In Silico Detection and Typing of Plasmids using PlasmidFinder and Plasmid Multilocus Sequence Typing

Definition of the PlasmidFinder database.A set of 559 complete plasmid sequences available in GenBank were first aligned against 39 DNA sequences of previously characterized replicons in plasmids of Enterobacteriaceae species (Table 1). Using these replicon sequences, a total of 263 plasmids from GenBank (262 large and 1 small, belonging to the X2 group; see pFL129 in Table S1 in the supplemental material) were successfully recognized, showing >95% nucleotide identity and >96% coverage with the reference replicon sequences (see Table S1 for the list of plasmids recognized by the probes in Table 1).

Of the 296 plasmids that were not recognized by the 39 existing PBRT probes, 72 and 224 were large and small plasmids, respectively. In the 72 large plasmids, the annotated replication protein sequences were analyzed by using BLASTp at the NCBI site. Forty of these plasmids showed replicons and replicase protein sequences matching those of plasmids previously classified in the FII, FIB, I, B/O/K/Z, N, A/C, L/M, X, and P groups (>85% amino acid identities). In particular, the replicase proteins showing the pfam02387 or pfam01051 conserved domains were assigned to the FII and FIB groups, respectively (31). Twenty-seven additional FII and FIB probes were therefore included in the PlasmidFinder database (Table 2). Thirteen probes were included for the I2-, B/O/K/Z-, N3-, A/C1-, L/M-, X-, and P-like replicons to recognize (>95% nucleotide identity) 58 of the 72 not-yet-classified plasmids (see Table S1 in the supplemental material for the list of plasmids recognized by the probes in Table 2). The last 14 plasmids encoded RepA proteins that were not clearly homologous to any of the previously assigned Inc groups; for these plasmids, 14 probes were included in the PlasmidFinder database, using as the name of the probe the name of the plasmid each derived from (Table 2; see Table S1).

The analysis of the replication initiation protein genes in plasmid data released in GenBank provided evidence that the majority of large plasmids were still classifiable by DNA homology into phylogenetically related families corresponding to the formerly defined Inc groups (Fig. 1). Therefore, PlasmidFinder has been conceived to translate plasmid DNA sequences obtained by WGS into the well-established Inc group-based classification framework, allowing the WGS results to be traced back into the scaffold of previous knowledge and literature on relevant resistance plasmids characterized by standard methods (PBRT or Inc typing by conjugation). For each replicon, the PlasmidFinder output is linked to an NCBI GenBank entry connecting to the whole sequence of the plasmid showing that respective replicon. This allows the identification of related plasmids that can be used as references for further, more extensive and accurate comparative analysis of the WGS data.

• Open in new tab
• Download powerpoint

FIG 1

Numbers of fully sequenced plasmids (y axis) classified into incompatibility groups occurring in the different bacterial species of the Enterobacteriaceae family. The collection of 335 large plasmids (>20 kb) (listed in Table S1 in the supplemental material) downloaded from the GenBank database was classified into Inc groups by BLASTn (using the criterion of >95% nucleotide identity) using reference replicon sequences (Table 1) and the novel probes in the PlasmidFinder database (Table 2).

Definition of a database for small plasmids in Enterobacteriaceae.The 224 fully sequenced small plasmids were classified into 23 families of homology. Because of the huge sequence variety within the replication controls of this group of plasmids, the number and length of replicon sequences recognizing these plasmids in the PlasmidFinder database were maintained at the minimum for using the >80% nucleotide identity and 96% coverage criteria (Table 3), to avoid the occurrence of multiple alignments on different replicon sequences, which would make interpretation of the output results difficult. This classification was based on multiple phylogenetic analyses of nucleotide alignments of one or more plasmid targets among the repA, RNAI, and other plasmid sequences. The largest homologous group was obtained using the RNAI of the small narrow host range (NHR) ColE1-like plasmids from Klebsiella pneumoniae (pIGMS31, pIGMS32, and pIGRK), which can be maintained in several closely related species of the class Gammaproteobacteria but not in members of the Alphaproteobacteria (32). In fact, one replicon sequence (probe ColRNAI_DQ298019) alone detected 146 of the 224 small plasmids and 1 large plasmid (K. pneumoniae p15S; see Table S1 in the supplemental material) at >80% nucleotide identity and >96% coverage. For the majority of the remaining plasmids, the replicon sequences were devised on the repA gene, since they lacked the RNAI. Two replicon sequences (probes Q1 and Q2) were devised to recognize the repA gene of 7 small IncQ-like plasmids, and one replicon sequence [probe P(6)] was devised to detect the small IncP plasmid pRIO-5 (see Table S1 for the list of plasmids recognized by the probes in Table 3).

In conclusion, a total of 77 novel replicon sequences, 54 and 23 recognizing large and small plasmids, respectively, were devised in this study and included in the PlasmidFinder database, together with the 39 sequences of previously studied replicons, thus obtaining a database of 116 specific plasmid replicon sequences.

Evaluating PlasmidFinder on complete resistance plasmid sequences.A collection of 24 previously characterized and fully sequenced plasmids, carrying different assortments of replicons of the FII, FIIK, HI1, FIB, FIA, X, I1, N, and A/C types and associated with the most relevant and diffused resistance genes, such as bla_NDM-1, bla_KPC-2, and bla_CTX-M-15,were used to test the ability of PlasmidFinder to correctly identify the replicons located on these plasmids. PlasmidFinder was able to recognize each replicon correctly at a 100% match.

Using PlasmidFinder on whole-genome sequencing data.A selection of 49 S. Typhimurium draft genomes were chosen in order to test the ability of PlasmidFinder to detect plasmid replicons in bacterial whole-genome sequencing data. These genomes originated from isolates from healthy pigs in Denmark, collected as part of the DANMAP program (33) in 2011, and represent an unbiased sample. An identity threshold of 80% was used in order to detect both large and small plasmids based on the criteria described above. Among the 49 S. Typhimurium isolates, none had more than five unique hits to replicons in the database, 1 isolate had five unique hits, 10 isolates had four hits, 8 isolates had three hits, 10 isolates had two hits, 14 isolates had one hit, and 6 isolates had no hits to any of the replicons in the database. Of the S. Typhimurium replicons identified through the PlasmidFinder Web tool, 97 sequence hits had a % ID score of 90% or above, whereas only six had % ID scores between 85% and 90%, and no hits were found with % ID scores below 85%. In 74 hits, the full length of the replicon sequence present in the database was found to be present in the draft genomes, while 29 hits had a length between the cutoff length of 60% and 100% relative to the matching sequence in the database.

The six plasmids with % ID below 90%, as well as all but one of the 29 hits with non-full-length sequences, all mapped to replicon sequences from plasmids predicted to belong to the small-size Col plasmid group (expected size, <20 kb).

Among the 116 replicon sequences in the PlasmidFinder database, 22 were found to match sequences in the draft S. Typhimurium genomes (Table 4) according to the criteria listed above. The most abundant hits were to the Col(RNAI) sequence (n = 21), the IncQ1 sequence (n = 18), and the FIB(S) and FII(S) sequences (n = 13).

The IncQ1 replicon is identical to the replicon of the broad-host-range RSF1010 plasmid from E. coli, containing the strA, strB, and sul2 resistance genes, which confer resistance toward streptomycin and sulfonamides. This was found to be present in 39% of the isolates and was in all cases located on sequence contigs with an approximate size of 4.5 kb. This element corresponds to the previously characterized IS26-ΔrepA, repC, sul2, strA, strB-IS26 composite transposon, in which part of the IncQ1 replicon of the RSF1010 plasmid has been mobilized by IS26, together with the streptomycin and sulfonamide resistance genes (34). This mobile element is widely disseminated on plasmids and bacterial chromosomes; for instance, it was previously reported to be present in 88% of S. Typhimurium isolates from an international collection of strains originating mostly from humans but also from various animals, including pigs (35). Interestingly, resistance gene analysis on the same draft genomes using the online search tool ResFinder to detect resistance genes (http://cge.cbs.dtu.dk/services/ResFinder/) also found the strA, strB, and sul2 genes to be located on this 4.5-kb fragment, as previously described in other Salmonella plasmids (34, 35). This analysis, therefore, enabled us to link antimicrobial resistance determinants directly to the replicon sequence, predicting the presence of the RSF1010-derived element in these strains.

The FIB(S) and FII(S) sequences (Table 1) present in 27% of the isolates were always found to be present together in the same isolates and were in more than half the cases actually located on the same DNA fragment (contig), with sizes between 20 kb and 93 kb and thereby, also on the same plasmid. These replicons are characteristic for the Salmonella virulence plasmids, which represent a specific subgroup within the large family of the IncF plasmids (4). Here, no antimicrobial genes could be identified on the contigs carrying the replicons for FIB(S) and FII(S) sequences, which could be explained either by the lack of such genes on the virulence plasmids or by the inability of the current sequencing and de novo assembly methods to assemble complete plasmid sequences.

As the actual plasmid content of the test strains examined was not known, it was not possible to verify that all plasmids present in the isolates were also detected within the WGS data. However, this also applies to the current in vitro PBRT method, which can fail in detecting replicons because of nucleotide mutations occurring at the primer site sequences (4, 8). Furthermore, the PlasmidFinder database can detect many more plasmid replicon sequences than PBRT, also including novel groups. However, to get an indication of the extent of underreporting from PlasmidFinder, we subjected the six isolates reported by PlasmidFinder not to contain any plasmids to in vitro plasmid analysis by S1-PFGE analysis and DNA electrophoresis of purified plasmids. Using these methods, we were not able to detect any plasmids present in the strains, thus confirming the PlasmidFinder result (data not shown).

In silico detection using PlasmidFinder and ResFinder on WGS data offers the opportunity to link replicons, as well as other genetic features, such as multiple replicons, antimicrobial resistance genes, and virulence genes, to the same DNA fragment, because these tools provide the exact position of the genes in the uploaded sequence data. However, with the limitations in the current sequencing technology, it cannot be concluded that these genetic elements identified on different contigs are not located on the same plasmid. To examine this, transfer experiments by conjugation or transformation of the relevant plasmid(s) combined with S1-PFGE to ensure transfer of a single plasmid is most likely still required. But as the next-generation sequencing technologies mature even further to produce longer read lengths or single-molecule sequencing of larger pieces of DNA than today and thereby increase the ability to assemble sequencing data into longer contigs, this linking of genes will at some point be generally achievable. Furthermore, novel plasmid replicon groups identified by in silico BLAST homology searches or traditional replicon cloning and subsequent sequencing can easily be compared to the existing sequences and eventually be added to the database if they are found to differ significantly from these, based on the criteria used to build the PlasmidFinder database (>95% ID for large plasmids and >80% ID for Col-like plasmids).

pMLST analysis of plasmids.Among the replicons identified in our S. Typhimurium collection, some belonged to four of the five incompatibility groups with available pMLST schemes. We detected strains carrying IncF, IncI1, IncHI1, and IncHI2 plasmids, while we did not detect any IncN plasmids as present in the S. Typhimurium WGS data.

Eighteen isolates contained an IncF replicon. In addition to the 13 with the IncFIB(S)/IncFII(S) virulence plasmid described above, this also included other IncF-related replicons, i.e., IncFIA(HI1), IncFIB(AP001918), IncFII(pRSB107), IncFIA, IncFII(pHN7A8), and IncFII(pCoo). In silico pMLST analysis using the FAB (FII:FIA:FIB) typing scheme for IncF plasmids suggested in reference 4 on the 13 sets of sequencing data carrying IncFIB(S)/IncFII(S) replicons showed 8 to belong to the S1:A⁻:B17 FAB type previously found to be associated with the S. Typhimurium virulence plasmid (4), while 5 belonged to a highly similar FAB type only differing in a single mutation within the IncFIB fragment (assigned to a new FIB allele 35, leading to an S1:A⁻:B35 FAB type).

A new FIA allele that was identified among the three isolates containing IncFIA(HI1) replicons (strains S25, S36, and S38) had only 88% identity to the closest match in the pMLST database (FIA allele 2). The FIA allele 8 was therefore assigned to this allelic variant, resulting in the F:A8:B⁻ FAB type. Strain S48 carried an IncFIB(AP0001918) and an IncFII(pHN7A8) replicon with an F40:A⁻:B20 FAB type, and strain S49 carried IncFII(pCoo) and IncFIA replicons, leading to a new FII allele closely related to allele 13 (assigned to FII allele 67) and an FIA allele 6, giving the FAB type F67:A6:B⁻.

Three strains contained replicons belonging to the IncI1 group. pMLST analysis showed them to belong to sequence type 25 (ST25) (strain S13), ST3 (strain S25), and ST36 (strain S31). In addition, two of these strains (strains S25 and S36), as well as a fourth strain (strain S14), contained IncHI1 replicons and one strain (strain S39) contained two IncHI2 replicons (HI2 and HI2A) located on the same DNA fragment after de novo assembly of the sequencing data. Isolates S25 and S36 both had perfect matches to all 6 alleles of the IncHI1 pMLST scheme; however, the combination of alleles was new (subsequently assigned to ST10), while isolate S14 produced six completely new alleles with 85% to 93% identity to the alleles already present in the pMLST database for IncHI1 plasmids. These alleles have been added to the pMLST scheme, and ST13 has been assigned to the IncHI1 plasmid from isolate S14. Interestingly, five of the six IncHI1 alleles in S14 were identical to the IncHI1 plasmid pNDM-CIT that was previously identified in a Citrobacter freundii isolate from a French patient with urinary tract infection returning from India (8). Finally, isolate S39 carrying two IncHI2 replicons only produced a result in one of the two alleles (smr0018 allele 1) of the IncHI2 scheme and is therefore not typeable by this method.

Based on the results presented above, the PlasmidFinder and pMLST Web tools present an opportunity for microbiologists without particular bioinformatics skills to analyze whole-genome data in both raw and assembled formats obtained from their own benchtop sequencers and retrieve plasmid replicon information to be used in clinical and epidemiological investigations. An advantage of using freely available Web-based services is that different investigators can use the same comprehensive curated database and standardized analytic settings, thus enabling more reproducible data for comparison of results between studies. At the moment, only one data set per each query session can be submitted, but a tool for batch upload of multiple data sets is currently under construction. Finally, correlation between plasmid replicons and antimicrobial resistance determinants is in some cases possible with the current technology; however, this is expected to further improve when next-generation sequencing technology and assembly eventually is able to produce long contigs and, ultimately, complete plasmid sequences.