The University Hospital Münster (UHM) is a 1,300-bed tertiary care center in Münster, Germany, admitting on average 55,000 patients per year. Bacterial samples were collected from hospitalized patients at the UHM for an 18-month time period from January 2022 to June 2023 as part of the routine hospital surveillance of multi-drug-resistant bacteria (MDRBs) according to national recommendations (Bundesgesundheitsblatt, 2012). Screening specimens were cultivated on chromID ESBL Agar (biomérieux, Marcy-l’Etoile, France) and incubated at 36°C ± 1°C for 18 to 24 h. Clinical specimens were cultivated on Columbia Blood agar with 5% Sheep Blood (BD, Heidelberg, Germany; aerobic incubation at 35°C ± 2°C ambient air for up to 3 days), and MacConkey selective agar for Gram-negative bacteria (MacConkey II Agar, BD, Heidelberg, Germany; aerobic incubation at ambient air 35°C ± 2°C for up to 2 days) if applicable. During the study period, >500 MDRB Enterobacteriaceae isolates were identified.

Species identification was performed by matrix-assisted laser desorption/ionization time-of-flight mass spectrometry (MALDI-TOF/MS) (MALDI-TOF MS, Biotyper^® Sirius one, Bruker, Bremen, Germany) with scores above 2.0.

Antimicrobial susceptibility testing was performed using VITEK^®2 automated system (biomérieux, Marcy-l’Etoile, France). In addition, Etests^® (biomérieux, Marcy-l’Etoile, France) for ceftazidime were performed on Müller-Hinton-Agar (Oxoid, Schwerte, Germany) to determine the minimal inhibitory concentration (MIC). All MICs were interpreted according to the 2023 European Committee on Antimicrobial Susceptibility Testing (EUCAST) clinical breakpoints version 13.1 (EUCAST, 2023).

DNA from bacterial isolates was extracted using the NEB Monarch Purification Kit (New England Biolabs, Ipswich, MA, USA) and sequenced on a PacBio^® Sequel IIe system (Pacific Biosciences, Menlo Park, CA, USA) as described previously (Effelsberg et al., 2021). A de novo assembly approach was used for the raw reads and analyzed using the SMRT^® Link software suite v. 10 or v. 11 with default parameters.

The phylogenetic tree was generated based on WGS data by the Genome BLAST Distance Phylogeny (GBDP) approach using the Type Strain Genome Server (Meier-Kolthoff and Göker, 2019). For genotyping based on multilocus sequence typing (MLST) and core genome MLST (cgMLST) and to determine the clonal relationship, the respective schemes in Ridom SeqSphere⁺ software version 9.0.9 (Ridom GmbH, Münster, Germany) were used (Jünemann et al., 2013). Isolates were rated as clonal when the distance of their cgMLST allelic profiles was ≤5 alleles. Antimicrobial resistance genes were determined using the NCBI AMRFinderPlus (Feldgarden et al., 2019) implemented in SeqSphere⁺. If duplicated AMR ESBL genes were identified, a more detailed analysis of the respective contigs was performed. Contigs were predicted as plasmids using MOB-Suite v3.1.4 (Robertson and Nash, 2018) and further characterized by pMLST Server 2.0 (Carattoli et al., 2014). Plasmids were completely annotated using DFAST v1.6.0 (including additional scans against TIGRFAM and COG database by NCBI) (Tanizawa et al., 2018). Next, it was searched for genetically matching plasmids with the same ESBL gene but without a gene duplication in the dataset by a mash-based approach (Ondov et al., 2016) implemented in SeqSphere⁺. Isolates containing these plasmids were used as references and were analyzed in the same way.

Progressive Mauve (Darling et al., 2004) was used to compare and visualize the similarity of plasmid sequences and to localize the gene duplication site of the transposable regions. In addition, all mobile genetic elements were annotated using ISFinder (Siguier et al., 2006) and MobileElementFinder with the default parameters (Johansson et al., 2021). Plasmid sequences were manually curated using SnapGene^® Viewer (Dotmatics; www.snapgene.com).

We identified four isolates (three E. coli and one K. pneumoniae) containing more than one plasmid-based ESBL gene copy within the isolates found by our routine molecular surveillance. For comparison, we matched these four isolates with more than one plasmid-based ESBL gene to four additional isolates (two E. coli and two K. pneumoniae) in the dataset containing genetically the same plasmids with only a single ESBL gene copy ( Supplementary Figures 1–4 ). These isolate matches are named pairs I–IV and originate from different or the same patient. Details of the isolates analyzed and the patients’ characteristics are listed in Table 1 . In total, bacterial isolates of six patients were included in the study: four female patients and two male patients. The median age of patients was 63 (32–94) years. According to their medical history based on the medical reports, all patients suffered from chronic or severe infections ( Table 1 ).

The phylogenetic relation of all isolates was calculated based on the WGS data to illustrate the phylogenetic similarity between the isolates in pairs III and IV and the species difference in pair II ( Figure 1A ). In case the matched isolates were the same species, cgMLST typing was performed to determine whether the pairs only share a similar plasmid or in addition have a clonal relation (on the chromosomal level). The isolates of pairs I and III did not show clonality in cgMLST analysis among the E. coli isolates ( Figure 1B ). Pair IV was identified as a clonal lineage of K. pneumoniae isolates found in one patient (patient 6). Here, the isolates were collected 3 weeks apart in different sample materials ( Table 1 ). The similar plasmids of pair II were shared between an E. coli isolate and a K. pneumoniae isolate that originated from the same specimen of patient 3 ( Table 1 ). There was no clonal relationship due to the two different bacterial species.

Using the MOB-suite software and AMRFinderPlus tool, we analyzed the plasmids harboring ESBL genes with respect to their sizes, replication and relaxase types, anticipated/predicted mobility, and encoded AMR genes ( Table 2 ). Overall, the replication type, the relaxase type, and the predicted mobility of the plasmids were identical within a pair. However, due to the ESBL gene duplication, the plasmid size varied. In addition to the duplicated ESBL-encoding genes, we only found an analog AMR duplication of qnrS1 in p1_Ec1.2 mediating quinolone resistance. Of note, the similar plasmids p2_Ec2.1 and p2_Kp2.2 are most likely directly related and caused by an intra-host transmission via horizontal gene transfer (HGT) from one species to another.

For an in-depth analysis of the plasmids, we annotated the whole plasmids with DFAST and used ISFinder and MobileElementFinder ( Supplementary Figures 1–4 ). All duplicated ESBL genes were part of a longer transposable DNA sequence, which was duplicated in full length ( Figure 2 ; Supplementary Table 1 ).

In pair I, the duplication occurred in p1_Ec1.2 as a tandem duplication with flanking IS26 elements ( Figure 2 ), notably with three IS26 elements in comparison to p3_Ec1.1 instead of two flanking IS26 elements for each duplication. The transposon was classified as a composite transposon and harbors the ESBL gene bla _CTX-M-15 as well as the quinolone resistance gene qnrS1. In addition to AMR genes, a WbuC family cupin fold metalloprotein and Tn3 transposase were encoded.

In p2_Ec2.1, the reference plasmid of pair II, we already detected an IS26 tandem duplication of a transposon harboring the ESBL gene bla _SHV-12. However, there was a difference in the sequence length of one of the transposons caused by an incomplete IS26 element in the second repeat ( Figure 2 ) and a deletion in the ESBL gene (bla _SHV-12/bla _SHV-family). A BLAST analysis revealed a deletion of three nucleotides in the ESBL gene coding originally for leucine 53 in bla _SHV-12. In addition, the transposon contains genes for a PAS domain-containing methyl-accepting chemotaxis protein and for the DNA-binding transcriptional repressor YgbI. While the reference plasmid already shows a duplication, we found a tandem triplication of the transposon in p2_Kp2.2. The coded genes remain the same as in the reference p2_Ec2.1. Nevertheless, one of the three bla _SHV genes shows a nucleotide deletion as well. The deletion corresponding to threonine 18 in bla _SHV-12 results in a frame shift and an early stop codon, thereby shortening the potentially encoded β-lactamase.

In pair III, we observed another duplication pattern. The ESBL target gene bla _CTX-M-14 in p2_Ec3.2 and the reference plasmid p2_Ec3.1 was carried in an ISEcp1-mediated unit transposon. ISEcp1 could be aligned with 100% and 99.94% sequence identities. In addition to the β-lactamase gene, only potential gene residues were transferred within the transposon. The comparison revealed that the point of insertion of the second repeat of the unit transposon in p2_Ec3.2 is the traU gene ( Figure 2 ).

The plasmid p1_Kp4.2 belongs to pair IV. A tandem duplication of IS26-mediated composite transposons could be detected. The one nucleotide difference in the first repeat in p1_Kp4.2 was caused by an insertion in the already truncated oligosaccharide MFS transporter gene as indicated in Figure 2 . The transposon contains the ESBL target gene bla _SHV-2 and a gene for an AAA family ATPase.

Taken together, IS26-mediated duplications were found to be tandem duplications, where one IS26 element is shared between two transposons. This mechanism was identified in three out of four described cases. An alternative mechanism was observed in pair II. Here, a unit transposon was duplicated in an ISEcp1-mediated mode.

To investigate the correlation between the phenotypic antimicrobial susceptibility and the presence of AMR resistance genes, we manually curated the VITEK^®2 data and used the NCBI AMRFinderPlus to identify matching resistance genes. Furthermore, we used Etests for ceftazidime as a reference substance to determine the cephalosporin resistance more specifically and compare it with the number of ESBL gene copies in the AMR plasmids. In total, AMRFinderPlus annotated 37 different AMR genes within the eight WGS datasets with four to 13 different genes per isolate; including duplications, six to 25 AMR-encoding genes were detected. The lowest numbers of resistance genes were found in Ec1.2 with the duplication of the ESBL gene bla _CTX-M-15 and the qnrS1, and cya_S352T and glpT_E448K. The broadest resistance spectrum showed the three K. pneumoniae isolates, with the highest number of AMR genes in Kp4.2. A complete overview of the resistome of the eight analyzed bacterial isolates, categorized into chromosome and plasmid-encoded genes, can be found in the Supplement ( Supplementary Table 2 ).

The phenotypic resistance pattern of the bacterial isolates was analyzed by comparing up to 12 different antibiotic substances ( Supplementary Table 3 ). All isolates were susceptible to carbapenems, and no carbapenemases were detected in the genomic data. In contrast, all isolates were resistant to ampicillin in line with the expression of at least one β-lactamase also found in the genotypic annotation ( Supplementary Table 2 ). To determine the MIC of ceftazidime, we performed Etests^®. The MICs within each pair were found to be more than two times higher for the isolates with the ESBL gene duplications compared to the references ( Figure 3 ). The elevated resistance to ceftazidime correlates in all pairs with the higher number of the ESBL target genes.

All presented cases show gene duplications within the plasmid. The association between plasmids, AMR genes, and a high number of plasmid-borne AMR genes is well known (Vrancianu et al., 2020). Despite the metabolic burden required to maintain a plasmid, most studies could demonstrate that the benefit of the encoded resistance genes is much higher for bacteria or is compensated by additional mutations (Loftie-Eaton et al., 2017; San Millan and MacLean, 2017). This seems plausible for the described cases, as all samples were collected from patients who frequently receive antibiotic treatment. Hence, all isolates have experienced selection pressure.

Interestingly, our study observed highly similar plasmids in different bacterial isolates that showed no genetic relation based on cgMLST (pairs I and III). This could be an example illustrating how plasmids are shared within a community of bacteria with different genetic backgrounds. This HGT is an important mechanism to exchange genetic information known for decades (Thomas and Nielsen, 2005) and well-studied in the adaptive evolution of bacteria (Arnold et al., 2022; Dimitriu, 2022). For the E. coli Ec2.1 and the K. pneumoniae Kp2.2 (pair II), we assume the HGT of the plasmid harboring the ESBL gene because the species borders have to be crossed. However, it is not possible to determine clearly the chronological sequence of the HGT and the duplication event. On note, even if we discuss predominantly the duplication event, the relation between the compared isolates could also be a loss of a transposable region except in pair IV, where the timeline is clear. Kp4.2 was isolated 3 weeks after Kp4.1 from the same patient.

IS-mediated transposition is often observed with accessory genes including AMR genes. All duplication events investigated used this mode of action but with either an IS26 or an ISEcp1 transposition. IS26 transposition events that could arrange transposons containing AMR genes with only three remaining IS26 elements were described before including the mechanistic details (He et al., 2015; Harmer and Hall, 2019; Harmer and Hall, 2021). Our data indicate that tandem duplication of the same transposon is also a quite common transposition event as observed in three out of four cases, i.e., in pairs I, III, and IV. In accordance with this observation, a few cases of IS26-mediated ESBL gene duplications were described before (Garza-Ramos et al., 2009; Tamamura-Andoh et al., 2021; van Almsick et al., 2022), and also the association of IS26 and bla _CTX-M genes was already described in Enterobacteriaceae in a systematic study (Shropshire et al., 2022). Interestingly, neither documented IS26 transposon with ESBL genes nor a high sequence similarity could be found in the databases of TnCentral (Ross et al., 2021) or Transposon Registry (Tansirichaiya et al., 2019) yet. In the fourth analyzed case (pair II), the transposition was ISEcp1 mediated. The detected unit transposon contains a bla _CTX-M-14 gene, which also was found in the E. coli isolates of Shropshire et al (Shropshire et al., 2022).

Gene duplication is a common mechanism in evolutionary biology to develop new traits and adaptations (Innan and Kondrashov, 2010; Vosseberg et al., 2021). A new variant to improve resistance by mutation might be observed in pair II annotated as bla _SHV genes in p2_Ec2.1 and p2_Kp2.2 ( Figure 2 ). Both investigated plasmids in this pair have different deletions of nucleotides within one gene copy of the ancestor bla _SHV-12 gene. However, it is unclear whether these mutated genes code for a functional and optimized β-lactamase. Schuster et al. found similar results; they identified a novel bla _CTX-M gene in this context (Schuster et al., 2022). In that study, the encoded β-lactamase only differed by one amino acid from the ancestor gene that was found in a similar gene array as the new gene. In accordance with this, the evolution of bla _SHV genes from bla _TEM with the high sequence similarities underlines this as a plausible mechanism for the generation of novel and better-adapted β-lactamases (Liakopoulos et al., 2016).

In addition, we hypothesize that the increased resistance level to ceftazidime correlates with the number of ESBL gene copies encoded on the plasmid. Only very few reported cases of ESBL gene duplications found this correlation between genotypic and phenotypic resistance so far (Tamamura-Andoh et al., 2021; van Almsick et al., 2022). However, in accordance, this observation has been described before for other AMR genes (Coppi et al., 2020; Han et al., 2021). Nevertheless, increased antibiotic resistance has also been linked to point mutations in resistance genes (Finci et al., 2022). To gain more insights into the genetic resistance patterns of Enterobacteriaceae, more genetic analyses based on studies with higher sample sizes are required.

One limitation of our study is the relatively small number of isolates included in the study. During the 18-month period, four cases of ESBL gene duplications were identified. While this result indicates that ESBL gene duplication is a possible mechanism for developing AMR in the clinical context, higher sample sizes are needed to estimate how frequently this phenomenon happens. Moreover, the comparison of different species and different clonal lineages with variable resistomes has an influence on MIC determination (Girlich et al., 2021). To consider these aspects, we listed additional β-lactamases encoded on other plasmids or integrated into the chromosome for all isolates ( Supplementary Table 1 ). Furthermore, the plasmid copy numbers and other regulatory mechanisms were not covered and are out of the scope of this study (Nelson et al., 2017).