Genomic Insights Into the Mechanism of Carbapenem Resistance Dissemination in Enterobacterales From a Tertiary Public Heath Setting in South Asia

Abstract Summary 10.6% patients were CRE positive. Only 27% patients were prescribed at least 1 antibiotic to which infecting pathogen was susceptible. Burn and ICU admission and antibiotics exposures facilitate CRE acquisition. Escherichia coli ST167 was the dominant CRE clone. Background Given the high prevalence of multidrug resistance (MDR) across South Asian (SA) hospitals, we documented the epidemiology of carbapenem-resistant Enterobacterales (CRE) infections at Dhaka Medical College Hospital between October 2016 and September 2017. Methods We enrolled patients and collected epidemiology and outcome data. All Enterobacterales were characterized phenotypically and by whole-genome sequencing. Risk assessment for the patients with CRE was performed compared with patients with carbapenem-susceptible Enterobacterales (CSE). Results 10.6% of all 1831 patients with a clinical specimen collected had CRE. In-hospital 30-day mortality was significantly higher with CRE [50/180 (27.8%)] than CSE [42/312 (13.5%)] (P = .001); however, for bloodstream infections, this was nonsignificant. Of 643 Enterobacterales isolated, 210 were CRE; blaNDM was present in 180 isolates, blaOXA-232 in 26, blaOXA-181 in 24, and blaKPC-2 in 5. Despite this, ceftriaxone was the most commonly prescribed empirical antibiotic and only 27% of patients were prescribed at least 1 antibiotic to which their infecting pathogen was susceptible. Significant risk factors for CRE isolation included burns unit and intensive care unit admission, and prior exposure to levofloxacin, amikacin, clindamycin, and meropenem. Escherichia coli ST167 was the dominant CRE clone. Clustering suggested clonal transmission of Klebsiella pneumoniae ST15 and the MDR hypervirulent clone, ST23. The major trajectories involved in horizontal gene transfer were IncFII and IncX3, IS26, and Tn3. Conclusions This is the largest study from an SA public hospital combining outcome, microbiology, and genomics. The findings indicate the urgent implementation of targeted diagnostics, appropriate antibiotic use, and infection-control interventions in SA public institutions.

Medical College Hospital (DMCH). The combination of whole-genome-based analysis with rigorous epidemiological data provides a powerful spatiotemporal assessment to explore the mechanisms and drivers of AMR in a Bangladeshi hospital setting.

Study Design, Hospital Setting, Participants, and Sampling
We performed an observational cohort study at DMCH, the largest public hospital setting of Bangladesh, containing 2600 allocated beds, from October 2016 to September 2017. This study was approved by the Ethical Review Committee of DMCH (Supplementary Methods) [14].
Specimens referred to the DMCH microbiology laboratory for culture and sensitivity based on local physicians' clinical judgment on suspected infections were included [15]. Participants' demographic and clinical information was recorded (Supplementary Methods). All isolates recovered at the DMCH microbiology laboratory were transferred to Cardiff University and investigated further (n = 643) ( Figure 1). Patients were enrolled for this study if at least 1 of their specimens was positive for Enterobacterales.
were considered as CRE cases. Any patient with a positive CSE culture was regarded as a CSE case.

Whole-Genome Sequencing
All Enteribacterales isolated were sequenced on the Illumina MiSeq platform (Illumina, Inc, San Diego, CA, USA) and a set of bla NDM-positive isolates by minION sequencing (Oxford Nanopore Technologies, Oxford, UK). Details of the bioinformatics analysis are described in the Supplementary Methods. Briefly, the Kmer database (v3.0.2) (available at the Center for Genomic Epidemiology [CGE]) was deployed for species identification, the Clermont phylotyping for Escherichia coli (v1.4.0), Kaptive (v0.7.3) for Klebsiella pneumoniae capsular typing. Comprehensive Antibiotic Resistance Database (CARD) and PlasmidFinder were deployed for antimicrobial resistance genes (ARGs) and plasmid replicon types with a cutoff of ≥95% coverage and ≥95% identity, respectively, using ABRicate (database for mass screening of contigs for antimicrobial resistance or virulence genes) (v0.9.7). Multilocus sequence type (MLST) was assigned based on 7 loci MLST databases in CGE (v2.0.0), where appropriate. Time-calibrated evolutionary analysis was performed using the BEAST package (v1. 10.4) to estimate the date of the most recent common ancestor (MRCA).

Scrutinizing "High-Risk" Clones for Carbapenem Resistance in Bangladeshi Hospital
The clonal relatedness of isolates was assessed using coregenome alignment following clustering for the presence and absence of genes, MLST profiling, and pairwise single nucleotide polymorphism (SNP) distances. A cutoff of ≤10 SNPs combining with epidemiological information was used to define possible clonal transmission clusters using the R library iGRAPH [19]. Clusters were removed if they did not contain any carbapenemase producer.

Statistical Analysis
Binary logistic regression was used to analyze the association of CRE with categorical variables of interest. The P values were  adjusted by the Benjamini-Hochberg procedure when repetitive variable types were tested. A Cox proportional hazards model was used to compare all-cause in-hospital 30-day mortality between CRE and CSE cases. Patients discharged alive or with an in-hospital mortality over 30 days were used as the competing variable for outcome analysis. Statistical analyses were conducted using IBM SPSS (v26; IBM Corporation) and Tableau (v2020.4).
Excluding patients discharged against medical advice, allcause in-hospital 30-day mortality was significantly associated with CRE cases, occurring in 50 of 180 (27.8%) CRE and 42 of 312 (13.5%) CSE cases (P = .001). Significant associations were also observed after adjusting for the confounders (Table 3). No significant association of mortality was observed with CRE for the cohort of patients with Enterobacterales bloodstream infections (BSIs) ( Table 4).
Based on the available data, only 27% (144/534) of the patients were prescribed at least 1 antibiotic to which their infecting pathogens were susceptible (Supplementary Table 10); however, data were not available regarding dosage, duration, and indication of antibiotics therapy.

Clonal Transmission of Carbapenem Resistance
We identified 5 clusters of E. coli (EC1, EC2, and EC4 to EC6), 12 of K. pneumoniae (KP1 to KP12), and 1 of E. cloacae complex (EnC1) using a 10-SNP threshold between isolates with common carbapenemase alleles and none of cluster-contained isolates outside DMCH included in our analysis ( Figure 6). We found a linkage between isolates in the clusters at a ≤2-SNP threshold using overlapping of patients' hospital stays and common wards of isolation. These connections were observed in the burns unit, fistula unit, ICU, neonatal ICU, and urology. The largest cluster was KP12 (ST23) followed by KP8 (ST15) ( Figure 6C). The dated phylogenetic tree of major clones revealed that the putative transmission clusters were predicted to have been introduced into DMCH between 2013 and 2016 (Supplementary Table 13). Values in parenthesis indicate row percentage. Patients who were discharged against medical advice (n = 40) and outlier cases (n = 2) (hospital stay .100 days from "time from infection" to outcome) were excluded from the outcome analysis. Abbreviations: CI, confidence interval; CRE, carbapenem-resistant Enterobacterales; CSE, carbapenem-sensitive Enterobacterales; SHR, subdistribution hazard ratio. a A Cox proportional hazards model was fitted with time points "time from Enterobacterales isolation to outcome" as "time-to-event" and "time from admission to Enterobacterales isolation" as covariates.
b Confounders such as age, gender, admission to burn unit and ICU, and exposure to amikacin, meropenem, levofloxacin, and clindamycin were adjusted to understand the changes in P values (level of statistical significance) in the adjusted model. Values in parenthesis indicate row percentage. Patients who were discharged against medical advice (n = 40) and outlier cases (n = 2) (hospital stay .100 days from "time from infection" to outcome) were excluded from the outcome analysis. A Cox proportional hazards model was fitted with time points "time from Enterobacterales isolation to outcome" as "time-to-event" and "time from admission to Enterobacterales isolation" as covariates.

Table 4. Cox Proportional Hazards Models to Analyze the Impact of Carbapenem Resistance and Mortality Among the Patients with Positive Blood Culture of Enterobacterales
Abbreviations: CI, confidence interval; CRE, carbapenem-resistant Enterobacterales; CSE, carbapenem-sensitive Enterobacterales; SHR, subdistribution hazard ratio.
NDM-1-positive plasmids were divided into 8 groups (designated as NIG1 to N1G8) based on the genetic environment around bla NDM-1 . The genetic structure Tn125 (bla NDM-1 -ble-trpF-dsbD-cutA-groES-groEL-IS91), bordered by intact ISAba125 at the upstream and downstream, was observed in plasmids of IncC (n = 5) (Figure 8). The insertion sequence ISAba125 was absent at the downstream of the plasmids of N1G2 to N1G8 and an incomplete ISAba125 was present at the upstream among plasmids of N1G3 to N1G8. Variation in genes in the conserved region was observed among the plasmids of N1G4 to N1G8. Plasmids of N1G4 were flanked by Tn3 at both 3 ′ and 5 ′ ends in the same orientation (Figure 8). tree generated from E. coli genomes belonging to ST167. The total number of isolates in this analysis was 97. Closely related isolates from other STs (ST10, ST1702, and novel allele) identified by core-genome phylogeny and pairwise SNPs count (if isolates differed by ≤100 SNPs from any isolate of ST167) were included in this analysis. (B) Phylogenetic tree generated from K. pneumoniae genomes belonging to ST15. The total number of isolates in this analysis was 54. Closely related isolates from a novel allele identified by core-genome phylogeny and pairwise SNP count (if isolates differed by ≤100 SNPs from any isolate of ST15) were included in this analysis. Putative transmission clades (0-10 SNP differences) are highlighted in green. MRCA and clock rate are stated in Supplementary Table 13. Isolates retrieved from the NCBI for the phylogenetic analysis in this figure are shown in Supplementary Table 17. Abbreviations: DMCH, Dhaka Medical College Hospital; ICU, intensive care unit; NCBI, National Center for Biotechnology Information; NDM, New Delhi metallo-beta-lactamase; NICU, neonatal intensive care unit; OPD, outpatient department; PSU, pediatric surgery; SNP, single nucleotide polymorphism; ST, sequence type.

DISCUSSION
AMR collaborators provide a sobering analysis on the burden caused by common MDR/extremely drug-resistant extensively drug-resistant (XDR) infections and a warning that, as a global community, we are rapidly surrendering any advantage we had on treating infections such as pneumonia and sepsis. Furthermore, AMR collaborators highlight significant gaps, not least from low-and middle-income countries (LMICs), and advocate the acute need for large LMIC clinical studies that combine detailed microbiology (and genomics) with recorded outcome data [20]. In this study, we present the largest dataset comprising clinical outcome, microbiology, and genomic data from SA and, in particular, based in a large public hospital where such datasets capture different socioeconomic cohorts from previous studies (Supplementary Table 2). Public hospitals in SA are grossly oversubscribed (typically, 4-5 times the number of inpatients/beds), antibiotics are delivered empirically, and a limited number of clinical specimens are only sent for culture sensitivity (Supplementary  Table 16) [21,22]. The problem of AMR is considerably higher in health settings with unsubscribed antimicrobial usage [23]. Our data revealed that bla NDM , bla OXA-181 /bla OXA-232 , and bla KPC-2 were main carbapenem resistance determinants in CRE in Bangladesh. New Delhi metallo-beta-lactamase (NDM)-positive plasmids except for IncX3 commonly coharbored multiple ARGs (Supplementary Figure 6). The association of CRE acquisition with the usage of multiple antibiotic classes may be explained by co-selection in hospital settings, both of MDR CRE clones and their key MDR plasmids (Table 2) [9,23,24]. Moreover, several factors, such as introduction of artificial devices, long-term antibiotic exposure, prolonged hospital stays, and clinical comorbidities, can be responsible for CRE acquisition among patients in burn units and the ICU [25][26][27]. The ancestral sequence at each node including the root was inferred using pyjar and the pairwise SNP distance between the roots and each isolate was calculated using pairsnp (v0.0.7). Pairwise SNPs between isolates were generated using pairsnp (v0.0.7). (C ) Diagram representing number of linkages among the isolates in the clusters by the 0-to 2-SNP threshold, aligning with epidemiological data. Isolates differed by 0 to 2 SNPs without overlapping of pertinent patients' hospital stay are represented as a common group in the figure using Tableau (v2020.4). Abbreviations: HDU, high dependency unit; ICU, intensive care unit; NDM, New Delhi metallo-betalactamase; NICU, neonatal intensive care unit; SNP, single nucleotide polymorphism; ST, sequence type.  Previous reports suggest that CRE was associated with a 3-fold greater mortality than CSE cases [2][3][4]. This study recorded substantially higher mortality with CRE (27.8%) than CSE cases (13.5%) (P = .001), including a cohort with positive culture of Enterobacterales from both sterile (blood) and nonsterile sampling sites (wound swabs, tracheal aspirates, etc).  Given the level of statistical significance in adjusted models, it was possible that comorbidities might influence mortality among patients in the ICU and burn units (Table 3) [25][26][27]. Worse patient outcomes are invariably associated with limited therapeutic options [2,4,28,29]. We demonstrated that 73% of the patients did not receive at least 1 appropriate antibiotic (Supplementary Table 10). However, this study was unable to conclude whether ineffective antibiotic therapy or any other confounders influenced mortality among the patients with CRE due to the limited number of BSIs and limited clinical information (eg, comorbidities or antibiotic therapy).
To date, the presence of bla NDM in different clonal lineages has mainly been reported from China, Europe, or the United States and, as such, there is a bias in the global reporting of their geographical distribution [30]. This study documented numerous prevalent clones with bla NDM (ST167, ST648, ST448, and ST405 of E. coli, and ST15, ST23, ST147, and ST16 of K. pneumoniae) (Figures 4 and 5, Supplementary Figures 1-5). Additionally, E. coli ST8346 was recognized as a newly emerging clone carrying bla NDM-1 (Supplementary Figure 2,  Supplementary Table 11).
Based on spatiotemporal analysis, it is possible that CRE clones had been established at DMCH over an extended period of time. The average substitution rates of these clones (3 SNPs/ genome/year) were in line with previous reports (Supplementary Table 13) [31,32]. We observed the presence of common carbapenemases among the isolates differing by ≤10 SNPs, some of which represented tight clades (0-2 SNP threshold), combined with evidence of a common ward of isolation and overlapping of patients' hospital stay (Figure 6), suggesting potential recent transmission or acquisition of clones from a common source. The number of such events was considerably higher among patients with K. pneumoniae infections compared with other species (Figure 6), indicating higher transmissibility of K. pneumoniae [33]. Of particular concern was the spread of bla NDM-5 via a highly virulent K. pneumoniae ST23 (KP12) clone having the KL1 locus [34].
This study documented the plasmid-mediated horizontal dissemination of bla NDM among different species of Enterobacterales, mostly by IncFII and IncX3 (Table 5, Supplementary Figures 9 and 13). IncX3 is widely spread throughout China and SA and associated with bla NDM , particularly, bla NDM-5 [35,36. However we found plasmids of a wide variety of Inc types (IncFII IncFIA IncR IncFIB and IncFII IncFIBpQil], and IncFIB(pQil) and IncFII) harboring bla NDM-5 , yet that the plasmids had identical conserved regions followed by a class 1 integron together flanked by IS26 (Figures  7 and 8). It can therefore be hypothesized that transposition of the IS26-flanked segment occurred via 2-step recombination where IS26 released a DNA segment from a donor plasmid and IS91 facilitated its insertion into the recipient plasmid by rolling circle replication [37,38].
A limitation of this study was the inability to capture all possible Enterobacterales infections due to the practice norms at DMCH in terms of microbiological sampling. These are typical of many public hospitals in SA. However, this study represents the most comprehensive report on the epidemiology and mechanism of CRE in Enterobacterales in SA. This study has also (1) shown the high burden of CRE compared with previously reported studies, (2) provided data on outcome and inappropriate antibiotic use that will inform better antibiotic stewardship programs including antibiotic access and affordability in the public sector, and (3) demonstrated genomic evidence on the clonal spread of virulent CRE, prioritizing infection-control programs in limited financial settings. While this study was conducted in the largest public hospital in Bangladesh, many of these findings can be extrapolated across SA, which encompasses a population of nearly 2 billion and signals the need for greater engagement and targeted investment.