The increasing rates of recovery of antimicrobial-resistant microorganisms in hospital and community settings are of growing concern (
1,
42). Resistance may emerge from a mutation in an intrinsic chromosomal gene or by acquisition of exogenous genetic material bearing resistance determinants. Resistance to antibiotics frequently reduces the fitness of bacteria in the absence of antibiotics; this is referred to as the “cost” of resistance (
38). In mathematical models, the fitness cost of resistance is the primary parameter that determines both the frequency of resistance at any given level of antibiotic use and the rate at which that frequency will change with changes in antibiotic use patterns (
3,
20,
21).
Restricted use of antibiotics is advocated not only to contain the dissemination of resistance but also to favor the nonexpansion and, finally, the disappearance of the resistant bacteria already present in human and environmental reservoirs (
3,
38). As a consequence of decreased use of antibiotics, rates of drug resistance usually fall but do not vanish, and stable rates of resistance in the apparent absence of direct selection pressure has been observed (
9,
12,
32). It is not clear whether this persistence of resistant bacteria is due to (i) low-level antibiotic contamination that maintains the selective pressure, (ii) selection by means other than antibiotics, or (iii) the stability of resistance genes.
Analogous to the resistance mediated by exogenous genetic elements (
13,
14,
19), chromosomal drug resistance-conferring mutations are commonly assumed to carry a fitness cost (
38). This is supported by the observation that some drug resistance mutations selected in vitro involve a significant decrease in bacterial fitness (
4,
20,
36); this fitness burden can subsequently be ameliorated by compensatory mutations (
4,
5,
36). However, for streptomycin resistance-conferring
rpsL mutations, a high level of selection for no-cost drug resistance mutations was suggested to exist in vivo (
6). In order to investigate whether this no-cost resistance mutation represents an isolated observation or points to a more general biological phenomenon, we examined the hypothesis that resistance mutations are costly.
MATERIALS AND METHODS
Bacteria.
Mycobacterium smegmatis rrnB represents a variant of
M. smegmatis mc
2155 SMR5 with a single rRNA allele;
M. smegmatis normally contains two rRNA operons (
33). Spontaneous drug-resistant mutants were selected on Luria-Bertani agar plates containing clarithromycin (50 μg/ml) or amikacin (20 μg/ml).
M. smegmatis rrnB rpsL3+ (strain 1682) is a streptomycin-sensitive derivative of mc
2155 with a single rRNA allele and was used to obtain streptomycin-resistant mutants by selection on brain heart infusion (BHI) agar containing 20 μg of streptomycin per ml. In brief,
sacB counterselection was used to inactivate the
rrnB operon, resulting in strain mc
2155
rrnB (strain 1434) with a single rRNA allele; subsequently, two additional
rpsL genes were introduced into the chromosome, resulting in mc
2155
rrnB rpsL3+ (
39).
Drug-resistant mutants were purified by streaking them on selective agar and were immediately frozen in 15% glycerol to prevent further genetic changes; further passages were done in the absence of antibiotics.
Recombinant DNA techniques.
Nucleic acids were analyzed by PCR-mediated sequencing with rRNA-specific primers. Genomic DNA was isolated as described previously; standard methods were used for restriction endonuclease digestion of DNA and other manipulations. Single rRNA mutations were introduced into integrative plasmid pMV361-H-rRNA (
34), which carries the complete
rrnB operon of
M. smegmatis, by PCR-mediated site-directed mutagenesis; plasmid pMV361 integrates once into the bacterial genome, thus providing a single copy of the mutated gene.
Transformation of mycobacteria.
Strains
M. smegmatis mc
2155
rrnB and
rrnB rpsL3+, which have single rRNA alleles, were used for the transformation experiments. Transformants were selected primarily on BHI plates containing hygromycin (50 μg/ml); subsequently, RecA-mediated gene conversion was used to obtain strains carrying homozygous mutant rRNA alleles (
30); mutants were passaged once on selective agar. The homogeneity of the mutations introduced was verified by manual sequence analysis with
32P-labeled CTP and Sequenase (U.S. Biochemicals); MICs were determined to verify the resistance phenotype.
Determination of bacterial fitness.
The cost of a resistance mutation was determined by direct competition against the drug-susceptible parental strain. Equal densities of drug-susceptible and drug-resistant strains were mixed and incubated in antibiotic-free BHI medium; every 24 h, 0.05 ml of the grown culture was transferred into 5 ml of fresh BHI medium for growth. Aliquots were plated every 24 h onto drug-free BHI agar to count the number of colonies. The number of drug-resistant bacterial cells was determined by plating the colonies on BHI agar containing the respective drug (streptomycin, 10 μg/ml; amikacin, 50 μg/ml; clarithromycin, 50 μg/ml); the number of parental drug-susceptible cells was calculated as the total number of bacterial cells minus the number of drug-resistant bacterial cells. The experiments were performed in triplicate with three independent cultures. Serial dilutions of each aliquot were plated three times, and a weighted mean according to the level of dilution was used for analysis.
The difference in fitness between two competing strains at time
t was computed by use of the following function:
where
rt and
st denote the absolute number of drug-resistant and drug-susceptible cells at a given time
t, respectively, and
rt−1 and
st−1 denote the number of drug-resistant and drug-susceptible cells at the preceding time point, respectively.
St is called the selection coefficient at time
t. The quotient of the ratios of the cell numbers was standardized with the exponent 1/8 because cell numbers were determined every eight generations.
The terms
rt/
rt−1 and
st/
st−1 give the growth rates for drug-resistant and drug-susceptible strains, respectively (
8). Hence,
S can be interpreted as the natural logarithm of the quotient of the growth rates of the competing strains.
S is equal to 0 if there is no difference in fitness between the competing strains,
S is negative if antibiotic resistance reduces bacterial fitness, and
S is positive if resistance increases bacterial fitness relative to that of the drug-susceptible competitor strain.
Relative bacterial fitness at time
t (fit
t) was calculated as
The cost per generation (cpg) was calculated as
The cost per generation can be interpreted as 1 minus the quotient of the growth rates. An analysis of variance was performed with
St as the dependent variable and the experiment as the random explanatory factor π
j (
35):
where α
0 is a nonrandom intercept,
Stj denotes the selection coefficient at time
t and experiment
j, and ε
tj is the normal distributed error term at time
t and experiment
j. The overall
St was estimated and tested against 0 on the basis of the null hypothesis (
H0) that α
0 is equal to 0. The data for different time points from one experiment were assumed to be independent. Given that the random factor “experiment” had no significant influence, subsequent analyses were performed without this factor and the data were assumed to be fully independent of each other.
To eliminate dilution errors and to determine the cost of resistance more precisely for no-cost and low-cost resistance mutations, additional experiments were performed. Aliquots from the competition assay were taken at time zero (t0; 0 h) and time 6 (t6; 144 h, corresponding to 48 generations) and plated onto nonselective agar; subsequently, >100 individual colonies were picked at random and the resistance phenotype was investigated for each colony individually to determine the ratio of susceptible and resistant cells (at least three independent experiments were performed for each mutant selected). This situation can be reflected by the following cross tabulation:
t0 t6r r0 r6(5)
s s0 s6where
rt and
st denote the absolute number of drug-resistant and drug-susceptible cells at time
t, respectively.
St is a monotonic transformation of the odds ratio (OR) of (
5)
Estimations of OR can be obtained by a logistic model (
17):
where
p is the probability that a drug-susceptible colony will be picked,
t is the time, γ
i is a random parameter to allow variability between the experiments, and β
0 and β
1 are the regression parameters. β
1 is an estimate for log OR, and a test for H
0 that β
1 is equal to 0 can be used to test if
St is 0. To test whether the bacterial fitness differed between bacteria with mutations introduced by site-directed mutagenesis and spontaneous drug-resistant mutants, an interaction term between time and the type of the mutation is added to the model (
7):
If the test for H
0 that β
2 (another regression parameter) is equal to 0 is not significant, the preceding model would be sufficient and no influence of the type of mutation on
St would be present. If the test is significant, an interaction term would be necessary for the model; i.e., the type of mutation would have a significant influence on
St. All statistical calculations were performed with SAS release 8.01.
Strains used for competition experiments.
The following strain combinations were used in the competition experiments: spontaneous streptomycin-resistant mutants of mc2155 rrnB (strains 1644, 1646, and 1674) versus drug-susceptible strain mc2155 rrnB (strain 1434); spontaneous streptomycin-resistant rrnB rpsL3+ mutants (strains 1592, 1630, 1632, 1634, 1636, and 1647) versus a drug-susceptible rrnB rpsL3+ strain (strain 1682); spontaneous amikacin-resistant rrnB mutants (strains 1181, 1183, 1184, 1185, 1187, and 1194) versus drug-susceptible rrnB strains (strains 1179, 1186, and 1193); and spontaneous clarithromycin-resistant rrnB mutants (strains 1082, 1086, and 1089) versus a drug-susceptible rrnB strain (strain 1179).
mc2155 rrnB rpsL3+ transformed with pMV361-H-rRNA2058 (strain 1691) was used as a drug-susceptible competitor strain (containing two wild-type 16S rRNA alleles, one chromosomal and one plasmid derived) for the strains into which the respective streptomycin resistance mutation was introduced by transformation with mutagenized plasmid pMV361-H-rRNA2058 and subsequent RecA-mediated gene conversion (two mutant 16S rRNA alleles, one chromosomal and one plasmid derived; strains 1683, 1684, 1687, 1688, 1689, 1699, and 1700).
An rrnB mutant transformed with pMV361-H-rRNA2058 (strain 1516) was used as a susceptible competitor for the strains into which the 1408A→G resistance mutation was introduced by transformation with mutagenized plasmid pMV361-H-rRNA2058 and subsequent RecA-mediated gene conversion (strains 1512, 1513, 1514-A, and 1515-B).
An rrnB mutant transformed with pMV361-H-rRNA (strains 2014, 2015) was the susceptible competitor for the strains into which the 2058A→G (strains 1998 and 1999) and 2059A→G (strains 2006 and 2007) resistance mutations were introduced by transformation with mutagenized plasmid pMV361-H-rRNA.
For a summary of the strains used in this investigation, see Table
1.
Frequency of resistance mutations in clinical isolates.
The literature was searched for published molecular biology-based analyses of mycobacterial drug resistance by use of the key words mycobacteria and drug resistance. Elimination of reviews and descriptions of clonal outbreaks and multiple isolates resulted in identification of 93 unique streptomycin-resistant strains of Mycobacterium tuberculosis for which the mechanism of resistance was characterized at the molecular level; 79 of these strains had one of the resistance mutations investigated here and were included in the analysis; 14 strains had a resistance mutation not investigated in this study (in RpsL at amino acid 88 and in rrn at positions 501, 912, and 913).
DISCUSSION
Competition experiments of the type used here are able to detect statistically significant fitness differences larger than 2% per generation (
4,
5,
36). Although smaller fitness differences might have some effects, our data suggest the following conclusions: (i) chromosomal resistance mutations found in laboratory (in vitro)-derived spontaneous mutants may be cost neutral; (ii) the failure to detect the costs of resistance in these spontaneous mutants, and probably also in mutants detected in nature (clinical isolates), is not due to compensatory mutations; and (iii) under natural conditions (clinical isolates), strong selection pressure seems to exist for those drug resistance mutations which impose little if any burden on fitness, i.e., mutations which do not confer a substantial cost in the absence of antibiotics.
As a starting point, we searched the literature for different resistance mutations occurring in a variety of pathogens; the experimental investigations, however, were conducted with
M. smegmatis, which was used as a model system. We suggest that the findings observed in our model system allow us to draw more general conclusions, as comparable costs of resistance were determined for mutations investigated previously in other microorganisms, i.e.,
rpsL mutations 42 Lys→Arg, Lys→Asn, Lys→Thr, which exhibited fitness costs of 0, 14, and 15%, respectively, in
E. coli,
S. enterica serovar Typhimurium, and
M. smegmatis (
5,
36) (Table
3).
Different constraints have been postulated to affect the translation machinery under various in vitro and in vivo conditions (
2,
25). The selection of mutant genotypes observed in vivo and determined in the in vitro competition model to be no-cost mutations suggests that the measurements obtained in vitro adequately reflect the measurements obtained in the in vivo situation. While determination of a mutation as no cost (within the limitations that accompany the use of a model organism) conflicts with the assumption that resistance mutations are costly, one cannot exclude a hypothetical condition in which some sort of difference in fitness would become manifest. However, the selection of no-cost resistance mutations observed in clinical isolates exposed to complex and fluctuating conditions as well as heterogeneous habitats makes this possibility unlikely.
In contrast to competitive environments such as clinical in vivo conditions, the use of a compartmentalized experimental environment, i.e., solid medium, for selection of drug-resistant mutants provides data on the frequency and type of resistant variants largely irrespective of a fitness cost (variants with all possible alleles capable of surviving the selection procedure will grow). The observation that clinically acquired resistance rarely involves mutations with a cost can be ascribed to two different mechanisms: (i) a priori, mutations that confer a substantial cost (e.g., mutations for resistance to aminoglycosides of the 2-deoxystreptamine type) do not arise even under in vitro conditions in the laboratory, as the functional constraints of the target molecule seem to allow only mutations that are both cost neutral and able to produce a resistance phenotype; and (ii) mutations that confer a cost occur under in vitro conditions in the laboratory (e.g., mutations for resistance to streptomycin), but there is selection in vivo for low-cost resistance mutations. This selection might be explained by fluctuating environments, i.e., expansion of mutants experiencing a low cost of resistance in the absence of antibiotics during periods in which selection for antibiotic resistance is removed.
While our findings do not deny the existence of compensatory mutations (
4,
5,
36), they indicate that under natural in vivo conditions these may be of minor relevance to the epidemiology of drug resistance. The likelihood that a costly resistance mutation is ameliorated by an additional compensatory mutation is far greater than the likelihood that a no-cost resistance mutation (occurring roughly at the same frequency as a costly resistance mutation) is ameliorated, making the scenario of costly but compensated resistance mutations unlikely in nature. In principle, resistance mutations that are acquired in vivo and that carry a cost might therefore be found only when a cost-neutral resistance mutation does not exist for a given drug (
5,
27). The eventual observation that a costly resistance mutation will emerge in a single patient (e.g., streptomycin resistance mutation RpsL 42 Lys→Thr, which was found in 1 of 79 streptomycin-resistant
M. tuberculosis isolates) probably reflects the stochastic probability of a resistance mutation in a bacterial population of limited size. It is under these circumstances that compensatory mutations that ameliorate the cost of resistance are likely to develop (
27,
37).
Two outcomes of resistance mutations are feasible: fixation or elimination (Table
5). For a costly resistance mutation, e.g., the streptomycin resistance mutation 523A→C, which carries a cost of resistance of 5% per generation, 359 generations would be required for the mutation to become fixed, i.e., for isolates of the compensated resistance genotype to become the dominant (>99%) drug-resistant population, assuming that the frequency of a compensatory mutation is 10
−6 (
5). These calculations are in line with the finding that the compensated resistance genotype is less frequent in vivo than in vitro (
27). On the other hand, it will be difficult to eliminate resistance mutations with a low fitness cost even in the absence of compensatory mutations. By assuming a population of 99% resistant cells and a cost per generation of 2%, more than 1,000 generations are required to lower the existing frequency of resistance to the frequency of resistance resulting from spontaneous mutations (frequency of spontaneous drug-resistant mutants, 10
−8). For
M. tuberculosis, which has an estimated generation time of 24 h, this would correspond to more than 3 years of total absence of the corresponding drug.
Our investigations focused on the fitness cost of chromosomal drug resistance. The epidemiology of drug-resistant pathogens is more complex and involves transmission and clearance of microbes from infected hosts. Although we hesitate to generalize our findings, the observation of no-cost or low-cost mutation types in clinical situations for resistance to different classes of antibiotics that act on the ribosome suggests that these observations may reflect a general biological phenomenon. Our results indicate the limitations of strategies for the containment of resistance based solely on restrictive drug use because, due to a missing fitness burden, negative selection pressure for such mutations does not seem to exist in an antibiotic-free environment. These results are consistent with the hypothesis that even if the rate of antibiotic consumption is reduced, in places where resistance is already common, the frequency of resistance will not decline rapidly, if at all (
1,
20).
Acknowledgments
We thank J. Buer and J. Colston for critical review of the manuscript; W. R. Jacobs, Jr., for providing M. smegmatis mc2155; C. K. Stover for plasmid pMV361; K. Ellrott for expert technical assistance; A. Goebel for initial help with the statistical analysis; and S. Maibom, K. Härri, and F. Mitterecker for typing the manuscript.
This study was supported in part by grants from the Deutsche Forschungsgemeinschaft (Schwerpunkt Ökologie bakterieller Krankheitserreger), the Swiss National Research Foundation (grant SRP 49), and the Niedersächsischer Verein zur Bekämpfung der Tuberkulose e.V.