A genomic epidemiology study of multidrug-resistant Escherichia coli, Klebsiella pneumoniae and Acinetobacter baumannii in two intensive care units in Hanoi, Vietnam

Study design, setting and participants

This prospective observational cohort study was conducted in two hospitals, the National Hospital for Tropical Diseases (NHTD) and Bach Mai Hospital (BMH) in Hanoi, Vietnam, between June 2017 to January 2018. All adult patients admitted to the ICUs of the two hospitals were eligible for inclusion in the study. Screening specimens (stool/rectal swabs, urine, skin/wound swabs and sputum/tracheal aspirates) were collected from ICU patients on admission, on discharge and weekly during their ICU stay. Environmental samples were collected on a monthly basis.

Laboratory methods

Specimens were cultured on selective media to identify extended-spectrum beta-lactamase producers and carbapenem-resistant organisms. Target organisms (Escherichia coli, Acinetobacter baumannii and Klebsiella pneumoniae) were identified using MALDI-TOF MS and stored at -80 °C prior to be being shipped to the University of Cambridge, United Kingdom, where they underwent antimicrobial susceptibility testing and DNA extraction.

Sequencing and bioinformatic analysis

DNA extracts were transferred to the Wellcome Sanger Institute for library preparation and sequencing. Further details of laboratory methods including read quality control, genome assembly, phylogenetic analysis, antibiotic resistance gene detection, multilocus sequence typing (MLST) and transmission cluster analysis are available in the Supplementary Materials.

Samples included in the study

Between June 2017 to January 2018, a total of 3,367 isolates were cultured, comprising Escherichia coli (n=765), Klebsiella pneumoniae (n=1,372) and Acinetobacter baumannii (n=1,230). Thirty-one isolates were excluded from the analysis because of poor assembly quality. A further 150 isolates were excluded because of suspected inter-species contamination, and 33 isolates were excluded because of suspected intra-species (strain-level) contamination (Supplementary Figure 2). Thus 3,153 isolates (93.6%), comprising 2,901 isolates from 369 patients, and 252 environmental isolates, passed quality filtering and were included in the final analyses.

Clinical characteristics

3,153 bacterial isolates and 369 patients were included in the study. The number and type of samples and the participant baseline characteristics and outcomes are summarised in Table 1 and Supplementary Figure 3. The number of samples collected was higher, and the median length of ICU stay were longer at NHTD than BMH.

All of the 369 patients were colonised or infected with one or more of the three species (E. coli, A. baumannii and K. pneumoniae): 146 patients (40%, 55 at BMH and 91 at NHTD) with all three species; 133 patients (36%, 66 at BMH and 67 at NHTD) with two of the three species; and 90 patients (24%, 61 at BMH and 29 at NHTD) only had one species detected.

Both E. coli and K. pneumoniae were isolated primarily from stool/rectal swabs (627/721 [87.0%] and 822/1316 [62.5%], respectively). K. pneumoniae was also isolated from other sites including sputum (325/1316 [24.7%]), urine samples (63/1316 [4.8%]) and skin swabs (17/1316 [1.3%]). In contrast, A. baumannii isolates were predominantly isolated from sputum (621/1116 [55.6%]), followed by stool/rectal swabs (247/1116 [22.1%]), urine (49/1116 [4.4%]) and skin swabs (36/1116 [3.2%]). A. baumannii also accounted for the highest number of environmental isolates (161/1116, [14.4%]), compared to 6.5% (85/1316) for K. pneumoniae and 2.2% (16/721) for E. coli.

Whole genome sequencing reveals dominant circulating lineages

Phylogenetic trees for each species were constructed to explore lineage diversity. The E. coli isolates were found to be highly diverse, with isolates spread over eight phylogroups and 80 sequence types (STs) (Figure 1). The most prevalent ST was ST648 (phylogroup A; 11.8%), followed by ST410 (phylogroup C; 9.7%), ST617 (phylogroup A; 9.2%), ST131 (phylogroup B2; 7.9%) and ST1193 (phylogroup B2; 7.4%). Overall, 33 of the 80 STs only had one representative isolate in this dataset.

• Download figure
• Open in new tab

Figure 1: Whole genome phylogenies for [A] E. coli, [B] A. baumannii, and [C] K. pneumoniae:

recombination-filtered core-SNP trees with mid-point root. Tree metadata includes (from left to right column beside trees): MLST, source and hospital. Outermost purple bars indicate environmental isolates. Branches corresponding to E. coli phylogroups are coloured accordingly. Main STs are highlighted in the image using the pale-yellow boxes.

In contrast, the K. pneumoniae and A. baumannii isolates appeared to have a limited number of dominant lineages. More than 80% of the K. pneumoniae isolates were from one of five STs, including ST15 (n=34%), ST16 (n=20%), ST656 (n=12%), ST11 (n=11%) and ST147 (n=7%). The majority of A. baumannii were global clone 2 (GC2) (n=832, 74.6%) (19) and mainly belonged to ST2 (n=48%) and ST571 (n=24%) (based on the Pasteur scheme). There did not appear to be any specific relationship between STs and hospitals, with all of the major ST lineages represented in both ICUs.

In order to gain broader insight into the lineages, we selected globally representative strains to contextualise our dataset. Addition of these global representatives into the E. coli phylogeny showed that most isolates belonged to a globally diverse set of STs that were not unique to Vietnam, but found across parts of North America, Europe and Asia (Supplementary Figure 4). Similarly, several of the major K. pneumoniae lineages were represented globally, particularly ST147, ST11 (mainly from China and the USA) and ST15 (mainly other Asian countries) (Supplementary Figure 5). However, it was also clear that local expansion was prominent, particularly among the ST656, ST16 and ST15 lineages. For A. baumannii, we focused primarily on GC2 isolates (Supplementary Figure 6). There was very little representation of global strains within our dataset, and those that were available consisted mainly of strains from other parts of Asia.

Closer inspection of the global representatives found several strains in each species that were closely related (<5 core SNPs) to isolates in our dataset (Supplementary Table 1). Most of these global representatives were also isolated in Asian countries, with agriculture implicated as a source, based on two closely related E. coli representatives originally isolated from poultry (biosample SAMEA104188722) and a farm worker in Vietnam (biosample SAMEA5277968). There were only two closely related representatives isolated outside Asia, from the United Kingdom and Australia (E. coli ST1193 and ST131, respectively, which are both well-characterised global lineages).

High prevalence of antibiotic resistance genes among majority of isolates

Almost all isolates carried acquired resistance genes belonging to at least three antibiotic classes, with 90% of E. coli, 97% of K. pneumoniae and 41% A. baumannii carrying genes across five antibiotic classes (Figure 2). There were no discernible differences based on sample source or hospital, with the exception of E. coli detected in wound/skin swabs (n=6) which appeared on average to carry resistance to more antibiotic classes. K. pneumoniae isolates tended to fall into one of three “peaks” (Figure 2). This was due to lineage-specific carriage of acquired resistance genes, where ST15 isolates tended to carry resistance to more classes, compared to ST16 which often carried the least. The other three main lineages (ST656, ST11, ST147) fell between these two peaks. The exception was in the environmental samples, where only two peaks were seen and was likely due to very few ST16 isolates being detected in the environment.

• Download figure
• Open in new tab

Figure 2:

Summary of isolates and the number of antibiotic resistance classes separated by species, hospital and sample type.

Resistance to antibiotics classes varied across the E. coli phylogeny, reflective of the diversity of strains within the dataset (Supplementary Figure 7). blaCTX-M genes were found in most E. coli, with blaCTX-M-15 (36%), blaCTX-M-27 (30%) and blaCTX-M-55 (17%) the most prevalent (Table 2). blaKPC-2 (13%) and blaNDM-[1,4,5,7] (24%) were present sporadically across the phylogroups, suggesting independent acquisitions events. Only 4% (n=28) of isolates carried mcr genes conferring resistance to colistin. Again, these seemed to be independent acquisitions, with the exception of an ST206 cluster (phylogroup A; n=11) involving three patients from NHTD.

Conversely, MDR gene presence across the K. pneumoniae isolates appeared consistent with the main lineages, suggesting clonal expansion rather than diverse sampling of the species (Supplementary figure 8). Similar to the E. coli, incidence of blaCTX-M-15 (37.5%) was high, but less so than blaKPC-2 (45%) and blaNDM (54%) (NDM-4 [27.9%], NDM-1 [24.7%] and NDM-5 [1.8%]) (Table 2).

Acquired AMR genes were overall less prevalent among the A. baumannii isolates. Similar to the K. pneumoniae, resistance to specific classes tended to be a feature of each distinct lineage, suggesting clonal expansion (Supplementary Figure 9). The carbapenemase gene blaOXA-23 was present in 83% of the dataset, with blaOXA-58 and blaOXA-72 present at much lower frequencies (5% and 0.2% respectively) (Table 2). The aminoglycoside resistance gene armA was also highly prevalent (76%).

Overall, 133 AMR genes were detected in BM and 154 were detected in NHTD. 49 genes were unique to either hospital (35 in NHTD, 14 in BM), but were only detected at a prevalence of less than 0.1%, suggesting sporadic cases. The remaining 129 genes were the same across both hospitals. Genes with at least 1% prevalence in each hospital were found to be almost identical. There were only five exceptions where the gene prevalence was >1% in BM but <1% in NHTD (blaNDM-4 (0.98%), dfrA12 (0.92%), rmtB1 (0.86%), qnrB6 (0.74%) and blaOXA-181 (0.56%)). We also attempted to discover possible mobile genetic element (MGE)-driven AMR transmission by monitoring fluctuation of AMR genes prevalence over time (further explained in the supplementary results, supplementary figure 10).

Most patients carried several AMR strains

For patients already infected or colonised with E. coli, K. pneumoniae and/or A. baumannii, over half had repeat isolation of the same bacterial species at a later timepoint during their stay (E. coli: 138/270 [51%], K. pneumoniae: 169/294 [57%], A. baumannii: 133/230 [58%]). Of these patients, 60 to 70% had different sequence types (ST) (E. coli: 67%, K. pneumoniae: 68%, A. baumannii: 64%) (Figure 3). For E. coli and K. pneumoniae, the majority of patients only had isolates detected in stool (80% and 54%, respectively). Conversely, most patients with A. baumannii were detected only in sputum (38%), or sputum and stool (32%).

• Download figure
• Open in new tab

Figure 3: Overview of strain diversity, recurrence and source among study patients:

“Patients” refers to the total number of patients in this study that had at least one isolate of that species. Within each species, we evaluated whether patients had (i) only a single isolate for that species, or (ii) multiple isolates. If only a single isolate, we determined whether it was collected on admission to the ICU or after. For multiple isolates, we determined if the patient’s isolates were the same ST (recurrent) or a different ST. We finally looked at how many patients had isolates from one site (urine, swab, stool or sputum) or mixed sites (any combination of sites).

Evidence of extensive transmission between ICU patients

Temporal observation of the isolates found no obvious association of any time point with any ST to suggest an outbreak of a specific lineage. In order to investigate potential clusters within the ICUs at a higher resolution, we examined plausible short-term transmission events using single nucleotide polymorphisms (SNPs).

To identify closely related strains that could indicate recent transmission, we evaluated clusters based on SNP distances across the core genome of each species for this dataset. Given the short sampling period, none of the three major species were likely to acquire more than 1 SNP while in the hospital. As such, we looked at samples with genomic evidence of most recent transmission; zero SNP clusters (follow-up with higher thresholds detailed below). Clusters were defined when they involved >1 patient. Clusters involving a single patient and environmental samples were not included.

Most clusters were detected in K. pneumoniae and A. baumannii isolates, with 71 and 74 clusters representing 38% and 52% of total isolates for that species, respectively (Figure 4). K. pneumoniae had some of the largest clusters, ranging in size from 2 to 79 isolates, while A. baumannii clusters were smaller, between 2 to 33 isolates. Only 22 clusters were detected in E. coli and were generally small (median 3 isolates, range 2 to 9 isolates), representing only 13% of the E. coli dataset.

• Download figure
• Open in new tab

Figure 4: Summary of 0 SNP clusters in all species:

[A] Clusters were defined as (i) multiple patients: samples were derived from at least two different patients, or (ii) same patient: isolates were derived from the same patient, or only a single patient and the environment. Epidemiological evidence to support clusters was defined as (i) confirmed patient overlap: all patient ICU stays overlap with another in the same cluster, (ii) some patient overlap: at least 2 patient ICU stays overlap, and (iii) zero patient overlap between all patients in cluster. [B] Environmental isolates in clusters: clusters were counted if an environmental isolate was found in that cluster. Y-axis = number of clusters. [C] Colonisation vs. infection: clusters were counted if they (a) had only isolates from stool (i.e. colonisation) or (b) had isolates from urine, swabs and/or sputum with or without isolates from stool (i.e. infection). Y-axis = number of clusters.

For all three species, the majority of clusters were detected between patients within a single ICU. Evaluating admission and discharge dates further confirmed patient overlap in these clusters (Figure 4, Supplementary Figures 12 to 17). Additionally, between-ICU clusters were also detected (discussed further below).

Compared to E. coli and K. pneumoniae, A. baumannii clusters were more often associated with environmental isolates (24/74 clusters, Figure 4). The largest A. baumannii cluster within the same hospital ICU involved 33 isolates from eight patients and eight environmental samples (Supplementary Figure 13; ST451, cluster number 13). Admissions for patients in this cluster overlapped with detection of the same strain in the environment, which was also detected in the hospital environment several months later. Only two E. coli clusters contained related environmental isolates. K. pneumoniae environmental isolates were more often found in within-hospital (ICU) clusters (n=11) compared to between-hospital clusters (n=4).

To broadly evaluate infection risk for each outbreak cluster, we determine whether clusters contained (i) stool sample / rectal isolates only, indicating colonisation or (ii) isolates from urine, skin swab or sputum samples, which could represent infection or colonisation at multiple sites. The majority of A. baumannii clusters contained isolates from non-stool / rectal swab samples (Figure 4). Conversely, colonisation-only clusters were common for E. coli; the largest E. coli cluster involving both hospital ICUs contained only 5 isolates from four patients, which were all isolated from stool (Supplementary Figure 14; ST617, cluster number 37). K. pneumoniae clusters were a mix, with both colonisation-only clusters and infection clusters.

In addition to suspected within-ICU transmission, we also detected a number of clusters involving patients from both hospital ICUs (Figure 4). The most pronounced example of this was a large ST15 K. pneumoniae cluster involving 79 isolates from 38 patients and 6 environmental samples (Supplementary Figure 18). Most were collected between July to September, with some late occurrences in October and November. All patients from NHTD between July to September had overlapping timelines, consistent with spread within the ICU (Supplementary figure 16; ST15, cluster number 15). Only one patient had no evidence of overlap (ND162) but did cluster with environmental isolates from the same timeframe, indicating a possible environmental source. Patients from BM appeared to have both patient overlap and consecutive acquisitions without patient overlap as time progressed. This suggests that patient-to-patient transmission and also transmission via other routes (e.g. inadequate cleaning before the next patient, transmission via healthcare workers) may have been important factors in the spread of this strain.

The identification of closely related isolates between independently operating ICUs suggested that there may have been a common source located outside the ICU e.g admission to the same location prior to admission to ICU. To determine whether certain lineages were more associated with acquisition within the ICU, we assessed diversity on arrival (i.e. the patients’ first sample) versus diversity within the ICU (all other samples). Based on ST alone, we found a slight increase in diversity in the ICU versus on arrival (Supplementary Figure 19). However, the unique STs recovered in either setting only represented a small portion of the isolates overall. All of the main lineages for each species were found on both admission and within the ICU (Supplementary Figure 19 and 20).

Identification of multiple transmission clusters per patients

Overall, there were 251 patients (representing 68% of the cohort) involved in 167 clusters across the three species collected over the course of this study. 112 patients were only involved in a single cluster during their time in the ICU (Figure 5). However, the remaining 139 patients were involved in at least two clusters, with one patient involved in 12 clusters. For patients with at least two clusters, 20 had clusters from all three species, 94 had clusters from two species and 25 had only one species. Overall, we saw a general trend towards more clusters in a single patient as they spent more time in the ICU ward.

• Download figure
• Open in new tab

Figure 5: Scatterpie showing the number of clusters in patients across all species:

y-axis represents patients from BM or NHTD involved in at least one 0 SNP cluster. X-axis represents length of stay for that patient; one pie is plotted per patient at the duration of their stay. Each circle represents 0 SNP clusters in a single patient. The size of the clusters corresponds to the number of clusters, while the colour corresponds to the species.

To determine if any of our (zero SNP) clusters were potentially derived from a single original cluster predating their time in the ICU, we looked at SNP distances between clusters of the same ST (Supplementary Figure 21). At a threshold of five SNPs, several of the prominent STs within each species formed large clusters, including ST804 in A. baumannii and ST16 in K. pneumoniae. At this threshold, we found 29 clusters in the A. baumannii dataset (originally 74), 23 clusters in the K. pneumoniae (originally 71) and 19 clusters in E. coli (originally 22). By readjusting our analysis per patient using these clusters, we found that slightly more patients (123 versus 112 previously) may have only been involved in a single cluster during their stay (Supplementary figure 22). 128 patients had 2 clusters, with the maximum number of clusters in a single patient being 7 (n=3 patients).