The Competitive Cost of Antibiotic Resistance in Mycobacterium tuberculosis

See allHide authors and affiliations

Science  30 Jun 2006:
Vol. 312, Issue 5782, pp. 1944-1946
DOI: 10.1126/science.1124410


Mathematical models predict that the future of the multidrug-resistant tuberculosis epidemic will depend on the fitness cost of drug resistance. We show that in laboratory-derived mutants of Mycobacterium tuberculosis, rifampin resistance is universally associated with a competitive fitness cost and that this cost is determined by the specific resistance mutation and strain genetic background. In contrast, we demonstrate that prolonged patient treatment can result in multidrug-resistant strains with no fitness defect and that strains with low- or no-cost resistance mutations are also the most frequent among clinical isolates.

Antimicrobial resistance has become a worldwide problem in many pathogens, incurring both economic costs and loss of human lives (1, 2). Antibiotic resistance is also often associated with a reduced competitive ability against antibiotic-sensitive strains, in the absence of the antibiotic (3). In a variety of model systems, it has been shown that this fitness “cost” depends on the specific drug resistance–conferring mutation and the strain genetic background and that it can be ameliorated by compensatory mutations (49). However, there are few studies linking the predictions of these model systems to epidemiological data from human populations (1012). Here, we connect the in vitro cost of resistance to the prevalence of clinically important mutants of Mycobacterium tuberculosis, the causative agent of human tuberculosis.

The emergence of multidrug-resistant strains of M. tuberculosis is threatening global disease control efforts (13). Multidrug-resistant tuberculosis (MDRTB) is defined as tuberculosis caused by organisms resistant to isoniazid and rifampin, the two most potent first-line anti-tuberculosis drugs. MDRTB represents a major public health burden, requiring prolonged treatment with more costly and less effective agents (13, 14). To date, most investigations of factors determining the spread of MDRTB have focused on the quality of tuberculosis-control programs and patient adherence to treatment. However, recent studies suggest that bacterial factors, such as the relative fitness of drug-resistant M. tuberculosis strains, are also important (1518). M. tuberculosis develops drug resistance by the sequential acquisition of chromosomal mutations (19), but there are only limited data on the effect of different drug resistance–conferring mutations on the relative fitness of M. tuberculosis (12, 20). Furthermore, there has been no attempt to investigate the effects of different genetic backgrounds on the competitive fitness of drug-resistant pathogens, including M. tuberculosis.

The relative competitive fitness of bacteria can be quantified experimentally with the use of competition assays originally developed for Escherichia coli (21). In these experiments, the drug-susceptible and the drug-resistant organisms compete for limited resources in a common environment. In M. tuberculosis, as in other bacteria, resistance to rifampin is mediated through missense mutations in rpoB, the gene encoding the β subunit of the RNA polymerase (19). To determine the impact of different rpoB mutations on the relative competitive fitness of rifampin-resistant M. tuberculosis, we selected a panel of spontaneous rifampin-resistant mutants by growing M. tuberculosis CDC1551 on media containing rifampin. CDC1551 is a clinical strain and part of a lineage of M. tuberculosis associated with Europe and the Americas (2224), and it is susceptible to currently used antituberculosis drugs. We sequenced known rifampin-resistance regions in the rpoB gene of all of the 52 colonies recovered and found that all of them had a missense substitution in rpoB (table S1). We chose nine mutants with different rpoB mutations and had them compete in vitro against their rifampin-susceptible ancestor, with the use of an assay that included lag, exponential, and early stationary phases of growth (21). All of the rifampin-resistant mutants had a statistically significant fitness cost compared with the susceptible ancestor, which is defined to have a competitive fitness of 1.0 (Fig. 1 and table S2). Among the different rpoB mutants, there were significant differences in competitive fitness, with values ranging from 0.58 to 0.91. We found a significant effect of mutation on fitness [analysis of variance (ANOVA), F8,38 = 12.1563, P < 0.0001]. The Ser531→Leu531 (S531L) mutant exhibited the lowest cost, whereas the Arg529→Gln529 (R529Q) mutant had the greatest fitness cost of all mutants (Fig. 1 and table S2). These results confirm that a wide range of rpoB mutations can have markedly different effects on the competitive fitness of M. tuberculosis in vitro, consistent with earlier work that examined the competitive cost of a small number of rpoB mutations (12, 20).

Fig. 1.

Relative competitive fitness of laboratory-derived rifampin-resistant mutants of M. tuberculosis. All mutants had a statistically significant fitness cost (error bars indicate 95% confidence intervals). This cost was less in rpoB S531L mutants than in other rpoB mutants, irrespective of the strain background. Light gray bars, CDC1551 mutants; dark gray bars, T85 mutants. Y, Tyr; W, Trp; P, Pro.

Next, we investigated the influence of the bacterial genetic background on the cost of rifampin resistance–conferring mutations. In the laboratory, we selected a second panel of spontaneous rifampin-resistant mutants in T85, another pan-susceptible isolate that is part of a distinct lineage of M. tuberculosis strongly associated with East Asia, commonly referred to as the Beijing strain (24, 25). All 63 rifampin-resistant colonies selected in T85 had a nonsynonymous substitution in rpoB (table S3). We measured the relative competitive fitness of four different mutants with rpoB mutations whose fitness costs had been measured in CDC1551. Similar to the results in the CDC1551 background, all T85 rifampin-resistant mutants had a significant fitness cost (Fig. 1 and table S2). We tested for an effect of genetic background by comparing the fitness costs of the four mutations found in each of the two backgrounds. We found an overall significant effect of genetic background when nested in mutation (nested ANOVA, F7,34 = 6.8962, P < 0.0001), and, as observed in the CDC1551 background, the rpoB S531L mutant had higher competitive ability than other mutations in the T85 background (Fig. 1 and table S2). The fitness of the S531L mutation did not significantly differ in the two genetic backgrounds (i.e., the confidence intervals overlap; Fig. 1 and table S2). However, genetic background did significantly alter the fitness of the His526→Asp526 (H526D) mutation (i.e., the confidence intervals do not overlap; Fig. 1 and table S2). Taken together, our data confirm that the in vitro relative fitness of rifampin-resistant M. tuberculosis depends on both the resistance-conferring mutation and the strain genetic background.

The relevance of empirically measured costs of laboratory-derived drug resistance to resistance that emerges in tuberculosis patients has not been established. Therefore, we measured the relative competitive fitness of rifampin-resistance in paired isolates (one that was drug resistant and one that was drug susceptible) from 10 patients who acquired rifampin resistance during antibiotic treatment (Fig. 2 and table S4). We used a standard DNA fingerprinting method to confirm that the paired isolates belonged to the same bacterial clone (26). In all 10 patients, both isolates had identical DNA fingerprinting patterns (table S5), which supports the assertion that the rifampin-resistant isolates recovered later were indeed derived from the corresponding rifampin-sensitive isolates. However, DNA fingerprinting methods are limited and may not reveal small genetic differences which could have been present between isolates. All of the clones were distinct but belonged to one of the two strain lineages represented by CDC1551 and T85 (tables S4 and S5). Five of the resistant isolates harbored the rpoB S531L mutation and five had other rpoB mutations (Fig. 2, table S4).

Fig. 2.

Relative competitive fitness of clinically derived rifampin-resistant mutants of M. tuberculosis. Four of the five mutants with the rpoB S531L mutation (light gray bars) had no fitness cost compared with their rifampin-susceptible ancestors. All mutants with other rpoB mutations (dark gray bars) had significant fitness defects (error bars indicate 95% confidence intervals).

All of the clinical isolates with mutations other than S531L had a relative fitness of less than 1.0, although the measured relative fitness was not always the same as that observed in the mutants selected in vitro (Fig. 2 and table S4). In contrast, four of the five clinical strains with the rpoB S531L mutation had a relative fitness of greater than or equal to 1.0. Notably, these four strains were the only rifampin-resistant mutants that had no fitness cost compared with their rifampin-susceptible ancestor.

Two alternative explanations could account for the high fitness we observed in clinical strains harboring the rpoB S531L mutation. First, this mutation may indeed be a no-cost (versus just a low-cost) mutation in these particular strains. Second, given that sufficient time had passed between the isolation of the susceptible and resistant members of each pair (table S4), compensatory mutations could have arisen and ameliorated the low initial fitness cost associated with this mutation, similar to what has been observed in other species of bacteria (4, 5, 8). In support of this possibility, we found that the clinically derived rpoB S531L mutants had a mean competitive fitness of 1.04 (95% confidence interval: 1.00 to 1.08). This was statistically significantly higher than the laboratory-derived rpoB S531L mutants, which had a mean competitive fitness of 0.93 (95% confidence interval: 0.90 to 0.96). Given that all clinical ancestors of the rpoB S531L mutants were already resistant to isoniazid or isoniazid and streptomycin (table S4), this increase in fitness could be due to amelioration of fitness costs associated with the other resistance mutations rather than specifically the rifampin-resistance mutation. In addition, general increases in fitness could have occurred during long-term infection, rather than specific amelioration of the cost associated with a resistance mutation. We cannot differentiate between these alternative hypotheses with our current data. Nevertheless, our results clearly demonstrate that clinical strains of multidrug-resistant M. tuberculosis can have a significantly higher fitness than their progenitors, either because of no-cost resistance mutations, compensatory evolution, or both.

We hypothesized that those drug-resistant strains with the least costly and potentially most easily compensated resistance mutation will be more likely to spread and become prevalent in human populations. This scenario is particularly likely given the recent observation that many patients harbor subpopulations of the same mycobacterial clone with different drug resistance–conferring mutations (27). Many studies have reported the frequency of different drug-resistance alleles in a variety of clinical settings (28), providing an opportunity to test our hypothesis. When we compared the relative fitness of the different rpoB mutants to their clinical frequency, an association became evident (Figs. 1 and 2 and tables S2 and S4). The rpoB S531L mutation, which exhibited the lowest fitness cost in laboratory-derived mutants and no fitness cost in clinical strains, is the most prevalent rifampin resistance–conferring mutation, overall accounting for 54% of rifampin-resistant isolates, even in parts of the world associated with phylogenetically distinct strain lineages (24, 28). In contrast, the rpoB R529Q mutant, which carried the highest fitness cost of all mutants, has never been observed in clinical settings (28). Taken together, these findings are consistent with the idea that drug-resistant strains harboring low- or no-cost mutations such as rpoB S531L are selected in patients during treatment and that such strains are more likely to spread in human populations.

Several mathematical models have been developed to predict the future of MDRTB epidemics. Some have assumed a universal fitness cost to drug resistance and concluded that MDRTB will remain a localized problem (15). More recent mathematical models have allowed for variable fitness among strains and have come to very different conclusions depending on the specific fitness assumed (17). Our work supports the more complex models and suggests that the heterogeneity in fitness is a function of the drug-resistance mutation, the strain genetic background, and potentially compensatory evolution as well. These more complex models suggest that resistance will not spread below a relative fitness of 0.7. Interestingly, all the rpoB mutants tested had a relative fitness >0.7, except for the one that has not been observed in clinical settings (Figs. 1 and 2). None of these models has allowed for changing fitness through time, either as a result of compensatory evolution or other increases in fitness due to within-host evolution. Future mathematical treatments of the MDRTB epidemic should include such changes.

Our work suggests that in vitro competition assays can be predictive of biological differences important to M. tuberculosis ecology and evolution. Our finding that genetic background is a factor in the measured cost of resistance suggests that the cost of a given mutation may vary from location to location worldwide owing to the phylogeography of this pathogen (24). The possibility that fitness changes can occur during infection of a single patient should be taken into account when applying evolutionary ecology to infectious disease treatments or to the prediction of epidemics. The availability of such predictive methodology permits the incorporation of fitness considerations into the design and deployment of new drug regimens for controlling tuberculosis.

Supporting Online Material

Materials and Methods

Tables S1 to S5


References and Notes

View Abstract

Navigate This Article