasmid sequences obtained from the NCBI repository. A circular coverage map was constructed by mapping the sequencing reads of E. coli isolates to fully assembled sequences of E. coli plasmids containing qnrS genes.

The 13 PMQR gene-carrying E. coli isolates were characterized. The phylogenetic group was determined by identifying the presence of the chuA, yjaA, and TspE4.C2 genes40 using blastn (similarity and coverage ≥ 90%). For the plasmid incompatibility group, the reference locus database was downloaded from PubMLST (https://pubmlst.org/). Plasmids were categorized according to the presence of a specific locus (similarity and coverage ≥ 90%). For example, SD13 was categorized as IncF because it contained the FII_95 and FIB_13 loci. Finally, the sequence type (ST) was annotated using seven ST-related genes (adk, fumC, gyrB, icd, mdh, purA, and recA) downloaded from EnteroBase (https://enterobase.warwick.ac.uk/). The ST number was determined using the seven ST-related genes.

When plotting BRIG, the E. coli plasmid H25, which was sequenced with the PacBio long-read sequencing platform, was used as a reference. The total length of the plasmid genome is 97,230 bp. The samples used as a query were 12 E. coli plasmid isolates sequenced by the Illumina HiSeqX platform. The option used for Blast was 1e−10.

Statistical analyses

Bray‒Curtis dissimilarity was measured when performing the principal coordinate analysis (PCoA) using the cmdscale function in the stats library. The clusters for microbial and ARG families were determined using DMM with the dmn function in the Dirichlet multinomial package. The optimal number of clusters was chosen using the Laplace index. To compare the distributions of ARGs between the two groups, a two-tailed Student’s t test was performed using SciPy. Statistical significance was set at P < 0.05.

Ethics approval and consent to participate

The study was conducted according to the guidelines of the Declaration of Helsinki. The study protocol was approved by the Institutional Review Board of Hanyang University Hospital (HYUH IRB 2017-06-001 and HYUH IRB 2016-05-031), and written informed consent was obtained from all the participants. All methods were carried out in accordance with relevant guidelines and regulations.