|Home | About | Journals | Submit | Contact Us | Français|
Understanding the survival of resistance plasmids in the absence of selective pressure for the antibiotic resistance genes they carry is important for assessing the value of interventions to combat resistant bacteria. Here, several poorly explored questions regarding the fitness impact of IncP1 and IncN broad host range plasmids on their bacterial hosts are examined; namely, whether related plasmids have similar fitness impacts, whether this varies according to host genetic background, and what effect antimicrobial resistance gene silencing has on fitness.
For the IncP1 group pairwise in vitro growth competition demonstrated that the fitness cost of plasmid RP1 depends on the host strain. For the IncN group, plasmids R46 and N3 whose sequence is presented for the first time conferred remarkably different fitness costs despite sharing closely related backbone structures, implicating the accessory genes in fitness. Silencing of antimicrobial resistance genes was found to be beneficial for host fitness with RP1 but not for IncN plasmid pVE46.
These findings suggest that the fitness impact of a given plasmid on its host cannot be inferred from results obtained with other host-plasmid combinations, even if these are closely related.
Antibiotic resistance is a serious threat to human and animal health and new ways to combat it are urgently needed. Broad-host range plasmids, such as those belonging to the IncN and IncP1 groups are important to the dissemination of antibiotic resistance due to their ability to replicate in a variety clinically relevant bacterial species and environments [1,2]. Indeed, both IncN and IncP1 group plasmids have been shown to encode clinically important resistance determinants such as blaCTX-M, blaIMP, blaNDM, blaVIM and qnr [3-8], whilst IncN plasmids have also been strongly implicated in the recent spread of blaKPC encoded carbapenemases .
Antimicrobial resistance can sometimes be accompanied by a reduction in biological fitness in the absence of antibiotic selection. Hence, less fit resistant bacteria may be outcompeted and displaced by fitter, susceptible bacteria in the absence of antibiotic use, leading to the suggestion that it may be possible to reduce the prevalence of antibiotic resistance by temporarily restricting prescribing. In practice, however, such approaches have enjoyed mixed success [10-14].
A fitness cost of antibiotic resistance has often been demonstrated in the case of chromosomal mutations conferring resistance, for example in the case of fusA mutations conferring resistance to fusidic acid  and gyrA mutations conferring resistance to fluoroquinolones . However, compensatory mutations can arise at secondary sites that reduce or eliminate this cost . In the case of acquired antibiotic resistance genes encoded on mobile genetic elements such as plasmids and transposons, the existence of a fitness cost is less clear. While early studies which often investigated cloning plasmids and/or laboratory strains demonstrated a cost to plasmid carriage [18-21], some more recent data using naturally-occurring plasmids and/or wild-type bacteria have failed to demonstrate significant costs and have sometimes shown a benefit. For example, the small sulphonamide and streptomycin resistance plasmid p9123 confers a 4% per generation fitness benefit in E. coli , and a benefit has also been demonstrated for some apramycin resistance plasmids isolated from bovine E. coli . A number of antibiotic resistance encoding plasmids and transposons conferred only a low fitness cost or were cost-neutral in the wild-type E. coli strain 345-2RifC in vitro and in the pig gut , whilst the resistance plasmid R751 and variants of it enhanced fitness under some growth conditions in E. coli . It is likely that the fitness cost a particular plasmid exerts on its host is variable depending on the plasmid as well as on the host itself. However, few studies have examined the fitness cost of a single plasmid on different strains of bacteria. The genetic factors, be they plasmid or host-encoded, that influence fitness are poorly understood, and it is not known whether related plasmids influence fitness in similar ways.
There are theoretically three ways in which a bacterial host can counteract the potential fitness cost exerted by antibiotic resistance genes carried on mobile genetic elements; the first is to acquire compensatory mutations, while the second is outright loss of the mobile genetic element. A third possibility is that bacteria could switch off the expression of resistance genes when they are not required whilst retaining the genes themselves in order to lower costs. We have previously demonstrated silencing of antibiotic resistance genes carried on the broad-host range plasmids pVE46 and RP1 by the wild-type E. coli strain 345-2RifC . Following passage through the pig gut, a small proportion (0.5%) of 345-2RifC(pVE46) colonies recovered lost expression of one or more of the four resistance genes encoded on the plasmid. Such isolates had retained the pVE46 plasmid and in most cases, intact, wild-type resistance genes and promoters were present, but no resistance gene mRNA was expressed. Similar results were found for three colonies of 345-2RifC(RP1) that also lost resistance following passage through the pig gut. Antibiotic resistance gene silencing appears to be restricted to only the plasmid with minimal effect on the remainder of the genome and is thought to be due to a mutation on the chromosome of E. coli 345-2RifC . Its precise mechanism is yet to be elucidated.
Here, we examine several unexplored questions regarding the fitness impact of broad host range IncP and IncN plasmids on their hosts; namely, the effect of the host background on fitness, whether related plasmids have similar fitness impacts and the fitness impact of antimicrobial resistance gene. To facilitate this task we also report the complete nucleotide sequence of the IncN plasmid N3.
The effect of host genetic background on the fitness impact of plasmid RP1 in the laboratory was investigated (Table (Table1).1). Five unrelated host strains representing all four E. coli phylogenetic groups were studied; E. coli 345-2RifC (group B1) and 343-9 (group D) of porcine origin, 99-24 (group D) and 99-40 (group B2) of human clinical origin (urine) and K12 (group A) JM109, a laboratory strain. Phylogenetic group B2, and to a lesser extent phylogenetic group D tend to be associated with extra-intestinal infections, whereas strains belonging to groups A and B1 are often commensals . There was considerable variation in the results obtained from different host backgrounds. The fitness impacts of RP1 on the strains of animal origin (343-9 and 345-8) were significantly lower than the costs imposed on those of human origin (JM109, 99-24 and 99-40) (p < 0.002 in all cases).
These results suggest that the fitness impact a particular antibiotic resistance plasmid confers on a given bacterial species is dependent on the genotype of the specific host strain that it is in. This conclusion is perhaps intuitive, but has to the best of our knowledge not been demonstrated for antibiotic resistance-encoding plasmids. One might expect this to be the case based on previous work by Dahlberg and Chao, who showed that amelioration of fitness costs conferred by the plasmids R1 and RP4 (very similar to plasmid RP1 used here) on E. coli K12 J53 depended on genetic changes in the host chromosome, thus implying a host genome component is involved in determining plasmid-encoded fitness cost . Similarly, the fitness cost and stability of the plasmid pB10 was highly variable in strains of different species [28,29]. Previous studies have also shown that target mutations leading to antibiotic resistance, for example gyrA mutations in Campylobacter jejuni or 23S rRNA mutations leading to clarithromycin resistance in Helicobacter pylori have different fitness effects in different host backgrounds [30,31]. It is not currently known which host genetic components may be important for determining the effect a plasmid will have on host fitness and it is likely that these will vary depending on the host-plasmid combination concerned. This finding has important implications for anyone wishing to use fitness cost as a parameter to model the spread or decline of a given plasmid in a bacterial population, perhaps in response to changes in antimicrobial selection, as it highlights the need to determine fitness in several different host genetic backgrounds. Similarly, recent work has also shown that fitness cost of antimicrobial resistance is variable depending on the growth conditions used in laboratory measurements [25,32], re-iterating the need for multiple measurements to obtain accurate fitness cost estimates.
Despite being a well-studied archetypal plasmid isolated in the 1960s, the DNA sequence of the IncN plasmid N3 has not previously been reported . Sequence analysis revealed that it is 54 205 bp in length, has a GC content of 51.1% and encodes 62 putative open reading frames (Table (Table2).2). It shares a common backbone with other IncN plasmids such as R46  and the recently described multiple antibiotic resistance plasmid pKOX105  (Figure (Figure1).1). The shared region comprises the plasmid's replication and transfer functions as well as genes encoding stable inheritance, anti-restriction and UV protection functions. N3 also encodes a class 1 integron and, in common with pKOX105 but lacking from R46, a type 1 restriction modification system. This characteristic and the high sequence identity shown between a number of proteins encoded by the two plasmids suggests pKOX105 may have evolved from a N3-like ancestor. N3 also encodes a unique region absent from other known IncN plasmids, bordered by IS26 elements. This comprises the tet(A) genes for tetracycline resistance, a putative bacA-like bacitracin resistance gene and seven novel genes. Several of the novel genes are predicted to have metabolic functions, most likely amino acid metabolism. Outside this region, the high similarity between N3 and other antibiotic resistance encoding IncN plasmids suggests that they have evolved from a common ancestor and diverged from each other relatively recently. The resistance region appears to have originated as a single class 1 integron initially carrying only an aadA1 cassette which has subsequently acquired further cassettes and/or insertions.
The fitness impacts of the related plasmids RP1 and pUB307 and R46 and N3 on E. coli 345-2RifC were compared. pUB307 is a derivative of RP1 which has lost the Tn1 transposon. The fitness impact of the Tn1 transposon itself has been demonstrated to be variable depending on the insertion site, with some insertion sites conferring a fitness benefit . Here, pUB307 had a small fitness cost of 1.9 ± 0.8% per generation, significantly lower than that of RP1 of -3.3 ± 0.9% per generation (students t-test p = 0.041). In animals, carriage of neither RPI nor pUB307 influenced the ability of E. coli 345-2RifC to colonize the pig gut compared to the plasmid-free 345-2RifC (ANOVA F value = 0.77, p = 0.471).
R46 was previously determined to confer a fitness cost of - 3.3 ± 1.7% per generation  in the laboratory, whilst no significant fitness cost in pigs was detected. In contrast, here, N3 was demonstrated to have a significantly higher fitness cost in the laboratory of 9.1 ± 1.8% per generation (students t-test p = 0.0002). In animals, 345-2RifC/N3 colonised the pig gut significantly worse than the plasmid free strain or 345-2RifC/R46 (ANOVA F value = 3.41, p = 0.035).
In the case of RP1 versus pUB307, these results suggest that the lower fitness cost of pUB307 compared to RP1 is related to the presence of less DNA. It is known that in single copy the Tn1 transposon does not itself have a detrimental effect on host fitness and can occasionally confer a benefit depending on the insertion site . Therefore, it can be assumed that in this case the advantage gained by deletion of Tn1 is due to the presence of less DNA and a lowered burden of gene expression as the TEM beta-lactamase encoded by the transposon is normally expressed at high levels. As RP1 is present in multiple copies, the burden of gene expression will be higher on the plasmid than in the case of Tn1 insertion at a single chromosomal site. Possible additional epistatic fitness effects due to the insertion site of Tn1 in RP1 will also be absent in pUB307.
The reason(s) why N3 and R46 have markedly different fitness costs is less clear, as the two plasmids are a similar size and share the same replication and conjugation functions. The marked fitness difference is therefore most likely due to accessory genes. The antibiotic resistance gene complement of the two plasmids is similar, although not identical (Figure (Figure1,1, Table Table2).2). The main differences are the presence of the arsCBADR on R46 and a Type 1 restriction system and a number of putative metabolic genes on N3. It is likely that one or more additional genes on N3 are responsible for the high fitness cost of N3 but this hypothesis requires experimental confirmation. Alternatively, a small mutation in the core plasmid genome may also be responsible.
... In addition to variable fitness costs brought about by different host-plasmid combinations, bacteria may influence the cost of plasmid carriage by modulation of gene expression. As antibiotic resistance can impose a fitness cost on the bacterial host in the absence of antibiotic selection, one might expect phenotypic silencing of plasmid-borne antibiotic resistance genes to confer a fitness advantage. The fitness costs of the plasmids pVE46 and RP1 on E. coli 345-2RifC had previously been established as moderate in vitro and non-detectable in vivo. Neither plasmid had a detectable cost in the pig gut . However, in both cases isolates that no longer expressed the resistance genes encoded on them but retained intact and wild-type resistance genes, were recovered during the pig gut colonisation experiments . Here, we investigated whether silencing of antibiotic resistance genes carried on pVE46 and RP1 had an effect on their fitness impact.
Three isolates with silent pVE46-encoded antibiotic resistance genes were investigated in vitro; L4, L5 and L7 (Table (Table3).3). Each isolate demonstrated variable degrees of antibiotic resistance gene silencing . Pair-wise growth competition assays were performed between silent isolates and the wild-type isolates expressing all antibiotic resistance genes. Isolate L5 had a slight in vitro cost of -2.1% ± 1.7% per generation whilst isolates L4 and L7 had slight fitness advantages of +1.1 ± 1.4% and +1.2% ± 0.5% per generation, respectively. However, the statistical significance of these results was low and overall the impact of silencing of pVE46 genes on fitness appeared negligible. The in vivo ability of isolate L5 to colonize the pig gut was found to be comparable to that of 345-2RifC(pVE46) (Figure (Figure22).
In contrast, antibiotic resistance gene silencing had a significant effect on the fitness of E. coli 345-2RifC(RP1). The silent isolates P1 and P2 (Table (Table3)3) both had fitness advantages of +2.5 ± 0.5% and +4.1 ± 3.7% in vitro, respectively. P2 was also able to colonize the pig gut better than 345-2RifC(RP1) (Figure (Figure22).
Surprisingly, antibiotic resistance gene silencing did not confer a fitness advantage on isolates carrying the pVE46 plasmid, in vivo or in vitro. This suggests that in this case antibiotic resistance gene silencing may have occurred by random chance that was fortuitously detected, or that if it exists, any fitness advantage only manifests itself under conditions not measured by our current assays. This observation may be explained by the fact that the initial cost conferred by carriage of pVE46 on E. coli 345-2RifC was moderate, 2.8 ± 0.9%, per generation. However, previous studies did show that pVE46-encoded antibiotic resistance genes were able to revert back to resistance at rates varying between 10-6 and 10-10 in vitro  suggesting that such strains may still pose a clinical threat.
In contrast, silencing of antibiotic resistance genes encoded on the plasmid RP1 conferred a significant fitness benefit both in vivo and in vitro. Such a strategy could be deemed beneficial for the bacterium, particularly if they were able to revert to antibiotic resistance again when challenged with antibiotic. However, this was not the case as none of the isolates with silent RP1 antibiotic resistance genes (P1, P2 or P3) were able to revert back to resistance in the laboratory. This suggests that the genetic event responsible for antibiotic resistance gene silencing of RP1 is not readily reversible, for example a transposon insertion or DNA deletion. Under such conditions one would expect the silenced DNA to eventually be lost, but until then it may act as an environmental reservoir of resistance genes.
In theory any fitness effects observed in silent isolates could also be attributed to unrelated mutations that may have arisen in the pig gut prior to their isolation. However, the silent isolate L5 is not known to carry any mutations compared to the wild-type 345-2RifC(pVE46) strain, whilst the possible role of unrelated mutations in the remaining isolates is yet to be determined (B.H. V.I.E and N.R.T, unpublished data).
Overall, the results presented here show that the fitness balance between the host genotype and a given resistance plasmid is extremely delicate and that even minor differences in the host or in the plasmid can have substantial effects on fitness. Future studies on the subject should therefore investigate multiple hosts in order to draw any general conclusions about a particular plasmid. Without better molecular understanding of the processes involved, it is difficult to predict the fitness impact of a given host-plasmid association, and hence difficult to make predictions about the spread or decline of associated antibiotic resistance phenotypes. It is therefore important to study molecular host-plasmid interactions. In the absence of such data one should preferably use a range of host strains and plasmids when studying the fitness of a particular resistance phenotype. As plasmids belonging to the IncN and IncP1 groups are broad-host range and conjugative they will likely move from host to host until they encounter one where costs are negligible and subsequently go on to thrive with that host. Thus, such plasmids may be of particular concern in the dissemination of novel antibiotic resistance phenotypes.
In addition, bacteria can sometimes "hide" their resistance genotype by silencing it. This can create a fitness advantage for the bacteria, whereas in other cases the silent phenotype is reversible, indicating that there is a risk of treatment failure in anyone infected by bacteria with silent genes and being treated by one of the affected antibiotics.
Based on these observations, further work should now concentrate on understanding the molecular mechanisms responsible so that the underlying process are understood and used to help develop better treatment and prevention and control strategies.
E. coli 345-2RifC, E. coli 345-8 and 343-9 are all commensal isolates of porcine origin. E. coli 345-2RifC is marked with a no-cost rifampicin-resistance mutation in RpoB (H526Y). Strains 99-24 and 99-40 are human urinary isolates, whilst E. coli K12 JM109 is a laboratory strain. Study strains were chosen on the basis that they did not carry acquired antibiotic resistance genes and that they exhibited good growth characteristics in laboratory media, with doubling ranging between 21 and 27 minutes in nutrient broth. Their phylogenetic group was determined as described previously . The relatedness of the isolates was investigated by randomly amplified polymorphic DNA (RAPD) PCR .
The broad-host range plasmids RP1, pUB307, R46, pVE46 and N3 were introduced into host strains by conjugation using the agar mating method . The 345-2RifC(pVE46) strain used was a variant passaged in the laboratory, the same from which silent isolates arose . Derivatives of 345-2RifC(pVE46) and 345-2RifC(RP1), carrying silent antibiotic resistance genes were as described previously . The characteristics of strains and plasmids used in this study are listed in Table Table33.
DNA of IncN plasmid N3 was prepared by alkaline SDS maxiprep and CsCl/EtBr density gradient centrifugation . The E. coli N3 plasmid was sequenced to approximately 37-fold shotgun sequence, totalling 1711 end sequences, from pUC19 (with insert sizes of 2-4 kb; 4-6 kb) genomic shotgun libraries that were sequenced using big-dye terminator chemistry on ABI3730 automated sequencers. The assembly was generated using phrap2gap. All repeat regions and gaps were bridged by read-pairs or end-sequenced polymerase chain reaction (PCR) products again sequenced with big dye terminator chemistry on ABI3730 capillary sequencers. The sequence was manipulated to the 'Finished' standard .
To assess the fitness impact of the plasmids upon E. coli host strains growth competition between plasmid-carrying and plasmid-free isogenic strain pairs was carried out as described previously in Davis minimal medium with 25 mg/ml glucose (DM25) . To estimate bacterial counts, competition cultures were diluted as appropriate and spread in triplicate onto IsoSensitest agar (Oxoid) and onto IsoSensitest agar containing the relevant antibiotic. For the competition between the silent strains L5 or L7 and 345-2RifC(pVE46) the agar contained tetracycline at 25 μg/ml, and for L4 it contained streptomycin at 25 μg/ml. For competition between 345-2RifC(RP1) and P1 or P2 agar contained ampicillin at 25 μg/ml. For competition between wild-type plasmids and their respective host strains it contained ampicillin for RP1 carrying strains, and tetracycline for the pUB307 and N3 carrying strains. Six replicates of each competition experiment were performed. Average per generation fitness (W) was calculated as W = 1 - b, where b is equal to t he gradient of the graph of ln(strain x count/strain y count) per transfer, divided by the number of generations per transfer (T). T was calculated as ln(dilution factor)/ln(2). The students t-test was used to estimate the statistical significance of results.
The recovery of resistance by isolates with intact but silent RP1 encoded resistance genes was investigated by spreading undiluted and serially diluted overnight nutrient broth cultures onto IsoSensitest agar containing the appropriate antibiotic (ampicillin, 25 μg/ml; kanamycin 30 μg/ml; tetracycline, 25 μg/ml). To calculate reversion frequencies, total cell counts were obtained following plating serial dilutions of the same culture onto antibiotic-free medium.
Animal experiments were carried out using a modified method of that described previously . For each experiment, six organic piglets from two litters of Saddleback-Duroc cross, weaned at five weeks of age, were housed as a single group for two weeks, to allow the animals to acclimatize to their surroundings. They were then randomly separated into two groups of three into pens with individual HEPA filtration and fed a standard organic feed (Organic feed company, grower/finisher pellets, UK) ad libitum. All procedures complied with the Animals (Scientific Procedures) Act 1986 and were performed under Home Office License.
Briefly, bacterial strains (E. coli 345-2RifC(pVE46), 345-2RifC(RP1), L5 and P1) were inoculated separately into six piglets as a single dose of 1010 cfu per animal by oral gavage. Faecal samples were collected from each animal by digital manipulation on day 3, 5, 7, 10, 12, 14, 17, 19 and 21 post-inoculation and analysed within 24 hours. One gram of faeces was suspended in nine millilitres of saline and plated at appropriate dilutions onto six MacConkey agar plates containing 50 μg/ml rifampicin (detection limit 2 cfu/g). They were incubated overnight at 37°C and colonies obtained replica plated onto MacConkey agar containing 50 μg/ml rifampicin with ampicillin (25 μg/ml), tetracycline (25 μg/ml), sulfamethoxazole (500 μg/ml) or streptomycin (25 μg/ml) for L5, and rifampicin with ampicillin, tetracycline or kanamycin (30 μg/ml) for P1, followed by replica plating onto MacConkey agar with rifampicin only.
The N3 DNA sequence has been submitted to EMBL under the accession number FR850039.
The authors declare that they have no competing interests.
BH, KB, MS, NRT and VIE performed the experimental work and data analysis. AAD and PMB participated in the study design. NRT, CMT, JMR and VIE co-ordinated the study and participated in the design. BH, NRT, CMT and VIE drafted the manuscript. VIE and PMB conceived the study. All authors read and approved the final manuscript.
This work was supported by grants OD2007 and OD2022 from the Department for Environment, Food and Rural Affairs (DEFRA) to P.M.B. and V.I.E, respectively, and grant 089222 awarded to V.I.E by the Wellcome Trust. We also thank The Wellcome Trust for their support of the Pathogen Genomics group under grant 076964.