Sharing of carbapenemase-encoding plasmids between Enterobacteriaceae in UK sewage uncovered by MinION sequencing

Dissemination of carbapenem resistance among pathogenic Gram-negative bacteria is a looming medical emergency. Efficient spread of resistance within and between bacterial species is facilitated by mobile genetic elements. We hypothesized that wastewater contributes to the dissemination of carbapenemase-producing Enterobacteriaceae (CPE), and studied this through a cross-sectional observational study of wastewater in the East of England. We isolated clinically relevant species of CPE in untreated and treated wastewater, confirming that waste treatment does not prevent release of CPE into the environment. We observed that CPE-positive plants were restricted to those in direct receipt of hospital waste, suggesting that hospital effluent may play a role in disseminating carbapenem resistance. We postulated that plasmids carrying carbapenemase genes were exchanged between bacterial hosts in sewage, and used short-read (Illumina) and long-read (MinION) technologies to characterize plasmids encoding resistance to antimicrobials and heavy metals. We demonstrated that different CPE species (Enterobacter kobei and Raoultella ornithinolytica) isolated from wastewater from the same treatment plant shared two plasmids of 63 and 280 kb. The former plasmid conferred resistance to carbapenems (blaOXA-48), and the latter to numerous drug classes and heavy metals. We also report the complete genome sequence for Enterobacter kobei. Small, portable sequencing instruments such as the MinION have the potential to improve the quality of information gathered on antimicrobial resistance in the environment.


INTRODUCTION
The global rise of carbapenemase-producing Enterobacteriaceae (CPE) over the last decade represents a major threat to public health [1][2][3]. CPE are often resistant to several additional classes of antibiotics, which may limit therapeutic options to drugs with a higher toxicity profile. Unsurprisingly, invasive infections caused by CPE are associated with poor clinical outcomes and carry an excess attributable mortality compared with those caused by carbapenemsusceptible isolates [4][5][6]. CPE are predominantly isolated in healthcare settings, but their spread to healthy humans, livestock and the environment has been reported [4,7]. The development of interventions that prevent further CPE dissemination requires delineation of their reservoirs and routes of spread, combined with accurate characterization of the mobile elements that transfer carbapenemase genes within and between bacterial species.
High-throughput whole-genome sequencing (WGS) has provided major insights into pathogen transmission and outbreak investigation based on core-genome phylogenetic analyses [8], but has not reached its potential to provide detailed characterization of plasmids and their transmission within specific environments. This is due to a methodological limitation of short-read data generated by Illumina technology, from which plasmids cannot be accurately assembled because of the presence of numerous repetitive regions. Technologies including the Pacific BioSciences RSII system (PacBio) and the Oxford Nanopore Technologies MinION system generate long-read data that can overcome this barrier. It has been demonstrated previously that PacBio long-read sequencing can be successfully used to elucidate transfer of carbapenem-resistancecarrying plasmids within hospital environments [9,10], and after years of on-going development the pocket-sized MinION is approaching a level of accuracy and usability at which it could be used to explore the central question of plasmid transmission. It has been used previously to detect the presence of antimicrobial-resistance genes and to track the transmission of viral outbreaks [11][12][13][14].
Our aim here was to apply the MinION and Illumina platforms to seek evidence for sharing of carbapenemase-encoding plasmids between bacterial species in human sewage. This reservoir contains a diversity of pathogenic and non-pathogenic bacteria that can exchange DNA (including genes encoding antimicrobial resistance) by horizontal gene transfer. CPE have been isolated from human sewage at wastewater treatment plants in Austria, Germany and Brazil, and from rivers and lakes in Switzerland, Portugal, Brazil and Vietnam [15][16][17][18][19][20][21][22], and carbapenemase-encoding plasmids have been found in wastewater in Germany [21], but definitive evidence of plasmid sharing and putative transmission is lacking.

METHODS
Wastewater processing and bacterial identification Samples of treated and untreated wastewater were obtained from each of 20 treatment plants. At each sampling point, two consecutive grab samples of 0.5 l each were collected and mixed in 1 l sterile bottles containing 18 mg sodium thiosulphate (Sigma-Aldrich). A single 1 l wastewater sample was obtained from the main septic tank at the Cambridge University Hospitals NHS Foundation Trust (CUH) facility. All samples were transported to the laboratory on ice packs in the dark and processed within 12 h. One millilitre of triplicate serial tenfold dilutions, 10 ml untreated wastewater samples and 100 ml treated wastewater samples were concentrated using the filtration technique onto 0.45 µm pore size filter membranes (S-Pak; Merck Millipore). Membranes were then placed onto the surface of ESBL Brilliance agar (Oxoid) and incubated for 24 h at 37 C in air. Duplicate membranes of the lowest dilution tested were also placed into tryptic soy broth with 2 mg imipenem l À1 and incubated for 24 h at 37 C in air, followed by subculture of 100 µl on chromID CARBA SMART plates (bioM erieux), and ESBL Brilliance and cystine lactose electrolyte deficient (CLED) agar with a 10 µg imipenem disc.
At least one colony for each bacterial colony morphology type suspected to be Enterobacteriaceae based on colour were picked and speciated using matrix-assisted laser desorption/ionization-time of flight mass spectrometry (MALDI-TOF MS; Biotyper version 3.1; Bruker Daltonics). Antimicrobial-susceptibility testing was determined using the N206 card on the Vitek 2 instrument (bioM erieux) calibrated against European Committee on Antimicrobial Susceptibility Testing (EUCAST) breakpoints. Minimum inhibitory concentrations (MICs) of meropenem, ertapenem, imipenem and colistin were determined using Etest (bioM erieux) for any isolate with reduced susceptibility to carbapenems on

IMPACT STATEMENT
The spread of carbapenem resistance among pathogenic Gram-negative bacilli is a major public-health threat. Studies to date have emphasized this problem in the clinical setting, largely ignoring the importance of environmental contamination and the significance of a One Health approach. The use of Illumina whole-genome sequencing to identify the genetic basis for resistance led to the discovery that several species from the same treatment plant carried the same carbapenem-resistance gene. Using the Oxford Nanopore MinION, we demonstrated putative sharing of several plasmids between different Enterobacteriaceae species (Enterobacter kobei and Raoultella ornithinolytica) in wastewater, with one plasmid carrying a carbapenemase gene conferring resistance to the carbapenem drugs, and the other containing resistance genes to numerous other drugs and heavy metals. To our knowledge, this is the first report of carbapenemase-producing Enterobacteriaceae being recovered from wastewater at UK sewage treatment plants. Importantly, we show that treatment plants do not prevent the release of these bacteria into the environment and this may provide a point of intervention. We also demonstrate the utility of a nextgeneration sequencing instrument, the MinION, to confirm the sharing of plasmids between bacterial isolates in wastewater, and have generated a complete genome sequence for Enterobacter kobei. The ease of use of this instrument would facilitate its implementation in future intervention studies that target environmental contamination with carbapenem-resistance genes. Vitek 2. Enterobacteriaceae with confirmed resistance to any carbapenem underwent further phenotypic testing using the KPC/MBL and OXA-48 Confirm kit (Rosco Diagnostica) to detect production of KPC, MBL and OXA-48 carbapenemases. All CPE (n=16) identified were de-duplicated for species and type of carbapenemase according to sample location, and further analysed using WGS (n=9).

IP
Illumina sequencing and bioinformatic analyses DNA extraction and library preparation were performed as previously described [23]. DNA libraries were sequenced using the Illumina HiSeq and MiSeq platforms (Illumina) to generate 100 and 150 bp paired-end reads, respectively. De novo multi-contig draft assemblies were generated using Velvet Optimiser and Velvet [24]. Contigs smaller than 300 bases were removed, the scaffolding software SSPACE was employed, and assemblies further improved using GapFiller. For isolates belonging to the Enterobacter cloacae complex, species identification was based on analysis of hsp60 and rpoB, as described elsewhere [25]. To detect acquired genes encoding antimicrobial resistance, a manually curated version of the ResFinder database (compiled in 2012) [26] was used. Assembled sequences were compared to this as described previously [27], and genes with 100 % match to length and >90 % identity match were classified as present and variants identified. To establish genetic relatedness between Raoultella ornithinolytica isolates, Illumina sequence reads were mapped to the reference genome B6 (GenBank accession no. CP004142) using SMALT v0.7.4 to identify single-nucleotide polymorphisms (SNPs). SNPs were filtered to remove those at sites with a SNP quality score below 30, and SNPs at sites with heterogeneous mappings were filtered out if the SNP was present in less than 75 % of reads at that site. Sequence data have been deposited in the European Nucleotide Archive (www.ebi.ac.uk/ena) under the individual accession numbers given in Table 1.

MinION sequencing and bioinformatic analyses
Isolates were streaked from frozen stock onto blood agar plates and incubated in air at 37 C overnight. A sweep of colonies was taken and DNA extraction carried out using the QiaAMP DNA mini kit (Qiagen) and quantified using the Qubit fluorometer and BR kit (Life Technologies). Genomic DNA was diluted in 10 mM Tris-HCl to a concentration of 1500 ng in 46 µl and sheared using G-TUBES (Covaris). Min-ION sample preparation and two-dimensional sequencing was carried out as previously described using the SQK-MAP-006 kit and R7.3 flow cells [28]. Basecalling was carried out using the Metrichor two-dimensional basecalling option using version 1.15.7 for VRES0269 and VRES0273, and version 1.20.5 for VRES0259.
Basecalled MinION reads were converted from FAST5 to FASTQ and FASTA formats using fast52fastq.py (https:// github.com/kim-judge/wastewater). Read mapping was carried out to assess the quality of data and coverage using the BWA-MEM algorithm of BWA v0.7.12 with the flag -x ont2d [29]. Output SAM files from BWA-MEM were converted to sorted BAM files using SAMtools v0.1.19-44428 cd [30]. Assembly using MinION data only was undertaken using Canu [31,32], followed by Circlator [33] 1.2.0. Canu version 1.0 was run using the commands maxThreads=8, max-Memory=16, useGrid=0, nanopore-raw. Hybrid assemblies were generated using SPAdes 3.6 [34] using the -nanopore and -careful flags, then filtered to exclude contigs of less than 1 kb. All plasmid sequences and the chromosomes were confirmed to be circular by the Circlator software 1.2.0. Assembly statistics were analysed using assembly-stats (https://github.com/sanger-pathogens/assembly-stats). All assemblies were annotated using Prokka 1.5 [35,36]. Assemblies were compared using the Artemis comparison tool [37] and further analysed using Artemis [38]. Plasmid sequences and other regions of interest were investigated using the BLASTN suite at National Center for Biotechnology Information (http://www.ncbi.nlm.nih.gov).

Manually finished genome
Assemblies were generated using Canu and SPAdes as described above. A gap5 database was made using corrected MinION pass reads from the Canu pipeline and the Illumina reads. Manual finishing was undertaken using gap5 [39] version 1.2.14 making one chromosome and nine plasmids. Icorn2 [40] was run on this for five iterations. The start positions of the chromosome and plasmids were fixed using Circlator [33] 1.2.0 using the command circlator fixstart. This assembly was annotated using Prokka 1.5 [35,36].  IMP-encoding genes were detected by in-house PCR and/or commercial microarray, which was performed by the AMR-HAI Reference Unit using published methods [41][42][43][44][45][46].

Isolation of CPE from wastewater
We undertook a cross-sectional study of 20 municipal wastewater treatment plants in the East of England between June 2014 and January 2015. Plant location was selected with reference to hospital waste, with ten plants situated immediately downstream of acute NHS Hospital Trust facilities (approximate median distance between wastewater treatment plant and respective hospital 5.3 km, range 3.3-9.6 km) and ten plants not connected to acute hospital effluent (Fig. 1). Paired samples of untreated and treated wastewater were obtained from each plant. The main septic tank at the CUH facility was also sampled in September 2014. CPE were isolated using filtration and culture procedures (see Methods).
CPE were isolated from four plants and the CUH sewer (see Fig. 1 for locations). All four positive plants directly received hospital effluent. A total of nine bacterial isolates belonging to six species were cultured, all of which may cause human infection (Table 1). E. coli (n=1) and R. ornithinolytica (n=1) were isolated from treated water that was destined for release into the environment, and Klebsiella pneumoniae (n=3), Klebsiella oxytoca (n=1), Enterobacter cloacae complex (n=2) and R. ornithinolytica (n=1) were isolated from untreated water. In one case, the same species (R. ornithinolytica) was isolated from pre-and post-treated water sampled from the same plant.

Phenotypic evaluation of antimicrobial resistance
To characterize the carbapenemases in the nine isolates, we first used a phenotypic method that detects the three main biochemical groups of carbapenemases in Enterobacteriaceae. This identified a class A serine carbapenemase in K. pneumoniae, class B metallo-b-lactamases (MbLs) in E. coli, K. pneumoniae and Enterobacter cloacae complex, and

Genetic characterization of antimicrobial resistance
The entire repertoire of resistance genes/mechanisms was characterized in the nine study isolates by performing WGS using Illumina short-read technology, followed by comparison of each genome to a comprehensive database of genes/gene variants that encode drug resistance (See Methods for details).  Table 1), which are known to be carried by mobile genetic elements. Three isolates with high imipenem resistance also had truncated versions of porin proteins due to insertion sequences, stop mutations and frameshift mutations identified in ompC and ompF genes ( Table 1). Disruption of these genes have been associated previously with high levels of carbapenem resistance [47]. Genes conferring resistance to multiple classes of antimicrobials were identified in all nine isolates, with gentamicin, quinolone and sulphonamide resistance being the most prevalent (Table 1).
Three CPE isolates were recovered from the same treatment plant: Enterobacter cloacae complex VRES0273 and R. ornithinolytica VRES0269 from pre-treated water, and R. ornithinolytica VRES0259 from post-treated water. VRES0273 was identified as Enterobacter kobei based on analysis of hsp60 and rpoB, as described elsewhere [25,48]. To assess the core genetic relatedness of the two R. ornithinolytica isolates, we aligned Illumina reads from both against the reference genome R. ornithinolytica B6 [49]. This identified that both genomes were over 18 000 SNPs different from the reference over a core genome length of 5 398 151 bp (equating to 99.7 % identity), but only 13 SNPs different between each other. This indicates a very high degree of relatedness. In addition, both R. ornithinolytica isolates had identical antibiotic-resistance patterns for 14/15 antimicrobials tested ( Table 2). The three CPE isolates from this plant all contained the bla OXA-48 gene conferring carbapenem resistance and the same non-b-lactam-resistance genes ( Table 1).

Sharing of resistance plasmids between Enterobacteriaceae in wastewater
We hypothesized that resistance was mediated by one or more plasmids that were shared between these three isolates. We first examined the Illumina data alone to determine whether a plasmid carrying the bla OXA-48 gene was the same in the three isolates, but predictably we failed to assemble complete plasmid sequence from this short-read data. From this, we were unable to conclude whether plasmids were shared, nor their specific content. The bla OXA-48 gene was located on contigs of 2.5 kb (VRES273), 3.6 kb (VRES269) and 3.4 kb (VRES259) in the Illumina assemblies. To overcome this, DNA from the three isolates was sequenced using the MinION system to generate long-read data and assembled as described in Methods. From this we assembled a manually finished genome sequence for Enterobacter kobei (VRES0273), which represents the first complete whole-genome sequence for this species, and highquality draft assemblies for the two R. ornithinolytica isolates. Tables 3 and S2 summarize the Illumina and MinION sequencing data. Using these assemblies, we were able to confirm the close genetic relationship between the two R. ornithinolytica isolates by mapping each against the other. These were 12 SNPs different over 5 344 247 aligned bases in a chromosome of 5 462 249 bp (VRES0269) and 5 334 936 aligned bases in a chromosome of 5 392 238 bp (VRES0259), corresponding to 99.9 % identity in both cases, and covering 97.8 and 98.9 % of the genome, respectively. One large genomic difference was present between the two isolates due to the absence of one of the four phages in VRES0259 (Fig. 3).
One contig from each isolate (74 kb in Enterobacter kobei, 69 kb in R. ornithinolytica VRES259 and 63 kb in R. ornithinolytica VRES269) contained the bla OXA-48 gene (Fig. 4a). Based on BLAST searches and assembly comparisons using the Artemis comparison tool, these contigs showed synteny and orthology to other widespread IncL/M-type bla OXA-48 plasmids in Enterobacter cloacae, R. ornithinolytica and K. pneumoniae (GenBank examples: JN626286.1, KP061858.1, NC_023027.1, CVRH01000036 to CVRH01000038 and LN864819.1) [41,[50][51][52][53][54]. The plasmids from the three isolates were identical to each other over a shared region of 63 kb, with zero SNPs difference. Variation in overall size was due to hypothetical proteins integrated via insertion sequence (IS) elements. There were several transposases, which were located around the bla OXA-48 gene and forming part of the Tn1999 transposon (Fig. 4a). Similar variation in plasmids encoding other carbapenemases has been identified in isolates from different species at a single hospital [9,10].

DISCUSSION
To our knowledge, this is the first report of CPE being recovered from wastewater at UK sewage treatment plants.
Our demonstration of the co-circulation of a diversity of pathogenic, carbapenem-resistant Gram-negative species carrying different resistance genes in sewage confirms that this is a complex and diverse reservoir. We also confirmed that treatment processes do not prevent the release of CPE into the environment, which will contribute to contamination of river and lakes, farmland, vegetable land and fisheries, with the potential for spread to humans and livestock. Human infection caused by bacteria such as those identified here would prove challenging to treat with available antimicrobial drugs.
All CPE isolates were recovered from plants located downstream of hospitals. The same bacterial species and carbapenemase gene (encoding NDM) was found in two hospitals and the downstream wastewater treatment plant, indicating that the resistance mechanisms found in wastewater are likely to mirror those associated with human disease. Currently, no national regulation nor the European Directive 91/271/EEC on urban wastewater treatment stipulates the use of disinfection of hospital wastewater in the UK, and as a result this may be contributing to the dissemination of CPE [57,58]. Further studies are required to fully ascertain the role of hospital effluent as a source of environmental contamination with CPE.
Phenotypic methods presumptively identify the presence of the most common CPEs (KPC, MBL, OXA-48), but are unable to identify variant types. We used genome sequencing to identify variants of carbapenemase-resistance genes, to detect sequence alterations in porin genes associated with decreased membrane permeability (an important factor contributing to carbapenem resistance [59]), and to define the full repertoire of resistance genes. heavy metals. Whilst the rate of mutation in plasmids may vary from that of the chromosome, our findings are consistent with the suggestion that a globally distributed bla OXA-48 plasmid was shared very recently between the two species (0 SNPs different), a figure that is contextualized by comparison with a published sequence of plasmid pOXA-48a [50], which was more than 100 SNPs different. This plasmid is known to be the origin of the widespread dissemination of bla OXA-48 and, similar to our findings, has a broad bacterial host range and does not encode additional antimicrobial-resistance genes [50,60]. In addition, a recent report from Findlay et al. in 2017 [3] identified OXA-48-like enzymes as being the second most frequently detected carbapenemases in the West Midlands region of the UK between 2007 and 2014 based on a study of 119 clinical isolates, including one environmental isolate from an endoscope camera head in a urology theatre. OXA-48-like enzymes were identified in 16/119 isolates from three different bacterial species (K. pneumoniae, E. coli and Citrobacterfreundii). Similar to the plasmids detected in our study, five IncL/M plasmids were identified using WGS, which exhibited >99 % identity to pOXA-48a with no additional resistance genes. These genome data were not publicly available and no genome comparison was possible with our data, but based on similarity to pOXA-48a and genetic content we postulate that a highly related plasmid encoding bla OXA-48 is found in both UK wastewater and associated with human infection. Similar to pOXA-48a, the bla OXA-48 gene and the transposases identified in our study are part of the Tn1999 transposon, indicating the mobility of this gene independent of the plasmid backbone (Fig. 4a). It has also been shown in an early characterization of this plasmid that it is highly identical and syntenic to the pCTX-M3 plasmid, except for the Tn1999 transposon from which it might have evolved [50].
The multidrug and heavy-metal resistance plasmid was highly similar in both species, although this acquisition may have occurred less recently based on a three SNP difference between the R. ornithinolytica and the Enterobacter kobei plasmid. Based on the published mutation rate for K. pneumoniae (2.7Â10 À6 SNPs per site per year, equating to 14 SNPs per genome per year) [61], this would suggest that the plasmids last shared a common ancestor more than 3 years ago. Identification of such resistance plasmids spreading within and between bacterial species is essential when trying to understand the epidemiology of CPE, as has been shown in studies using PacBio sequencing [9,10]. A study by Sheppard et al. [10] particularly emphasizes that a similar detailed picture of plasmid transfer and variability is not available when only studying short-read data, which may prove misleading.
The main limitation of this study is the lack of quantification of CPE load per sample. This was due to technical limitations relating to the incomplete selectivity of currently available CPE media (which results in overgrowth of organisms other than CPE), combined with the low prevalence of CPE in samples. This leads to a low positive predictive value for detection of a true CPE from the growth obtained on the plate and makes quantitation extremely challenging. In order to enrich for low numbers of CPE in a very high bacterial background, we used an additional selective enrichment step (tryptic soy broth with imipenem), which increased the yield of CPE, but ruled out quantitation. Furthermore, the study was conducted in a region with very low CPE prevalence and may not be generalizable across the UK [62]. Future studies including longitudinal data from additional regions and hospitals will provide further insights into the community spread of CPE.
In conclusion, the combined use of Illumina and MinION technologies revealed sharing of plasmids carrying multiple antimicrobial-resistance genes between different bacterial species in wastewater, including the plasmid encoding OXA-48 that confers carbapenem resistance. Whilst phenotypic testing resolved the general class of carbapenemase and short-read data identified the specific carbapenemase genes involved, long-read data was essential to resolve the plasmid architecture and for accurate comparisons. Small, portable sequencing instruments, such as the MinION, have the potential for use in real-time genomic surveillance in a One Health approach that includes genetic material from wastewater, animals and hospitals to monitor the effectiveness of treatment systems, and to contribute to the development of interventions to limit the dissemination of antimicrobial resistance.