Plasmids are extra-chromosomal genetic elements ubiquitous in bacteria, and commonly transmissible between host cells. Their genomes include variable repertoires of ‘accessory genes,’ such as antibiotic resistance genes, as well as ‘backbone’ loci which are largely conserved within plasmid families, and often involved in key plasmid-specific functions (e.g., replication, stable inheritance, mobility). Classifying plasmids into different types according to their phylogenetic relatedness provides insight into the epidemiology of plasmid-mediated antibiotic resistance. Current typing schemes exploit backbone loci associated with replication (replicon typing), or plasmid mobility (MOB typing). Conventional PCR-based methods for plasmid typing remain widely used. With the emergence of whole-genome sequencing (WGS), large datasets can be analyzed using in silico plasmid typing methods. However, short reads from popular high-throughput sequencers can be challenging to assemble, so complete plasmid sequences may not be accurately reconstructed. Therefore, localizing resistance genes to specific plasmids may be difficult, limiting epidemiological insight. Long-read sequencing will become increasingly popular as costs decline, especially when resolving accurate plasmid structures is the primary goal. This review discusses the application of plasmid classification in WGS-based studies of antibiotic resistance epidemiology; novel in silico plasmid analysis tools are highlighted. Due to the diverse and plastic nature of plasmid genomes, current typing schemes do not classify all plasmids, and identifying conserved, phylogenetically concordant genes for subtyping and phylogenetics is challenging. Analyzing plasmids as nodes in a network that represents gene-sharing relationships between plasmids provides a complementary way to assess plasmid diversity, and allows inferences about horizontal gene transfer to be made.

Plasmid typing can provide insights into the epidemiology and transmission of plasmid-mediated antibiotic resistance. The principal plasmid typing schemes are replicon typing and MOB typing, which utilize variation in replication loci and relaxase proteins respectively. Previous studies investigating the proportion of plasmids assigned a type by these schemes (‘typeability’) have yielded conflicting results; moreover, thousands of plasmid sequences have been added to NCBI in recent years, without consistent annotation to indicate which sequences represent complete plasmids. Here, a curated dataset of complete Enterobacteriaceae plasmids from NCBI was compiled, and used to assess the typeability and concordance of in silico replicon and MOB typing schemes. Concordance was assessed at hierarchical replicon type resolutions, from replicon family-level to plasmid multilocus sequence type (pMLST)-level, where available. We found that 85% and 65% of the curated plasmids could be replicon and MOB typed, respectively. Overall, plasmid size and the number of resistance genes were significant independent predictors of replicon and MOB typing success. We found some degree of non-concordance between replicon families and MOB types, which was only partly resolved when partitioning plasmids into finer-resolution groups (replicon and pMLST types). In some cases, non-concordance was attributed to ambiguous boundaries between MOBP and MOBQ types; in other cases, backbone mosaicism was considered a more plausible explanation. β-lactamase resistance genes tended not to show fidelity to a particular plasmid type, though some previously reported associations were supported. Overall, replicon and MOB typing schemes are likely to continue playing an important role in plasmid analysis, but their performance is constrained by the diverse and dynamic nature of plasmid genomes.