On June 25, 2025, a total of 72,556 complete plasmid sequences were extracted from the Plasmid database (PLSDB) (version 2024_05_31_v2), a comprehensive plasmid repository of the National Center for Biotechnology Information (NCBI). PLSDB applies systematic quality control procedures, including sequence filtering and deduplication, ensuring that each record represents a unique complete plasmid sequence.

AMR genes were detected in the 72,556 plasmid sequences using the AMRFinderPlus tool (version 3.12.8), based on standard parameters of at least 90% identity and 60% coverage. Among these, 28,836 sequences presented at least one AMR gene. The resulting data were subsequently integrated with data from mobile genetic elements, allowing analysis of the association between AMR genes and the mobilome of plasmid sequences.

The identification of MGEs was performed using an integrative approach that combined two complementary tools. Initially, we performed the search using the TnCentral BLAST library, integrating TnCentral, INTEGRALL, and ISFinder¹ with ABRicate² (version 1.0.1) to apply these tools to our sequences. To increase detection sensitivity and identify elements not recognized by the TnCentral database (Ross et al., 2021), we additionally used the ISEScan (version 1.7.2.3) (Xie and Tang, 2017). Subsequently, all results were compiled into a single standardized annotation table using an R-based integration workflow. The complete integrated dataset is provided as Supplementary Table S7 (Merged_MGE_Annotations), and the R script used for data integration is available as Supplementary File S1.

The association between AMR genes and MGEs was performed with a maximum proximity of 5 kilobases (kb), as defined by the VR Profile2 tool and described by Shang et al. (2025). Furthermore, to identify potential compound transposition events, we considered carbapenemase genes flanked by identical insertion sequences at a maximum distance of 5 kb from each gene boundary. This criterion was applied to infer genomic contexts compatible with classical compound transposon structures. This approach allows the detection of possible genomic arrangements consistent with classical compound transposon structures, thus increasing the accuracy of the inferred mobilization mechanisms (Wang et al., 2022; Shang et al., 2025).

Plasmid Inc group data were obtained from the PLSDB, which provides curated plasmid sequence metadata, including replicon typing results from the PlasmidFinder database. These annotations were used to identify plasmid replicons across deposited sequences accurately.

For each plasmid carrying at least one mobilization element associated with carbapenemase genes, we quantified (i) the number of distinct mobilization elements and (ii) the number of unique carbapenemase genes present. To explore which plasmids combined the greatest mobilization capacity with resistance diversity, we computed the simple product between these two variables:

This score was used exclusively as an operational criterion within this study to highlight plasmids that accumulated higher values, interpreted as having a greater potential to disseminate carbapenemase genes. Plasmids with the highest scores were therefore referred to as “supervectors,” reflecting their combined mobilization and resistance-loading characteristics. Finally, plasmids were grouped according to their incompatibility (Inc) families, allowing us to identify which plasmid groups contained the greatest concentration of high-scoring plasmids.

Among the 28,836 resistant plasmid sequences, 6,017 (20.9%) contained at least one gene encoding carbapenem resistance. The most frequently resistance genetic variants were bla_KPC-2 (n = 1,513, 25.1%), bla_NDM-1 (n = 1,006, 16.7%), bla_NDM-5 (n = 770, 12.8%), bla_OXA-48 (n = 672, 11.2%), bla_KPC-3 (n = 454, 7.5%), bla_VIM-1 (n = 187, 3.1%), bla_OXA-181 (n = 169, 2.8%), bla_IMP-4 (n = 151, 2.5%) and bla_IMP-1 (n = 143, 2.4%).

In the analysis of carbapenemase genes, which encode carbapenemase enzymes of different classes, class B enzymes were the most frequent, present in (n = 2,677, 44.5%) of the sequences, followed by class A (n = 2,039, 33.9%) and class D (n = 1,305, 21.7%) enzymes. Individual plasmid resistance load varied from one to two distinct carbapenemase genes, with the median showing a predominant carriage of only one distinct gene per sequence. While a small fraction of plasmids carried two distinct genes, the co-occurrence of multiple carbapenemase genes on the same plasmid was uncommon (n = 136, 2.2%).

Analysis of plasmid sequences identified 173 IS, 68 Tn, and 118 In elements. Of these, 90/173 (73.4%) IS, 58/68 (85.3%) Tn, and 84/118 (71.2%) In elements were found in proximity to carbapenem resistance determinants, based on predefined association criteria (Figure 1a). The high frequency of association of carbapenemase genes with these mobile genetic elements demonstrates how these modules contribute to maintaining carbapenem resistance in plasmids.

The transport capacity of different carbapenem resistance variants was evaluated among the mobile genetic elements present in the plasmid sequences (Figure 1b), showing that IS carries a greater diversity of genes per element compared to Tn and In (Figure 1b). The medians confirmed that IS had the highest transport capacity, although the differences between categories were relatively small.

The distribution of mobilization categories was observed among the 30 most frequent carbapenemase gene variants (Figure 1c). Among the most prevalent genes, bla_OXA-48 stood out due to its high frequency of association with Tn, followed by bla_KPC-3. In contrast, different patterns were observed for bla_KPC-2, bla_NDM-1, and bla_NDM-5, which were preferentially associated with IS. Conversely, bla_IMP-4 and bla_IMP-1 showed a higher frequency of association with In. It can be observed that some genes, even belonging to the same subclass, can opt for different mobilization mechanisms and, consequently, different dissemination pathways.

The distribution of the top 10 MGEs was analyzed among the most prevalent carbapenemase gene variants, including bla_KPC-2, bla_NDM-1, bla_NDM-5, bla_OXA-48, bla_KPC-3, bla_VIM-1, bla_OXA-181, bla_IMP-4, and bla_IMP-1, which together represent approximately 80% of all identified variants (see Figure 4).

To comprehensively map the mobilome landscape in carbapenemase dissemination, we performed a systematic paired-association analysis on 18,878 plasmid sequences, that carried at least one antimicrobial resistance (AMR) gene located within a genomic distance of ≤5 kb from a mobile genetic element, irrespective of resistance class. This dataset comprised 845 carbapenemase genes and 209 mobile genetic elements (MGEs) distributed among integrons (In), insertion sequences (IS), and transposons (Tn). This large-scale analysis identified 1,323 significant associations (FDR <0.05), revealing remarkable hierarchies in both association strength and mechanistic specificity. The strongest associations between carbapenemases and MGEs exhibited extraordinary mobilization potency, with odds ratios reaching very high magnitudes.

Among the top 30 associations (Figure 5a), the aac(6′)-Iae + INTEGRALL-aacA28 integron partnership showed an odds ratio of 717.041 (FDR = 3.66 × 10⁻³²), while dfrA27 co-occurred with its cognate integron cassette at OR = 702.763 (FDR <10⁻³⁰⁰). Notably, it dominated with high mobilization strengths, claiming 19 of the top 30 positions. IS4_84, carrying the aminoglycoside resistance gene aph(3′)-IIa, emerged as the most potent IS-mediated association (OR = 94.153, rank 8), demonstrating that although IS actively participates in mobilization, it operates at characteristically lower association magnitudes than integron-based systems. This classification revealed a clear mechanistic hierarchy: Integrons specializes in ultra-resistant and cassette-specific mobilization, while IS facilitates broader, but statistically weaker, associations.

We next interrogated whether MGEs exhibit specialist versus generalist strategies in carbapenemase gene mobilization. Classification of 102 MGEs by their association breadth revealed a striking polarization (Figure 5b). Most elements (63/102, 61.8%) operated as specialists, forming significant associations with only one carbapenemase gene. However, a small cadre of three IS elements, IS6_292, IS91_50, and IS30_173, exhibited extreme generalist behavior, each associating with 10–13 distinct carbapenemase genes. IS6_292 emerged as the apex generalist, maintaining 13 significant associations while still preserving high individual association strength (maximum OR = 83.6 with IMP-6). This generalist behavior appeared exclusive to IS; no In or Tn achieved associations with eight or more carbapenemase genes. The specialist-generalist dichotomy was not evenly distributed across MGE categories: IS accounted for all three generalists and showed the highest proportion of moderate-breadth elements (9/45, 20.0%), while In and Tn were predominantly specialist (27/34 and 14/23, respectively). This fundamental difference in mobilization strategy, between high-fidelity specialists and IS as generalists, has profound implications for predicting dissemination routes and designing surveillance strategies.

Network topology analysis identified critical bridge elements whose removal would fragment the carbapenemase mobilome (Figure 5c). Among 159 network nodes (90 carbapenemase genes and 69 MGEs), 10 IS elements dominated the highest betweenness centrality positions, with IS6_292 exhibiting the most extreme hub properties (betweenness = 3815.87, degree = 13, PageRank = 0.024). This element served as the obligate connection point for multiple carbapenemase families, making it a structural linchpin in the dissemination network. IS30_173 (betweenness = 1656.35) and IS1_316 (betweenness = 1461.15) occupied secondary hub positions, together forming a triumvirate of IS that disproportionately control network connectivity. Transposon-mediated connections, while present, exhibited markedly lower betweenness values (maximum 604.78 for TnPMLUA4), indicating their role as peripheral rather than central connectors. The concentration of hub functionality in three IS elements suggests that targeted disruption or monitoring of these specific sequences could have a disproportionate impact on the global spread of carbapenemase genes.

Direct comparison of association strength across MGE categories revealed profound mechanistic differences in mobilization efficiency (Figure 5d). Integrons elements demonstrated the highest median odds ratio (107.2), with associations concentrated in a narrow, high-potency range (Q₁–Q₃: 27.1–1113.8, n = 47). Transposons exhibited comparable median strength (92.7) but with dramatically higher variance (Q₁–Q₃: 14.2–855.5, n = 38), indicating a more heterogeneous mobilization landscape. IS elements, despite their numerical dominance (n = 125 significant associations), operated at characteristically lower association strength (median OR = 17.3, Q₁–Q₃: 7.1–103.4). Statistical comparison confirmed these differences as highly significant: In versus IS (Wilcoxon test, p = 7.50 × 10⁻⁷, ****), and IS versus Tn (p = 4.25 × 10⁻³, **). Critically, In and Tn showed no significant difference in median strength (p = 0.329, ns), revealing a bimodal distribution in the mobilome: high-strength, low-promiscuity systems (In and Tn) versus moderate-strength, high-promiscuity systems (IS).

The most frequent Inc groups were IncX3 (n = 744), IncL (n = 661), IncR (n = 658), IncFII(pHN7A8) (n = 485), IncFII(K) (n = 382), IncC (n = 370), IncN (n = 363), IncFII (n = 310), IncFIB(pQil) (n = 264), and repB(R1701) (n = 228). In plasmid sequences carrying carbapenemase genes, we investigated the distribution of different mobilization elements across Inc groups. For the analysis, we used unique plasmid sequences and assessed how many sequences within each Inc group contained at least one mobilization element associated with carbapenemase genes. Among the 10 most frequent plasmid Inc groups, we observed distinct patterns in the distribution of mobilization elements. Groups such as IncX3 [IS (n = 733, 98.5%), Tn (n = 152, 20.4%), In (n = 3, 0.4%)], IncR [IS (n = 656, 99.7%), Tn (n = 132, 20.1%), In (n = 7, 1.1%)], repB(R1701) [IS (n = 217, 95.2%), Tn (n = 80, 35.1%), In (n = 2, 0.9%)], IncFII(pHN7A8) [IS (n = 485, 100%), Tn (n = 5, 1.0%)] and IncFII [IS (n = 309, 99.7%), Tn (n = 29, 9.4%)] were clearly dominated by IS.

In contrast, IncL showed a balanced distribution between IS (n = 615, 93.0%) and Tn (n = 623, 94.3%). A similar pattern was observed in IncFII(K) [IS (n = 374, 97.9%), Tn (n = 285, 74.6%), In (n = 4, 1.0%)] and IncFIB(pQil) [IS (n = 261, 98.9%), Tn (n = 189, 71.6%), In (n = 2, 0.8%)]. Finally, IncN and IncC plasmids exhibited greater heterogeneity, including a relatively high proportion of In in IncN [IS (n = 338, 93.1%), Tn (n = 95, 26.2%), In (n = 55, 15.2%)] and a mixed distribution in IncC [IS (n = 301, 81.4%), Tn (n = 146, 39.5%), In (n = 26, 7.0%)].

For each Inc group of plasmids, the three most frequent mobilization elements per category were identified, when available. Not all groups contained three elements per category; in such cases, only the elements present were considered. This analysis identified the predominant elements in each group, enabling us to determine the vectors involved in the dissemination of carbapenem resistance genes (Table 1).

A co-occurrence pattern analysis was performed to identify plasmids with high mobilization potential, defined as “supervectors.” The most representative groups in terms of plasmid number were IncX3 (744 plasmids), IncL (661 plasmids), and IncR (658 plasmids). The mobilization-resistance score allowed us to classify replicons with a high capacity for gene acquisition and dissemination (defined as the product of the number of unique mobilization elements by the number of unique carbapenemase genes), displayed differences between the groups. IncFIB(pQil) had the highest average score (≈2.89), indicating that, on average, its plasmids have greater diversity of genes and mobilization elements. Conversely, IncR had the highest maximum score (22), indicating that this group contains individual plasmids with a very high potential for disseminating carbapenemase genes, although the group’s average score is lower than that of IncFIB(pQil).

Other groups, such as IncFII(pHN7A8), IncFII(K), and IncX3, also presented high mean indices (≈2.59–2.70), combining a relatively high plasmid frequency with good supervector potential. In contrast, groups such as IncC and IncN presented lower mean indices (≈2.20), indicating lower diversity of genes and mobilization elements per plasmid, despite having a considerable number of plasmids (see Table 2).

Tn4401, a member a member of the Tn3 family, remains the most frequent platform responsible for bla_KPC mobilization (Cuzon et al., 2011; Migliorini et al., 2021) and was frequently associated with both bla_KPC-2 and bla_KPC-3 in our dataset. In contrast to previous reports suggesting isoform-specific Tn4401a and Tn4401b preferences (Migliorini et al., 2021; Fortini et al., 2016), our large-scale analysis revealed a balanced distribution of these variants, with both genes commonly associated with the Tn4401a isoform, reinforcing its continued epidemiological relevance.

Beyond this classical framework, our data highlight an expanded repertoire of mobilization platforms for bla_KPC. In particular, Tn7247 emerged as an alternative transposon associated with bla_KPC-2, supporting recent evidence that dissemination of this gene is no longer restricted to the Tn4401 family. This observation indicates an ongoing diversification of genetic vehicles capable of capturing and propagating bla_KPC’s (Forero-Hurtado et al., 2023).

Notably, a substantial fraction of bla_KPC-2 occurrences were embedded within compound transposition contexts mediated by IS6 family elements (Figure 3b), independent of Tn4401-associated structures. These IS6-mediated pseudo-compound transposons represent a flexible mobilization strategy, enabling the capture and rearrangement of resistance loci through cointegration-driven mechanisms rather than classical transposition (Harmer and Hall, 2019; Harmer et al., 2020). The prominence of IS6 in these configurations underscores its role as a generalist driver of bla_KPC dissemination, complementing the high-fidelity but more restricted mobilization mediated by dedicated transposons. Collectively, these findings demonstrate that the contemporary spread of bla_KPC-2 is shaped by the interplay between conserved specialist platforms and highly adaptable IS-mediated architectures.

The mobilization landscape of bla_NDM exhibited a distinct evolutionary pattern when compared to other carbapenemase genes. Consistent with previous genomic studies, Tn125 remained the predominant platform associated with bla_NDM-1, supporting its role as the ancestral transposon responsible for the emergence and early dissemination of this gene (Poirel et al., 2012; Bogaerts et al., 2012; Partridge and Iredell, 2012; Acman et al., 2022). Nevertheless, accumulating evidence indicates that this classical architecture is no longer exclusive.

In our dataset, compound transposition contexts involving blaNDM-1 were frequently mediated by insertion sequences, with IS91 emerging as the main flanking element associated with these events (Figure 3b). This observation is consistent with recent reports describing a progressive replacement of Tn125-derived structures by IS-mediated pseudocomposite transposons, which confer increased architectural flexibility to resistance loci (Acman et al., 2022).

A clear divergence was observed between bla_NDM variants. While bla_NDM-1 retained strong associations with Tn125-related contexts, the bla_NDM-5 variant was predominantly linked to the Tn2 transposon, a well-established mobilization platform historically associated with β-lactamase genes (Bailey et al., 2011). In addition, bla_NDM-5 frequently co-occurred with multiple IS families, including IS6, IS30, and IS91, consistent with previous observations that IS-mediated rearrangements facilitate the adaptation of resistance genes to diverse plasmid backbones (Lee et al., 2016; Varani et al., 2021; Harmer and Hall, 2024).

Together, these findings support an evolutionary transition in bla_NDM mobilization, from an early reliance on a single specialized transposon toward increasingly IS-driven and modular dissemination strategies. This shift toward greater structural plasticity likely contributes to the rapid global expansion and epidemiological success of contemporary bla_NDM variants.

The dissemination of bla_OXA-48 displayed a highly conserved mobilization architecture. Consistent with previous reports, Tn1999.1 emerged as the predominant transposon associated with this gene, reinforcing its role as the principal vehicle driving the global spread of bla_OXA-48 among plasmids and across diverse bacterial hosts (Li et al., 2022; Pitout et al., 2019; Peirano and Pitout, 2025).

In our dataset, bla_OXA-48 showed an exceptionally strong and specific association with IS4 family elements, reflected by a Jaccard index of 0.95, underscoring the near-obligate linkage between this resistance determinant and Tn1999-derived structures. This high degree of structural conservation contrasts sharply with the more heterogeneous mobilization patterns observed for other carbapenemase genes.

Beyond the canonical Tn1999.1 platform, we identified Tn2696 as an alternative mobilization structure associated with bla_OXA-48. Originally described in Escherichia coli isolates from Italy, Tn2696 represents a derivative of Tn1999 in which the bla_OXA-48 gene is flanked by IS elements of the IS1 family rather than IS4 (Giani et al., 2012). Although less frequent, the presence of this variant highlights that, despite its overall structural conservation, bla_OXA-48 can be accommodated within distinct but closely related composite transposon architectures.

We defined “supervectors” as plasmids with a high mobilization potential, characterized by the presence of multiple carbapenemase genes associated with various MGEs. These elements act synergistically, enabling recombination, integration, and horizontal transfer of resistance determinants, thereby facilitating their dissemination among different bacterial species (Partridge et al., 2018). Although the IncX3, IncL, and IncR Inc groups were the most frequently detected among plasmids carrying carbapenemase genes associated with mobilization elements, the IncFIB(pQil) and IncR plasmids exhibited the highest mean and maximum mobilization-resistance score, indicating superior potential for the spread and acquisition of new resistance determinants.

The high average score observed for IncFIB(pQil) plasmids indicates that they combine a high diversity of mobilization elements with several carbapenemase genes, functioning simultaneously as stable reservoirs and active platforms for gene transfer. The role of these plasmids as supervectors can be explained by two complementary characteristics: (i) their ability to co-integrate with other replicons, forming multireplicon structures, increases the presence of Tn and In, favoring the acquisition and dissemination of multiple resistance genes (Hanke et al., 2024); (ii) their modular plasticity allows them to rearrange genes, integrate new elements, and interact with other plasmids, increasing their adaptability and persistence in different contexts, including hospital settings under high selective pressure from carbapenem use (Wang et al., 2021). These factors make IncFIB(pQil) plasmids highly efficient in propagating critical resistance genes, such as bla_KPC, contributing to the emergence and maintenance of multidrug-resistant strains in clinical settings.

Interestingly, although less prevalent, IncR plasmids exhibited a genetic profile consistent with supervectors frequently associated with bla_NDM and bla_KPC genes. Previous studies have demonstrated a strong association between IncR and bla_KPC-2, suggesting its role as a preferential platform for the dissemination of this variant among different carbapenem-resistant bacteria (Feng et al., 2019). Despite lacking a complete conjugation system and therefore not being inherently self-transmissible, IncR plasmids often coexist with other conjugative or mobilizable plasmids, which enables their transfer through cointegration or plasmid-assisted mobilization (Calia et al., 2022). The identification of plasmids that can act as supervectors in the dissemination of carbapenemases demonstrates that the importance of the replicon is not limited to the prevalence of specific Inc groups, but rather to the modular complexity of plasmids, the coexistence of multiple mobilization elements, and their ability to interact with conjugative systems of different origins.

A limitation of this study is the use of a fixed genomic proximity threshold (≤5 kb) to define functional AMR-MGE associations. While this conservative criterion was intentionally adopted to minimize possible associations driven by simple co-residence on plasmid backbones, it may underestimate biologically meaningful mobilization events involving larger genetic architectures. In particular, composite transposons, nested transposition units, or plasmid-level mobilization processes may enable the co-dissemination of carbapenemase genes and their mobilization machinery without satisfying a short-range proximity definition. Consequently, our analytical framework is biased toward detecting tightly linked mobilization events and may undercount broader transposon- or plasmid-mediated dissemination structures.

Despite the large scale and global scope of this analysis, we acknowledge that our reliance on publicly available complete plasmid genomes may introduce representational bias. Curated plasmid databases such as PLSDB are inherently enriched for plasmids derived from clinical and surveillance-focused studies, while plasmids circulating in environmental, agricultural, and non-clinical reservoirs remain comparatively underrepresented. This bias reflects current sequencing priorities rather than a limitation of the analytical framework itself and is common to most large-scale plasmid-centered studies. Consequently, the mobilome architectures described here should be interpreted primarily in the context of well-characterized plasmids, particularly those associated with human and animal health.

Importantly, emerging metagenomic approaches targeting plasmid-associated sequences, often referred to as the plasmidome or metaplasmidome, offer promising avenues to overcome current database limitations. Recent studies applying long-read metagenomics and graph-based reconstruction have revealed extensive and previously inaccessible plasmid diversity across non-clinical ecosystems, including the human gut and environmental reservoirs (Yu et al., 2024; Debroas, 2025). Integrating such approaches with plasmid-centric analyses may allow future studies to evaluate whether the specialist versus generalist mobilome architecture identified here also governs carbapenemase dissemination outside clinical settings, or whether distinct dissemination regimes operate in environmental reservoirs.