SEARCH

SEARCH BY CITATION

Keywords:

  • antibiotic resistance;
  • epistasis;
  • experimental evolution;
  • multiple drug resistance;
  • pleiotropy

Abstract

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References

The spread of bacterial antibiotic resistance mutations is thought to be constrained by their pleiotropic fitness costs. Here we investigate the fitness costs of resistance in the context of the evolution of multiple drug resistance (MDR), by measuring the cost of acquiring streptomycin resistance mutations (StrepR) in independent strains of the bacterium Pseudomonas aeruginosa carrying different rifampicin resistance (RifR) mutations. In the absence of antibiotics, StrepR mutations are associated with similar fitness costs in different RifR genetic backgrounds. The cost of StrepR mutations is greater in a rifampicin-sensitive (RifS) background, directly demonstrating antagonistic epistasis between resistance mutations. In the presence of rifampicin, StrepR mutations have contrasting effects in different RifR backgrounds: StrepR mutations have no detectable costs in some RifR backgrounds and massive fitness costs in others. Our results clearly demonstrate the importance of epistasis and genotype-by-environment interactions for the evolution of MDR.


Introduction

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References

The evolution of bacterial antibiotic resistance poses an important threat to human health, increasing mortality rates and monetary costs associated with bacterial infections worldwide (Levy, 1998; Levy & Marshall, 2004; Evans et al., 2007). In recent years, the incidence of infections attributable to multiple drug-resistant (MDR) strains of many bacterial species, including Mycobacterium tuberculosis, Staphylococcus aureus, Pseudomonas aeruginosa and several Enterococcus spp. has increased, and it is currently estimated that, depending on the species in question, between 20% and 80% of bacterial infections worldwide are caused by MDR strains (McCormick et al., 2003; Verdier et al., 2006; Wright et al., 2006). Given the elevated costs and difficulties associated with treating MDR bacterial infections (Levy, 1998), it is important to understand the evolutionary mechanisms that constrain and promote MDR.

One factor likely to constrain the evolution of MDR bacteria is the cost of antibiotic resistance. Many studies have shown that resistance mutations are associated with a pleiotropic fitness cost; in terms of reduced fitness in the absence of the antibiotic to which they confer resistance (reviewed by Andersson & Levin, 1999). The most commonly occurring resistance mutations in clinical isolates are often those that incur the lowest fitness cost in the laboratory (Andersson, 2006; Gagneux et al., 2006), suggesting that the costs of resistance play an important role in determining patterns of antibiotic resistance emergence and spread. Despite the apparent importance of fitness costs in the evolution of drug resistance, few studies have quantified the costs associated with carrying multiple resistance mutations (Enne et al., 2004; Lindgren et al., 2005; Rozen et al., 2007). This is an important limitation, as the cumulative cost of carrying multiple resistance mutations will depend on the cost of each individual mutation, as well as any interactions between them. For example, an antagonistic interaction between resistance mutations would reduce the relative fitness cost of resistance, whereas a synergistic interaction would increase the relative fitness cost of resistance.

The objective of this study was to quantify the fitness costs associated with carrying MDR mutations, such that we were able to test the hypothesis that epistatic interactions between independent mutations conferring resistance to different antibiotics determine the fitness cost of MDR. In order to do this, we measured the fitness cost of streptomycin resistance (StrepR) mutations in six rifampicin-resistant strains of P. aeruginosa, each with a different rifampicin resistance (RifR) mutation in RNA polymerase (rpoB), by calculating the growth rate of StrepR mutants relative to their streptomycin-sensitive (StrepS) ancestral strains. As a negative control, we also measured the cost of streptomycin resistance mutations in a rifampicin-sensitive genotype (PAO1). A diagram illustrating the history of the strains tested, from their common ancestor, can be found in Fig. 1.

image

Figure 1.  History of the Pseudomonas aeruginosa test strains used in our experiment, from their common ancestor PAO1. A schematic diagram of our experimental design.

Download figure to PowerPoint

Our results clearly demonstrate that although the cumulative costs of multiple resistance mutations tend to exceed the cost of a single resistance mutation, the cost of carrying resistance mutations to a second antibiotic, in this case streptomycin, depends on the genetic background in which resistance evolves (i.e. on the resistance mutations that are present at other loci) and the interaction between genetic background and the environment in which the cost is expressed.

Materials and methods

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References

Strains

Seven isogenic strains of P. aeruginosa PAO1 were used for this investigation. Six of these strains (D521G, H531R, Q518R, Q152L, Q152R and S517L) each had a different nonsynonymous mutation in rpoB conferring resistance to the antibiotic rifampicin (R. C. MacLean and A. Buckling, unpublished data). The name of each strain refers to the amino acid change in the RNA polymerase protein associated with each mutation. The seventh strain was the wild-type P. aeruginosa strain (PAO1) from which the six RifR mutant strains were derived. The use of isogenic lines ensured that streptomycin resistance could only evolve via the appearance of novel chromosomal mutations (Perron et al., 2006). We confirmed that all ancestral strains were fully sensitive to streptomycin prior to the beginning of our experiment.

Selecting for streptomycin resistance

We used a fluctuation test (Luria & Delbruck, 1943) to generate libraries of independent streptomycin-resistant mutants in each genetic background. A culture of each ancestral strain, derived from a single clone that had been incubated overnight at 37 °C in M9KB, was diluted 10−6 into M9KB, and 200 μL of this diluted culture was inoculated into each well on two 96-well microtitre plates (initial titre≈1000 CFU per well). The resulting 14 plates were then incubated for 24 h at 37 °C to produce stationary-phase cultures. To select for streptomycin-resistant mutants, 4 μL of a 100× dilution of each culture was immediately pipetted on to an M9KB agar plate containing streptomycin at 240 μg mL−1, which was the minimum concentration of streptomycin required to inhibit the growth of PAO1 during preliminary experiments. One agar plate was used for each microtitre plate; so there were two plates for each ancestral strain. These plates were incubated for approximately 36 h. Following incubation, 24 independent streptomycin-resistant (StrepR) mutants of each ancestral strain were randomly chosen (12 from each plate).

To generate libraries of cultures that could be easily assayed, we set up microplates containing cultures of streptomycin-resistant strains alongside their appropriate controls. Each microplate contained six cultures of a common control strain (PAO1), six cultures of an ancestral strain and two cultures of each of the 24 streptomycin-resistant mutants derived from the given ancestral strain. These cultures were placed in the inner 60 wells on a 96-well microplate to avoid any edge effects. Outer wells were filled with sterile culture medium.

Fitness assay

In order to assay the fitness of StrepR mutants and their ancestral strains, we measured the growth rate of all genotypes when cultured in antibiotic-free Pseudomonas culture medium (M9KB: 10 g L−1 glycerol, 20 g L−1 proteose peptone #3, 12.8 g L−1 Na2HPO4·7H2O, 3 g L−1 KH2PO4, 0.5 g L−1 NaCl, 1 g L−1 NH4Cl) and in medium supplemented with rifampicin (M9KB + 64 ug mL−1 rifampicin). Overnight microplate cultures containing streptomycin-resistant genotypes and their streptomycin-sensitive controls were diluted by a factor of 102 into microplates containing 200 μL of M9KB per well, and by a factor of 106 into microplates containing 200 μL of M9KB + rifampicin per well. Fewer cells were transferred into wells containing media with rifampicin to minimize the possibility of transferring spontaneous PAO1-derived rifampicin-resistant mutants into the cultures. All cultures were then incubated at 37 °C. We used an automated microplate reader to measure the optical density (OD 600 nm) of these cultures at approximately hourly intervals until all cultures had reached stationary phase. The assay was then repeated and the results from replicate assays pooled, so that we assayed the growth of four replicate cultures of each StrepR mutant, 12 cultures of each StrepSRifR ancestral strain and 96 cultures of PAO1 in each assay environment.

Measuring fitness

We defined fitness as the exponential growth rate of each culture. This was calculated by estimating the linear rate of change in optical density during the log phase (Vmax) – i.e. when population growth is exponential. Optical density is proportional to the log of population size: Vmax therefore provides an estimate of the growth rate, r, of each culture, such that Nt Nert.

To estimate the fitness cost of a mutation conferring streptomycin resistance in M9KB and in M9KB + rifampicin, we subtracted the mean growth rate (i.e. absolute fitness) of each streptomycin-sensitive (StrepS) ancestral strain (n = 12 or 96 replicates per strain) from the mean growth rate of each StrepR mutant derived from the StrepS ancestral strain (n = 4 replicates per strain) (Hartl & Clark, 2007), such that negative values represent a cost of acquiring streptomycin resistance.

Sequencing

In many bacteria, including Escherichia coli, streptomycin resistance arises as a consequence of a small number of mutations in the 30S subunit of the ribosome (rpsL; Björkman et al., 1998, 1999). To determine if ribosomal mutations were responsible for streptomycin resistance in our StrepR mutants, we sequenced the rpsL gene in 16 StrepR mutants derived from PAO1. Genomic DNA was isolated from each colony using a Wizard Genomic DNA extraction kit (Promega, Southampton, UK) as per the manufacturer’s instruction. We amplified rpsL using primers rpsLup (5′-TGTCGTAAGACAATCAGTGGAGC-3′) and rpsLdown (5′-CCGACCTTACTCTTATCGACTC-3′), which are external to the 5′- and 3′-ends of the gene respectively. Reaction mixtures consisted of BIOTAQ polymerase (Bioline, London, UK), 1 mm dNTPs, 16 nm (NH4)2SO4, 62.5 mm Tris–HCl (pH 8.8), 0.01% Tween 20, 2 mm MgCl2, and each primer at a concentration of 0.2 pm. Amplification reactions began with incubation at 94 °C for 5 min, followed by 35 cycles of 94 °C for 30 s, 60 °C for 30 s and 72 °C for 30 s, followed by a final incubation at 72 °C for 10 min. PCR products were enzymatically purified using ExoSap (USB Europe, Staufen, Germany). Purified PCR products were sequenced with both forward and reverse primers using BigDye 3.1 sequencing (Applied Biosystems International, Warrington, UK), followed by ethanol/EDTA precipitation of sequencing products.

Results

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References

Fitness interactions in the absence of antibiotics

To investigate interactions between RifR and StrepR mutations, we measured the fitness effect associated with acquiring streptomycin resistance in the six rifampicin-resistant genotypes and in a rifampicin-sensitive control (PAO1). Streptomycin resistance carries a fitness cost in the absence of antibiotics in all seven of the genetic backgrounds that we investigated (Fig. 2a,b).

image

Figure 2.  Costs of multiple drug resistance in the absence of antibiotics. Panel (a) shows the absolute fitness of streptomycin-sensitive strains of Pseudomonas aeruginosa and the mean fitness (±SEM; n = 24) of StrepR mutants in each genetic background. Panel (b) shows the fitness cost of StrepR mutants in each genetic background. Box plots show the mean (dashed line), median (solid line) and 10th, 25th, 75th and 90th percentiles of the distribution of fitness costs in each genetic background. We used two-tailed t-tests with 23 degrees of freedom to test the null hypothesis that the fitness cost of streptomycin resistance is equal to 0 in each genetic background (NS: > 0.05; *< 0.05, **< 0.01; ***< 0.001). We used a Bonferonni test to correct for multiple comparisons (n = 7).

Download figure to PowerPoint

If the fitness costs of StrepR and RifR mutations are additive, we would expect the cost of streptomycin resistance to be the same in both RifR and RifS genetic backgrounds. A diminished cost of StrepR in RifR backgrounds would indicate antagonistic epistasis between RifR and StrepR mutations, whereas an elevated cost of StrepR in RifR genetic backgrounds would indicate synergistic epistasis between RifR and StrepR mutations. To discriminate between these possibilities, we first compared the average cost of acquiring streptomycin resistance in RifR and RifS genetic backgrounds using pooled data on the cost of StrepR from all six RifR backgrounds. The average cost of acquiring StrepR in a RifR background (average cost −0.95, SEM 0.081, n = 144) is less than the cost of acquiring StrepR mutations in a RifS background (average cost −1.47, SEM 0.20, n = 24): this difference is significant using a t-test (= −2.42, d.f. = 166, = 0.017). To test for differences in the cost of resistance between individual RifR backgrounds and the RifS control, we used a Dunnett’s test. The only significant difference between a RifR background and the RifS control (PAO1) that this test identified was in the H531R background (average cost −0.56, SEM 0.13, n = 24, = 0.0074).

Fitness interactions in the presence of antibiotics

To test the possibility that interactions between StrepR and RifR mutations vary between environments, we measured the fitness of RifRStrepR mutants in culture media supplemented with rifampicin. In the presence of rifampicin, the cost of streptomycin resistance varies extensively between RifR genotypes, as determined by a one-way anova (Fig. 3a and b; F5,138 = 9.26, < 0.0001).

image

Figure 3.  Costs of multiple drug resistance in the presence of rifampicin. Panel (a) shows the absolute fitness of streptomycin-sensitive strains of P. aeruginosa carrying different RifR mutations and the mean fitness (±SEM; n = 24) of StrepR mutants in each RifR background. Panel (b) shows the fitness cost of StrepR mutations in each RifR background. Box plots show the mean (dashed line), median (solid line) and 10th, 25th, 75th and 90th percentiles of the distribution of fitness costs. We used two-tailed t-tests with 23 degrees of freedom to test the null hypothesis that the fitness cost of streptomycin resistance is equal to 0 in each genetic background (NS: > 0.05; *< 0.05, **< 0.01; ***< 0.001). We used a Bonferonni test to correct for multiple comparisons (n = 6).

Download figure to PowerPoint

StrepR mutations have no detectable average effect on fitness in four of six RifR backgrounds but entail significant and sometimes large costs in the remaining two RifR genetic backgrounds (Fig. 3b). If interactions between RifR and StrepR mutations are additive in the presence of rifampicin, the average fitness effect of StrepR mutations should be the same across all RifR backgrounds. The clear differences observed in the effects of StrepR mutations among RifR backgrounds imply that epistatic interactions must occur between RifR and StrepR mutations in the presence of rifampicin. However, the absence of a suitable negative control makes it impossible to determine the relative prevalence of antagonistic and synergistic epistasis.

Fitness effects across genetic backgrounds and environments

To test for differences in fitness effects of StrepR mutations across environments and genetic backgrounds, we fitted a model with main effects of genetic background (G), assay environment (E) and G × E interaction to our fitness cost data (i.e. Figs 2b and 3b) for all six RifR backgrounds (Fig. 4). The average cost of StrepR mutations varies significantly among assay environments (F1,276 = 31.9, = < 001), because the cost of resistance tends to be low in the presence of rifampicin. The cost of resistance varies significantly between genetic backgrounds (F5,276 = 5.14, = 0.0002), because StrepR mutations tend to have small costs in some RifR backgrounds, such as H531R, and large costs in other backgrounds, such as Q152L. Moreover, there is a significant effect of the interaction between genetic background and assay environment on the fitness cost of StrepR mutations (F5,276 = 10.92, = 0.0077). This interaction arises because the effect of environmental variation on the cost of StrepR mutations varies between genetic backgrounds (Fig. 4). For example, the presence of rifampicin in the environment substantially alleviates the cost of resistance in most genetic backgrounds, except in S517L, where it increases the cost of StrepR mutations.

image

Figure 4.  Costs of streptomycin resistance across genetic backgrounds and environments. This figure shows the average fitness cost of StrepR mutations in each RifR genetic background, in both the presence and absence of rifampicin.

Download figure to PowerPoint

Genetic basis of streptomycin resistance

To determine if ribosomal mutations generate streptomycin resistance, we sequenced the entire rpsL locus in 16 StrepR mutants derived from PAO1 using both forward and reverse primers. We failed to find evidence for rpsL mutations in any of the genotypes that we sequenced.

Discussion

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References

Summary

This study demonstrates that the evolution of MDR from single drug resistance in P. aeruginosa tends to be associated with an increased pleiotropic fitness cost of resistance (Figs 2–4). This cost is manifested as a decrease in the fitness of mutants resistant to streptomycin and rifampicin compared with singly resistant strains, both in the presence and absence of antibiotics. Intriguingly, we find that fitness interactions between resistance mutations are common, and these interactions are critical for determining the costs associated with carrying multiple resistance mutations.

Antagonistic epistasis and the cost of multiple drug resistance

The cost of streptomycin resistance in the absence of antibiotics is lower in RifR genetic backgrounds than in an isogenic RifS control (Fig. 2b). This result demonstrates a tendency for antagonistic epistasis in the cost of carrying MDR mutations. This finding is not entirely surprising, given that antagonistic epistasis seems to be a prominent feature of the genetic architecture of bacterial genomes (Bohannan et al., 1999; Elena & Lenski, 2001; Sanjuán & Elena, 2006) and at least one previous study has found evidence for antagonistic epistasis between resistance mutations (Rozen et al., 2007).

It is worth noting that, in the absence of antibiotics, doubly resistant mutants that do not pay an increased cost were observed in five of six rifampicin-resistant genetic backgrounds. The existence of these clones represents the worse possible scenario for the reinvasion of susceptible clones following drug removal, and suggests that MDR bacteria will be able to persist in the environment at the expense of susceptible strains even after antibiotic deployment has ceased. However, we argue that this finding must be interpreted with caution. First, we measured only four replicates of each RifRStrepR mutant, implying that we had low statistical power to estimate the fitness of individual mutants. Second, we were unable to identify the mutations responsible for streptomycin resistance and there is a possibility for nonindependence among streptomycin-resistant clones in each genetic background. Hence, we have performed a conservative analysis looking at the average cost of StrepR mutants in each genetic background without placing any emphasis on the differences in fitness between StrepR clones in each genetic background.

At the molecular level, it is known that streptomycin resistance can arise via a large number of mechanisms in P. aeruginosa. These include enzymatic inactivation of the drug by phosphotransferases and adenyltransferases, reduced drug uptake due to membrane impermeability or active efflux pumps, and ribosomal mutations that alter the target of streptomycin (reviewed in Poole, 2005). Two lines of evidence are consistent with the idea that mutations in many genes were responsible for streptomycin resistance in our experiment. First, the mutation rate towards StrepR in P. aeruginosa is very high (≈5 × 10−5 per replication; H. Ward & R. C. MacLean, unpublished data) implying that a large number of mutations must be able to generate this phenotype. Second, in a sample of 16 StrepR isolates that we sequenced, we failed to find any evidence for ribosomal mutations.

Our inability to identify the mutations responsible for the StrepR phenotype is a weakness of this study because it makes it impossible to determine the mechanistic basis of antagonistic epistasis. However, it is important to note that this limitation is not unique to our study: most experimental studies of epistasis in microbes, including very extensive experiments in E. coli, viruses and yeast, have failed to determine the mechanistic basis of epistasis (Elena & Lenski, 1997; Bonhoeffer et al., 2004; Sanjuan et al., 2004; Jasnos & Korona, 2007). The implication that streptomycin resistance has been acquired by a wide variety of genetic mechanisms demonstrates that the tendency towards antagonistic epistasis between StrepR and RifR mutations we observed does not depend on interactions between very small numbers of mutations, supporting the conclusion that fitness interactions between resistance mutations are common.

Genotype-by-environment interactions and the cost of resistance

It is often assumed that the deleterious effects of mutations will become more pronounced in stressful environments. In this system, however, this is clearly not the case. Although all clones had lower growth rates in M9KB + rifampicin than in M9KB, the average cost of StrepR was reduced in the stressful environment containing rifampicin compared with the benign environment lacking antibiotics (see Kishony & Leibler (2003) for similar data in E. coli). However, environmental stress caused by the presence of rifampicin has contrasting effects in different genetic backgrounds: environmental stress alleviates the costs of StrepR in most genetic backgrounds and enhances the cost of StrepR in the S517L background (Fig. 3). Although we lack a mechanistic understanding of the causes of these G × E interactions, these data provide a very clear demonstration that the deleterious effects of mutations, in this case the cost associated with StrepR mutations, are sensitive to both the environmental and genetic context in which they are expressed.

Applications and outlook

Previous studies have shown the fitness costs of resistance to be an important determinant of the prevalence of resistance alleles in a clinical environment (Andersson & Levin, 1999; Gagneux et al., 2006). Consequently, in the absence of antibiotics, the presence of antagonistic interactions between independent resistance mutations, which lower the cost of MDR, could enable MDR bacteria to persist at higher frequencies in the environment and may help to explain why MDR is so widespread in a clinical context. Future studies of interactions between mutations conferring resistance to different antibiotics in clinically isolated bacteria would be useful to determine whether antagonistic interactions between mutations in hospital infections help to maintain MDR in a clinical setting. Given that a small increase in fitness cost can translate into a large competitive disadvantage (Andersson, 2006), we expect that MDR genotypes displaying antagonistic epistasis between mutations will frequently be observed. Combinatorial therapies using drugs that have synergistic costs of resistance are likely to be most therapeutically effective, and the methods presented in this study could be extended to screen for such combinations on a large scale.

Acknowledgments

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References

We thank two anonymous reviewers for their comments on a previous version of this manuscript. This work was supported by funds from the MSc course in integrative bioscience (University of Oxford) to HW and by a research grant from the Royal Society to RCM.

References

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References