Don't overlook the little guy: An evaluation of the frequency of small plasmids co-conjugating with larger carbapenemase gene containing plasmids

KPC -plasmids were transferred. In all of the 143 transconjugants, multiple plasmids, both large and small, transferred with each mating. When two bla KPC- plasmids were present in the host, frequently (87%; 98/113) both would be transferred during mating. p3223 is found in a wide range of bacterial hosts that harbor AMR genes; p1916 has been identi ﬁ ed in only a limited number of publicly available sequences to date. From our evaluation, there is still much to learn about SCPs, and the high rate of co-transfer of multiple plasmids from real-world carbapenemase-producing Enterobacteriales.


Introduction
Carbapenemase-producing Enterobacteriales (CPE) are considered an urgent threat to modern medicine because of increased mortality in infected patients due to the lack of therapeutic options and their rapid emergence globally over the last decade (Center of Disease Control, 2013). Carbapenemase genes in Enterobacteriales most often reside on mobile genetic elements (MGE), such as plasmids, which can be easily shared across many species (Sheppard et al., 2016a). Klebsiella pneumoniae carbapenemase (KPC, encoded by bla KPC ) is an Ambler Class A serine carbapenemase, first isolated in the United States in 1996 (Yigit et al., 2001), and now accounts for the largest proportion of global CPE infections (Barría-Loaiza et al., 2016;Partridge et al., 2015;Paul et al., 2015;Pulcrano et al., 2016;van Duin and Doi, 2017). bla KPC is located on transposon Tn4401, which is about 10 kb in size. Tn4401 is capable of mobilization at a high frequency (Cuzon et al., 2011), and is almost always located on a plasmid which increases the likelihood of spread of bla KPC .
After the introduction of KPC-Enterobacteriales (KPCE) to our institution in 2007, and in the context of on-going transmission, the screening of high-risk patients for asymptomatic carriage of KPCE was implemented in 2009 to try and reduce colonization and infection rates (Mathers et al., 2014). Unlike many institutions, which at the time described clonal transmission of KPC-producing K. pneumoniae, we witnessed sustained multispecies transmission of KPCE (Sheppard et al., 2016a). In 2013, there was recognition that wastewater premise plumbing was colonized with KPCE and we began environmental surveillance and mitigation strategies (Mathers et al., 2018). Hospital wastewater premise plumbing is increasingly recognized as a reservoir for CPE (Decraene et al., 2018;Kizny Gordon, 2014), and an ideal niche T for the horizontal gene transfer of antimicrobial resistance (AMR) genes (Conlan et al., 2014).
For plasmids to be considered conjugative, they need to encode the necessary machinery. Conjugation apparatus includes a type IV secretion system (T4SS) to form the mating channel, and a mobility (MOB) module, which includes an origin of transfer (oriT), a relaxase, and a type IV coupling protein (Cabezon et al., 2015). Traditionally, for a plasmid to be characterized as mobile, it only needs an oriT and a relaxase, as it can hijack T4SS systems encoded by other MGEs within the same bacterial cell (Smillie et al., 2010). Until recently, if a plasmid has neither, it was considered non-transmissible. It is now understood that non-transmissible plasmids can undergo relaxase-in trans, where a relaxase from another plasmid in the same cell can be utilized to mobilize the non-transmissible plasmid (Moran and Hall, 2018;Ramsay and Firth, 2017). In most cases, the non-transmissible plasmid must have a highly similar oriT, called a mimic oriT, for this to work (O'Brien et al., 2015). This was shown to not be an absolute requirement if there was a relaxosome accessory factor (RAF) that could alleviate the need for a close oriT match for mobilization to occur (Moran and Hall, 2017). Nontransmissible plasmids have also been theorized to propagate via conduction (Clark and Warren, 1979), or by transduction (Smillie et al., 2010).
Plasmids impose a cost on their host organism, so their long-term existence is deemed the plasmid paradox (Harrison and Brockhurst, 2012) as the beneficial genes should be incorporated into the chromosome or the costly plasmid should not propagate and persist (Baltrus, 2013). Plasmid persistence can occur by multiple mechanisms such as host cell adaptations and addiction systems (Bustamante and Iredell, 2017;Loftie-Eaton et al., 2017). Small antimicrobial resistance (AMR) plasmids that co-infect hosts with large AMR plasmids have increased stability when the larger plasmids are present (San Millan et al., 2014). Plasmids that do not carry any AMR genes, nor any other genetic information that is advantageous to the host, but are maintained due to a high copy number can be deemed small, abundant cryptic plasmids (SCPs) (Burian et al., 1997). The experimental data on SCPs remains limited, and generally, the incidence of co-transfer of SCPs is likely underreported as it may easily be overlooked in experimental evaluations if there are no identified genes of interest on the SCPs.
The application of next generation sequencing techniques to KPCE from our institution revealed that the majority of isolates contained multiple plasmids in addition to bla KPC -plasmids (Sheppard et al., 2016a). Although little directed effort has been focused on the mobility of non-resistance plasmids from "real-world" highly drug-resistant Gram-negative bacteria, previous work to understand plasmid populations has demonstrated that co-existence and interactions amongst plasmids in a single bacterial cell is common and that these interactions play an important role in plasmid stability and mobilization (Christiansen et al., 1973;Laufs and Kleimann, 1978;San Millan et al., 2014). Co-transfer of plasmids was proposed to be a non-independent event that is limited primarily by the plasmid with the lowest conjugation efficiency (Gama et al., 2017). Thus, once conjugation is initiated, multiple conjugative plasmids from the same cell can transfer jointly. Other work has shown that strains carrying multiple plasmids may be more fit than strains carrying single plasmids (Silva et al., 2011), which can help explain why multiple plasmids might persist in a host over generations. Although conjugative plasmids do not make up the majority of plasmid types (Smillie et al., 2010) many laboratory studies regarding the interaction or transmission of multiple plasmids have been focused solely on conjugative plasmids. Small plasmids are often non-conjugative and rely on a high copy number to avoid segregational loss (Summers, 1998); they frequently depend on other plasmids to facilitate their horizontal transfer (Ramsay et al., 2017).
To better understand the conjugation frequency of small and large plasmids from "real world" CPE isolates, we used bla KPC -positive Citrobacter freundii collected from patients and the wastewater environment at our institution. C. freundii is a recognized nosocomial pathogen (Samonis et al., 2009) and is the fourth most common species to carry bla KPC at our institution (Sheppard et al., 2016a). We describe the frequency and patterns of transfer of two SCPs commonly seen in C. freundii with larger plasmids carrying bla KPC . Improving our understanding of the dissemination of MGEs and AMR genes in patients and wastewater niches is of importance in developing future approaches to reducing their emergence and spread in the healthcare environment.

Materials and methods
2.1. DNA extraction, whole genome sequencing, assembly and annotation DNA was extracted utilizing a commercial kit (QuickGene DNA Tissue Kit S, Fuijifilm, Japan) as previously described (Stoesser et al., 2013) and sequenced by both long read (Pacific Biosciences, CA) and short read (HiSeq 2000, Illuminia, San Diego, CA) sequencing technologies. A hybrid, complete assembly was generated for each isolate using methods previously described (Sheppard et al., 2016b). Assemblies were annotated using Serial Cloner 2.6.

Conjugation
Single colonies of J53 rif Escherichia coli and bla KPC -positive C. freundii were grown at 37°C overnight in Luria broth (LB) without and with meropenem (3.3 μg/mL) respectively. Overnight cultures were added to a sterile nitrocellulose membrane atop a sheeps' blood agar plate (Thermo Scientific Remel, San Diego CA) at a 1:2 donor to recipient ratio and incubated at 37°C for 8 h. The filter was aseptically removed from the agar plate with forceps and placed in 5 mL of Phosphate Buffer Solution (PBS) and then vortexed to dislodge the bacteria. 200 μL of the liquid suspension was plated on meropenem (1 μg/mL) and rifampicin (600 μg/mL or 250 μg/mL) LB plates. Single colonies were selected and a biochemical Indole test (Becton, Dickerson, and Co, Sparks, MD) that can differentiate between C. freundii and E. coli was done to ensure that all transconjugant colonies were E. coli and not C. freundii.

Real-time PCR to assess plasmid presence in transconjugants
Based on the hybrid assemblies, primers were designed (Eurofins Genomics, Louisville, KY) to target stable regions of each study plasmid that were unique when compared with other plasmids in the host cell (Table 1). The PCR master mix was made using PowerUp SYBR Green Master Mix (Applied Biosystems, Austin TX) according to the manufacturer's directions for a 20 μL reaction. Each bacterial colony underwent a crude boil prep extraction (95°C for 10 min) (Mathers et al., 2011) and 2 μL of the supernatant was added as the template. The PCR assay was run on the BioRad CFX96 Real Time PCR System using the following parameters: 50°C for 2 min, 95°C for 2 min, and cycling 95°C for 15 s, 60°C for 15 s, and 72°C for 20 s for 40 cycles with a melt curve step (65-95°C, 0.5 increments for 5 s). Each transconjugant was then tested for the presence/absence of all potential plasmids individually by PCR.

Southern gel electrophoresis
Selected isolates were grown overnight in a shaking culture at 37°C in 50 mL LB with a 1μg/mL meropenem concentration and then processed using the Qiagen CompactPrep Plasmid Midi Kit (Qiagen, Hilden, Germany) following the manufacturer's instructions. The extracted DNA was then quantified (Nanodrop, Thermofisher Scientific).
All samples were loaded on a 0.7% 0.5× TBE gel and run at 70 V overnight at 4°C. Purified bla KPC PCR product, was added for the last 10 min at 100 V to use as a positive control. The gel was then processed, blotted, labelled, hybridized, and underwent detection using the Amersham ECL Direct Nucleic Acid Labelling and Detection Systems (GE Healthcare, United Kingdom) according to the manufacturer's protocol. The same primers that were described in Table 1 were used to run PCR on the parent strain, and the PCR product was purified using the Qiagen PCR Purification Kit (Qiagen, Hilden, Germany) according to the manufacturer's instructions. The purified PCR product was used to create probes for the blotting.

NCBI query
Using nucleotide BLASTn, both the sequences for p3223 and p1916 were queried in the NCBI database using > 99% identity and across > 80% coverage as thresholds for a match.

Alignment of plasmid sequences
Comparative analysis between plasmids carried by CAV1321, CAV1741 and CAV1857 was performed using progressive MAUVE (Darling et al., 2004) with a match seed weight of 15 and a minimum LCBscore of 30,000.

In silico tools used for characterizing isolates
All plasmid sequences were imported into the web based program, PlasmidFinder (Carattoli et al., 2014) 2.0, for in silico plasmid typing. The plasmids were typed using the Enterobacteriaceae database, with minimum thresholds of 80% identity and 60% coverage. As recommended, small plasmids (< 20 kb) were typed with greater stringency at > 80% identity and > 96% coverage (Carattoli et al., 2014).
The putative oriTs for the target plasmids were identified in silico using the web based tool oriTfinder (Li et al., 2018) and named from the oriTDB. All putative oriT sequences identified were then compared using ClustalW (https://www.genome.jp/tools-bin/clustalw).

Estimating genomic resemblances between plasmids
The genomic resemblances between circularized plasmids from the 3 C. freundii isolates CAV1321, CAV1741 and CAV1321 was estimated using the MinHash algorithm implemented in MASH (Ondov et al., 2016).

Data availability
Complete genome assemblies have been deposited in the GenBank database under BioProject PRJNA246471.

Strain and plasmid characteristics
271 bla KPC positive C. freundii (90 from patients and 181 from the environment) were identified and sequenced (Illumina) between 2007 and 2017 and some were also selected for subsequent long-read sequencing (PacBio) to generate closed assemblies. Three C. freundii isolates with closed assemblies were selected for conjugation experiments because they had highly related chromosomes (< 100 SNVs) and harbored the two same SCPs, but varied with their remaining plasmid content and organization ( Fig. 1 and Supplement Table S1). All 3 strains are sequence type (ST)22; they had no apparent direct epidemiologic link. The 3 C. freundii strains were as follows; CAV1321 (patient isolate November 2010), CAV1741 (patient isolate October 2012), and CAV1857 (environmental intensive care room sink isolate December 2013). Combined the three isolates carried 22 total circularized structures (Fig. 1). It should be noted that there appears to be a high amount of recombination between plasmids and thus traditional typing methods targeting plasmid backbone regions do not provide enough resolution to describe the variability here. All three strains harbor identical SCP structures, namely p3223 (3223 bp; accession no. NZ_CP011652.1) and p1916 (1916 bp; accession no. NZ_CP011651.1), which have not undergone recombination. p3223 is a mobilizable plasmid that encodes for a MobC gene, replication initiation protein, DNA polymerase, and a small multidrug resistance (SMR) transporter (Fig. 2). p1916 is annotated almost entirely of hypothetical proteins besides a TM2 domain-containing protein. Due to the lack of an identified relaxase, it could be considered traditionally non-transmissible (Smillie et al., 2010). Each of the SCPs has a putative oriT according to in silico analysis. p3223 has oriT_ pH205 and p1916 has oriT_pNPO1. pCAV1741_KPC and pKPC_CAV1857-85 also have a recognizable putative oriT region (oriT_pCTXM360), while the other bla KPC plasmids in this study do not have a known oriT that has been previously described. Isolate CAV1321 harbors nine plasmid structures, two of which contain bla KPC , namely pKPC_CAV1321-45 (accession no. CP011608.1) and pKPC_CAV1321-244 (accession no. CP011611.1) (Fig. 1). Isolate CAV1741 harbors six plasmid structures and only one of those plasmid carries bla KPC , namely pKPC_CAV1741 (accession no. CP011656.1). Isolate CAV1857 harbors seven plasmid structures, with pKPC_CAV1857-85 (accession no. CP037737.1) and pKPC_CAV1857-43 (accession no. CP037739.1) both carrying bla KPC . Both pKPC_CAV1321-45 and pKPC_CAV1857-43 are homologous to the plasmid pKPC_UVA01 (accession no. NZ_CP017937.1) which was our original index case plasmid from 2007 (Sheppard et al., 2016b). Amongst the three strains, there are both homologous regions of the plasmids that are shared, and some plasmid regions that are unique to one parent strain ( Fig. 3 and Supplementary Table S1).

Conjugation frequency of p3223 and p1916
A total of 143 bla KPC -PCR positive transconjugants (59 J53/ CAV1321, 30 J53/CAV1741, and 54 J53/CAV1857) were generated from three separate mating experiments which were done multiple times to gather an adequate number of transconjugants for analysis and to minimize pseudoreplication effects. The transconjugants underwent

Fig. 1. Plasmid Fragment Map
There is a wide amount of genetic variability amongst the plasmids within the 3 Citrobacter freundii isolates. Common colors illustrate areas of homology amongst plasmid fragments of each isolate. Fragments are initially colored using CAV1321 plasmid fragments as a reference, followed by CAV1741 then CAV1857.

Plasmid movement when parent strain has multiple bla KPC -plasmids
Isolates CAV1321 and CAV1857 each have two separate bla KPCplasmids. In the transconjugants created from those parent strains, both SCPs and both bla KPC-plasmids had co-mobilized in 84% (95/113) cases. The other 18 transconjugants can be broken down into the follow categories: only one bla KPC -plasmid transferred (n = 7), neither bla KPC -plasmid backbone transferred, but bla KPC was detected (n = 8), or both bla KPC -plasmids co-mobilized but p1916 did not (n = 3).

Transposition
An interesting phenomenon occurred in 14% (8/59) of the J53/ CAV1321 transconjugants, where bla KPC was detected in the transconjugants but the PCR primer target locations on the plasmid backbones observed to originally carry bla KPC in the donor were not found. To help us understand this phenomenon, a second area of each plasmid was also targeted. However, all results for PCR of another region of the original bla KPC -plasmids were also negative. Southern blotting was performed to investigate if the Tn4401 transposon mobilized onto another plasmid in the donor cell prior to conjugation. A southern blot of all eight transconjugants was probed for bla KPC as well as the aforementioned plasmid area targets for pCAV1321-71 and all five small plasmids in CAV1321. The results showed various plasmids sizes from the transconjugants consistent with a transposition event occurring in vitro. We were able to get increased resolution on the transposon movement in 88% (7/8) of the transconjugants in question. The plasmids where the bla KPC probe was co-localizing with on the southern blot were pCAV1321-71, pCAV1321-4938, and pCAV1321-3820.

NCBI query
In order to provide a global context for the prevalence of p3223 and p1916, the sequences were queried against the NCBI database using BLASTn. As of October 2018, there were 20 strains of eight species; K. pneumoniae (9), C. freundi (2), Serratia marcescens (2), Enterobacter  cloacae (2), E. coli (2), Salmonella enterica Serovar Typhimurium (1), Raoutella ornithinolytica (1) and Enterobacter hormaechei (1) that had a match to p3223. All of the matches were in isolates harboring other AMR genes. Specifically, 95% (19/20) of the hits were associated with isolates also carrying plasmids encoding beta-lactamases including bla CMY-2 , bla KPC , bla NDM , bla OXA , bla CTX-M-15 , and bla TEM . Thirty-five percent (7/20) were alignments with published isolates from our own collection. The BLASTn query for the p1916 sequence had four hits -two of which were isolates from our institution and represented in this study, namely CAV1321 and CAV1741. The other two matches were to an E. cloacae (accession no. CP030349.1) and an E. coli (accession no. KF992024.1) sequence. All of these isolates also carried beta-lactamase genes: bla KPC , bla NDM , bla OXA , bla VIM , and bla TEM on other plasmids.

Discussion
Plasmids are diverse MGEs that can be readily shuttled between host strains through horizontal gene transfer; however, much of the focus has been on plasmids which carry identifiable genes which alter host function (i.e. antimicrobial resistance or virulence). The co-transfer and persistence of SCPs in Enterobacteriales has yet to be fully explained and remains largely overlooked. It was recently recognized that SCPs may not require their own unique relaxase and can transfer by utilizing another relaxase if the oriT demonstrates enough similarity or has a RAF that can help accommodate the differences in oriT (Moran and Hall, 2018). The oriTs of the SCPs (oriT_pNPO1 and oriT_pH205) and the bla KPC plasmids pCAV1741_KPC and pKPC_CAV1857-85 (or-iT_pCTXM360) are not homologous, therefore the traditional relaxase in trans method does not explain the mobilization of p1916. As previously mentioned, pKPC_CAV1321-45 and pKPC_CAV1857-43 are homologous to pKPC_UVA01, all which have no recognizable putative oriT or relaxase. pKPC_UVA01 has been shown to be transmissible but has lower conjugation efficiency in vitro than that observed epidemiologically (Hardiman et al., 2016). In our study, we have demonstrated very high rates of co-transfer of almost all targeted plasmids in" real-world" isogenic C. freundii isolates with a diverse plasmid content including unique bla KPC -plasmids, albeit under laboratory conditions. Small plasmids can impose a fitness cost to the host that is similar to the high cost of a large plasmid (San Millan et al., 2009). If a plasmid imposes a high cost to the host, does not provide any benefit, and is without an addiction system, it should not persist. Since neither p3223 nor p1916 carry any AMR genes, identifiable addiction systems or any other obvious genes of benefit to the host bacterium and theoretically cannot mobilize on their own, their continued existence across numerous species and strains is hard to understand. They were not shown to be essential for bla KPC plasmid conjugation in our experiments as they were not found in every transconjugant, but they did co-transfer with high frequency. We speculate that these SCPs frequently transfer with conjugative plasmids carrying AMR genes because they potentially could have a helper role, but this would require further testing outside the scope of this initial evaluation.
During in vitro conjugation of bla KPC -plasmids from 3 C. freundii isolates with a J53-E. coli recipient, the larger bla KPC -plasmids never transferred alone. Even when multiple large plasmids carried the same resistance gene, both large plasmids would transfer to the recipient the majority of the time (87%). Although this phenomenon has been previously recognized (Ramsay and Firth, 2017) it is not thought to occur so frequently largely limited by the idea that each plasmid contains a unique oriT and relaxase (Zechner et al., 2017). It has been hypothesized that coexisting AMR plasmids in a cell can produce positive epistatic effects on their host, thus minimizing the costs of carrying multiple plasmids (San Millan et al., 2014). This is concerning as it increases the ability of AMR plasmids to persist in the absence of selective pressure (San Millan et al., 2014). Consistent with this, we demonstrate that multiple plasmids (of varying size and diversity) transferring during conjugation were the norm rather than the exception. Ultimately, we saw a high rate of transfer of SCPs as well as multiple large plasmids with redundant resistance (Fig. 4).
The amount of genetic diversity within the mobile elements (n = 22 plasmids) across three chromosomally similar strains from a single institution all with the same AMR gene is striking. There has not been widespread transmission of bla KPC -C. freundii (Jimenez et al., 2017) but local outbreaks have been identified. We hypothesize that all of these isolates were locally acquired as this strain has been identified longitudinally in both patients and premise plumbing. Postulating about the evolution of this strain at our institution, three possibilities are: (1) the C. freundii strain acquired a single bla KPC plasmid ancestor with subsequent genetic rearrangement, (2) there were multiple occasions where the bla KPC was acquired, (3) the strain acquired multiple plasmids in a single transfer. Prior to our in vitro work described here, the first and second explanation may have seemed more likely, but based on our findings, the third explanation is also plausible. It is most likely a combination of genetic rearrangement events and uptake of multiple plasmids that explain the evolution of this strain. Such variability in plasmid content within the same strain over time has not been well characterized previously but we propose it can occur with high frequency. Additional work characterizing the mechanisms of co-transfer and persistence would be important to understanding horizontal gene transfer in a non-laboratory setting.
Another important finding was the relative high frequency with which in vitro transposition of a characterized transposon with a clinically relevant carbapenemase gene potentially occurred (Cuzon et al., 2011). It seemed to occur in 5% (8/143) of the transconjugants, but only when CAV1321 was the parent strain. CAV1321 is less susceptible to rifampicin than the other C. freundii strains used and thus the conjugation had to be done under a greater concentration of rifampicin (600 μg/mL vs 250 μg/mL). Increased antibiotic pressure could be the cause of why visible transposition events occurred in only the J53/ CAV1321 transconjugants. Overall the frequency of transposition was likely underestimated because our experimental assay would not have picked up additional transposition events within the bla KPC plasmids.
There are several shortcomings to the manuscript. First, we have only evaluated conjugation from a single strain and species. This was done intentionally in part to control the contribution of the host bacterium and focus solely on the plasmid DNA of clinical and environmental "real-world" strains. In addition, the SCPs were shown to encode a significant number of hypothetical proteins which could perform beneficial functions, but were not evaluated in our study. Also, though the BLASTn query in NCBI only returned a finite amount of hits, this number may actually be underrepresenting the abundance of p3223, and other potential important small plasmids, in the environment. As more sequence data are uploaded to public repositories we may see greater evidence of the presence of SCPs. Of note, with Pacific Biosciences SMRT sequencing, the library preparation frequently size selects for larger DNA fragments to optimize sequencing, thus potentially excluding small plasmid structures from the sequencing process which can only be identified with a hybrid assembly (George et al., 2017).
We have observed p1916 and p3223 being shared across many species of Gammaproteobacteria in both our isolate collection and NCBI's global collection. p3223 has found its way into multiple species carrying bla KPC demonstrating that the finding of multiple plasmid transfer of SCP with large AMR plasmids can occur in vivo as well. We propose further evaluation needs to be done on the potential beneficial/ synergistic relationship between large KPC plasmids and the SCPs during horizontal gene transfer as this could be crucial in the dissemination of antimicrobial resistance genes.

Conclusion
As more sequencing work is done globally, it is becoming more evident that bacterial isolates can harbor numerous plasmids and that experimental models need to account for interplay between plasmids in a host. Further work is needed to understand the relevance and function of SCPs, and their role in the dissemination of AMR plasmids.

Funding
This work was supported by the Centers for Disease Control and Prevention Broad Agency (BAA 200-2017-96194), the Health Innovation Challenge Fund (grant HICF-T5-358 and WT098615/Z/12/ Z) and funding through the NIHR Oxford Health Protection Research Unit on Healthcare Associated Infection and Antimicrobial Resistance (HPRU-2012-10041).

Declarations of interest
None.