CDC & FDA AR Bank assemblies were downloaded from NCBI (BioProject accession PRJNA294416). Only assemblies with complete genomes were downloaded to ensure quality of putative plasmid sequences. This included sequences from isolates in the following ten panels: “Ceftazidime/avibactam,” “Ceftolozane/tazobactam,” “Enteric Pathogen Diversity,” “Enterobacteriaceae Carbapenem Breakpoint,” “Enterobacteriaceae Carbapenem Diversity,” “Gram Negative Carbapenemase Detection,” “Isolates with New or Novel Antibiotic Resistance,” “Pseudomonas aeruginosa,” “Staphylococcus with Borderline Oxacillin Susceptibility,” and “Vancomycin Intermediate Staphylococcus aureus.” After assemblies were downloaded, sequences labeled as ‘plasmid’ were used for further analysis.

Open reading frames (ORFs) for each recovered plasmid sequence were obtained using Prodigal v2.6.3 (Hyatt et al., 2010). A protein database was then created from these ORFs and used for BLASTp v2.13.0 to identify replication, virulence, partitioning, toxin-antitoxin, SOS inhibition and mating-pair stabilization proteins. Rep protein sequences were obtained from NCBI as previously described (Shintani et al., 2015). Virulence protein sequences were obtained from the VFDB (Liu et al., 2022). Toxin-antitoxin sequences were obtained from the TADB (Xie et al., 2018). Due to some antitoxins being RNA opposed to protein, a nucleotide database from the plasmid sequences was created and BLASTn was used to identify the RNA antitoxin sequences. Finally, partitioning, SOS inhibition, and mating-pair stabilization protein sequences were obtained from UniProt by downloading the UniRef50 clusters associated with well-studied proteins associated with these functions (The UniProt Consortium, 2023). Supplementary Table S2 describes all the reference proteins used for these analyses. Replication and toxin-antitoxin protein hits were called with the following parameters: e-value <10⁻⁵, > 50% identity, and > 50% query coverage as previously described (Shintani et al., 2015). For all other proteins, the threshold of query coverage was raised to >70% as previously described (Shintani et al., 2015). For partitioning, SOS inhibition, and mating-pair stabilization proteins, the percent identity threshold was raised to >80% since the databases used for these proteins included protein sequences that were not experimentally verified. The Comprehensive Antibiotic Resistance Database’s (CARD) Resistance Gene Identifier v5.2.1 was used to identify ARGs in the plasmid sequences (Alcock et al., 2023). Finally, rmsFinder was used to determine if a plasmid contained methyltransferase sequences (Oliveira et al., 2016). True hits were defined as sequences that met rmsFinder’s built-in coverage threshold, an e-value <10⁻⁵, and > 50% identity.

The previously described MOB-Suite v3.1.0 was used to determine the presence of relaxase genes, T4SS genes, and oriT, as well as a prediction of both mobility and host-range (Robertson and Nash, 2018; Robertson et al., 2020). All recovered plasmid sequences were run through MOB-Suite’s “MOB-typer” tool with its default settings and results for relaxase and T4SS genes were verified by BLASTp as described above. Plasmids were then classified as “Conjugative,” “Mobilizable,” or “Non-mobilizable.”

Average nucleotide identity (ANI) was calculated for each pair plasmid sequences, which were then clustered using an ANI threshold of 95% through the “ANIm” tool in PyANI v0.2.12 (Pritchard et al., 2016). Plasmid sequences were then annotated using prokka v1.14.5 to compare annotated vs. unannotated genes in each cluster (Seemann, 2014).

We identified 439 putative plasmid sequences from 199 closed genomes. After searching for conjugation-related sequences, we found that 332 of these putative plasmids are predicted to be transferable through conjugation due to the presence of an origin of transfer. From the transferable plasmids, 206 are predicted to be conjugative, or self-transmissible, by encoding for a relaxase and T4SS proteins, while 126 are predicted to be mobilizable, containing an origin of transfer, but lacking T4SS machinery. Sequence length varied from 911 bp-439 kb, with a median length of 73.4 kb. We note that the shorter end of this range is more likely to be plasmid-derived sequences rather than complete plasmids themselves. Interestingly, we observed that mobilizable (mob) plasmids tend to be longer than non-mobilizable (non-mob) ones (Kruskal-Wallis test, followed by Dunn’s multiple comparison test, P_KW-D = 6.02×10⁻⁴), and that conjugative (conj) plasmids tend to be longer than those that are only mobilizable (P_KW-D = 5.57×10⁻⁹) (Figure 1A). These results are in accordance with previous data that showed a relationship between plasmid length and mobility (Smillie et al., 2010). It has been suggested that plasmid GC content is an important factor in determining host-range (Shintani et al., 2015). Therefore, we also analyzed the plasmid sequences’ GC content and found that, consistent with other reports (Shintani et al., 2015; Che et al., 2021), conjugative plasmids have comparably higher GC content than either mobilizable or non-mobilizable plasmids (mob P_KW-D = 0.034, non-mob P_KW-D = 0.029) (Figure 1B).

There were 135 unique antibiotic resistance genes in the plasmid sequences. We observed a relationship between presence or absence of an ARG and plasmid mobility (non-mob vs. mob P_KW-D = 2.83×10⁻⁶, non-mob vs. conj P_KW-D = 1.56×10⁻¹⁷), confirming previous results (Che et al., 2021; Newbury et al., 2022). The three ARGs with the most occurrences in the dataset were the sulfonamide resistance gene sul1, the antiseptic resistance gene qacEdelta1, and TEM-1, which encodes a beta-lactamase, with 104, 94, and 82 occurrences, respectively. Plasmids containing TEM-1 were especially of note since 88.3% (68/77) of plasmids containing at least one copy of the gene were conjugative or mobilizable. Interestingly, genes that provide resistance to aminoglycosides were the most prevalent when grouped by antibiotic class (Supplementary Figure S1).

After obtaining the distribution of plasmids and ARGs in the dataset, we focused on classifying the mobilizable and conjugative plasmids by incompatibility type, relaxase type (also known as MOB type), and T4SS type (also known as MPF type). Incompatibility (Inc) type varied within our dataset, with 239 mobilizable or conjugative plasmids being identified under one of 31 known Inc. types (Figure 2A). The most common Inc. type in our dataset was IncF, with 118 plasmids, which is consistent with previous data that have shown that IncF plasmids are commonly associated with extended spectrum beta-lactamases (Rozwandowicz et al., 2018).

MOB and MPF types are commonly used to classify mobilizable and conjugative plasmids based on similarity between their relaxase or T4SS proteins, respectively (Smillie et al., 2010). We identified relaxase gene sequences in 276 plasmids, with representation of all major MOB types (Figure 2B). MOB_F was the most common type of relaxase found with 120 plasmids containing this type. This type of relaxase is commonly associated with IncF plasmids (Rozwandowicz et al., 2018), so this result is not surprising. Similarly, we observed that the most common MPF type in the conjugative plasmids of this dataset was MPF_F, with 123 plasmids encoding MPF_F T4SS proteins (Figure 2C). It should be noted that since our analysis did not include integrative conjugative elements in chromosomes, we did not observe any putative plasmid sequence containing MPF_G type T4SS proteins.

In order to identify candidate plasmids with broad host-ranges, we analyzed the presence of various proteins that can influence host-range in the mobilizable and conjugative plasmids: toxin-antitoxin systems (T-AT), partitioning proteins (Par), plasmid SOS inhibition (Psi), mating-pair stabilization (MPS), and methyltransferases (MTases) (Figure 3). T-AT systems, also known as post-segregational killing systems, are commonly found in mobile genetic elements and improve plasmid stability by ensuring that all daughter cells contain a copy of the plasmid after segregation (Jurėnas et al., 2022). Daughter cells that do not receive a copy of the plasmid are typically killed by the present toxin due to the lack of constant expression of its cognate antitoxin (Jurėnas et al., 2022). We found 167 plasmids encoding toxins from T-AT systems, but only 105 of these also encoded its antitoxin pair. It has been shown that toxins from T-AT systems can also be used as secreted effectors for virulence (Triplett et al., 2016), so it is not surprising to find plasmids that solely contain the toxin gene. The most common T-AT pair was ccdA:ccdB, first found in F plasmids (Bernard and Couturier, 1992). We then searched for partitioning proteins, which are known to aid in plasmid stability by directing plasmid DNA to daughter cells during cell division (Baxter and Funnell, 2014). We found 102 plasmids with Par genes, the most common of which was sopB, the F plasmid centromere-binding protein, occurring in 52 plasmids. Although plasmid SOS inhibition and mating-pair stabilization are two relatively less well-studied processes, with only a few experimentally verified proteins that perform each function, we decided to include them in our search due to their potential importance in determining host-range (Neil et al., 2021a). SOS inhibition prevents RecA binding to the single-stranded plasmid DNA when it is transferred to a recipient, while MPS prevents the interaction between donor and recipient to be interrupted during DNA transfer (Neil et al., 2021a). Surprisingly, we found 98 plasmids encoding Psi genes and 89 plasmids encoding MPS genes. The most common Psi gene was psiB, while the most common MPS gene was traN. Both of these proteins were first described in the F plasmid (Klimke et al., 2005; Pérez-Mendoza and de la Cruz, 2009). Finally, we decided to search for MTase genes in the plasmid sequences as a proxy for restriction avoidance, one of the ways plasmids evade host defenses during conjugation (Shaw et al., 2022). It has previously been shown that plasmids and other MGEs carry solitary MTases as a way to prevent restriction from recipient hosts (Oliveira et al., 2014). We found 37 plasmids encoding MTases from 14 different restriction-modification systems.

To obtain putative broad host range plasmids, we chose sequences that contained at least three of the five characteristics analyzed earlier (T-AT systems, Par proteins, Psi proteins, MPS proteins, and MTases), resulting in 82 potentially broad host-range plasmid sequences. To validate host-ranges for these putative BHR plasmids, we used MOB-suite’s prediction of “host-range rank” (Robertson et al., 2020) and found that all plasmids are predicted to transfer at the order-level or above, with 3 plasmid sequences predicted to transfer at the phylum-level (Supplementary Table S1). We clustered these candidate sequences by average nucleotide identity (ANI) and found that many sequences had (>95%) ANI and clustered together (Supplementary Figure S2). This resulted in 22 different candidate plasmid types (Supplementary Table S3). The presence of very similar plasmids in distinct isolates suggests previous transfer of these, confirming their capabilities as mobile delivery vectors.

In order to learn more about our putative BHR plasmids, we analyzed their gene content in search of virulence factor (VF) genes, a potential limitation to the use of these plasmids in animal communities. Our VF gene search showed that out of the 22 candidate plasmid types, 13 contained a plasmid that did not have known VF genes. We obtained 27 plasmid sequences that did not encode for a VF (Supplementary Table S3). Additionally, we annotated the 82 putative BHR plasmid sequences and determined the number of coding sequences that were labeled as ‘hypothetical protein’. This allowed us to determine how many features of the putative BHR plasmids were similar to those of previously characterized plasmids. The median number of features that could not be annotated reliably per plasmid was 57.35% (Supplementary Table S3), emphasizing that these plasmids potentially encode for proteins which we cannot predict function. These proteins could be related with maintenance of a broad host-range and are potential targets for further investigation.