We had previously conducted a retrospective cohort study involving all six multi-disciplinary acute-care public hospitals, providing approximately 80% of inpatient medical care in Singapore (estimated population size, 5.5 million in 2015). From September 6, 2010 to April 28, 2015, 1251 CPE isolates (from 791 patients) confirmed by both genomic and phenotypic methods were collected across participating study sites. Hybrid assembly reconstructed whole genomes for 1088 isolates sampled from 705 patients, for which 1115 closed carbapenemase-producing plasmids were identified (Supplementary Fig. 1). Of the 1088 isolates, 20 isolates carried two different carbapenemase genes on separate plasmids, and 2 isolates carried two different carbapenemase genes on the same plasmid. Details on carbapenemase gene co-carriage can be found in Supplementary Data 1. The majority of isolates were K. pneumoniae (n = 486, 44.7%), followed by Escherichia coli (n = 339, 31.2%) and Enterobacter cloacae (n = 145, 13.3%). One hospital accounted for 67.5% (n = 753) of the isolates, in which K. pneumoniae (n = 323, 42.9%), E. coli (n = 232, 30.8%), and E. cloacae (n = 106, 14.1%) were similarly the predominant three species. The complete distribution of isolates by species and hospital is provided in Supplementary Table 1. E. coli ST131 (n = 64), K. pneumoniae ST14 (n = 50), and E. cloacae ST93 (n = 32) were the three most common sequence-types (ST), accounting for 18.9%, 10.3% and 22.1% of the respective species (Supplementary Table 2).
Differences in epidemiology between study sites could potentially be due to varying capacity per site, ranging from 300 to 1600 beds. Two hospitals were academic medical centers with solid organ and stem cell transplant units and four were teaching hospitals with academic affiliations; these factors, in addition to location of the study site (e.g., proximity to mature housing estates, city centre, or industrial areas), could have some influence on inpatient demographics and case mix.
Across 1115 closed plasmids, 15 carbapenemase genes were detected, most commonly bla (n = 477, 42.8%), followed by bla (n = 434, 38.9%), bla (n = 60, 5.4%), bla (n = 41, 3.7%), bla (n = 22, 2.0%), bla (n = 21, 1.9%), bla (n = 20, 1.8%), bla (n = 18, 1.6%), bla (n = 6, 0.5%), bla (n = 4, 0.4%), bla (n = 4, 0.4%), bla (n = 4, 0.4%), bla (n = 2, 0.2%), bla (n = 1, 0.1%), and bla (n = 1, 0.1%). bla and bla were distributed across all six study sites (Supplementary Table 3). Classification of the 1115 closed plasmids by replicon typing assigned majority (n = 898, 83.8%) of the plasmids to a single plasmid incompatibility (Inc) group; the remaining plasmids were found to carry multiple replicons. The predominant replicon types identified were IncU (n = 475, 42.6%), IncN (n = 304, 27.3%), IncC (n = 100, 9.0%), and IncL/M (n = 79, 7.1%) Fig. 1A and Supplementary Table 4). Majority of the plasmids were relaxase-encoding (n = 1005, 90.1%), for which MOB-typing identified the predominant classes to be MOBP (n = 503, 45.1%), MOBF (n = 381, 34.2%), and MOBH (n = 104, 9.3%) (Fig. 1B and Supplementary Table 5).
Carbapenemase-encoding plasmids were clustered based on pairwise k-mer (21 bp) similarity. To assign plasmids to clusters, we built an undirected similarity network in R with igraph (v1.6.0): each plasmid was represented as a node and an edge was drawn between any two plasmids whose 21-mer Jaccard similarity was ≥ 0.90. Clusters correspond to the connected components of this network (single-linkage grouping). Using this method, 92.1% (n = 1027) of the 1115 plasmids were grouped into 48 distinct clusters named according to size, with the largest plasmid cluster named PC1 and the smallest, PC50 (Fig. 1C). For consistency in the analysis, plasmids that contained two CP genes (bla/bla and bla/bla, n = 2) were included as duplicates to accurately reflect CP-gene specific analyses. Therefore, PC33 and PC47 are not listed as true plasmid clusters, and these four plasmids are categorized as unclustered (Table 1). PC1 comprised 389 closely-related bla-positive plasmids carrying the IncU replicon and accounted for 34.9% of all plasmids and 81.3% of all bla-positive plasmids. Although by replicon typing PC1 is classified as IncU based on the in silico identity of a partial replicon sequence, previous in vitro experiments have shown that the functional replicon of PC1 is trfA and the plasmid is classified as IncPe1. PC2 comprised IncN plasmids that were predominantly bla-positive with the exception of one bla-positive plasmid, and accounted for 284 (25.5%) of all plasmids. PC2 accounted for 67.2% of all bla-positive plasmids. In comparison, only 37 (3.3%) plasmids were represented in the third-largest cluster, PC3 (Table 1). We applied the same plasmid clustering approach to a different, publicly-available dataset of circularized bla plasmids (n = 154) from an institutional collection of mostly clinical carbapenem-resistant isolates systematically collected from 2002 to 2020. This dataset included some environmental isolates but did not include patient surveillance isolates, and showed a similar pattern in which a subset of highly conserved plasmid clusters dominate the cohort (Supplementary Fig. 2).
PC1 was distributed across 10 species, predominantly K. pneumoniae (42.9%, n = 167) and E. coli (33.2%, n = 129). Among K. pneumoniae isolates carrying PC1, there were 60 assigned STs, most commonly ST231 (6.7%, n = 11). Among E. coli isolates carrying PC1, there were 57 assigned STs, the most common being ST131 (11.9%, n = 14). PC2 was also distributed across 10 species, predominantly E. coli (38.7%, n = 110) and K. pneumoniae (27.5%, n = 78). Among E. coli isolates carrying PC2, there were 31 known STs, with ST131 (43.2%, n = 45) being the most common. Among K. pneumoniae isolates carrying PC2, there were 28 known STs, the most common being ST34 (15.8%, n = 12) (Supplementary Table 6).
To contextualize the plasmids in our dataset in terms of known plasmid diversity, we performed BLAST on all 48 plasmid clusters against the PLSDB database. Thirty-five plasmid genotypes matched (weighted average identity >99%) a previously described plasmid outside of our dataset (Supplementary Table 7), and 13 were putative novel plasmids. We also compared our plasmid clustering approach with other approaches, such as using MOB-suite to assign MOB-suite clusters and COPLA to define plasmid taxonomic units. Our clustering approach demonstrates good agreement with MOB-suite, which also utilizes an alignment-free method. All of our plasmid clusters are assigned to a single MOB-suite cluster (with the sole exception of PC11), but MOB-suite clusters could be further subdivided into multiple closely-related plasmid clusters, suggesting that our method is higher in resolution (Supplementary Fig. 4). COPLA was unable to classify 35.7% of our clustered plasmids into plasmid taxonomic units, suggesting previously unsampled diversity in our dataset (Supplementary Fig. 5).
To investigate the relative impact of vertical versus horizontal transmission on the dissemination of carbapenemase-encoding plasmids in the study population, we analysed isolates carrying bla-positive PC1 plasmids and bla-positive PC2 plasmids, as designated by our k-mer based clustering method. A carbapenemase-encoding plasmid was considered vertically acquired by an isolate if the host met pairwise clonal linkage criteria with an earlier isolate. Briefly, a pair of isolates were determined to be clonally linked if they shared the same ST-cluster, same carbapenemase gene allele and had a pairwise single-nucleotide polymorphism (SNP) count (based on the recombination-filtered core gene alignments) below the BEAST-derived mutation rate threshold, assuming a Poisson distribution for the accumulation of mutations. A carbapenemase-encoding plasmid was considered horizontally acquired by an isolate if the host was not clonally linked to another isolate.
Of the 389 isolates carrying a bla-positive PC1 plasmid, 236 (60.7%) putatively acquired the plasmid via horizontal transmission, whereas 153 (39.3%) putatively vertically inherited the plasmid from a parent cell. Vertical transfer was unable to account for the spread of bla-positive PC1 plasmids across the five institutions where it was detected. Of 60 clonal lineages, most were limited to a single hospital, and only five were found across two hospitals (Fig. 2A and Supplementary Table 8).
Of the 284 isolates carrying a bla-positive PC2 plasmid, 168 (59.4%) putatively acquired the plasmid via horizontal transfer, whereas 115 (40.6%) putatively vertically inherited the plasmid from a parent cell. Similarly, vertical transfer was unable to account for the spread of bla-positive PC2 plasmids across all six study sites. Of 38 clonal lineages, only seven spread beyond one hospital: six spanned two hospitals and one spanned three (Fig. 2B and Supplementary Table 9).
Similar trends were observed for isolates carrying plasmids from the next two largest clusters, bla-positive PC3 (n = 37) and bla-positive PC4 (n = 31), as well as from PC7 (n = 25), the next-largest bla-positive plasmid cluster after PC1, and PC5 (n = 29), the next-largest bla-positive plasmid cluster after PC2, in which clonal lineage-dependent vertical transmission limited carbapenemase gene spread across fewer institutions compared to plasmid-mediated horizontal transmission of carbapenemase genes (Supplementary Fig. 6). Five E. coli ST162 isolates (from three unique patients and two hospitals) were found to co-carry bla-positive PC1 and bla-positive PC2 plasmids.
Our data suggest that plasmids are important drivers in the mobilization of carbapenemase genes between species and between geographic niches (different institutions), and also in the persistence of carbapenemase genes over time in the population. Based on the temporal trends of plasmid clusters that each accounted for >1% of all plasmids (n = 12), many plasmid genotypes are stably maintained in the population for years, although apart from PC1 and PC2, none of the other plasmids showed signs of potentially moving towards hyperendemicity within our period of study (Supplementary Fig. 7).
PC1 and PC2 were considered evolutionarily successful plasmids due to their high prevalence, with each accounting for more than 25% of plasmids in the cohort (Table 1). The PC1 core genome of 97 genes that were present in 95% of PC1 plasmids was found to be highly conserved (> 90%) in PC7 and PC43 compared to other bla-positive IncU plasmid genotypes (Fig. 3A). The PC2 core genome of 52 genes that were present in 95% of PC2 plasmids was found to be most highly conserved (> 90%) in PC20 and PC49 compared to other bla-positive IncN genotypes (Fig. 3B). Gene cluster organisation of representative PC1 (Fig. 3C) and PC2 (Fig. 3D) plasmids, visualized by Clinker, emphasizes the structural conservation of a distinct backbone of core gene loci, and reveals divergence from the PC1 and PC2 genetic settings through insertion events.
The bla-associated PC1 genotype, a hybrid of pSA20021456.2-like plasmids (GenBank accession no. CP030221) and pKPCAPSS-like plasmids (GenBank accession no. KP008371), appears to be unique to Singapore to date, and was previously characterized as pKPC2_sg1 (GenBank accession no. MN542377). pKPC2_sg1 was demonstrated in vitro to impose low fitness costs, have high conjugation frequencies and high retention rates in multiple Enterobacterales species. PC7 and PC43 also appear to be unique to our dataset (Supplementary Table 7).
In contrast, bla-associated PC2 has been reported previously as pNDM-ECS01 (GenBank accession no. KJ413946) and has been widely linked to high-risk clones such as E. coli ST131 and K. pneumoniae ST11 and ST15. PC20 and PC49 have also been previously identified outside of Singapore (Supplementary Table 7).
To investigate if there were any specific genes that could be associated with the likelihood or failure to achieve hyperendemicity, we compared the full gene complement of PC1 (34.9% of all plasmids) with that of plasmids with high core genome similarity but significantly lower prevalence (Supplementary Fig. 3), such as PC7 (2.2% of all plasmids) and PC43 (0.2% of all plasmids) (Fig. 4A). Likewise, we compared the full gene complement of PC2 (25.5% of all plasmids) with that of PC20 (0.4% of all plasmids) and PC49 (0.2% of all plasmids) (Fig. 4B). It could be ruled out that the less prevalent plasmids were poorly represented in the cohort as a result of late emergence towards the end of the sampling time frame (Supplementary Fig. 8).
PC7 plasmids harbored additional genes not found in PC1 (Supplementary Fig. 9). Comparison of representative PC1 (71,855 bp) and PC7 (87,271 bp) plasmids show two distinct inserted genomic regions in PC7 (Fig. 4A). Notably, the 8020 bp region (Inserted Region 1) encoded genes belonging to the Mer operon involved in mercury detoxification and the 6574 bp region (Inserted Region 2) encoded genes belonging to the FrmRAB operon, involved in formaldehyde detoxification (Supplementary Table 10). Similarly, plasmids in PC43 (97,516 bp) contained the arsR, copG, and srpC genes involved in heavy metal detoxification (Supplementary Table 11).
Likewise, PC20 and PC49 plasmids also contained accessory genes not present in PC2 (Supplementary Fig. 9). Comparison of representative PC2 (41,183 bp) and PC20 (50,744 bp) plasmids show two distinct inserted genomic regions in PC20 (Fig. 4B), although only one of the regions (8494 bp) encoded known genes -- belonging to the Ars operon and involved in arsenic detoxification (Supplementary Table 12), as well as genes related to integration and excision. Comparison of representative PC2 and PC49 (59,567 bp) plasmids show a single inserted genomic region in PC49 (Supplementary Table 13).
Based on 85% coverage and identity, of isolates carrying PC1, the mercury detoxification region was found in the chromosome of 0.26% (n = 1) and in the carbapenemase-negative plasmid compartment of 0.51.% (n = 2). The formaldehyde detoxification region was found in the chromosomes of 12.3% (n = 48) and in the carbapenemase-negative plasmid compartment of 9.5% (n = 37) of PC1 isolates. Similarly, among isolates carrying PC2, the arsenic detoxification region was found in the chromosome of only 12.3% (n = 35) and in the carbapenemase-negative plasmid compartment of only 16.5% (n = 47), suggesting that these detoxification genes were not essential for the survival of bacteria harboring the predominant carbapenemase plasmids. The presence of heavy metal and organic contaminants in the hospital environment has been documented. The detoxification genes found in patient isolates could have been acquired horizontally from environmental bacteria, where it may have conferred a survival advantage. Biocide and metal resistance genes are also known to co-occur with antibiotic resistance genes on the same plasmid.
We next examined plasmid evolution along the same clonal lineage between patients and across multiple species within the same patient. bla-positive plasmids were grouped into 10 plasmid clusters based on k-mer similarity (Table 1). A clonal transmission cluster comprised index isolates that met pairwise clonal linkage criteria with at least one earlier index isolate from another patient. An index isolate was defined as the first-detected isolate carrying a unique carbapenemase gene in a patient during the study period. More than one index isolate (carrying different carbapenemase genes) could be associated with a patient if they shared the same date of culture. Only index isolates were considered for construction of clonal transmission clusters to ensure that clusters reflect between-patient transmission events. PC1 plasmids were associated with 14 clonal transmission clusters across four species, while the remaining bla-positive plasmid clusters were collectively associated with only six clonal transmission clusters across two species. bla-positive plasmids were grouped into 20 plasmid clusters (Table 1). PC2 plasmids were associated with 19 clonal transmission clusters across six species, compared with 10 clonal transmission clusters across three species for all other bla-positive plasmid clusters. We found evidence for carriage of bla and bla genes on multiple plasmid genotypes within the same clonal lineage, reflecting dynamic alterations to plasmid structures on which carbapenemase genes are localized, or the transfer of carbapenemase genes to non-carbapenemase plasmids. However, most of the clonal transmission clusters with more than three isolates arose from an isolate carrying either bla on a PC1 plasmid or bla on a PC2 plasmid, suggesting that PC1 and PC2 genotypes are well-adapted for stable propagation of bla and bla, respectively, during inter-patient clonal spread (Fig. 5A, B). This is exemplified in the largest bla clonal transmission cluster (Clonal cluster 1) comprising 19 E. cloacae ST93 isolates, where putative recombination-driven changes to the bla plasmid resulting in its departure from the predominant PC1 genome were not preserved, and bla reverted to localization on a PC1 backbone in subsequent clones.
Twenty-two individuals (of 705 patients) were determined to be the source of five or more isolates. In 20 of 22 patients, the same plasmid was found in two or more different species; in five of 22 patients, the same plasmid was found in three or more different species. The median number of species in which PC1 and PC2 were found were 2 (IQR, 2-2 and IQR, 1-2, respectively). For remaining PCs, the median number of host species was 1 (IQR 1-1). Our data suggests that in the context of a singular human host, bla and bla are more stably maintained over time and across multiple species (and STs) in PC1 and PC2 plasmids, respectively, compared to other genetic settings (Fig. 6).