Epidemiology, genetic dynamics, and transmission of the mcr-1 gene in China Authors: Boqian Wang, Zhong Li, Mingliang Chen, Yuxin Wang, Buaijier Aimaiti, Jingjing Fu, Kexin Li, Zhixi Peng, Rui Zhang, Tianyi Wang, Honghao Wang, Bin Gu, Hongbin Song, Hongguang Ren, Xiaofeng Hu Published: 2025 ABSTRACT ABSTRACTColistin is a broad-spectrum antibiotic that kills bacteria through the disruption of the cell membrane, making it effective against various bacterial infections. Since the first global report of mobile colistin-resistant (mcr-1)-positive bacteria, which were discovered and described by Chinese scientists in 2015, monitoring and research on these bacteria have continuously been conducted. In this study, we analyzed 70 mcr-1-positive strains detected by surveillance hospitals across three different provinces of China, as well as 671 strains screened from the NCBI database that contain the mcr-1 gene. The epidemiological analysis indicated that the mcr-1 gene had circulated undetected in China for over three decades prior to its first report in 2015 and extensive agricultural use of polymyxins likely initiated the emergence of mcr-1 and blaNDM co-harbouring strains in poultry and livestock, with subsequent fecal-environmental transmission leading to human infections. The genetic dynamics and transmission analysis provides the first detailed elucidation of the adaptive evolutionary migration of mcr-1, which enhanced its dissemination across diverse bacterial hosts and facilitated the emergence of super-resistant Enterobacterales strains and IncHI2(A) super-plasmids co-harbouring mcr-1 and blaNDM. Introduction Polymyxin had been widely employed in both human and veterinary medicine for several decades [1], but was gradually replaced in clinical practice by aminoglycoside antibiotics due to its serious side effects, particularly nephrotoxicity [2]. It was not until the emergence of bacteria resistant to β-lactam and carbapenem antibiotics that polymyxins had to be reintroduced to the clinic. The bacteria with extended-spectrum β-lactamases (ESBLs) severely limit the available antibiotic options for treating infections caused by carbapenemase-producing bacteria, particularly in “superbugs” carrying the New Delhi Metallo-β-lactamase (blaNDM) gene [3]. Unfortunately, after years of inadequate regulation, bacterial resistance to polymyxin caused by mcr-1 gene has also emerged. More alarmingly, the co-occurrence of mcr-1 and blaNDM in pathogenic bacteria threatens to precipitate a crisis of untreatable infections. Therefore, it is crucial to investigate the evolution and transmission of the mcr-1 resistance gene and its relationship with blaNDM. Bacteria carrying mcr-1 were first discovered and are most widely distributed in China [4,5]. It follows that mcr-1-positive bacteria in China constitute a primary research focus. By exploring their early origins and evolutionary trajectories and tracing their dissemination patterns, this study aims to establish a methodological and theoretical foundations for understanding their development. However, current Chinese surveillance data are dominated by isolated reports on specific strains or incidents [4,6,7], leaving a critical gap in systematic epidemiological and molecular analyses. In this study, we established a dataset comprising 70 newly sequenced strains from our monitoring agency and 671 Chinese mcr-1-positive bacteria from the National Center for Biotechnology Information (NCBI). Through systematic analysis of these samples, this study aims to delineate the epidemiology of mcr-1-positive bacteria in China, including their spatial and temporal distribution, bacterial species, host sources, sample origins, and resistance gene subtypes. These data will establish an essential foundation for tracing the origin and evolution of colistin resistance and enabling precise surveillance and early warning of resistant pathogens. Furthermore, we will characterize key genetic structures carrying the mcr-1 gene, such as transposons and plasmids, to investigate its molecular evolution and dissemination dynamics. In contrast to earlier studies relying on fragmented or localized data, our large-scale genomic approach allows for reconstructing a more comprehensive and refined model of its evolution, thereby providing a theoretical basis for source control and transmission interruption of resistant pathogens. Additionally, by integrating the genetic and evolutionary profiles of blaNDM, we will focus on analyzing the evolutionary linkage between mcr-1 and blaNDM, as well as the mechanisms facilitating their co-occurrence in bacterial hosts. This effort is expected to offer scientific insights for preventing the emergence of co-resistant strains that may lead to untreatable infections. Results Epidemiology of mcr-1-positive bacteria in China The database comprises 741 mcr-1-positive bacterial strains isolated in China, each with complete genome sequences and basic epidemiological information. Among these, 70 novel strains were identified by our surveillance system, primarily collected from sentinel hospitals monitoring antibiotic resistance in three Chinese provinces (SF. 1). The remaining 671 strains were obtained from NCBI (SF. 2). The provinces with higher numbers of mcr-1-positive bacteria are mainly concentrated in coastal areas, such as Guangdong, Zhejiang, Shanghai, Jiangsu, and Fujian, as well as the inland province of Sichuan (Figure 1a). Together, these regions account for 82.14% of all reported strains. The majority of these bacteria belong to the order Enterobacterales, with Escherichia, Salmonella, and Klebsiella collectively representing 97.3% of cases, suggesting a foodborne transmission route (Figure 1b). Notably, the prevalence of mcr-1-positive bacteria increased sharply between 2010 and 2015, then began to decline steadily after 2017 (Figure 1c), which is temporally consistent with the Chinese government's policy prohibiting the use of colistin as an animal growth promoter, effective from April 30, 2017. It is also noteworthy that the mcr-1 gene had been circulating undetected for many years before its first identification in 2015 – with the earliest collected sequence dating back to 1984. Figure 1.Epidemiological statistics of 761 mcr-1-positive isolates. (a) Geographic distribution of mcr-1-positive assemblies. The self-collected strain and NCBI datasets are distinguished by different colours. From the perspective of geographic distribution, provinces with a greater number of collected strains are shown in darker colours on the map, with the font in green, blue, or red depending on the increasing number. Self-collected strains are in light blue. The map was generated from coordinates using R software, and the main provinces are highlighted. (b) Distribution of mcr-1-positive isolates at the genus level. (c) Distribution of isolate collection years and the cumulative number of genera carrying the mcr-1 gene. (d) Distribution of mcr-1-positive bacteria in various hosts. (e) Distribution of mcr-1-positive bacteria in various sample sources. (f) Distribution of mcr-1 subtypes. Epidemiological statistics of 761 mcr-1-positive isolates. (a) Geographic distribution of mcr-1-positive assemblies. The self-collected strain and NCBI datasets are distinguished by different colours. From the perspective of geographic distribution, provinces with a greater number of collected strains are shown in darker colours on the map, with the font in green, blue, or red depending on the increasing number. Self-collected strains are in light blue. The map was generated from coordinates using R software, and the main provinces are highlighted. (b) Distribution of mcr-1-positive isolates at the genus level. (c) Distribution of isolate collection years and the cumulative number of genera carrying the mcr-1 gene. (d) Distribution of mcr-1-positive bacteria in various hosts. (e) Distribution of mcr-1-positive bacteria in various sample sources. (f) Distribution of mcr-1 subtypes. Homo sapiens, avians, and swine serve as the primary host sources (Figure 1d). As avians and swine represent major livestock species in China, cross-species transmission of mcr-1-positive bacteria between animals and humans is likely. Most samples originated from fecal and intestinal sources, consistent with the predominance of mcr-1-positive bacteria in the order Enterobacterales (Figure 1e). The main subtype remains mcr-1.1, including five strains harbouring two copies of the gene (SF. 2). The other six subtypes collectively accounted for only about 9% of the total (Figure 1f). Additionally, we identified 125 strains (17%) carrying both the mcr-1 and blaNDM genes, of which 123 strains are Escherichia and the other two are Klebsiella (SF. 3). The distribution of other antibiotic resistance genes (ARGs) was also analyzed (SFig. 1 & SF. 4). Among these, other types of carbapenem resistance genes have also been found to coexist with the mcr-1 gene, such as blaKPC (11), blaIMP (1) and blaOXA-48-like (1). Besides, CTX-M (229) confers resistance to cephalosporins and mphA (316) to azithromycin – both of which are WHO-recommended treatments for Enterobacteriaceae infections. Genetic dynamics of mcr-1 genes Nowadays, the mcr-1 gene comprises 34 subtypes, designated mcr-1.1 to mcr-1.34. All these subtypes arise from point mutations, with a maximum nucleotide sequence divergence of 2.4%. The classic mcr-1 transposon is Tn6330 – ISApl1-mcr-1-pap2-ISApl1 [8] (Figure 2a). We defined their corresponding position as P_-1, P_0, P_1 and P_2 and analyzed genes diversity across the 741 strains (Figure 2b). P_1 is predominantly occupied by pap2 (83.6%), indicating long-term co-transmission of these two genes. In contrast, the genetic elements at P_-1 and P_2 are more complex, with various IS/Tn elements accounting for only 36.6% and 8.0%, respectively. Further analysis was performed regarding the IS/Tn elements upstream and downstream of mcr-1. The results revealed that ISApl1 is the predominant element, almost exclusively concentrated at positions P_-1 and P_2 (Figure 2c). In addition, a newly observed Tn3-mcr-1-pap2-Tn3 structure was also observed (2/43), which may represent a rare or potential variant that requires further investigation, potentially in a manner analogous to the Tn6452 of mcr-5 [9,10]. Figure 2.Statistical analysis of mcr-1 upstream and downstream gene diversity data. (a) The classic structure of mcr-1 in the gene; that is, P_-1 is ISApl1, P_1 is pap2, and P_2 is ISApl1. (b) Proportions of genes in the P_ -1, P_1, and P_2 positions. 3) Number of IS/Tn in the P_ -1, P_1, and P_2 positions. (c) Heatmap of the IS/Tn distribution relationship upstream and downstream of the corresponding position of mcr-1. (d) The Tn/IS pair containing the mcr-1 gene was separated into two major groups: one included the same Tn or IS in the Tn/IS pair, and the other included different Tn or IS in the Tn/IS pair. Statistical analysis of mcr-1 upstream and downstream gene diversity data. (a) The classic structure of mcr-1 in the gene; that is, P_-1 is ISApl1, P_1 is pap2, and P_2 is ISApl1. (b) Proportions of genes in the P_ -1, P_1, and P_2 positions. 3) Number of IS/Tn in the P_ -1, P_1, and P_2 positions. (c) Heatmap of the IS/Tn distribution relationship upstream and downstream of the corresponding position of mcr-1. (d) The Tn/IS pair containing the mcr-1 gene was separated into two major groups: one included the same Tn or IS in the Tn/IS pair, and the other included different Tn or IS in the Tn/IS pair. To further analyze the diversity of genes in each strain, an algorithm [11] was applied to cluster the complete mcr-1 upstream and downstream genes (Figure 3a-d), where the frequently occurring genes were labelled and the other genes were left blank (Figure 3e). The classic Tn6330 structure accounted for less than 10%, and the other dominant structures (excluding hyp.ORF and accounting for over 1% across all samples) are ISApl1-mcr-1-pap2-VWA (12%), ISApl1-mcr-1-pap2-terF (3%), and DUF2806-mcr-1-pap2-DUF2726 (15%). Figure 3.Clustering of genes upstream and downstream of mcr-1 based on similarity. (a) Genes upstream and downstream of mcr-1 in four strains. (b) A fully connected network, with each line representing the similarity of the two strains connected. The lines are gradually eliminated by a certain restriction until all strains are isolated. (c) A clustering tree is established according to the results of step b, where the thickness of each branch represents the number of strains. (d) Strain clustering method. (e) Details of mcr-1 upstream and downstream genes. Clustering results of genes upstream and downstream of mcr-1. A tree was constructed according to the algorithm described in step d. The same gene is labelled with the same colour. Clustering of genes upstream and downstream of mcr-1 based on similarity. (a) Genes upstream and downstream of mcr-1 in four strains. (b) A fully connected network, with each line representing the similarity of the two strains connected. The lines are gradually eliminated by a certain restriction until all strains are isolated. (c) A clustering tree is established according to the results of step b, where the thickness of each branch represents the number of strains. (d) Strain clustering method. (e) Details of mcr-1 upstream and downstream genes. Clustering results of genes upstream and downstream of mcr-1. A tree was constructed according to the algorithm described in step d. The same gene is labelled with the same colour. Regarding each dominant structure above, point mutations are detected (SFig. 2). Subsequently, we utilized BEAST [12] to infer the earliest emergence time of each structure and explore the evolutionary process of mcr-1 context genes. The results under different growth models are consistent (Figure 4): DUF2806-mcr-1-pap2-DUF2726 →ISApl1-mcr-1-pap2-ISApl1 → ISApl1-mcr-1-pap2-terF/VWA. Figure 4.Molecular dating of mcr-1 mobilization. The occurrence of ancestral sequences was estimated using Bayesian cutting-edge dating. Three different population growth priors were used in the case of BEAST2 (tMRCA: time to most recent common ancestor): (a) coalescent: Bayesian skyline, (b) coalescent: constant size, and (c) coalescent: exponential growth, given by the colour scheme and legend on the right. The median estimate, with the highest density interval of 95%, is provided in the panel legend. (d) Trend of the proportions of the four structures over time. Molecular dating of mcr-1 mobilization. The occurrence of ancestral sequences was estimated using Bayesian cutting-edge dating. Three different population growth priors were used in the case of BEAST2 (tMRCA: time to most recent common ancestor): (a) coalescent: Bayesian skyline, (b) coalescent: constant size, and (c) coalescent: exponential growth, given by the colour scheme and legend on the right. The median estimate, with the highest density interval of 95%, is provided in the panel legend. (d) Trend of the proportions of the four structures over time. Transmission of mcr-1-carrying plasmids We further studied mcr-1-carrying plasmids via PlasmidFinder [13] among the 741 strains. In contrast to plasmid types such as IncFIB and IncFIC, which are predominant in bacterial strains, mcr-1 is mainly located in IncI2 (80.6%), IncX4 (86.6%) and IncHI2(A) (49.4% and 31.4%) (Figure 5a). IncI2 and IncX4 primarily carry the mcr-1-pap2 gene structure (Figure 5b) and IncHI2(A) mainly carries the ISApl1-mcr-1-pap2 structure (Figure 5b). Figure 5.Statistics of plasmid distribution in positive bacteria. (a) The number of plasmids in each type of mcr-1-positive bacteria, the number of plasmids carrying mcr-1, and the proportion of the latter in the former. (b) The proportion of different mcr-1 upstream and downstream structures in each type of the four selected plasmid types carrying the largest number of mcr-1 genes. Statistics of plasmid distribution in positive bacteria. (a) The number of plasmids in each type of mcr-1-positive bacteria, the number of plasmids carrying mcr-1, and the proportion of the latter in the former. (b) The proportion of different mcr-1 upstream and downstream structures in each type of the four selected plasmid types carrying the largest number of mcr-1 genes. Plasmid transmission between bacterial strains relies on the type IV secretion system (T4SS), which is related to the genes oriT, relaxase, T4CP and T4SS. Using the online analysis platform oriTfinder [14], we screened these genes among these major plasmid types (SFig. 3). The complete T4SS is primarily present on IncHI2(A) plasmids. Finally, based on plasmid similarity and epidemiological information (SFig. 4, see Methods for details), we further explored the transmission pathways of major mcr-1-carrying plasmids across China (Figure 6). These pathways were primarily concentrated in provinces with intensive poultry/livestock farming and/or high population density. For instance, one IncHI2(A) transmission route (grey) originated from Guangdong and Henan, eventually converging in highly populated provinces such as Zhejiang, Sichuan, and Shandong (Figure 6a). Another example is IncI2, for which a transmission path (purple) started in Henan, passed through Jilin, and eventually spread to most regions of northern China (Figure 6b). Figure 6.Geographical transfer of three plasmid types. Associated transmission pathways are indicated by the same colour (grey, orange, purple or red). (a) Geographical transfer of IncHI2A, which may have originated in Guangdong and then spread to Shandong, Zhejiang and Jilin Provinces, with two-way transmission between Zhejiang and Fujian Provinces. (b) Geographical transfer of IncI2, which presumably originated in Henan and then spread to Jilin for the first time, and then from Jilin to Heilongjiang, Sichuan and Zhejiang. (c) Geographical transfer of IncX4, which originated in Fujian Province and first spread to Sichuan, Hubei and Liaoning Provinces. Geographical transfer of three plasmid types. Associated transmission pathways are indicated by the same colour (grey, orange, purple or red). (a) Geographical transfer of IncHI2A, which may have originated in Guangdong and then spread to Shandong, Zhejiang and Jilin Provinces, with two-way transmission between Zhejiang and Fujian Provinces. (b) Geographical transfer of IncI2, which presumably originated in Henan and then spread to Jilin for the first time, and then from Jilin to Heilongjiang, Sichuan and Zhejiang. (c) Geographical transfer of IncX4, which originated in Fujian Province and first spread to Sichuan, Hubei and Liaoning Provinces. Discussion Colistin currently serves as the last-line agent against carbapenem-resistant Enterobacteriaceae (CRE) and other multidrug-resistant Gram-negative bacterial infections [15]. However, extensive agricultural misuse has driven the global dissemination of the mobile colistin resistance gene mcr-1, raising serious concerns over the emergence of colistin-carbapenem co-resistant strains that could compromise last-resort treatment options [16,17]. To investigate this issue, we analyzed 741 mcr-1-positive isolates from China, comprising 70 newly sequenced clinical strains and 671 publicly available genomes from NCBI. Among these, 36 isolates were collected prior to the first report of mcr-1 in 2015 [4] – the earliest dating back to 1984 (Figure 1c) – indicating that the gene had circulated undetected for over three decades [18,19]. The majority of these mcr-1-positive bacteria belong to the order Enterobacterales (Figure 1e). Additionally, we also identified 125 strains carrying both the mcr-1 and blaNDM genes, of which are Escherichia (123) and Klebsiella (2). Although we identified five bacterial strains harbouring two copies of the mcr-1.1 resistance gene, unfortunately, we were unable to obtain the corresponding polymyxin resistance testing results for these strains. Moreover, factors such as the expression level of the resistance gene and the presence of other resistance mechanisms (e.g. efflux pumps) in the strains can also influence resistance phenotypes. Therefore, elucidating the relationship between the copy number of resistance genes and the corresponding resistance phenotype represents a complex yet worthwhile research question, which deserves an intriguing direction for our future studies. Our analysis also reveals that the extensive use of polymyxins in agriculture may have led to the initial emergence of these Escherichia and Klebsiella co-resistant strains in poultry and livestock, which then transmitted via fecal contamination into the environment, ultimately resulting in human infections (SF. 3). Moreover, although mcr-1 was initially confined to narrow-host plasmids such as IncI2 and IncX4 in the form of a DUF2806-mcr-1-pap2-DUF2726 structure, which are predominantly found in E. coli within the Enterobacteriaceae family and relatively uncommon in other enterobacterial species [20], limiting its spread largely to E. coli. It gradually migrated via the Tn6330 transposon into the broad-host IncHI2(A) plasmid (Figures 4, 5), which is distributed widely across Enterobacteriaceae and potentially even beyond this family [21]. There, it further evolved into a stabilized genetic structure, ISApl1-mcr-1-pap2-terF/VWA. Since the blaNDM gene is predominantly found in Enterobacter, Klebsiella, and Escherichia, and is often carried by plasmids including IncHI2, IncX3, and IncF [22]. This transition of mcr-1 not only enabled its dissemination across a broader range of bacterial species but also expanded the reservoir of bacterial strains co-harbouring mcr-1 and blaNDM and facilitated the emergence of super-resistant IncHI2(A) plasmids. Currently, China is the country with the highest reported number of mcr-1-positive bacteria [5,23]. The past widespread use of colistin in livestock likely facilitated the transfer of these mcr-1 genes from natural sources to Enterobacteriaceae, making livestock the main reservoir for mcr-1 [1,24,25] and transmitting through foodborne infections among avians, swine, and humans [7,15,26]. These conclusions are largely consistent with our analytical results. Furthermore, leveraging large-scale data analysis and incorporating information on blaNDM resistance genes, our study provides a more in-depth elaboration on the potential transmission process (animal–environment–human), primary transmission vehicle (feces), and major bacterial carriers (Escherichia and Klebsiella) of co-resistant strains. The trends in the prevalence of colistin-resistant bacteria in China align closely with the production and usage patterns of colistin: both peaked around 2015 and have declined annually since [27]. These findings underscore the importance of rational antibiotic use in mitigating the spread of ARGs [28]. To combat the dissemination of bacteria carrying the mcr-1 resistance gene, a national antimicrobial resistance surveillance network encompassing livestock farms and healthcare facilities must be established, facilitating early warning through the implementation of rapid molecular diagnostics. For the livestock, stringent enforcement of antimicrobial use reduction initiatives is essential, including the prohibition of extra-label polymyxin applications and enhanced management of veterinary drug prescribing practices coupled with a comprehensive traceability system. Furthermore, by reinforcing biosecurity protocols in agricultural settings and optimizing quarantine supervision during animal transportation, an integrated “surveillance-control-containment” framework should be developed to disrupt transmission pathways across the animal-environment-human continuum. Gene transfer between plasmids is primarily mediated by transposons [29]. The mobilization of the mcr-1 gene has largely depended on the Tn6330 transposon structure [30–32]. However, in-depth clustering analysis of the mcr-1 context genes revealed that only 42 structures contained complete transposons, accounting for less than 10% of the total (Figure 3e). The majority consisted of stable configurations incapable of gene transfer. This strongly suggests that throughout its decades-long evolution, mcr-1 has adapted to optimal plasmid vectors for dissemination between bacteria, reducing its dependence on transposon-mediated transfer [1,5,23,33]. Although this evolutionary selection process has been briefly mentioned in some studies, it has not been fully elucidated. Based on diverse model systems and large-scale genomic analysis, we have for the first time accurately reconstructed this evolutionary trajectory. Initially, mcr-1 was present in a non-transposon structure – DUF2806-mcr-1-pap2-DUF2726 – within narrow-host plasmids IncX4 and IncI2, predominantly in E. coli. At this stage, only the pap2 gene was consistently associated with mcr-1 and may contribute to its expression [34]. Subsequently, mcr-1 acquired the Tn6330 transposon structure [8], enabling transfer between plasmids and facilitating the selection of more suitable vectors [30]. Ultimately, mcr-1 stabilized primarily on the broad-host IncHI2(A) plasmid as its key vehicle for interbacterial dissemination [35]. This was followed by the loss of downstream insertion sequences or their replacement with genes such as terF and VWA, which likely enhance plasmid stability [36,37]. Notably, terF and VWA are implicated in DNA replication and bacterial interaction processes, potentially supporting the replication and spread of mcr-1-carrying plasmids [38]. Through this stepwise adaptation, mcr-1 has achieved stability and broader dissemination across an expanding range of bacterial species inside or even outside of Enterobacteriaceae. Simultaneously, its widespread dissemination across densely populated regions in China (Figures 1a, 6) increases the risk of co-infection with blaNDM-positive bacteria in settings such as farms and hospitals. This facilitates the emergence of co-resistant superbugs and even hybrid IncHI2(A) super-plasmids. Due to the complex hospital environment, there is a high risk of cross-infection with different drug-resistant bacteria, potentially leading to co-resistant strains [32,39]. Therefore, hospitals must strictly implement infection prevention and control measures, enforce contact isolation for patients carrying drug-resistant bacteria and enhance admission screening. Regular sequencing and traceability analysis should be conducted for key drug-resistant bacteria, with particular emphasis on the blaNDM and mcr-1 resistance genes located on IncHI2(A) plasmids. Additionally, proactive environmental monitoring should be carried out in key departments such as the intensive care unit and burn wards to block the pathways of nosocomial cross-transmission. Materials and methods Database construction The NCBI database (Isolates Browser) was comprehensively searched (gene = “mcr-1”) to identify all bacterial strains carrying mcr-1, and all genomic sequences of sufficient assembly level (≥scaffold) were downloaded by NCBI tools (datasets and dataformat), including files in “fna” and “gbff” formats. We performed a comprehensive comparison of all currently known mcr-1 variant gene sequences using DIAMOND (-sensitive -evalue 0.00001 -outfmt 6) to analyze their sequence variations. Our findings revealed that all variants demonstrated sequence coverage exceeding 1623 base pairs and sequence similarity above 97%. Using these parameters, the mcr-1.1 subtype was selected as a reference to rigorously screen the downloaded database with the criteria of coverage ≥1623 bp and identity > 97%. To ensure the reliability of the analysis results, we ensured that each sequence downloaded from NCBI had a unique identifier to prevent redundant data. Secondly, we subjected all sequences to quality screening using CheckM (-t 16 -x fna –nt –tab_table) and rigorously removed low-quality sequences. According to the “Minimum Information about a Metagenome-Assembled Genome” (MIMAG) standards, we selected middle-quality data, specifically sequences with completeness >50% and contamination 10%, to construct the database. Finally, we supplemented missing epidemiological information by referring to the original publications of the sequences. For cases where the information could not be confirmed, we labelled them as “unknown” and appropriately decrease their significance in subsequent analyses. The identification criteria for the self-isolated strains were as follows: strains exhibiting a polymyxin-resistant phenotype were collected, and the presence of the mcr-1 resistance gene was confirmed by screening their whole genomes against the mcr-1 reference gene. Our bacterial genomes were collected by PLA Center for Disease Control and Prevention and extracted from cultured bacteria using the QIAamp DNA Mini Kit (Qiagen, Inc., Valencia, CA, USA) and sequenced by Novogene Co., Ltd. (Beijing, China) using the Illumina HiSeq 2500 platform (Illumina, Inc., San Diego, CA, USA) with an insert size of 350 bp. The genome was assembled de novo using the Unicycler genome assembler [40]. Scaffolding and gap filling were performed using SSPACE and GapFiller, respectively [41,42]. Detection of drug resistance genes We first downloaded the CARD database [43], based on which we constructed a resistance gene database index by BLAST (command makeblastdb) [44]. Then, we searched each sequence against the resistance gene database index (-evalue 0.00001 -max_target_seqs 5 -outfmt 6) and selected the results under the criteria of 70% identity (SFig. 1a, b) and 100% identity (SFig. 1c, d), respectively (both with coverage > 98%). Finally, we performed statistical analysis of the distribution of resistance genes based on the number of occurrences of each resistance gene (SFig. 1a, c) and the number of resistance genes carried in each bacterial strain (SFig. 1b, d). Annotation of mcr-1 context genes The mcr-1 context genes of the data downloaded from NCBI were automatically extracted from the corresponding gbff annotation files (python code: https://github.com/wr-sky/NDM/blob/main/read_gbff_file_to_extract_down_upstream_genes_of_NDM.py). To ensure consistency with the annotation method used for NCBI data, the mcr-1 context genes in the self-collected data were annotated by using NCBI's BLAST function to search for identical plasmids. Based on comparative searches of gene nomenclature and sequence alignments with reference genes, we manually reviewed and standardized the names of upstream and downstream genes of mcr-1 to ensure consistent and accurate designation, particularly when some sequences used abbreviated gene names while others used full names. Clustering of mcr-1 context genes According to the clustering algorithm in the related work [45] and the core idea of the BLAST sequence alignment algorithm, we designed the clustering algorithm in this paper (python code: https://github.com/wr-sky/NDM/blob/main/score_fully_connected_network_tree.py [22]). The original algorithm in the reference was designed for continuous nucleotide sequences, whereas our algorithm targets gene modules. Therefore, we introduced the scoring logic from the BLAST sequence alignment process, where successfully continuous matches lead to a stepwise increase in scores [46]. The specific clustering process is as follows: The initial score is set to 1 when two genes are identical, and the score increases stepwise with continuous similarities (Figure 3a);Each sequence is treated as a node to construct a fully connected network. The connection value between any two sequences is their similarity score, and clustering is performed following the method described in reference [45] (Figure 3b);The clustering results are represented as a clustering tree (Figure 3c). The initial score is set to 1 when two genes are identical, and the score increases stepwise with continuous similarities (Figure 3a); The initial score is set to 1 when two genes are identical, and the score increases stepwise with continuous similarities (Figure 3a); Each sequence is treated as a node to construct a fully connected network. The connection value between any two sequences is their similarity score, and clustering is performed following the method described in reference [45] (Figure 3b); Each sequence is treated as a node to construct a fully connected network. The connection value between any two sequences is their similarity score, and clustering is performed following the method described in reference [45] (Figure 3b); The clustering results are represented as a clustering tree (Figure 3c). The clustering results are represented as a clustering tree (Figure 3c). Sequence alignment and temporal tracing We utilized Mega [47] and the MUSCLE algorithm (with default parameters) to align the plasmids with identical plasmid types and manually corrected the results. Based on the aligned nucleotide sequence files, we inferred the time of ancestral node appearance via BEAST software [12]. We used the BEAUti [12] programme from the BEAST software suite to set the parameters (parameter details: https://github.com/wr-sky/NDM/). We utilize the GTR substitution model and the uncorrelated relaxed clock type with a Gamma relaxed distribution model. The length of the chain was set to 1.0E + 08. We performed multiple runs separately using Bayesian skyline, constant size, and exponential growth tree prior models. The results of the runs were combined and filtered using LogCombiner [12], and results where the effective sample size (ESS) values for each parameter were greater than 200 were selected to ensure high confidence in the results. Finally, we extracted the tMRCA distribution from the log files. Plasmid similarity evaluation and transmission tracing This analysis was conducted based on our algorithm model to evaluate plasmid similarity and predict transmission (SFig. 4) [11,48]. The algorithm is largely inspired by the SpreaD3 model [49], which identifies the most phylogenetically similar nodes and determines the direction of their transmission based on timestamps to ensure high confidence in the results (python code: https://github.com/wr-sky/NDM/blob/main/plasmid_similarity_evaluation.py). Supplementary Material SFig2.pdf SFig3.pdf SF3 125 Strains with blaNDM gene.xlsx SF2 Basic information of 671 NCBI stains.xlsx SF4 All ARGs of 741 strains.xlsx certificate.pdf SFig4.pdf SFig1.pdf SF1 Basic information of 70 novel strains.xlsx