Genomic analysis of carbapenemase-producing Enterobacteriaceae in Queensland reveals widespread transmission of bla IMP-4 on an IncHI2 plasmid

Carbapenemase-producing Enterobacteriaceae (CPE) are an increasingly common cause of healthcare-associated infections and may occasionally be identified in patients without extensive healthcare exposure. bla IMP-4 is the most frequently detected carbapenemase gene in Enterobacteriaceae within Australia, but little is known about the mechanisms behind its persistence. Here we used whole genome sequencing (WGS) to investigate the molecular epidemiology of bla IMP-4 in Queensland, Australia. In total, 107 CPE were collected between 2014 and 2017 and sent for WGS on an Illumina NextSeq500. Resistance genes and plasmid types were detected using a combination of read mapping and nucleotide comparison of de novo assemblies. Six isolates were additionally sequenced using Oxford Nanopore MinION to generate long-reads and fully characterize the context of the bla IMP-4 gene. Of 107 CPE, 93 carried the bla IMP-4 gene; 74/107 also carried an IncHI2 plasmid, suggesting carriage of the bla IMP-4 gene on an IncHI2 plasmid. Comparison of these isolates to a previously characterized IncHI2 plasmid pMS7884A (isolated from an Enterobacter hormaechei strain in Brisbane) suggested that all isolates carried a similar plasmid. Five of six representative isolates sequenced using Nanopore long-read technology carried IncHI2 plasmids harbouring the bla IMP-4 gene. While the vast majority of isolates represented ﻿ E. hormaechei , several other species were also found to carry the IncHI2 plasmid, including Klebsiella species, Escherichia coli and Citrobacter species. Several clonal groups of E. hormaechei were also identified, suggesting that persistence of bla IMP-4 is driven by both presence on a common plasmid and clonal spread of certain E. hormaechei lineages.


InTRoDuCTIon
Carbapenemase-producing Enterobacteriaceae (CPE) are a growing burden worldwide, and infections with these organisms are often associated with significant morbidity and mortality [1,2]. The family Enterobacteriaceae includes bacterial species such as Escherichia coli, Klebsiella pneumoniae and the Enterobacter cloacae complex, which are responsible for the majority of infections caused by gram-negative bacteria [3]. CPE typically carry a variety of antibiotic resistance genes, leaving few effective treatment options [4,5]. CPE transmission occurs most frequently in clinical environments [6]; however, community-acquired CPE are increasingly recognized, posing a great threat to public health [6,7].
CPE genes endemic to certain geographical locations have been described, such as bla KPC in the United States and parts of Europe [8,9], bla NDM in India and China [10], OXA-48like carbapenemases in Mediterranean countries and the Middle East [9], and bla IMP in Australia and the Asia-Pacific region [11,12]. CPE may be maintained in the population by the transmission of carbapenemase genes circulating on plasmids or via clonal expansion [6]. Few studies have used whole genome sequencing (WGS) on large geographically related datasets, enabling in-depth analyses into the genomic context of carbapenemase genes and their associated plasmids [13,14].
This was a prospective cohort study of patients admitted to any Queensland Health facility served by the state-wide microbiology laboratory network (Pathology Queensland), who were colonized or infected with CPE. The aim was to describe the clinical features of patients with CPE, define the range of CPE species, identify the dominant carbapenemase genes, and elucidate patterns of clonal or plasmid transmission using WGS.

Setting
Pathology Queensland is a network of 35 laboratories serving all public hospitals in Queensland, Australia. According to laboratory protocols, suspected CPE or carbapenem-resistant Enterobacteriaceae are referred to the central laboratory in Brisbane for further analysis.

Isolate selection and CPE detection
Susceptibility testing was performed using Vitek 2 (bioMérieux), with minimum inhibitory concentrations (MICs) for meropenem also determined using MIC gradient tests (Etest; bioMérieux). Carbapenemase production was detected using colorimetric methods (RAPIDEC CARBA-NP; bioMérieux; or β CARBA; Bio-Rad) and chromogenic agar (chromID CARBA SMART Agar; bioMérieux). Any Enterobacteriaceae with a meropenem MIC >0.25 mg l −1 by Vitek2 or ≥0.125 mg l −1 by Etest [15], a positive colorimetric test for carbapenemase or growth on chromID CARBA SMART agar was submitted for molecular confirmation of CPE status using a multiplex PCR assay targeting NDM, IMP-4-like, VIM, KPC and OXA-48-like carbapenemase genes [16][17][18][19]. Any Enterobacteriaceae isolated between January 2014 and May 2017 with PCR confirmation of carbapenemase genes, or suspected carbapenemase production by phenotypic methods, was characterized further by WGS. Only a single CPE isolate per patient was included in the WGS analysis. However, additional CPE sequences were included if a different CPE species was subsequently isolated or PCR demonstrated the presence of a different carbapenemase gene as compared with the initial organism.

Clinical data
Clinical variables of patients with CPE were collected from the electronic medical record. This included ward of admission at the time of CPE detection, admitting clinical service, body site of initial specimen with CPE, the Charlson Comorbidity score [20] at admission, and other significant risk factors for infection (end-stage renal failure requiring dialysis, solid organ or haemopoietic stem cell transplant, cytotoxic chemotherapy, monoclonal antibody therapy or other immunosuppression). The following were recorded if

Impact Statement
Gram-negative pathogens belonging to the family Enterobacteriaceae, such as Escherichia coli, Klebsiella pneumoniae or Enterobacter species may acquire genes that encode carbapenemases; these enzymes diminish the efficacy of carbapenem antibiotics. Carbapenemaseproducing Enterobacteriaceae (CPE) are difficult to treat, but have been historically rare in Australia. More recently, CPE have caused outbreaks within hospitals and are now a major infection control concern. Previous work has demonstrated bla IMP-4 to be the most common carbapenemase in Queensland, but little is known about transmission within the population. We prospectively used whole genome sequencing of CPE collected from the main public referral laboratory in Queensland. The dominant carbapenemase was bla IMP-4 , mainly found in species within the Enterobacter cloacae complex, in association with a widespread IncHI2 plasmid. The majority of bla IMP-4 were carried on a very similar IncHI2 plasmid which was found to be present across different clonal groups of Enterobacter and other species such as Escherichia coli, Klebsiella or Citrobacter species. Other carbapenemase genes were infrequent and almost exclusively associated with healthcare exposure overseas. The persistence of bla IMP-4 in Queensland is probably maintained by both clonal expansion of certain Enterobacter lineages as well as a common IncHI2 transmissible plasmid. present within 12 months prior to CPE detection: any history of travel or healthcare exposure overseas, prior admission to hospital or regular healthcare exposure (e.g. haemodialysis), intensive care unit (ICU) admission, previous colonization or infection with an extended-spectrum β-lactamase (ESBL)producing Enterobacteriaceae, methicillin-resistant Staphylococcus aureus (MRSA) or vancomycin resistant enterococci (VRE). The presence of any of the following was recorded if present within 1 month prior to CPE detection: surgical procedures, endoscopy, any episode of neutropenia and antibiotic exposure. Any directed antibiotic therapy for the CPE and its duration was recorded, as was the outcome (died or survived) up to hospital discharge.

DnA extraction and sequencing
DNA was extracted from pure bacterial colonies after overnight incubation on horse blood agar at 37 °C, using the DSP DNA Mini Kit on the QIAsymphony SP instrument (Qiagen). Libraries were prepared using the Nextera XT DNA preparation kit (Illumina) and sequencing was performed on the NextSeq 500 (Illumina) with 2×150 bp chemistry, NextSeq Midoutput kit v2.5. Isolates for sequencing with the Oxford Nanopore MinION were plated on Luria Bertani (LB) agar and incubated overnight at 37 °C. DNA was extracted using a Qiagen DNeasy UltraClean Microbial kit (as per the manufacturer's instructions). Library preparation for eight isolates was done using the 1D sequencing by ligation kit (SQK-LSK108) with the native barcoding expansion kit (EXP-NBD103) to multiplex all isolates on a single FLOW-MIN106 R9.4 flow cell. Sequencing on the MinION with local base-calling ran for ~40 h, generating 65 993 reads.

Assembly
Trimmed Illumina reads were de novo assembled using SPAdes v3.11.1 [23] under default settings. De novo assembly metrics are provided in the Supplementary Data (available in the online version of this article). Isolates CQS2 and CQS52 were removed from the study as more than 10 % of the assembled genome was found to be <10× coverage and in contigs <100 bp. Filtered Nanopore reads were de novo assembled using Canu v1.7 [24] with default settings. CQS39 and CQS33 were removed from the longread analysis due to low throughput and batch problems with the barcoding kit.

Phylogenetic analyses
Parsnp (v1.2) [32] was used to create a phylogeny using the draft assemblies of all isolates against the Enterobacter hormaechei strain MS7884 chromosome (gbk: CP022532.1) under default settings (67 457 SNP positions). MS7884 was originally isolated in south-east Queensland and was chosen as the reference to represent the majority of isolates in this dataset, which were part of the E. cloacae complex. Metadata was added to the resulting tree (mid-point rooted) using Phandango [33]. SNP distances were determined by mapping trimmed reads from E. hormaechei isolates to representative draft assemblies for each ST using Bowtie2 v2.3.4.2 (as implemented through Nesoni v0.132: https:// github. com/ Victorian-Bioinformatics-Consortium/ nesoni) (see Supplementary Data for reference genomes and ST groups). SNV (Single nucleotide variant) relationship matrices (Nways) produced by Nesoni were interrogated manually to remove low-confidence SNPs, specifically (1) those that were inconsistent between mapping and assembly of the same strain, and (2) those that had an ambiguous allele call (and therefore possible mismapping to a repetitive region in the genome). The relationship matrices were built both manually and using the Neighbor-Joining function (Hamming distance, Saitou-Nei criterion) in Phyloviz v2.0 [34] based on SNPs detected using Nesoni.

RESuLTS
From 2014 to 2017, a total of 94 CPE from Queensland were identified from 81 patients (nine patients had more than one CPE isolate sequenced) (Fig. 1). During this period, 179 suspected CPE isolates were submitted to the central laboratory for PCR testing. Cumulative antibiogram data for all public hospitals served by Pathology Queensland in 2017 demonstrated that <1 % of all Enterobacteriaceae were resistant to meropenem. An additional four CPE strains cultured from endoscope surveillance samples were also included in the analysis, as well as nine CPE (from two additional patients) sequenced during an outbreak in 2015 [35], bringing the total number of strains to 107 from a total of 83 patients (
Overall, non-Enterobacter species carried on average 35 % of the MDR region (based on comparison of non-redundant CDS regions). Interestingly, those that carried the bla IMP-4 gene carried twice as many CDS from this region compared to those that did not (24 % vs 48 % on average) (see Supplementary Data). These results suggest that bla IMP-4 has been mobilized via an integron or other transposable element from a pMS7884A-like plasmid.
Based on these findings, we propose that the prevalence of bla IMP-4 is mainly facilitated by the dissemination of a pMS7884A-like plasmid among a diverse range of Enterobacter STs around Queensland. The increased prevalence of certain STs among the dataset also suggests probable expansion of specific Enterobacter clonal lineages, such as ST90 and ST830 (Fig. 2).

Long-read oxford nanopore minIon sequencing further resolves the context of bla ImP-4 in a range of bacterial species
To further understand the stability and context of the bla IMP-4 carbapenemase gene in other species, we selected six isolates from this dataset for long-read sequencing with Oxford Nanopore MinION. The six isolates represented E. hormaechei, Escherichia coli, K. aerogenes, C. freundii and C. amalonaticus. Plasmids carrying bla IMP-4 were assembled from the MinION long-read data for six isolates. Five of the six isolates all carried the bla IMP-4 carbapenemase gene on an IncHI2 plasmid with >98 % ANI to pMS7884A (isolated from an E. hormaechei strain in 2015) (Fig. 3). The ~55 kb MDR region was mostly conserved, except for a small ~6.6 kb region containing the catA2 and strAB genes that was not present in CQS18 (K. aerogenes) and CQS53 (E. hormaechei). Isolate MS7925 (Escherichia coli) was found to have acquired an additional ~4.4 kb transposon carrying bla SHV-12 at the 3′ end of the ~55 kb MDR region. CQS53 contained a small inversion within the ~55 kb MDR region. CQS51 contained a large inversion that encompassed part of the MDR region and also the IncHI2 plasmid backbone (Fig. 3).
The additional E. hormaechei isolate, CQS89, carried both an IncL/M and an IncHI2 plasmid. The IncHI2 plasmid backbone was highly similar to pMS7884A, but the bla IMP-4 gene, as well as the In809 integron, were located on the IncL/M plasmid, flanked by IS26 (Fig. S10). The IncL/M plasmid also carried an ~16 kb region with genes relating to tunicamycin, macrolide and chromate resistance.

IncL/m plasmids as possible vector for bla ImP-4 and bla oXA-48 carbapenemases
The nine Enterobacter isolates that did not carry an IncHI2 plasmid, but retained bla IMP-4 , were found to have IncFIB (n=8), IncFII (n=7), IncL/M (n=6) and IncR (n=1) plasmid types. As carriage of bla IMP-4 has previously been described on IncL/M-type plasmids in Australia [36], we compared all IncL/M-positive isolates to the previously described IncL/M plasmid pEl1573 isolated in Sydney, Australia (GenBank: NC_019368.1 [36]). Of 23 isolates, we found 10 that appeared to carry a very similar plasmid to pEI1573, including the bla IMP-4 gene within a large region comprising the majority of other resistance genes (Fig. S9, Table S3). Four isolates appeared to have retained the IncL/M plasmid backbone, but lost the entire resistance region, which was consistent with the lack of bla IMP-4 and other resistance genes. These four isolates [CQS3 (K. pneumoniae), CQS4 (K. pneumoniae), CQS5 (K. aerogenes) and CQS6 (E. hormaechei)] all carried a bla OXA-48 gene instead, suggesting an interchangeable resistance region on the IncL/M plasmid. All four isolates were collected from the same patient, demonstrating the transmissibility of this suspected blaOXA-48-carrying IncL/M plasmid to all species within this patient. The remaining isolates (n=9) appeared identical to pEI1573 except for the loss of an ~5 kb region containing ISCR1, qnrB2 (responsible for resistance to fluoroquinolones), qacEdelta1 (responsible for resistance to quaternary ammonium compounds) and sulI (responsible for resistance to sulphonamides).

Analysis of E. hormaechei ST groups reveals possible long-term transmission
A number of ST830 E. hormaechei isolates were recovered from similar geographical locations, indicating a possible clonal relationship and ongoing transmission within the area. To investigate this, we analysed the SNP differences between E. hormaechei ST830 isolates, which revealed two distinct groups (Fig. S11). By mapping reads from the ST830 isolates to references within each group, we found that group 1 isolates were on average 6 SNPs away from the next closest isolate (range=2-11 SNPs), while in group 2 there were on average 11 SNPs distant (range=7-16 SNPs) (Fig. 4). Based on these SNP distances, it is unlikely that these strains are related by direct recent transmission. However, the clustering of similar strains primarily within two hospitals over a period of 18 and 13 months (for groups 1 and 2, respectively), as well as the identification of environmental isolates, suggests undetected transmission over a long period of time (Fig. 4).
To identify other possible transmission events, we analysed an additional nine E. hormaechei ST groups containing two or more isolates from different patients using the same readmapping method as with the ST830 isolates (see Supplementary Data). Comparison of the isolates within these groups identified possible transmission events for E. hormaechei ST66, ST90, ST108 and ST133 (Fig. 4).
Two separate clusters were detected in the ST66 group, both confined to specific hospitals. The hospital A cluster (purple) was not deemed to reflect transmission as both isolates were taken from the same patient. All isolates from the hospital E cluster (light blue) were predicted to be related, with possible transmission detected in two groups; the first group of isolates (CQS78, CQS73, CQS29, CQS30 and CQS32) differed by less than 4 core SNPs overall and were collected in either March or October 2016. The second group (CQS26 and CQS80) was collected almost a year apart but only differed by 3 core SNPs, indicating probable persistence in the environment. The 13 core SNPs separating these two groups also suggests that direct transmission between these groups of isolates is unlikely, and that transmission from an intermediate source (environmental or unsampled patients/staff) is a more probable explanation.
The ST90 cluster has been described previously [35]. No additional ST90 isolates related to this outbreak were detected in this dataset (CQS46 and MS14389 in reference [35] are derived from the same frozen stock culture in Pathology Queensland). ST108 was found to have three closely related isolates from the same hospital (G; pink), all collected within a 3-month window and probably indicative of transmission. The ST133 group had a cluster of three identical isolates all obtained on the same day (two from the same patient), followed by two more distantly related isolates all from the same hospital. This is likely to indicate short-term transmission between the two patients with identical isolates, with additional long-term transmission or a common reservoir in the environment for at least 10 months.
Four of the five remaining ST groups (Supplementary Data) were found to be unrelated at the core SNP level, suggesting no recent transmission events. This conclusion is also supported by the distinct geographical locations where these isolates were originally obtained. The exception was the ST45 group, which had two isolates from the same hospital. However, due to the core SNP distance (>100 SNPs) and the time between hormaechei STs that had two or more isolates were analysed to determine their relatedness and any possible transmission events. Four ST groups (ST830, ST66, ST108 and ST133) were found to have evidence of recent and/or long-term transmission. Each figure represents core SNP differences between isolates within that ST group: isolates within the same circle are identical at the core genome level, while the black lines represent the number of SNPs different between isolates. Colours represent the hospital where the isolate was sourced. There was no evidence of transmission within the ST90 group outside of what has already been reported [35].
isolation (almost a year apart) they were deemed unlikely to be related to ongoing transmission.

DISCuSSIon
In this study we used WGS to explore the epidemiology of CPE in Queensland. CPE are a major concern for hospitals globally due to their high levels of resistance to antibiotics [37]. Reporting incidences of CPE has become a key priority in combating the spread of these bacteria and understanding their distribution within specific geographical regions. The use of WGS in epidemiological surveillance and nosocomial outbreak investigations has become well established due to its high discriminatory power and reduced costs [38][39][40]. Short-read sequencing is by far the most commonly used technology due to its high throughput and accurate read quality. However, a major caveat of short-read data is the inability to fully characterize complete plasmid sequences and definitively determine plasmid transmission networks [41].
Here we show how short-read data with selective long-read sequencing was able to characterize the epidemiology of an important plasmid circulating in Queensland.
In our study we found that the majority of CPE in Queensland carry the carbapenemase gene bla IMP-4 , which is consistent with previous reports describing its high prevalence in Australia [11]. This is in contrast to North America, Europe and parts of South East Asia, where bla KPC and bla NDM have a much higher prevalence [6,42]. The only cases of uncommon (i.e. non-bla IMP-4 ) β-lactamase acquisition in our dataset were associated with recent travel to Greece (for a single bla KPC isolate) or India (for all isolates with bla NDM variants). These results reflect what has previously been reported by AURA and the CARAlert system [43], where IMP carbapenemases appear to be most common in New South Wales and Queensland, while Victoria has a more even proportion of IMP, NDM and OXA-48-like carbapenemases. This disparity has also been recently characterized in a broader study of CPEs within Victoria, where bla IMP-4 and bla KPC-2 were most commonly identified in their cohort [44].
In addition to the high prevalence of bla IMP-4 , presumably carried by a common IncHI2 plasmid, there was also evidence that clonal groups of E. hormaechei are simultaneously driving the spread of this carbapenemase gene in Queensland. This was found to be particularly prevalent in intra-hospital transmission events, which could eventually result in community or other healthcare facility spill-over events. While we could not in all cases discern direct transmission events, we found that the number of core SNP differences between isolates with presumed transmission agreed with the SNP threshold defined in a Victorian dataset of CPE, which was ≤23 SNPs for 'highly likely' or 'probable' local transmission [44]. In this dataset we did not define a specific SNP threshold. This was mainly because of the size of our dataset (<100 isolates), which left few isolates per transmission cluster to draw a conclusive threshold. Defining SNP thresholds can also be problematic as biological variance (e.g. differing environmental pressures), sampling bias (e.g. missing intermediate isolates) and analytical variance (i.e. SNP calling process) can all introduce large inconsistencies in how the thresholds are defined. Future work in the area of transmission clusters is likely to move away from SNP thresholds and towards probabilistic methods of transmission inference [45].
While previous analyses based on traditional molecular techniques were able to observe presumed in vivo transfer of bla IMP-4 [46], the exact mechanism remained unclear. Based on our analysis we suggest that the high prevalence of bla IMP-4 is likely to be maintained by the widespread dissemination of an IncHI2 plasmid similar to pMS7884A. We also propose that this IncHI2 plasmid has pervaded a broad host range, including all species presented in this study as well as other species, such as Salmonella enterica Typhimurium [35,47]. This adaptability has probably assisted the continued persistence of this plasmid within Australia. Further work is required to determine if this plasmid is able to transfer directly between the species investigated here, or via an intermediate.
Another study investigating the presence of carbapenemase genes in silver gulls in south-east Australia revealed a high prevalence of bla IMP-4 carrying Enterobacteriaceae associated with IncHI2 plasmids, presumed to be from gulls feeding on human refuse [48]. This significant environmental reservoir coupled with potentially cyclical re-introduction of this plasmid into human populations could help to explain the endemic levels of bla IMP-4 . Further analysis of human, animal and environmental microbiome and isolate genome datasets from around Australia is needed to fully resolve the missing links in IncHI2/bla IMP-4 dissemination.
Six isolates from the study were selected for long-read sequencing with Oxford Nanopore MinION to determine the true context of the carbapenemase gene bla IMP-4 . Five isolates were found to carry an IncHI2 plasmid similar to pMS7884A, confirming that this plasmid is probably circulating within Queensland and driving the dissemination of the carbapenemase bla IMP-4 among different organisms. It is also likely that this plasmid has propagated successfully and widely for a number of years as a consequence of it being able to survive in several different Enterobacteriaceae, making it hard to eradicate from environmental reservoirs or detect in patients. Large inversions, insertions or deletions of particular regions appeared to be mainly mediated by insertion sequences.
While it was not explored in this study, these genomic rearrangements provide valuable information when discerning plasmid transmission between closely related isolates that may not be detected using short-read sequencing alone. For example, all isolates from the hospital E cluster of ST66 E. hormaechei were missing a Tn3-family transposon as well as the tetracycline resistance region. Additionally, all ST830 E. hormaechei were missing the streptomycin resistance region, further supporting probable transmission of a single clone between multiple hospitals ( Fig. 1 and Supplementary Data). In most cases, the entirety of the ~55 kb MDR region described in pMS7884A was intact, except for CQS51 (where an inversion has caused the region to split) and CQS89 (where an IncL/M plasmid contained some of the MDR region). Several of the isolates were also shown to have gained or lost certain portions of the MDR region, highlighting the plasticity of this region and how easily additional resistance genes could be acquired.

Conclusion
Using WGS of 107 isolates collected between 2014 and 2017, we show that persistence of the bla IMP-4 gene in Queensland is associated with the spread of a large IncHI2 plasmid with a broad host range. Long-read sequencing of six example isolates revealed the true context of the bla IMP-4 gene within IncHI2 plasmids in different species backgrounds and highlighted the ease with which additional mobile regions containing resistance genes are acquired and lost. This genetic plasticity suggests that there is potential for this endemic plasmid to become an even more compelling threat to public health through the acquisition of more potent resistance genes.