Genomic and clinical characterisation of multidrug-resistant carbapenemase-producing ST231 and ST16 Klebsiella pneumoniae isolates colonising and causing disease in hospitalised patients at Siriraj Hospital, Bangkok, Thailand from 2015 to 2017

Background: Infections caused by carbapenemase-producing Enterobacteriaceae (CPE) have continually grown as a global public health threat, with signicant mortality rates observed across the world. We examined the clinical data from patients with CPE infections and their outcomes, concentrating on Klebsiella pneumoniae isolates. We analysed the clinical information, performed antimicrobial susceptibility testing, and conducted molecular epidemiological and genomic analyses on the isolates to identify patterns in the data. Methods: The clinical characteristics of 33 hospitalised patients with conrmed CPE, including patient-related factors associated with the development of CPE infections, were examined. Patients were divided according to whether they were “colonised” or “infected” with CPE and by the timing and frequency of their rectal swab collections, from which 45 swabs were randomly selected for analysis. CPE isolates were puried, and antimicrobial susceptibility tests performed. Whole genome sequences of these isolates were determined and analysed to compute bacterial multilocus sequence types and plasmid replicon types, infer phylogenetic relationships, and identify antimicrobial resistance and virulence genes. Results: Altogether, 88.9% (40/45) of the CPE were most carbapenemase gene pneumoniae was bla OXA-232 , with bla NDM-1 additionally identied 19 them. CPE bla NDM-1 resistant meropenem, but only 40 from 45 CPE-infected patients CPE-colonised patients CPE infections study standard colistin-based combination therapy. Phylogenetic analysis the polyclonal spread of carbapenemase-producing K. pneumoniae within the patient population, with the following two major subclades


Background
Infections caused by carbapenemase-producing Enterobacteriaceae (CPE), which have increased worldwide in number to become a signi cant clinical problem over the last decade, are associated with high morbidity and mortality (1). An early longitudinal study from Asia (2000-2012) revealed the prevalence of CPE was extremely low with average rates between 0.6-0.9% (2). However, a survey from the National Antimicrobial Resistance Surveillance, Thailand from 2008-2016 revealed that carbapenemase-producing Klebsiella pneumoniae (CPKP) had increased in prevalence from 0.4% in 2008 to 5.4% in 2016 (3). A recent retrospective cohort study from a 1,200-bed university hospital in Bangkok reported on an increased incidence of CPE from 3.37 per 100,000 patient-days in 2011 to 32.49 per 100,000 patient-days between 2011 and 2016 (4). The resistance mechanism for CPE is attributed to the following Ambler molecular classes of carbapenem-hydrolysing beta-lactamases: class A (KPC), class B (IMP, NDM, VIM), and class D (OXA-48) (5). There is insu cient data from Thailand on the distribution of beta-lactamase (bla) genes (6,7). However, a recent study from a university hospital in Bangkok revealed that bla NDM was the most common such gene followed by bla OXA-48-like alleles (e.g., bla OXA-48 , bla OXA-181 , and bla OXA-232 ) (7). Among all Enterobacteriaceae, CPKP is commonly associated with numerous antimicrobial resistance (AMR) genes and virulence determinants (8). CPKP with its plasmid-encoded carbapenemases (e.g., bla NDM and bla KPC ) and its multiclass antibiotic resistance is associated with hospital-acquired infections and treatment challenges (9,10). Hypervirulent K. pneumoniae (hvKP) can also cause invasive diseases such as liver abscesses and metastatic infections (11). In addition to having the K1 capsular serotype, hvKP encodes other virulence determinants (e.g., yersiniabactin, aerobactin, and salmochelin siderophores), and rmpA1/rmpA2 genes, which upregulate capsule expression and are associated with more invasive infections (8,12,13).
Although genetic diversity in carbapenemases has previously been reported in Thailand (7,14), the epidemiology and characteristics of CPKP and its virulence determinants are not as well understood.
Information is also lacking on the role played by hypervirulent strains of CPKP in hospital-acquired infections and how speci c virulence determinants are associated with AMR pro les, disease severity and outcomes among hospitalised patients. The aim of the present study was to investigate the molecular epidemiological features of CPKP and its association with the clinical presentations of CPKP-infected patients. We also aimed to identify virulence determinants in the CPKP strains isolated from patients in this study.

Patient population and specimen collection
Eligible patients included all hospitalised patients aged ≥ 18 years who had CPE recovered from clinical specimens submitted to the microbiology laboratory. The participants were classi ed as CPE-colonised patients or CPE-infected patients at enrolment. CPE-colonised patients were de ned according to whether or not CPE carriage was found, but its presence was not associated with any symptoms or clinical disease. Rectal swabs were collected from the patients within 48 hours of enrolment and then once a week until the specimens were CPE-negative for 3 consecutive weeks (at the time of study there was no routine CPE screening procedure). One hundred and nineteen patients met the eligibility criteria, including 69 CPE-colonised patients and 50 CPE-infected patients at enrolment. Forty-ve randomly selected and deidenti ed rectal swab samples from 33 patients who were admitted for treatment between December 2015 and April 2017 used for bacterial isolation and subsequent whole genome sequencing. Clinical information from each patient was collected, including information on patient demographics, clinical diagnosis, as well as their treatment during hospitalisation and outcome. The clinical outcomes of the CPE-infected patients at the end of treatment were classi ed as either 'favourable response' (absence or improvement of all clinical signs and symptoms of CPE infection) or 'unfavourable response' (worsening or persistence of clinical signs and/or symptoms of CPE infection, superinfection, or death).

Bacterial isolates and antimicrobial susceptibility testing
Rectal swabs were inoculated onto MacConkey agar (BD, USA) supplemented with ceftriaxone (4 mg/L) and the plates incubated at 37°C for 18 h. Bacterial identi cation was performed using Biotyper MALDI-TOF MS (Bruker Daltonics, Germany) according to the manufacturer's protocol. Colonies identi ed as Enterobacteriaceae were tested for antimicrobial susceptibility using standard methods and following the guidelines for the disk-diffusion method (15). Con rmation of suspected carbapenemase production in Enterobacteriaceae-positive specimens was performed using a modi ed carbapenem inactivation method (16). Phenotypic screening for the presence of carbapenemases was performed using a doubledisc synergy approach with phenylboronic acid or ethylenediaminetetraacetic acid with meropenem as previously described (17). Colistin resistance was tested using the broth microdilution method with cation-adjusted Mueller-Hinton II broth (16). Susceptibility to tigecycline was not tested at Siriraj Hospital during the study period.

Results
Clinical characteristics of the study participants Thirty-three patients had their CPE isolates whole genome sequenced for this study. Of these, 15 patients were found to be colonised with CPE, while the others had a con rmed CPE infection at the time of enrolment. Nineteen of the 33 patients (57.6%) were female. The mean age of all the patients was 62.8 years (IQR, interquartile range: 47-81 years) ( Table 1). Most CPE-colonised patients (n=13/15) had experienced prolonged hospitalisation before CPE was detected in their rectal swabs (median stay: 21 days, IQR 0-34 days), and 39% of them were admitted to the intensive care unit. Thirty-two patients had underlying conditions such as diabetes mellitus or chronic kidney disease and had received antibiotic treatment within the month prior to enrolment, of which two thirds (n=23) received carbapenems (Supplementary Table 2 Table 2). We still included these three patients in the CPE-colonised group. Among the CPE-infected patients, ventilator-associated pneumonia (VAP) was the most common consequence of CPE infection, followed by urinary tract infection and primary bacteraemia (Supplementary Table 2). Only one CPE-infected patient received colistin monotherapy; all 17 of the other patients received colistinbased combination therapy, with the median duration of antibiotic treatment lasting 11 days (IQR: 7-14 days) (Supplementary Table 2). Colistin-fosfomycin was the most common antimicrobial combination regimen (45%, n=9/20) followed by colistin-piperacillin/tazobactam (15%, n=3/20) (Table 1). Colistinfosfomycin was the rst treatment option for patients with carbapenem-resistant infections because this combination has been shown to afford higher microbiological eradication rates than colistin monotherapy in Siriraj Hospital (29). Regarding the local antibiogram, because CPE was more susceptible to piperacillin-tazobactam than to imipenem and meropenem, the second most common combination regime was colistin-piperacillin/tazobactam. Unfavourable clinical outcomes were observed in 52.4% of all the CPE-infected patients (n=21), including three who were initially colonised with CPE but later developed CPE infections, and seven of these patients experienced superinfections with different bacterial species at the end of their antibiotic regimes (Table 1). There was no statistically signi cant mortality observed between the CPE-infected patients (47.6%, n=10/21) vs. the CPE colonised ones (33.3%, n=4/12; chi-square test; p = 0.43).
Antimicrobial susceptibility patterns detected in the CPE isolates We isolated 39 K. pneumoniae, four Escherichia coli, and one isolate each of Enterobacter hormaechei subsp. steigerwaltii and K. quasipneumoniae subsp. similipneumoniae from the 33 patients in our study ( Table 1). All 45 isolates displayed meropenem resistance, only 20% of them (n=9) were susceptible to amikacin, and 17.8% (n=8) were susceptible to fosfomycin ( Table 1). All of the CPE isolates were resistant to cipro oxacin, cefoxitin, ceftriaxone, ceftazidime, piperacillin-tazobactam, ertapenem and imipenem (Table 1). Only ve of the isolates, all K. pneumoniae, showed resistance to colistin with MIC values ranging between 32 and 64 mg/L (Table 1). No signi cant differences between the antimicrobial susceptibility patterns of isolates from CPE-colonised patients and those from CPE-infected patients were observed, indicating that colonising and disease-causing strains show very similar AMR pro les, although this nding may also be attributed to the relatively small sample size available.

High diversity in AMR genes and plasmids in the CPE isolates
Our genomic analysis showed that bla OXA-232 was the most dominant carbapenemase gene family and was found in 34 of 39 K. pneumoniae and two of the four E. coli isolates we sequenced (Supplementary Table 1). The two most common sequence types (STs) identi ed in K. pneumoniae were ST16 (n=15) and ST231 (n=14), from which 12 ST16 isolates carried bla OXA-232 and bla NDM-1 , whereas almost all of the ST231 (n=13) isolates carried only bla OXA-232 (Figure 1, Supplementary Table 1). In addition, all of the CPE isolates carrying β-lactamase genes also carried genes encoding other AMR genes, including aminoglycosides (aac(6)-I, aph (3)), uoroquinolones (qnrB, qnrS), and fosfomycins (fosA6, UhpT) (Supplementary Table 1). None of the ve colistin-resistant isolates harboured mcr-genes, although they were highly resistant to colistin (Table 1), and mutational changes in mgrB and pmrB were also detected (Brinkac et al., manuscript in preparation).
We identi ed the following range of incompatibility (Inc) plasmid groups in the CPE isolates: FIA, FIB (pQil), FII, HI2B, N2 and R (Figure 1). We were particularly interested in the presence of IncFIB and the small-sized Col plasmid group in our CPE dataset because these two plasmid groups are reported to be most commonly found in clinical samples and are associated with the spread of AMR genes (30). Interestingly, in our dataset, all cases where ST231-CPKP was present (n=14) and nearly half of those with ST16-CPKP (n=7) contained an IncFIB(pQil)-like plasmid ( Figure 1). Additionally, genomic analysis indicated that all bla OXA-232 -containing CPE isolates were predicted to contain ColKp3 plasmid replicons ( Figure 1).

K. pneumoniae isolates carry genes associated with hypervirulence
We searched the K. pneumoniae genomes for the virulence genes previously found in hvKP strains, including those encoding siderophores for the biosynthesis and uptake of iron (ybt, iuc and iro) and genes for the regulator of mucoid phenotype (rmpA1/rmpA2) (13). The ybt locus, encoding the siderophore yersiniabactin, was present in 38/39 of the CPKP genomes. The most common allele, ybt14 (located on ICEKp5), was identi ed in 19 isolates, while the second most common allele, ybt9 (located on ICEKp3), was identi ed in 17 isolates, and the rest two isolates had ybt8(located on ICEKp9) and ybt10(located on ICEKp4), respectively ( Figure 2). Notably, iuc5, encoding the siderophore aerobactin, was only detected in ST231 (n=14). We detected six distinct K locus (KL) types among 39 CPKP isolates, the most frequent ones being KL51 (n = 28), KL2 (n =5), and KL17 (n = 3) (Figure 2). Virulence plasmidassociated loci such as iro, encoding the siderophore salmochelin, colibactin and rmpA1 and rmpA2 were not present in the investigated CPKP genomes. We also found that wzi50 was more common in the ST16 isolates, whereas wzi104 was only found in the ST231 isolates, and capsular antigen KL51 was found in ST16 and ST231 isolates alike.

Associations between the patients' clinical data and the CPE isolates
We identi ed nine patients who had more than one CPE isolate isolated throughout their hospital stay ( Table 1). Six of them, despite receiving appropriate treatment, had > 2 follow-up isolates that were the same bacterial species with the same sequence type and similar antibiogram pattern (Table 1). Genomic analysis also con rmed that the bacterial isolates from the same patient were identical with only 0-1 single-nucleotide polymorphism (SNP) difference.
We noted that the bla OXA-232 carbapenemase-encoding ColKP3 plasmid was present in different strains of K. pneumoniae as well as in E. coli (Figure 1), indicating the possibility of horizontal interspecies spread of this plasmid and possibly resulting in a polyclonal outbreak within our hospital. Although ST231 and ST16 were the two main clones associated with invasive disease and poor outcomes in our study, we did not identify any particular STs that were found only in CPE-colonised patients or only in CPE-infected patients.

Discussion
To the best of our knowledge, this is the rst study to document detailed molecular bacterial isolate information on carbapenem resistance, plasmid replicons, and virulence determinants in relation to the clinical characterisation of hospitalised patients in Thailand. Of the 25 CPE-patients with follow-up rectal swab cultures, the mean time to culture negativity was 37.7 days in our study. This nding is consistent with previously reports that 54% of CPE rectal carriers remained CPE carriers for 30 to 60 days after their initial screening, 28% remained as such after six months to one year, and 14% remained as such after one year (31,32). In our study, approximately 16.7% of the asymptomatic rectal carriers developed a clinical infection with a median duration of 20 days. The incidence of CPE infections in the CPE-colonised patients in our study was as high as that seen previously (33,34), and there are several possible explanations for this. One explanation is that we began to observe patients who already had CPE colonisation at sites other than the gut, which might be a risk factor for them developing clinical infections (35). Another explanation is that most of the patients had multiple comorbidities (e.g., diabetes mellitus and renal diseases) resulting in prolonged hospitalisation, possibly predisposing them to CPE colonisation and subsequent CPE infection.
Antimicrobial susceptibility testing in our study con rmed resistance to piperacillin/tazobactam, cipro oxacin and meropenem in all the CPE isolates. Moreover, only 17.8% of the CPE strains isolated from the patients were susceptible to fosfomycin. Colistin is presumably the most active agent against up to 89% of the CPE isolates from our study. The evidence from a cohort study (26) and systematic review (13) on antibiotic therapy in CPE infections revealed that combination therapy is probably more effective than monotherapy. Therefore, the antibiotic therapy recommendation for CPE infections at Siriraj Hospital is combination therapy, with colistin acting as the backbone of the regimen.
In our study, the mortality rate was 47.6% for CPE-infected patients and 33.3% for patients colonised with CPE. The difference was not statistically signi cant. However, attributable mortality is di cult to assess because both groups already had high overall mortality and the sample size was small.
Although KPC-producing Enterobacteriaceae are reported to have spread rapidly over the last decade, their prevalence in Thailand remains very low (6,7). Notably, the CPE prevalence was 1.4%, and bla  and bla IMP-14 were the only carbapenemase genes detected among the CPE isolates identi ed at Siriraj Hospital during 2009 to 2011 (6). However, the incidence of CPE bacteraemia has signi cantly increased from < 1% in 2011 to 3.8% in 2017 (6,36). The main CPE identi ed herein was CPKP, which carried one carbapenemase gene (bla NDM-1 or bla  ) and at least one other bla gene. bla OXA-48 -like genes were the most common carbapenemase genes, with bla OXA-232 detected in 78% of the isolates. We also found that 46% of the bla OXA-232 isolates also carried bla NDM-1 , a nding consistent with that reported previously in Thailand (7). This highlights that isolates with bla OXA-48 -like genes continue to be a problem in Thailand.
The previously reported cases of bla OXA-232 -harbouring K. pneumoniae were mainly serotypes ST14 and ST231 (37)(38)(39), while ST16 and ST231 were the dominant epidemic serotypes in our study. Thus, ST231 may be a high-risk, carbapenem-producing K. pneumoniae clone actively disseminating across Southeast Asia, with related outbreaks being reported in Switzerland (40). We found that all 36 bla OXA-232 -harbouring isolates were present on a small ColKP3 plasmid in our dataset of E. coli and K. pneumoniae genomes, a nding concordant with that from a previous report (37). Interestingly, IncFIB(pQil) plasmids were identi ed in all ST-231 K. pneumoniae isolates in our study, and both ColKP3 and IncFIB(pQil) are known to carry bla OXA-232 and bla TEM-1 (41,42). These ndings con rm that both plasmids, IncFIB(pQil) and ColKP3, are often found in clinical isolates and contain multiple AMR genes, as has been previously reported (30).
Among our CPKP isolates, we found two virulence loci that have been previously associated with invasive diseases: ybt and iuc, encoding the siderophores yersiniabactin and aerobactin, respectively (8,12,43). Ybt was found in almost all of our CPKP isolates (97.4%), and all ybt loci detected in the CPKP genomes were associated with an ICEKp structure located in a chromosomal region (12). ICEKp, an integrative conjugative element, is self-transmissible and occasionally contains virulence factors such as ybt and iro (12). Thus, ICEKp is considered to be an important mediator of pathogenicity in K. pneumoniae (12).
Regarding, VAP was the most common CPKP infection and all isolates from patient with VAP had ybt, these ndings raise the interesting possibility that yersiniabactin siderophore can promote respiratory tract infection as previous studies (44,45). This is the rst identi cation of iuc5 in ST231-CPKP isolates in Thailand and Southeast Asia; otherwise, iuc5 has only been found in ST231-CPKP from India (46). Some KL types (e.g., KL1, KL2, KL5 and KL57) are considered to be hypervirulent variants of K.
pneumoniae and are associated with invasive diseases (47). Our results show that there were at least seven distinct Klebsiella capsule genes/loci present among the 39 isolates, from which KL51 was the most common. However, only 5 out of 39 of our isolates were KL2 types and all of them belonged to ST14, a non-hypervirulent clone usually encountered in hospital-acquired infections (48). Our results also revealed that wzi alleles were associated with the expected MLSTs more than with KL types.
Of note, when we integrated the clinical information with the bacterial genomic data, we identi ed ST231-CPKP as the most common pathogen in CPE-infected patients, 6 out of 11 of which had invasive diseases such as primary bacteraemia and pneumonia. According to our analysis of virulence determinants, ST231-CPKP had the highest virulence score ( Figure 2) and contained iuc5. The iuc locus has been increasingly detected in hvKP over the last couple of years and is considered to be one of the most prominent features of invasive isolates (46,49). Second, four out of ve of the colistin-resistant CPKP isolates belong to ST16. ST16-CPKP with colistin resistance was found in CPE-infected patients presenting with VAP and three of these patients died while in hospital. Therefore, ST16-CPKP is considered to be one of the more clinically signi cant clones in our study, as was also reported elsewhere (50). Third, the core SNP-based phylogenetic tree suggests the possibility of a polyclonal outbreak of CPKP, predominantly involving ST231 and ST16 CPKP in Siriraj Hospital between 2015 and 2017. We identi ed two major subclades of CPKP: ST231 (n = 15) and ST16 (n = 14). Lastly, ST101 and ST14 were identi ed among the CPE-infected patients, something previously reported in South and Southeast Asia (46,51).
Several limitations in our study require mentioning. First, it was not possible to identify the risk factors potentially associated with poor outcomes because of the small sample size that was available in this study. We have probably overestimated the true prevalence of CPE colonisation because there was a lack of routine screening for CPE in patients on admission during the study period. Nevertheless, our results suggest some clinical correlations between the clinical outcomes of patients with CPE infections and the genomic analysis of the organisms responsible, and also provide essential epidemiological data that could be used to guide empirical treatment and infection control strategies for CPE patients.

Conclusion
Our study represents the rst report of a genomic epidemiological investigation on CPE among hospitalised patients in Thailand. By analysing resistance and virulence genes in combination with clinical patient information and bacterial genetic diversity, our approach provides important information that can be used to promptly track the emergence and spread of clinically signi cant isolates, suggest empirical antibiotics, assess mechanisms of drug resistance, and guide infection control strategies for CPE. Future larger-scale studies are needed to determine the true prevalence of CPE and to identify the risk factors for CPE acquisition and their impact on treatment outcomes in Thailand. Authors' contributions AB, VT, and DF conceived the study. AB, LB CG, and KN undertook data analysis with input from KL and TT. EJ provided additional input into the framing of the results. AB, EJ, VT, and DF produced the rst draft of the manuscript. All authors contributed to the nal draft.