Introduction

The cationic polypeptide colistin (also known as polymyxin E) is used as a last-line treatment for life-threatening infections caused by multidrug-resistant and carbapenem-resistant Gram-negative bacteria1,2. Colistin works by disrupting the lipopolysaccharide layer of the cytoplasmic membrane, causing increased permeability and leakage of intracellular contents and consequently leading to cell death3. Discovered in 19474, colistin was abandoned from clinical use in the 1970s because of its nephrotoxicity, adverse neurological side effects, and low renal clearance5. However, it has been reintroduced in the 2000s as a last-resort or salvage treatment for severe infections caused by multidrug-resistant bacteria6, aggravated by the dearth of alternative antibiotic options. Colistin has also remained in use for therapeutic and prophylactic care in livestock farming and aquaculture settings for many decades7. However, increasing reports of colistin resistance, especially in many members of the Enterobacterales8,9, is troubling.

Bacteria employ multiple resistance mechanisms to defend themselves from colistin toxicity. Traditionally, colistin resistance has been associated with chromosomal mutations and regulatory changes affecting genes such as pmrAB, phoPQ, ccrAB, lpxACD, or mgrB, which modify the outer lipopolysaccharide layer and reduce colistin binding10,11,12,13. However, a major shift occurred in 2016 when plasmid-mediated colistin resistance was first documented in Escherichia coli isolated from intensively farmed pigs in China14. Since then, numerous allelic variants of the first mobile colistin resistance (mcr) gene mcr-1, a phosphoethanolamine-lipid A transferase14, have been reported worldwide15. Currently, there are ten different gene families of mcr (mcr-1 to mcr-10) described in the literature16, with mcr-9 and mcr-10 being the most recently discovered. Different mcr gene families exhibit a non-uniform distribution worldwide, largely influenced by their geographical and ecological sources17,18.

The mcr-9 gene was first reported in 2019 in an in silico screening of Salmonella enterica genomes19. It was identified in S. enterica serovar Typhimurium and carried on an IncHI2 plasmid19. The strain was isolated from a human patient in the United States in 2010 and conferred resistance when introduced into a colistin-susceptible E. coli strain19. In E. coli, subinhibitory concentrations of colistin has been shown to induce the expression of the mcr-9 gene, leading to higher minimum inhibitory concentration (MIC) levels20. This induction is associated with the two-component regulatory system in the outer lipopolysaccharide layer, qseB/C20. In contrast, several tested S. enterica and E. coli strains harboring mcr-9 were susceptible to colistin21. Colistin susceptibility was associated with the absence of the regulatory genes qseB/C in mcr-9-bearing Gram-negative bacteria22. On the other hand, colistin-resistant strains carrying mcr-9 have also shown variable expression levels of the qseB/C regulatory system23. Other factors such as the serotype of harboring strain and changes in mcr-9 sequences have been suggested to potentially play a role in colistin resistance in mcr-9-bearing strains21. Altogether, these previous reports indicate that the precise mechanism of mcr-9 in conferring colistin resistance remains unclear. The mcr-9 gene has now been reported across disparate ecological sources worldwide24. Its high prevalence and expression in colistin-resistant clinical strains present a major challenge to antimicrobial resistance (AMR) management and infection control efforts24.

The gene mcr-10 was first identified in a clinical isolate of Enterobacter roggenkampii recovered from a Chinese hospital in 201625. It was borne on an IncFIA plasmid and is associated with a 4-fold increase in colistin MIC in cloned strains25. Since its initial identification, mcr-10 has been detected in diverse bacteria from various human, animal and environmental sources26,27,28. It has been postulated that mcr-9 and mcr-10 may have originated from a common ancestor based on similarities in their secondary protein structures25.

Several studies have investigated mcr-9 and mcr-10 genes in the context of their host bacterial strains. A longitudinal study on carbapenem-resistant and carbapenem-susceptible strains of the Enterobacter cloacae complex found a high prevalence of mcr-9 and mcr-10 genes, often co-occurring with genes encoding carbapenemases in their genomes23. A 10-year study at a tertiary hospital demonstrated the dominance of mcr-9 and mcr-10 in both colistin-resistant and colistin-susceptible Enterobacter species27. However, investigations of the genetic context of the plasmids that harbor mcr-9 and mcr-10 across bacterial taxa remain limited.

Here, we aim to elucidate the characteristics of mcr-9 and mcr-10 in a local population of Enterobacter species isolated from bloodstream infections in the Dartmouth Hitchcock Medical Center (DHMC), New Hampshire, USA. We carried out short-read sequencing of 59 Enterobacter isolates from DHMC, of which nine isolates carrying mcr-9 or mcr-10 were further analyzed using long-read sequencing. We contextualize these findings with a global collection of plasmid sequences bearing mcr-9 and mcr-10 (n = 319). Our results reveal that variants of mcr-9 and mcr-10 are widely disseminated via diverse Inc plasmid types that also carry other AMR genes. Understanding the dissemination pathways of mcr-9 and mcr-10 is critical in implementing meaningful public health strategies to minimize the broader dispersal of high-risk colistin-resistant strains at both local and global scales.

Methods

Ethical approval and bacterial isolation

The Committee for the Protection of Human Subjects of DHMC and Dartmouth College granted ethical approval (CPHS # STUDY00031572). Samples used in the study were subcultured bacterial isolates that had been previously archived as a routine part of clinical laboratory operations. No patient blood or other specimens were used in this study. Patient-protected health information was not collected or used in this study. Hence, the study protocol was deemed not to be human subjects research, and therefore informed consent was not required for this study.

Bacterial isolates from unique pediatric and adult patients with bloodstream infection were obtained from clinical blood cultures submitted to the Department of Pathology and Laboratory Medicine at DHMC from January 2017 to January 2022. Species identification was initially performed using either the FilmArray Blood Culture Identification (BCID) panel for samples collected between January 2017 and May 2021 or the FilmArray Blood Culture Identification 2 (BCID2) panel for samples from June 2021 onwards. In total, 59 isolates were identified as belonging to the E. cloacae species complex. All isolates were stored at −80 °C in dimethyl sulfoxide solution.

Antimicrobial susceptibility testing (AST) was carried out using the automated MicroScan Walkaway 96 Plus (Beckman-Couter, La Brea, California, USA). Two FDA (Food and Drug Administration)-cleared AST panels were used: the NUC62 Panel from January 2017 to May 2019, followed by the Neg MIC Panel 46 from June 2019 onwards. Each panel underwent verification study before use. For the NUC62 Panel, manufacturer breakpoints approved at FDA clearance were applied. When switching to the Neg MIC Panel 46, off-label breakpoints for carbapenems and certain cephalosporins were validated to match the Clinical Laboratory Standards Institute M100 S28 guidelines29. We tested the following antimicrobial compounds: amikacin, gentamicin, sulfamethoxazole/trimethoprim, ertapenem, meropenem, cefoxitin, cefazolin, cefepime, cefotaxime, cefotetan, ceftazidime, ceftriaxone, cefuroxime, ciprofloxacin, levofloxacin, aztreonam, ampicillin, ampicillin-sulbactam, amoxicillin-clavulanic acid, piperacillin/tazobactam. AST results are presented in Supplementary Data 1.

DNA extraction, library preparation, and whole genome sequencing

Subcultured isolates were grown onto commercially prepared tryptic soy agar with 10% sheep red blood cells (Remel, Lenexa, Kansas, USA) and in brain heart infusion broth (BD Difco, Franklin Lakes, New Jersey, USA) at 37 °C for 24 h. For short-read sequencing, we extracted and purified DNA from broth cultures using the QuickDNA Fungal/Bacterial Miniprep Kit (Zymo Research, Irvine, California, USA) in accordance with the manufacturer’s procedure. DNA libraries were prepared with the Illumina DNA Prep Kit and IDT 10 bp UDI indices following the manufacturer’s procedure. An Illumina NextSeq 2000 sequencer was used to sequence DNA samples as multiplexed libraries following the manufacturer’s guidelines. Illumina bcl-convert (v.3.9.3) was employed for demultiplexing, sequence quality control, and adapter trimming of the Illumina short reads. Sequencing resulted in 151-nt paired-end reads.

For long-read sequencing of the nine isolates that harbored mcr genes, DNA was extracted using the Quick-DNA high molecular weight MagBead Kit (Zymo Research, Irvine, California, USA) and prepared using the Oxford Nanopore Technologies (ONT) SQK-LSK114 native barcoding kit following the manufacturer’s procedure. Sequencing was performed on the ONT GridION platform with a FLOW-MIN114 Spot-ON Flow Cell, R10 version and employing a translocation speed of 400 bps. Base calling was done using Guppy v.7.0.9 with the option super-accurate mode. DNA concentrations were measured with the Qubit fluorometer (Invitrogen, Grand Island, New York, USA). Sequencing was carried out at the SeqCenter (Pittsburgh, Pennsylvania, USA).

Genome assembly, sequence quality check, and species confirmation

Illumina short reads were de novo-assembled into adjoining sequences using the SPAdes v.3.14.1 assembler30 implemented in Shovill v.1.1.0 (https://github.com/tseemann/shovill) (Supplementary Data 2). Adapter sequences were trimmed by enabling the -trim option. Adapter sequences on the ONT long reads were deleted using porechop v.0.2.4 (https://github.com/rrwick/Porechop). Further filtering of high-quality reads was carried out with Filtlong v.0.2.1 (https://github.com/rrwick/Filtlong) by discarding reads <1 kb and 10% of the worst reads. The Illumina short reads and ONT long reads were then combined and hybrid genome assemblies were generated from them using Unicycler v.0.5.031. Genome sequence quality was determined using CheckM v.1.1.332 and QUAST v.5.0.233. Genome completeness ranged between 98.40 and 99.61% (mean = 99.29%) and contamination ranged between 0.41 and 2.02% (mean = 0.99%) (Supplementary Data 1). All hybrid genome assemblies were of high quality comprising ≤7 contigs and a mean N50 of 4.75 Mbp (Supplementary Data 3)34.

Bactinspector (https://gitlab.com/antunderwood/bactinspector) was used to assign species identity of the assembled genomes. Bactinspector searches NCBI’s RefSeq database using mash to find the best-matching genomes. Furthermore, we uploaded our genome assemblies through the Type Strain Genome Server35 to confirm their taxonomy and results were similar to the Bactinspector outcomes. We also calculated the genome-wide average nucleotide identity (ANI) for every possible pair of genomes using fastANI v.1.3236. ANI refers to the mean nucleotide identity of all orthologous pairs of genes that are shared between a pair of microbial genomes. We used the ≥ 95% ANI threshold to delineate species boundaries36.

Phylogenetic tree reconstruction

The draft genomes were annotated using Prokka v.1.14.637 and the pan-genome characterized using Panaroo v.1.2.738. Sequence alignments of the 3022 core genes (i.e., gene families present in 99% of the genomes) were concatenated to create the core genome alignment, and from which single-nucleotide polymorphisms (SNPs) were obtained using SNP-sites v.2.5.139. The core SNP alignment was then used to reconstruct a maximum likelihood phylogeny in RAxML v.8.2.1240, implementing a generalized time reversible41 model of nucleotide substitution and Gamma distribution of rate heterogeneity. We used figtree v.1.4.4 (http://tree.bio.ed.ac.uk/software/figtree/) and Interactive Tree of Life (iTOL) v.6.8.142 for phylogeny visualization and annotation.

In silico sequence typing and identification of AMR genes in genome and plasmid assemblies

We used MLST v.2.19.0 (https://github.com/tseemann/mlst) to identify STs in our dataset. Multi-locus sequence typing is done by comparing query alleles with previously deposited allele combinations of the E. cloacae species complex compiled in the PubMLST database (https://pubmlst.org/organisms/enterobacter-cloacae)43. ST designation is based on allelic variation in seven single-copy housekeeping genes (dnaA, fusA, gyrB, leuS, pyrG, rplB, rpoB)44. Genome assemblies with novel ST configurations were curated on PubMLST43. ABRicate v1.0.1 (https://github.com/tseemann/abricate) was used to screen the hybrid genome assemblies to identify plasmid replicons using PlasmidFinder45 (Supplementary Data 4). The presence of clinically relevant AMR genes was determined using the NCBI AMRFinderPlus v.3.10.23 and its accompanying NCBI-compiled database46. The genetic environment of mcr-9.1 on plasmid and chromosomes was manually inspected using the Geneious Prime software v.2022.2.1. We further ran the chromosomal sequences of genomes EXB47 and EXB59 through the MobileElementFinder database (https://cge.food.dtu.dk/services/MobileElementFinder/) to determine if there were ICE (Integrative and Conjugative Elements) and IME (Integrative and Mobilizable Elements) associated with chromosomal mcr-9.1. Clinker v.0.0.25 (https://github.com/gamcil/clinker) was used to determine the sequence similarity of mcr-9.1 detected in chromosomes and plasmid genome assemblies. Plasmid genomes were visualized using Proksee47. Within Proksee with default settings applied, we used mobileOD-db v.1.1.3 to identify coding sequences related to replication/recombination/repair, transfer, integration/excision, and phages elements. CARD-RGI v.1.3.0 was used to annotate AMR genes (https://card.mcmaster.ca/analyze/rgi).

In silico analysis of globally distributed plasmids carrying mcr-9 and mcr-10

All publicly available sequences in the PLSDB database of complete bacterial plasmids48 were screened for the presence of mcr-9 and mcr-10 genes using AMRFinderplus v.3.10.23 as described above. The PLSDB contains an extensive set of complete bacterial plasmids from the NCBI’s RefSeq and international nucleotide sequence database collaboration (which includes DDBJ, EMBL-EBI and GenBank). Associated metadata of the global plasmid sequences were also retrieved (Supplementary Data 5) and used for further downstream analysis. Plasmid incompatibility types of the mcr-9 and mcr-10 harboring plasmids were determined using the PlasmidFinder database as described above. Plasmid mobility was predicted and categorized as mobilizable, non-mobilizable or conjugative using MOB-suite v.3.1.949.

Resistance gene co-occurrence in plasmids carrying mcr-9 and mcr-10

We used ReGAIN v.1.0.5 to elucidate on probabilistic patterns of AMR gene co-occurrence50. We selected resistance genes that occur in 5 – 95% of the global plasmid dataset to exclude ubiquitously occurring genes and reduce noise in creating the Bayesian network. For every pair of AMR genes, we calculated the conditional probability, relative risk, and bidirectional probability score. We carried out 500 bootstrap replicates. A 0.5 significance threshold was used to filter out weakly supported gene pairs. We also used the multivariate analysis (MVA) module of ReGAIN to visualize the structure of the underlying resistance gene data. MVA was constructed by applying Jaccard distances on the gene presence–absence matrix and principal component analysis. A k-means clustering algorithm was used to incorporate ellipses at 95% confidence around clusters.

Statistical analysis

All statistical analyses were carried out using the stat_compare_means on ggpubr v.0.4.0 package (https://rpkgs.datanovia.com/ggpubr/) in RStudio v.2022.02.1 + 46151. Kruskal-Wallis test was used to determine statistical differences between the distribution of plasmids grouped by their plasmid replicons and their AMR gene content and genome sizes. Wilcoxon signed rank test was used to compare the number of AMR genes in multi-replicon and single-replicon plasmids. We calculated Jaccard distances using the vegdist function on the vegan 2.6-6.1 package on RStudio 2024.04.2 + 76451. Results were considered significant when p < 0.05.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Results

Four species of bloodstream-derived Enterobacter carry mobile mcr-9 and mcr-10

We recovered 59 isolates belonging to the E. cloacae species complex from bloodstream infections in unique patients at the DHMC from December 2016 to February 2022. Majority of the genomes were phenotypically resistant to ampicillin (n = 58 isolates), whereas 57, 49 and 37 isolates were resistant to cefazolin, ampicillin-sulbactam and cefuroxime, respectively (Supplementary Data 1). Analysis of the 59 short-read genome sequences revealed that they belong to species Enterobacter hormaechei (n = 30 genomes), Enterobacter asburiae (n = 8), E. roggenkampii (n = 8), E. cloacae (n = 4), Enterobacter ludwigii (n = 4), Enterobacter kobei (n = 3), and one genome each belonging to Enterobacter bugandensis and Enterobacter mori (Supplementary Data 2). Across the eight species, we identified a total of 49 unique sequence types (ST), including 11 novel STs assigned in this study (ST3335 to ST3342, and ST3344 to ST3346) based on their unique configurations of seven housekeeping genes (dnaA, fusA, gyrB, leuS, pyrG, rplB, rpoB)44. The most common STs were ST141 and ST69 in E. hormaechei with three representative genomes each (Supplementary Fig. S1).

The short-read genome assemblies were examined for the presence of clinically relevant AMR genes. We identified nine genomes representing four species that harbor mcr-9.1 or mcr-10.1 (Fig. 1a). The four mcr-carrying species were E. asburiae (isolates EXB38 and EXB59), E. kobei (EXB6 and EXB315), E. hormaechei (EXB34, EXB45 and EXB47), and E. roggenkampii (EXB16 and EXB40). Other AMR genes were also present in the nine genomes and included those that confer resistance to beta-lactams (blaACT present in seven genomes and blaMIR present in two genomes), fosfomycin (fosA present in five genomes), and phenicol/quinolone (oqxAB present in all genomes) (Supplementary Data 3).

Fig. 1: Nine Enterobacter isolates carrying mcr-9 and mcr-10 from bloodstream infections in Dartmouth-Hitchcock Medical Center, New Hampshire, USA.
figure 1

a A maximum likelihood phylogenetic tree based on 573,405 single nucleotide polymorphisms (SNP) extracted from the 2.99 Mb sequence alignment of 3022 core genes. The tree is rooted at the midpoint. Tree scale indicates the number of nucleotide substitutions per site. Colored strips next to the tree show the Enterobacter species, sequence type (ST), year of isolation, and mcr gene variants. The distribution of different plasmid replicon types is indicated by the green (present) or white (absent) circles. b Genetic neighborhood of mcr-9.1 detected in chromosomes (EXB47 and EXB59) and plasmids (pEXB40_2 and pEXB45_1). Genes are represented by arrows. Colors of arrows represent different functions, with the same colors indicating homologous genes. Grayscale strips connecting the arrows from different cellular locations represent the sequence similarity (%) between them. Only those regions with 80 – 100% sequence similarity are shown. Genomic characteristics of the nine Enterobacter isolates are presented in Supplementary Data 3. Features of the plasmid assemblies are presented in Supplementary Data 4.

Full size image

The evolutionary relationships of the nine genomes can be inferred from the maximum likelihood phylogeny built from 573,405 single nucleotide polymorphisms (SNP) that were extracted from 3022 core genes. The nine genomes represented eight STs: ST484 and ST807 in E. asburiae; ST68, ST145, ST536 in E. hormaechei; ST125 in E. kobei; ST655 and ST1168 in E. roggenkampii. We observed that the mcr-containing genomes were markedly different in terms of STs compared to other closely related non-mcr-containing genomes within the entire Enterobacter population in this study (Supplementary Fig. S1). However, the mcr-10 positive E. hormaechei ST68 isolate EXB34 differed by 334 and 370 core genome SNPs from the non-mcr ST68 isolates EXB2 and EXB54, respectively.

To elucidate the likely mode of mobility of the mcr-9.1 and mcr-10.1 variants, we extracted the plasmid contigs from the hybrid assemblies derived from short- and long-read sequencing data. We identified a total of 14 plasmids representing nine unique plasmid replicons in genomes from this study (Fig. 1a). All plasmid genome assemblies, including those carrying mcr-9.1 and mcr-10.1, have a circular topology (Supplementary Fig. S2).

The mcr-10.1 gene was identified only in plasmid assemblies from five Enterobacter genomes. The mcr-10.1-bearing plasmids have replicon types IncFIB(K) (pEXB34_1), IncFIB(pECLA) (pEXB16_1), IncFIA(HI1)--IncFIB(K) (pEXB38_1), and IncFIB(K)--IncFII(pECLA) (pEXB6 and pEXB31). Sizes of the plasmids harboring mcr-10.1 ranged from 105 kbp (pEXB38_1) to 172 kbp (pEXB34_1) (Supplementary Data 4 and Figure S2).

The mcr-9.1 gene was present in either plasmid or chromosome of four Enterobacter genomes. It was detected in pEXB40_2 (122 kbp) and pEXB45_1 (149 kbp) containing the IncFIB(pECLA)--IncFII(pECLA) replicons (Supplementary Data 2). In isolates EXB47 and EXB59, the mcr-9.1 gene was integrated in the chromosomes. An examination of the genetic environment of mcr-9.1 in plasmids and chromosomes showed notable sequence similarities (Fig. 1b). The mcr-9.1 gene was flanked by several insertion sequences, IS5 family transposase, IS6-like family transposase, IS110 family transposase, and an integrase. The qseC/B genes, which encode a putative two-component system involving a histidine kinase sensor (QseC) and its cognate partner (QseB), were located upstream of the mcr-9.1 gene. These associated genes are important in inducing the expression of mcr-9.1 in E. coli in the presence of subinhibitory concentration of colistin, thus causing elevated MIC20. Our results are consistent with previous reports of the integration of mcr-9 into chromosomes24,27,52.

Variants of mcr-9 and mcr-10 are borne on plasmids in a globally distributed, diverse dataset of bacterial species

We sought to place our Enterobacter plasmid genome assemblies in a broader context with other plasmids carrying mcr-9 and mcr-10 to elucidate their global distribution and genetic relationships. All 59,895 complete plasmid assemblies available from the PLSDB database48 (as of January 2024) were screened for the presence of mcr-9 and mcr-10 genes. A total of 319 (0.49%) of the entire plasmid collection harbored alleles of mcr-9 and mcr-10 genes, which originated from 13 genera (Fig. 2a and Supplementary Data 5). These plasmids were recovered from 33 countries, of which majority were from China (n = 90), United States (n = 44), Spain (n = 25), Australia (n = 18), Japan (n = 15), and 12 plasmids each from Switzerland and the Netherlands. These plasmids were isolated between 2008 to 2023. The number of plasmids isolated per year ranged from one (in years 2008 and 2013) to 98 (in year 2022).

Fig. 2: Global distribution of plasmid assemblies carrying mcr-9 and mcr-10 (n = 326) across bacterial host species.
figure 2

The global dataset includes the seven plasmids from bloodstream infections in DHMC sequenced in this study and 319 plasmids from the PLSDB database48. a Pie chart showing the number of species identified in the global dataset. b Distribution of mcr-9 and mcr-10 alleles among bacterial species. c Number of replicons per plasmids. d Distribution of replicon types among bacterial species. Only those replicons represented by three or more plasmids are shown for visual clarity. Red asterisk in color legends of panels C and D represent plasmid genome assemblies sequenced in this study. Features of the 319 global plasmids are presented in Supplementary Data 5.

Full size image

Together with the seven complete mcr-harboring plasmids from the DHMC population that we sequenced, the global dataset of plasmids bearing mcr-9 and mcr-10 (n = 326) came mostly from the genera Enterobacter (n = 195, 59.81%), Salmonella (n = 37, 11.34%) Klebsiella (n = 35, 10.73%), and Citrobacter (n = 22, 6.74%), which altogether comprised 88.65% of the global dataset (Fig. 2b). We identified variants of mcr-9 and mcr-10 in this global dataset: mcr-9.1 (n = 198, 60.73%), mcr-9.2 (n = 60, 18.40%), mcr-10.1 (n = 47, 14.41%), mcr-10.2 (n = 1, 0.30%) mcr-10.4 (n = 1, 0.30%). A total of 19 mcr genes did not have a perfect match with known mcr variants, which we subsequently referred to as simply mcr-9 (n = 15) or mcr-10 (n = 4) after their closest matching gene family names.

A total of 33 unique plasmid types defined based on the unique combination of incompatibility replicon genes were detected in the global dataset (Supplementary Data 5). We observed that a large subset of the plasmids contained two or more fused replicons (n = 300, 92.02%) present in a plasmid assembly (Fig. 2c). Plasmids harboring three fused replicons were most frequent (n = 234, 71.77%) across the global plasmid dataset and showed no preference in the host bacterial species. In contrast, the DHMC genomes sequenced in this study harbored either one or two plasmid replicons. The 3-replicon plasmid IncHI2A--IncHI2--pKPC-CAV1321 was most abundant in the global dataset (n = 234 plasmid assemblies, 71.77%) (Fig. 2d). This plasmid replicon type was detected in plasmid assemblies from 30 countries and present in at least one member of every genus included in the global dataset study except Hafnia paralvei (Supplementary Data 5).

The global dataset was examined to assess whether the replicon combinations identified in our local DHMC population were also present elsewhere. The replicons IncFIB(K)--IncFII(pECLA) that were detected in plasmids pEXB6_1 and pEXB31_1 appeared to be unique to our study. In contrast, the combination IncFIB(pECLA)--IncFII(pECLA) that was detected in pEXB40_2 and pEXB45_1 was also detected in 11 other plasmids from Enterobacter spp. originating from China (n = 6), Spain (n = 3) and one each from Taiwan and the USA. The IncFIB(pECLA) replicon identified in pEXB16_1 was also present in plasmids isolated from three Enterobacter spp. from China, Brazil and an unknown source. Additionally, IncFIB(K) identified in pEXB34_1 was also present in two mcr-10 harboring plasmids isolated from Enterobacter spp from Japan and Switzerland. Finally, the replicon combination IncFIA(HI1)--IncFIB(K) in pEXB38_1 (E. asburiae) was identified in a plasmid from another E. asburiae isolate in the United Kingdom.

Using MOB-suite49, we inferred the likely mobility of 83.4% (n = 273) of the global plasmid dataset as conjugative (n = 243, 89%), non-mobilizable (n = 22, 8.05%) or mobilizable (n = 8, 2.93%) (Supplementary Data 5). Our results show that a higher percentage of predicted conjugative plasmids was associated with multi-replicon plasmids (98.77%, n = 240) whereas mobilizable plasmids were mostly in single-replicons forms (n = 7) with one multi-replicon case. Non-mobilizable plasmids exhibited a moderate skew towards multi-replicons (n = 17) compared to single replicons (n = 5).

Plasmids carrying mcr-9 or mcr-10 vary in length and AMR gene carriage

Using the global dataset of plasmid assemblies bearing mcr-9 and mcr-10 (n = 326 plasmids), we analyzed the sequence variation of plasmids grouped by their replicon type and only those represented by three or more genomes. Our findings revealed significant variation in plasmid sizes across different replicon types (p = 2.2 e−16, Kruskall-Wallis test; Fig. 3a). Plasmids belonging to IncHI2A--IncHI2--IncR--pKPC-CAV1321 group had largest median size of 349 kbp (range: 282–381 kbp, 95% confidence interval: 277–378 kbp), followed by IncHI2A--IncHI2--pKPC-CAV1321 with 295 kbp (range: 192–477 Kbp, 95% confidence interval: 297–306 kbp), and IncFIB(K)--IncFII(pKP91) with 202 kbp (range: 124–313 kbp, 95% confidence interval: 146–284 kpb) (Supplementary Data 6A). The smallest median sizes were observed in single-replicon plasmids IncFIA(HI1) (68 kbp, range: 58–90 kbp) and IncFIB(K) (90 kbp, range: 70–172 kbp).

Fig. 3: Replicon types and antimicrobial resistance (AMR) gene carriage of the global dataset of plasmids carrying mcr-9 and mcr-10 (n = 326).
figure 3

a Comparison of plasmid sizes grouped according to incompatibility replicon types. Colored dots represent the mcr alleles present in the plasmids. Black diamonds show group means and black error bars show the 95% confidence intervals. b Comparison of the number of AMR genes detected in each plasmid grouped according to incompatibility replicon types. Only replicon groups represented by three or more genomes are shown for visual clarity. White diamonds show group means and white error bars show the 95% confidence intervals. c Comparison of the number of AMR genes present in genomes of multi-replicon and single-replicon plasmids. Black dots represent individual plasmids. For the box plots in (A, C), the box represents the interquartile range, horizontal line in the middle of the box represents the median, and whiskers represent the minimum and maximum values. Details on the global plasmids and replicon groups are presented in Supplementary Data 3 and 6.

Full size image

We observed differences in the plasmid distribution of specific mcr genes or mcr gene family alleles among the plasmid replicon groups (Fig. 3a and Supplementary Data 6B). For instance, the largest plasmid groups in this study, i.e., those containing IncHI2A--IncHI2--IncR--pKPC-CAV1321 replicons, which were detected only in Enterobacter spp. from China, Spain and unknown sources, harbored the mcr-9.1 exclusively. In contrast, plasmids with fused replicons IncHI2A--IncHI2--pKPC-CAV1321 (isolated from bacterial species representing 12 genera and across 30 countries) contained mcr alleles belonging to the mcr-9 gene family, including mcr-9.1 (n = 166 plasmid assemblies), mcr-9.2 (n = 56 plasmid assemblies) and mcr-9 (n = 12 plasmid assemblies).

Several plasmid replicons bore only mcr-10.1 as the only mcr gene in their genomes. These include IncFIB(K)--IncFII(pKP91) [isolated from Klebsiella spp. and Raoutella ornithinolytica in China, Myanmar, Netherlands and Vietnam], IncFIA(HI1)--IncFII(Yp) [isolated from E. coli and Enterobacter spp. in China and Czech Republic], and IncFIB(K) [from Enterobacter spp. in Japan, Switzerland and USA, including those we sequenced]. All plasmids belonging to the IncFIA(HI1) replicon [Enterobacter spp. from China and an unknown source] contained an mcr-10.1, except a plasmid from Enterobacter spp. in Vietnam which contained the mcr-10 (Supplementary Data 6B). Other plasmid replicon groups contained either mcr-10 or mcr-9 alleles in their genomes. These include IncFIB(K)--IncFII(Yp) [mcr-9.1 = one genome, mcr-10.1 = ten genomes], IncFIB(pECLA)--IncFII(pECLA) [mcr-9.1 = seven genomes including two genomes sequenced in this study, mcr-10.1 = four genomes, mcr-10.2 and mcr-10.4 = one genome each], IncFIA(HI1)--IncFII(pECLA) [mcr-9.1 = one genome, mcr-10.1 = 3 genomes], and IncFIB(pECLA) [mcr-9.1 = two genomes, mcr-10.1 = two genomes including one genome sequenced in this study].

We found significant differences in the number of AMR genes present in each plasmid replicon group (p = 2.9e−12, Kruskal–Wall test; Fig. 3b and Supplementary Data 6C). The median number of AMR genes per plasmid genome assembly varied among the multiple copy plasmid replicon groups, with IncHI2A--IncHI2--pKPC-CAV1321, IncHI2A--IncHI2--IncR--pKPC-CAV1321, IncFIB(K)--IncFII(pKP91) and IncFIA(HI1)--IncFII(pECLA) carrying 10 (range: 1–23), nine (range: 7–19), six (range: 1–12) and two (range: 1–4), respectively. Plasmid replicon groups IncFIB(pECLA)--IncFII(pECLA), IncFIB(K)--IncFII(Yp) and IncFIA(HI1)--IncFII(Yp) had a median of one AMR gene in their genomes. Among the single-copy plasmids, IncFIB(K)(pCAV1099-114), IncFIA(HI1), IncFIB(pECLA) and IncFIB(K) carried a median number of AMR genes per genome of eight, five, one, and one, respectively. The number of AMR genes per plasmid genome assembly also varied between multi-replicon and single-replicon plasmids (p = 3.3e−4, Wilcoxon rank-sum test; Fig. 3c).

We examined the plasmid genome assemblies to determine co-carriage of mcr gene and other mobile genetic elements (MGE). The plasmid genome assemblies were furnished with an assortment of insertion sequences, transposons and integron elements (Supplementary Data 7). We observed that the majority of plasmid genome assemblies (n = 174, 53.37%) carried a combination of transposons and integrons in their genomes. Additionally, 61 plasmids (18.71%) harbored at least one representative of all three MGE types (insertion sequences, transposons, and integrons) (Supplementary Fig. S3). Sixty-five plasmids contained only insertion sequences, while three harbored only transposons (Supplementary Fig. S3).

Co-occurrence of plasmid-borne AMR genes with mcr-9 or mcr-10 alleles

We examined the co-occurrence of AMR genes in plasmids carrying mcr-9 and mcr-10 to identify genes that frequently associate with specific mcr alleles. For the gene co-occurrence analysis, we included only AMR genes present in 5–95% of the global plasmid dataset and the mcr alleles mcr-9.1, mcr-9.2, and mcr-10.1.

First, we carried out a multivariate Principal Coordinate Analysis (PCA) calculated using Jaccard distances to visualize the similarities and differences among the mcr-bearing plasmids in terms of the AMR genes they carry. The two principal components from the PCA analysis explained 38.87% of the variance in the study (Fig. 4a). The AMR genes grouped into three clusters that represented distinct sets of AMR genes that tend to co-occur within the same plasmid. The spatial separation suggests varying strengths in their likelihood of co-occurrence. Additionally, the clustering of AMR genes within the ellipses may also reflect not only genomic proximity but suggest likely distinct selective pressures on each cluster shaping their persistence and spread. However, such selection pressure is unclear with the available data from this study. All AMR genes were distinctively positioned within one of the three clusters, except qnrB2 and catA1, which were associated with clusters containing both mcr-9.1/mcr-10.1 and mcr-9.2/mcr-10.1, respectively. Each of the three mcr alleles was represented in at least one of the three distinct clusters. This was expected as all the plasmid genome assemblies in the study are unique representatives of an mcr allele. Notably, we observed that mcr-9.1 clustered closely with qnrB4 (quinolone resistance), blaDHA-1 (cephalosporin resistance) and aph(3’)-Ia (aminoglycoside resistance), whereas mcr-9.2 was in closer proximity with blaCTX-M-9 (cephalosporin resistance) and ant(2’”)-Ia (aminoglycoside resistance).

Fig. 4: Co-occurrence of mcr-9 and mcr-10 with antimicrobial resistance (AMR) genes in the global plasmid dataset (n = 326).
figure 4

a Multivariate principal component analysis (PCA) showing the distribution of AMR genes harbored by plasmids carrying mcr-9 and mcr-10. Ellipses represent 95% confidence with three specified clusters. The PCA plot was generated using the Jaccard measure of distance. Variants mcr-9.1, mcr-9.2, and mcr-10.1 are labeled in blue for visual clarity. b Heatmap matrix showing the conditional probabilities of AMR gene co-occurrence calculated for all pairs of genes using ReGAIN50. Conditional probability refers to the probability of observing gene A given the presence of gene B50. Antimicrobial class of every AMR gene is indicated by colored blocks on the left and top of the matrix. (c) Association network showing the co-occurrence of mcr alleles and AMR genes based on relative risk. Relative risk is the ratio of the conditional probability of observing gene A given gene B to the conditional probability of only observing gene A in the absence of gene B50. Nodes represent AMR genes and are colored by antimicrobial class. Edges are colored by relative risk values. Details of the co-occurrence analysis are presented in Supplementary Data 8.

Full size image

We further investigated the probabilistic relationships of AMR gene co-occurrence in plasmids carrying mcr-9 and mcr-10 using a Bayesian network analysis50. For every pair of AMR genes, we calculated the conditional probability and relative risk of their co-occurrence. Conditional probability refers to the probability of observing gene A given the presence of gene B expressed on a scale of 0 to 150 (Fig. 4b). In all, we generated 1560 gene-gene comparisons, excluding self-queries (Supplementary Data 8). Results showed that certain AMR gene pairs and AMR-mcr pairs tend to co-occur more often than expected by chance. Among the three mcr alleles, mcr-9.1 had the highest conditional probability values with other AMR genes, including aac(3)-IIg, aac(6’)-IIc, arr, dfrA19, and ereA, all of which have conditional probabilities of at least 0.8 with mcr-9.1 (i.e., there is at least 80% probability of observing the AMR gene given the presence of mcr-9.1).

The relative risk refers to the ratio of the conditional probability of observing gene A given gene B to the conditional probability of only observing gene A in the absence of gene B50. A relative risk value of 1 suggests that gene A and gene B are likely independent of each other, while a value > 1 suggests that it is more likely to observe gene A in the presence of gene B50. Similarly, a value less than 1 indicates that gene A is more likely to be observed in the absence of gene B. While conditional probability focuses on the absolute likelihood of one event occurring given another event, relative risk emphasizes how much likely one event is to occur in the presence of another compared to its absence50. A high relative risk indicates a strong relative association but does not necessarily mean the absolute probabilities are high. Here, we calculated the relative risk for every mcr allele and every AMR gene (Fig. 4c).

Our results indicate that mcr-9.1 showed a strong probabilistic relationship with other AMR genes, but low relative risk (Fig. 4c). For instance, the conditional probability of observing the trimethoprim resistance gene dfrA19 given mcr-9.1 [P (dfrA19∣mcr-9.1)] was 0.82 (relative risk = 1.75). Similarly, we observed an equal and strong positive probabilistic relationship with the rifamycin resistance gene arr and the macrolide resistance gene ereA in the presence of mcr-9.1 (conditional probability = 0.81, relative risk = 1.65). The aminoglycoside resistance genes showed strong, though slightly varying, probabilistic relationship with mcr-9.1: P (aac3-IIg ∣ mcr-9.1) = 0.80, relative risk: 1.64; P (aac6-IIc∣mcr-9.1) = 0.80, relative risk: 1.64; and P (aph3-Ia∣mcr-9.1) = 0.79, relative risk =1.62. Thus, if gene A is already very common in the population (high baseline probability), the presence of mcr-9.1 might not significantly increase its probability, leading to a low relative risk even if conditional probability is high.

In contrast to mcr-9.1, the gene mcr-9.2 exhibited low conditional probability with AMR genes but high relative risk values (Fig. 4c). The strongest positive probabilistic relationships were with the trimethoprim resistance gene dfrA16 (conditional probability = 0.76, relative risk = 5.96) and the tetracycline resistance gene tetA (conditional probability = 0.67, relative risk = 7.14). This means that if gene A is rare in the population (i.e., low baseline probability), even a moderate increase in probability due to the presence of mcr-9.2 can lead to a high relative risk.

The gene mcr-10.1 showed low conditional probability scores with other plasmid-borne AMR genes. However, the small sample size of mcr-10 in our dataset likely influenced our results. Lastly, we also found instances where an AMR gene is associated with at least two mcr alleles, as in the case of genes conferring resistance to aminoglycoside (aac6-1b3, aph6-1c), sulfonamide (sul1, sul2), tetracycline (tetA, tetD), rifamycin (arr-3), beta-lactam (blaACC-1), methicillin/teicoplanin (msrR), and quinolone (qnrB2).

Discussion

Colistin resistance is a formidable global health challenge. It jeopardizes our dwindling reserve of last-resort antibiotics that are intended to be used only in cases of dire need, i.e., severe multidrug-resistant infections and after exhausting all other antibiotic options. Infections therefore become more life-threatening and medical treatments riskier. Understanding the evolution and dissemination of colistin resistance is critical to addressing this threat. Here, we present our analysis of colistin resistance genes mcr-9 and mcr-10 at local and global scales. We combined long- and short-read sequence data of bloodstream Enterobacter at DHMC (n = 14 plasmids from nine genomes) contextualized against a global collection of plasmid sequences (n = 319).

We highlight two major findings. First, allelic variants of mcr-9 and mcr-10 are widely disseminated via diverse Inc plasmid types. These plasmids, differing in terms of their replicon types and genetic structure, are present across diverse bacterial taxa and geographic locations. Of particular concern is that the majority of the plasmid genome assemblies were predicted to be conjugative or mobilizable, highlighting their potential to transfer the colistin resistance gene to other bacteria of clinical importance and spread to other ecological niches across the One Health spectrum53 (e.g., between animals and humans, between environmental and clinical settings, throughout the food production system). Gram-negative bacteria can therefore quickly evolve to thwart the effects of colistin via both integration into diverse plasmid backbones and the continuous allelic diversification of mcr genes. These two genetic innovations enable bacteria to keep pace with the intense evolutionary pressures of antimicrobial treatments. For instance, misuse and overuse of human and veterinary antibiotics are critical factors known to impose selective pressure towards the evolution of colistin resistance54,55. We can therefore consider the evolutionary dynamics of colistin resistance as embodying the Red Queen hypothesis (i.e., to avoid extinction, a species must continually evolve in the face of evolutionary changes in their enemies and other organisms they interact with)56. Hence, bacteria must constantly gain new adaptive traits to maintain their relative fitness in a competitive, fluctuating environment57, i.e., multiple strategies for colistin resistance to develop (via multiple mcr gene families and allelic variants) and disseminate (via multiple plasmid backbones and co-occurrence with other AMR genes). Ultimately, alternative public health strategies that will sustain a decrease in mcr diversity and transmission opportunities may be needed to disrupt the genetic arms race against colistin.

Our second key finding is that plasmid-borne genes conferring resistance to other antimicrobial agents, such as aminoglycoside, tetracycline and trimethoprim, tend to co-occur with mcr-9.1 and mcr-9.2 alleles. This creates a concerning scenario because the co-occurrence, potentially reflecting co-selection or linked functions58, of specific mcr alleles with other AMR determinants may complicate available treatment options. We observed variation in the co-occurrence of certain AMR genes, which certainly must be considered when elucidating the origins of multidrug resistance. For example, dfrA19 given mcr-9.1 has a high conditional probability but low relative risk. We interpret this as dfrA19 being frequently observed in the presence of mcr-9.1, but its presence might not be strongly dependent on mcr-9.1 due to its high baseline prevalence. In contrast, tetA given mcr-9.2 has low conditional probability but high relative risk, suggesting that the tetA is much more likely in the presence of mcr-9.2 compared to its absence, even if the absolute probability is lower. Moreover, our observation that some AMR genes co-occur with at least two mcr alleles warrants a deeper investigation, including whether such instances result in increased levels of colistin resistance and the ecological drivers that promote such associations.

According to the British Society for Antimicrobial Chemotherapy (BSAC) Resistance Surveillance Programme, annual colistin resistance rates among E. cloacae complex isolates increased from 4.4% to 20% between 2011 and 2017, notably higher than rates observed in Klebsiella species and E. coli59. While our DHMC dataset is small and only consists of Enterobacter species, our results are consistent with previous observations that mcr-9 and mcr-10 are the predominant mcr gene families in clinical Enterobacter 23,27,60. These two genes are known to have a unique propensity for transmission among Enterobacter species24,61, although the reason for this remains obscure. Future investigations should probe whether distinct mcr gene families exhibit species specificity in terms of the bacterial host that harbors them, as this may clarify the frequency, epistatic effects, and opportunities for inter-species transfer of mcr.

In the global dataset, mcr-9 and mcr-10 genes were frequently carried on multi-replicon plasmids and were significantly associated with a higher number of AMR genes. The presence of multi-replicon plasmids is increasingly observed in members of Enterobacterales62,63,64. Sykora’s 1992 hypothesis suggests that in plasmid co-integrates (i.e., multi-replicon plasmids), one replicon remains highly conserved, while the other is no longer under strong selective pressure, allowing it to accumulate mutations more freely65. This hypothesis highlights plasmid co-integration as a potentially vital mechanism for the diversification and proliferation of mcr-carrying plasmids in pathogen populations. Yet it also complicates plasmid classification, which can obfuscate the evolutionary relationships of plasmids and their replicons66. As such, plasmid co-integrates may confer several advantages. For instance, multi-replicon plasmids may be able to essentially mitigate issues related to incompatibility, thereby fostering stability within the host67, and facilitating interaction with a wider range of bacterial hosts68. Additionally, plasmid co-integrates can retain genetic elements from their parent plasmid, enabling them to act as important vectors for the dissemination of AMR, virulence and other genes that enhance cell fitness69. Future work is needed to elucidate the processes that give rise to multi-replicon plasmids and their contributions to promoting multidrug resistance.

This study is not without limitations. We acknowledge that analysis of the DHMC isolates would benefit from susceptibility testing to determine MIC of colistin but this was not carried out due to limited resources. The DHMC isolates also only consisted of those recovered from bacteremia cases and not from other kinds of infections. Hence, other plasmid types in the local population that may harbor mcr-9 or mcr-10 were left unexplored. In the global analysis, the mcr-9 and mcr-10 plasmids in the PLSDB database48 may not accurately reflect their true prevalence in healthcare settings. This is because long-read sequencing to resolve plasmid genome assemblies often arises from studies favoring the detection of multidrug resistance in bacteria and/or MGEs. Hence, our dataset likely underestimates the extent of spread of mcr-9 and mcr-10 among bacterial taxa and geographical regions, and unsampled reservoirs remain unidentified. Finally, while we rely on genomic attributes in the plasmid genome assemblies to predict their potential mobility, no conjugation experiments were conducted to ascertain this phenotypically. Nonetheless, our work provides a solid foundation for future genome-based epidemiological investigations of mcr genes. The growing number of mcr gene families and allelic variants in recent years16,17 reflects their rapid evolution under selective pressures, and it is not inconceivable that new mcr genes will continue to be discovered. In fact, mcr-11 has recently been added to the Reference Gene Catalog of the National Center for Biotechnology Information (NCBI), although to the best of our knowledge, no publication has described this novel gene yet.

In summary, our study revealed that the local and global dissemination of mcr-9 and mcr-10 is facilitated by diverse Inc plasmid types that mediate their mobility between bacterial taxa. The co-existence of mcr-9 and mcr-10 with other AMR genes in plasmids, which enable shuffling into different gene combinations, introduces grave challenges in combating the global health threat of colistin resistance and multidrug resistance. Findings from this study enhance our understanding of the plasmid backgrounds of mcr-9 and mcr-10. This work will be useful in current efforts to design effective public health strategies and targeted interventions against resistance to our last line of antimicrobial defense. This knowledge may be critical to inform scalable and effective public health interventions aimed at preserving the efficacy of colistin.