Persistence of antibiotic resistance in bacterial populations


  • Editor: Fernando Baquero

Correspondence: Dan I. Andersson, Department of Medical Biochemistry and Microbiology, Uppsala University, Box 582, SE-751 23 Uppsala, Sweden. Tel.: +46 18 471 4175; fax: +46 18 471 4673; e-mail:


Unfortunately for mankind, it is very likely that the antibiotic resistance problem we have generated during the last 60 years due to the extensive use and misuse of antibiotics is here to stay for the foreseeable future. This view is based on theoretical arguments, mathematical modeling, experiments and clinical interventions, suggesting that even if we could reduce antibiotic use, resistant clones would remain persistent and only slowly (if at all) be outcompeted by their susceptible relatives. In this review, we discuss the multitude of mechanisms and processes that are involved in causing the persistence of chromosomal and plasmid-borne resistance determinants and how we might use them to our advantage to increase the likelihood of reversing the problem. Of particular interest is the recent demonstration that a very low antibiotic concentration can be enriching for resistant bacteria and the implication that antibiotic release into the environment could contribute to the selection for resistance.


Scope of the review

Antibiotic concentrations in the environment are the main driving force for the selection of resistant bacterial populations. Antibiotic resistance mechanisms are usually associated with fitness costs for bacteria and should reduce the frequency of resistance in the absence of antibiotic selection, but compensatory evolution can potentially restore fitness without loss of acquired resistance. Thus, the continued maintenance of antibiotic resistance in a bacterial population is expected to be a function of the rates and magnitudes of competing forces selecting for and against resistance. In addition, there is a possibility that some resistance mechanisms might be cost-free. This review will consider and evaluate examples for which there is experimental evidence relevant to the reversibility and persistence of antibiotic resistance in bacterial populations. We will consider the roles played by compensatory evolution, cost-free mutations, genetic coselection and plasmid-specific mechanisms in slowing the reversibility of antibiotic resistance.

Framing the problem of antibiotic resistance

The introduction of antibiotics in the mid-20th century was arguably the single most important medical event in recent history with regard to reducing human morbidity and mortality. However, the subsequent and continuing intensive use of antibiotics, both in medicine and in agriculture, estimated to total several million tons worldwide since their introduction (Wise, 2002; Andersson & Hughes, 2010), has helped to select a huge increase in the frequency of resistance among human pathogens. High frequencies of resistance significantly reduce the possibility of effectively treating infections. This increases the risk of complications and fatal outcome (Helms et al., 2002; Depuydt et al., 2008), increases the economic burden on health care systems (Cosgrove, 2006; Sipahi, 2008; Roberts et al., 2009) and may ultimately threaten a postantibiotic era (Guay, 2008; Lew et al., 2008; Woodford & Livermore, 2009).

Questions of persistence and reversibility

A rational approach to dealing with antibiotic resistance problems requires detailed knowledge of the different biological and nonbiological factors that affect the rate and extent of resistance development. On the one hand, this will require quantitative information on antibiotic selective pressures and on bacterial population structures and how the interplay between these influences rates of resistance development. On the other, it will also be necessary to quantify the influence of resistance on bacterial fitness and to determine the relationships between fitness costs and changes in the frequency of resistant pathogens. It is noteworthy that acquired antibiotic-resistance traits are frequently found in bacteria isolated from humans and animals that are not known to have been subjected to significant antibiotic exposure and that live in remote parts of the world (Pallecchi et al., 2008). In this context, one of the most important issues is to understand how resistance is stabilized in bacterial populations, in particular under environmental conditions where antibiotic selective pressure may be very low or absent for significant periods of time. In theory, resistance might be stabilized by (1) compensatory mutations that restore fitness without loss of resistance, (2) by the occurrence of rare cost-free resistance mutations or (3) by genetic linkage and coselection between resistance mutations and other selected genetic markers (e.g. virulence factors or resistances) as well as by other factors (Fig. 1). There is experimental evidence that supports the relevance of each of these stabilization mechanisms and we will discuss some examples in later sections. There are also a number of theoretical studies modeling the relationship between the biological cost of resistance, the rate by which resistance develops and the steady-state frequency of resistance at a given antibiotic pressure as well as the rate by which resistance is predicted to disappear if antibiotic use is reduced (Levin et al., 1997; Andersson & Levin, 1999; Austin & Anderson, 1999; Austin et al., 1999; Levin et al., 2000; Levin, 2001, 2002; Andersson, 2003). In addition, there are factors that will confound these models (e.g. clonal shifts and changes in population structure unrelated to antibiotic selection) affecting the expected relationships and making predictions less certain (Andersson & Levin, 1999; Austin & Anderson, 1999; Austin et al., 1999; Levin et al., 2000; Levin, 2001; Andersson, 2003). Finally, we will describe the outcomes of laboratory studies and clinical interventions to reduce the frequency of resistant bacteria at the community level and explain why reversibility, if it occurs at all, will occur so slowly that it in most cases, it is unlikely to be of practical importance.

Figure 1.

 Various selective forces and mechanisms that will act to either cause loss or persistence of resistance.

Driving forces for reversibility

When a resistance problem has been generated and the frequency of resistance has reached a level that prohibits the successful use of empirical therapy, any potential for reversibility of the problem will be determined by essentially two factors – dilution and the fitness cost – depending on whether one considers open or closed systems. Examples of an open system would be a hospital or an animal farm where there is a continuous in- and out-flow of humans or animals and where the incoming population has a lower frequency of resistance than the resident population (which is generally exposed to a higher antibiotic pressure). If antibiotic use is reduced in the hospital/animal farm, the driving force for reversibility is essentially a dilution effect where the incoming population reduces the frequency of resistance in the hospital/farm. This notion is supported by mathematical modeling (Austin & Anderson, 1999; Lipsitch et al., 2000; Bonten et al., 2001) as well as by clinical interventions where quite rapid (weeks to months) reductions in resistance can be observed after the use of specific antibiotic has been decreased (McGowan, 1986; Mahamat et al., 2007; Aldeyab et al., 2008). In a closed system, for example in a community where migration is limited, the main factor determining whether antibiotic-resistant populations can be displaced with antibiotic-susceptible ones is the biological fitness cost of resistance, i.e. any effect the resistance mechanism has on reducing the ability of the pathogen to reproduce and spread in the host population. Both theoretical arguments (Levin et al., 1997, 2000; Levin, 2001; De Gelder et al., 2004) and clinical data (Bottger et al., 1998; Sander et al., 2002; Cohen et al., 2003) support the notion that the larger the fitness cost, the lower the frequency of resistant bacteria in the population at a given antibiotic pressure. Similarly, if antibiotic use is reduced, the resistant mutants with the highest fitness costs are expected to be outcompeted and disappear the fastest, but unfortunately, if no fitness cost is present, the resistant mutants will remain even if antibiotic use is discontinued.

Factors that will cause persistence

1. Direct selection for resistance marker in the presence of high drug concentrations [above the minimal inhibitory concentration (MIC)]. The rate of appearance of antibiotic-resistant bacteria in a susceptible population will be a function of the combined rates of de novo mutation, which need not be high (Örlén & Hughes, 2006), and of the horizontal gene transfer (HGT) of resistance determinants. However, the more important measure is the rate at which existing resistant variants increase in frequency as a function of the level of drug exposure in pathogen populations (Bergman et al., 2004, 2009). The level of exposure in clinical settings will be associated with pharmacokinetic and pharmacodynamic aspects of the drug in patients that affect clearance of the pathogen or selection of resistant variants within the patient (Olofsson et al., 2007), and also with hygienic and transmission control measures in clinics and hospitals. In addition, the volume and distribution of antibiotics released into the wider physical environment can be expected to contribute to the selection of resistance and transfer of resistance by HGT. This latter point is important because the large reservoir of resistance genes, even within the human microbial community, could potentially serve as donors for transfer of resistance to human pathogens (Sommer et al., 2009).

2. Sub-MIC selection. While it is evident that antibiotic concentrations above the MIC of the bacteria can select for resistant variants, it still largely remains unclear whether levels far below MIC can be selective. Because many natural environments and body compartments will be exposed to antibiotic concentrations ≪MIC, due to anthropogenic pollution (Thiele-Bruhn, 2003; Cabello, 2006; Li et al., 2008; Kummerer, 2009) or antibiotic treatments (Drlica, 2003; Drlica & Zhao, 2007), there is in theory also ample opportunity for resistance to be selected under sub-MIC conditions (Baquero et al., 1997; Negri et al., 2000) because of very small reductions in the growth rate of the susceptible strains (Fig. 2). Recent results suggest that this process might be more important than previously realized. Thus, we have observed for several well-defined, clinically common mutants of Escherichia coli and Salmonella enterica with resistance to three different classes of clinically used antibiotics (tetracyclines, fluoroquinolones and aminoglycosides) that selection of resistant bacteria can occur at antibiotic concentrations up to several hundred-fold below the MIC of the susceptible bacteria, corresponding to absolute antibiotic concentrations in the μg mL−1 to pg mL−1 range (E. Gullberg et al., unpublished data). These results were obtained using competition experiments between pairs of susceptible and resistant strains that were genetically tagged with green fluorescent protein (GFP) and counted by fluorescence-activated cell sorting. This experimental setup allowed the detection of growth rate differences as small as 0.3% (Lind et al., 2010). Furthermore, using a different experimental setup, recent results from Liu et al. (2011) report similar findings. In conclusion, these results imply that the ultralow (≪MIC) antibiotic concentrations found in many natural environments are, in contrast to common belief, sufficiently high to confer the selection and persistence of antibiotic resistance.

Figure 2.

 Schematic representation of growth rates as a function of antibiotic concentration. The sub-MIC and traditional selective windows where the resistant strain will outcompete the susceptible strain are indicated. MICsusc, minimal inhibitory concentration of the susceptible strain; MICres, minimal inhibitory concentration of the resistant strain; and MSC, minimal selective concentration where the resistant mutant is enriched over the susceptible strain.

3. Coselection between resistance markers (within genetic elements and within clones). A resistance marker could increase in frequency by virtue of its genetic linkage to a selected marker in the population and coselection is a common feature of resistance that is acquired by HGT. This is clearly of importance with regard to plasmids, transposons and integrons, but it also applies to the carriage of resistance in the context of any multiresistant clone. Because of genetic linkage, the frequency of resistance to a particular antibiotic could in theory remain stable or even increase in environments where the antibiotic might not currently be in use. The causes governing the dynamics of an unselected resistance marker in a bacterial population could be many. Selection might be for another resistance determinant or for a determinant of clonal fitness not directly related to antibiotic resistance. Regardless of the particular cause, the implication of coselection is that those strains that increase in the population do so because they are more fit in the presence of the antibiotics that are in use. There are examples in the literature where there is little or no correlation between changes in the consumption of a particular antibiotic and the observed changes in the frequency of resistance and this has in some cases been tentatively attributed to coselection. For example, a recent intervention study in Kronoberg County in Sweden, where the use of trimethoprim-containing drugs was decreased by 85% during the 24-month period of the study, reported only a marginal effect on trimethoprim-resistance levels in E. coli (Sundqvist et al., 2009). The authors calculated that it would take a sustained intervention for 12 years just to bring the level of resistance from the 12% measured at the start of the intervention back to the level of approximately 8% measured in 1991. The reasons for this disappointingly small effect were reasoned to be due to a combination of the small fitness cost measured for trimethoprim-resistance together with a strong coselection for other antibiotics (e.g. mecillinam, furantoin, fluoroquinolones, cephalosporins, etc.), which remained in use during the intervention period. Similarly disappointing results were also recorded in an earlier study from Great Britain when the use of sulfonamides was reduced. The consumption of cotrimoxazole in the United Kingdom was reduced by 97% between 1991 and 1999, but this did not result in a reduction in sulfamethoxazole resistance (Enne et al., 2001). A follow-up study based on additional data from the same area in 2004 showed that sulfamethoxazole and streptomycin resistance in E. coli had remained stable despite a continued low use of these drugs (Bean et al., 2005). The absence of an effect on resistance was attributed in part to the surprising observation that at least one of the clinical plasmids carrying sulfonamide resistance actually increased fitness for E. coli in laboratory tests (Enne et al., 2004). An additional contributing factor to the observed coselection was that sulfamethoxazole resistance was genetically linked to other resistance genes that continued to be selected during the study period. In two earlier nationwide studies carried out in Finland and Iceland, the effects of reduced antibiotic use on resistance were analyzed retrospectively (Kristinsson, 1997; Seppäläet al., 1997). In both studies, decreases in antimicrobial resistance frequencies were noted. A reduction in the use of macrolides in Finland was followed by a significant decrease in erythromycin-resistant Streptococcus pyogenes while in Iceland a decrease in Streptococcus pneumoniae resistant to penicillin followed a decrease in the use of antibiotics in children a couple of years earlier. However, later reanalysis of the data suggested that for both studies, the reduction might have been caused by clonal replacements unrelated to the reduction in antibiotic use (Kataja et al., 1999; Arason et al., 2002). The actual driving force for this clonal replacement is not known, but logic suggests that it is likely to be a consequence of the different fitness profiles of the different clones involved; however, this has not been tested.

4. Cost-free resistances. To determine whether a resistance mechanism is truly cost-free for a bacterium is difficult because it can always be argued, in spite of repeated demonstrations that a resistance appears cost-free under the tested conditions, that there is some environmental condition where a cost might be present. In addition, limitations of the experimental setups will also make claims of cost-free resistances difficult to confirm. By single culture/animal growth experiments, differences usually have to be >5%/several-fold to be reliably detected. The resolution can be increased considerably by performing competition experiments, and by utilizing GFP tagged and pre-adapted strains, differences as small as 0.3% can be identified (Lind et al., 2010). However, differences smaller than 0.3% are extremely difficult to detect due to the phenomenon of periodic selection, where one of the competitors acquires an adaptive mutation during the competition, resulting in a noninterpretable experiment. Thus, claims that have been made that certain mutations appear to be cost-free need to be evaluated with the above limitations and caveats in mind (Ramadhan & Hegedus, 2005; Criswell et al., 2006). One example of a mutation that has been suggested to be cost-free, based on in vitro and animal experiments, is streptomycin resistance caused by the substitution K42R in ribosomal protein S12 (Tubulekas & Hughes, 1993; Kurland et al., 1996; Bjorkman et al., 1998; Enne et al., 2005). This substitution is also found at a high frequency in streptomycin-resistant clinical isolates of Mycobacterium tuberculosis (Bottger et al., 1998; Sander et al., 2002). However, highly sensitive competition experiments (see Driving forces for reversibility) show that this mutation does confer a small cost (about 3%) that has gone undetected previously. Whether this cost is medically relevant with regard to the short/long-term persistence of the mutation in a pathogen population is unclear, but theoretically, one would expect a selection coefficient of −0.03 for a mutation to provide a significant counter-selective force in most bacterial populations.

5. Fitness-increasing resistances. Although it is normally observed that antibiotic resistance is associated with a fitness cost for bacteria in the absence of drug selection, it is in principle possible that resistance could increase fitness in drug-free environments. An example of this type has recently been demonstrated with regard to resistance to fluoroquinolones in E. coli (Marcusson et al., 2009). Resistance to fluoroquinolones develops in E. coli in a multistep process, typically including the acquisition of efflux mutations and several different mutations altering drug-target proteins (Heisig, 1996; Komp Lindgren et al., 2003, 2005). In addition, there is increasing evidence that a large variety of fluoroquinolone-resistance genes and mechanisms can be acquired by HGT events (Martinez-Martinez et al., 1998; Robicsek et al., 2006; Perichon et al., 2007; Jacoby et al., 2008). Each individual resistance mutation, gene or mechanism contributes only to a very small increase in MIC and clinically relevant levels of resistance always require the accumulation of several alterations. In step-wise laboratory selections, the accumulation of up to five fluoroquinolone resistance mutations in E. coli was shown to be generally associated with significantly reduced fitness (Komp Lindgren et al., 2005). However, in a few of the evolved lineages, this monotonous relationship was broken and an increase in resistance was found to be associated with an increase in fitness at a late selection step (Komp Lindgren et al., 2005). A reversal of the expected fitness–resistance relationship was also noted in constructed strains of S. pneumoniae carrying one or two fluoroquinolone resistance mutations (Rozen et al., 2007). To establish the relationship between cause and effect, a set of isogenic E. coli strains were constructed carrying, in different combinations, up to five fluoroquinolone-resistance mutations (Marcusson et al., 2009). Within this set of constructed isogenic strains, low-fitness triple-mutants were identified, where the addition of a fourth resistance mutation significantly increased fitness in vitro and in vivo while at the same time dramatically decreasing drug susceptibility (Marcusson et al., 2009). The largest effect was observed when a parC mutation (topoisomerase IV) was added to a low-fitness strain carrying resistance mutations in gyrA (DNA gyrase) and marR (upregulating drug efflux). Increased fitness was accompanied by a significant change in the level of gyrA promoter activity. When the same low-fitness triple mutants were subjected to selection and competition experiments performed in the absence of drug, spontaneous parC mutations that improved fitness and reduced drug susceptibility were also selected. The implication of these data is that natural selection for improved growth in low-fitness bacteria with a low-level resistance to fluoroquinolones can select for reduced drug susceptibility in the absence of drug selection (Marcusson et al., 2009).

6. Compensatory evolution reduces costs and allows maintenance even without selective pressure. The fitness costs of drug resistance can be reversed, at least partially, by compensatory mutations without any significant loss of resistance (Andersson & Hughes, 2010). Thus, several in vitro and in vivo studies show that when resistant bacteria are serially passaged, compensatory mutations are easily selected, although the degree of restoration varies and is usually below that of the original susceptible strain. However, in principle, by the reiteration of cycles of selective compensation, resistant clones could become as fit as susceptible clones. Accordingly, genetic compensation could act to stabilize and maintain resistant populations even in the absence of antibiotic selection. The details of which mutations will be associated with genetic compensation for any particular drug–bacteria combination should depend on the interplay of mutation rates, fitness of different mutants and bottleneck size during serial transfer (Levin et al., 2000; Maisnier-Patin et al., 2002; Handel et al., 2006). This general expectation is supported by experimental data from studies of fusidic acid-resistant fusA mutants and streptomycin-resistant rpsL mutants in S. enterica ser. Typhimurium (Johanson et al., 1996; Bjorkman et al., 1998, 1999, 2000; Nagaev et al., 2001; Maisnier-Patin et al., 2002). Thus, genetic compensation occurs much more frequently than reversion. This bias in favor of new mutations is also expected to be relevant for natural populations, when population bottlenecks are present. Although in principle new mutations could restore fitness by a variety of different mechanisms, direct restoration of the impaired function is, according to the available data, the most common mechanism (Johanson et al., 1996; Bjorkman et al., 1998, 2000; Nagaev et al., 2001).

Here, we outline four illustrative examples of different fitness compensatory mechanisms. The first example concerns a case of apparent reversion to the wild-type genotype that actually occurs by a process of gene conversion. In Staphylococcus aureus, resistance to linezolid caused by a mutation altering a 23S rRNA gene was observed in one case to be lost (Meka et al., 2004b) and in the other case to be considerably reduced (Meka et al., 2004a), after the removal of the antibiotic selective pressure. In both cases, linezolid-resistance carried a fitness cost and the loss or the reduction in the resistance phenotype was caused by gene conversion between the multiple copies of 23S rrn genes where at least one copy had remained wild type in sequence. In the above examples, selection for increased fitness is leading to reversion of resistance. However, gene conversion works both ways and in an environment where selection favored increased resistance, for example in the presence of low levels of antibiotic, gene conversion would provide a rapid genetic and phenotypic solution, providing at least one gene copy carried a pre-existing resistant allele. A similar gene conversion mechanism could explain the phenomenon of heterogenous macrolide resistance observed in pneumococci (Wolter et al., 2006). Although it is not currently in clinical use, resistance to kirromycin, an antibiotic that targets the translation elongation factor EF-Tu, is also strongly influenced by gene conversion because EF-Tu is produced in many bacteria from duplicate tufA and tufB genes (Hughes, 1990). Depending on the type of selection pressure that is imposed, either for increased drug resistance, or for increased fitness, the process of tuf gene conversion can either copy a resistance mutation to the second tuf gene or it can restore a wild-type sequence to both tuf genes (Abdulkarim & Hughes, 1996; Hughes, 2000). This makes the development of resistance to drugs that target the products of repetitive genes (such as rRNAs and EF-Tu) highly unstable. On the one hand, resistance is very unlikely to arise in one step because it is recessive and would require multiple mutations, but on the other, the effects of a single small effect mutation may be rapidly amplified by gene conversion.

In the second example, environment and epistasis have been shown to play significant roles in determining the magnitude of fitness costs associated with antibiotic resistance. The fitness costs of streptomycin resistance (SmR) caused by mutations in rpsL (encoding ribosomal protein S12) is partly caused by reduced rates of protein synthesis affecting growth rates in vivo and in vitro (Bjorkman et al., 1999), but it can also depend on environmental and epistatic effects (Paulander et al., 2009). Thus, the Salmonella typhimurium SmR mutants K42N and P90S have impaired growth on a rich medium, but in media with poorer carbon sources, the same SmR mutants grow faster than the wild type. The reason appears to be that the drug-resistant mutants do not induce the stress-inducible sigma factor RpoS (σS), a key regulator of many stationary phase- and stress-inducible genes, and thus escape the growth inhibition that this induction imposes on drug-susceptible wild-type cells under conditions of environmental stress and starvation (Paulander et al., 2009). In another example, a mutation in gyrA that reduces susceptibility to ciprofloxacin in Campylobacter jejuni enhances the fitness of the mutant strain in competition against an isogenic drug-susceptible strain in a chicken infection model, in the absence of drug selection (Luo et al., 2005). However, the same mutation in a different strain of C. jejuni imposes a fitness cost. These examples show that fitness costs can be influenced by the environment in which they are measured and raise the hope that by imposing suitable environmental conditions, it might be possible to increase the selection against some drug-resistant pathogens. The influence of epistatis in modulating the magnitude of fitness costs may be a relatively common feature of drug resistance that is currently under-appreciated simply because of a lack of relevant research.

Modulating the potential costs of resistance by the appropriate regulation of gene expression may be of critical importance in maintaining resistance genes in the absence of direct drug selection. Drug resistance plasmids usually carry a fitness cost in terms of a reduced growth rate, at least at the stage when they enter naïve bacteria (Lenski & Bouma, 1987; Bouma & Lenski, 1988; Smith & Bidochka, 1998; Johnsen et al., 2002; Dahlberg & Chao, 2003). The nature of these costs may vary and, in most cases, has not been determined, but it is likely that the costs are strongly associated with the disruptive physiological effects of introducing multiple new biochemical pathways into the host bacteria. Compensatory evolution following plasmid acquisition by naïve E. coli has been shown experimentally to occur both on the plasmid and on the chromosome and to create genetic combinations for which fitness would be reduced if the plasmid were subsequently lost (Dahlberg & Chao, 2003). In such situations, the plasmid carrying drug-resistant genes may be maintained stably in the bacteria even in the absence of selection with antibiotics. An example relevant to both epistasis and the importance of gene regulation concerns the AmpC-type β-lactamase. The gene for ampC is not intrinsic to the chromosome of Salmonella spp. and the AmpC-resistance phenotype was rarely found in clinical isolates. When ampC was experimentally transformed on a plasmid into Salmonella, it reduced the growth rate and invasiveness, suggesting reduced fitness as a plausible reason for its rarity (Morosini et al., 2000). However, when both ampC and its regulator ampR were introduced (making β-lactam resistance inducible rather than constitutive), these fitness costs were eliminated (Morosini et al., 2000). In this example, the genetic context of the resistance gene, and the physiological costs of expressing the gene, determined the relative fitness cost associated with its acquisition. However, this fitness barrier has not prevented the spread of ampC-β-lactamases to Salmonella. Plasmid-borne AmpC-type enzymes are now found widely distributed in veterinary and clinical isolates; reviewed in (Jacoby, 2009). Interestingly, not all plasmids bearing ampC carry the regulator gene ampR and experimental measurements suggest that the biological cost of high-level AmpC production can be compensated by other plasmid-encoded functions (Hossain et al., 2004). Another example in which appropriate gene regulation has been shown to reduce the cost of resistance is illustrated by VanA-type glycopeptide resistance in S. aureus (Arthur et al., 1996). Resistance is acquired by HGT of a vanA resistance operon and results from the synthesis of alternative cell wall precursors (ending in d-Ala-d-Lac) with a low affinity for glycopeptides and the elimination of the normal susceptible precursors (ending in d-Ala-d-Ala) to which vancomycin binds (Bugg et al., 1991). Resistance is induced in response to glycopeptides (Arthur et al., 1992). When it was shown that the vanA gene had crossed the genus boundary from vancomycin-resistant enterococci to methicillin-resistant S. aureus (MRSA), it was feared that it would spread rapidly, representing an immediate threat to patient care and with the possibility of spreading to other, more virulent pathogens (Willems et al., 2005). However, VanA-positive S. aureus have not spread in the clinical setting as initially feared and the number of MRSA strains that are also resistant to vancomycin remains very small (Perichon & Courvalin, 2009). Pairwise comparisons of isogenic MRSA strains with and without three different clinical isolates of the VanA resistance operon showed that when induced, the fitness cost was large and reduced the growth rates by 20–38% (Foucault et al., 2009). In contrast, in the absence of induction, the fitness cost of carrying the VanA operon was only 0.04–0.3%. Thus, VanA-type resistance can persist because it has a low cost in the absence of the drug and although it is costly in absolute terms when induced by the drug, it is also essential for bacterial survival under those conditions. This example clearly illustrates the importance of gene regulation and physiology in considering the potential and real costs of antibiotic resistance for bacteria. It is likely that the costs initially associated with many plasmids and other HTG sequences are related to the disruption they cause to growth physiology by inappropriate patterns of gene expression and that to survive in the absence of drug selection, they must adapt to the host bacteria by restoring an appropriate physiological balance to their gene expression.

Finally, the genetic flexibility associated with genetic duplication and amplification (GDA) events may be important because it increases the probability of certain resistances arising and also because it can serve as a crude way of regulating the costs of resistance in the absence of selection. By increasing genetic target size GDA increases the probability of occurrence of mutations that can permanently alter the resistance and/or fitness phenotypes (Andersson & Hughes, 2009; Sandegren & Andersson, 2009). GDA can also reversibly alter resistance and fitness phenotypes. For example, Salmonella mutants resistant to the peptide deformylase inhibitor actinonin carry mutations in either of two genes required for the formylation of methionyl initiator tRNA (tRNAi): fmt and folD. In the absence of antibiotic selection, these mutations reduce fitness. It has been shown experimentally that approximately one-third of the extragenically compensated fmt mutants carried amplifications of the tandemly repeated metZ and metW genes, encoding the initiator tRNA (Nilsson et al., 2006). The increase in metZ and metW gene copy number was by up to 40-fold, increasing tRNAi levels and compensating for the lack of methionyl-tRNA formyltransferase activity. The advantages of GDA mechanisms are that they are very frequent, they facilitate the selection of relevant mutations by increasing target size and they are reversible in situations where selection pressure is subsequently relaxed.

7. Plasmid persistence. Plasmids typically carry genes that are conditionally beneficial and as a result plasmids are expendable under many conditions. Therefore, the continued maintenance of plasmids requires either a continuous selection (e.g. for antibiotic or heavy metal resistance genes) or the presence of other mechanisms that ensure persistence. Apart from the previously mentioned processes (1–6), which can confer persistence to both chromosome- and plasmid-encoded resistance mechanisms, there exist several plasmid-specific mechanisms that ensure plasmid maintenance even in the absence of direct selection. For example, several interesting cases have been described where plasmids confer increased fitness in the form of faster growth rates in the absence of any selective pressures. Thus, both conjugative and nonconjugative plasmids (carrying various antibiotic resistance genes) can substantially increase bacterial exponential growth rates, implying that the plasmids are maintainable without drug selection (Enne et al., 2004; Dionisio et al., 2005; Yates et al., 2006). The plasmid functions responsible for this fitness increase are unknown at present. Apart from systems that directly increase host fitness, plasmids also often encode stabilization functions, including plasmid multimer resolution systems, plasmid partitioning system and various types of postsegregational systems (Bahl et al., 2009), which ensure stable plasmid maintenance in the host cell. Another potential maintenance mechanism, even when the plasmid has a detrimental effect on growth, is infectious transfer, where plasmid transmission is efficient enough to overcome negative selection and plasmid loss during host growth. Mathematical modeling has shown that such conditions may exist when conjugative transfer functions are derepressed (at least transiently) and bacterial densities are high (Lundquist & Levin, 1986).

Conclusions and perspectives

One major conclusion emerging from the past decade or more of studies into antibiotic resistance fitness costs and their relationship with the dynamics of resistance development is that the rate of reversibility in clinical settings will be, at best slow, and possibly, nonexistent. The major reason is that, even if a fitness cost is associated with resistance, compensatory genetic evolution will rapidly reduce this cost and thereby reduce the probability of reversal. In addition, if there is coselection between the resistance mechanism and other selected markers (as there often is for plasmid-borne antibiotic resistance), this will tend to reduce the rate of reversibility driven by fitness costs. This first conclusion could encourage despair. However, a second major conclusion emerging from the recent research into the interplay between resistance and fitness is that we should be able to apply this knowledge to develop strategies to address current resistance problems and to reduce the likelihood of resistance development, in particular for any novel drugs that may be introduced. One obvious possibility is to exploit our expanding knowledge of genomics, physiology and resistance mechanisms, to direct the choice of novel drugs, and drug targets, for which actual or potential resistance mechanisms would be associated with a high fitness cost and where the expected rate and/or the extent of genetic compensation would be low. A second possibility is to apply our growing understanding of the physiological basis for fitness costs associated with resistance, to design novel therapies, possibly involving drug combinations, to target any Achilles heel associated with a particular resistance mechanism. Finally, the knowledge we are accumulating of the dynamics and mechanisms of antibiotic resistance development should be integrated into mathematical models that can predict the effects of different drug-dosing regimens on efficacy and resistance development. To the extent that we succeed in developing good predictive models, we will have the possibility to shape the outcome of this medically important evolutionary experiment.


Our work is supported by the Swedish Research Council, Vinnova, Strategic Research Foundation (SSF) and European Commission in the 7th framework program (PAR project).