To obtain a general framework for understanding selection of antibiotic-resistant mutants, allelic diversity was examined with about 600 fluoroquinolone-resistant mutants of mycobacteria. Selection at low fluoroquinolone concentration produced many low-level resistance mutants. Some of these contained mutations that conferred unselected antibiotic resistance; none contained alterations in the quinolone-resistance-determining region of the GyrA protein, the principal drug target. As selection pressure increased, a variety of GyrA variants became prevalent. High concentrations of antibiotic reduced the variety to a few types, and eventually a concentration was reached at which no mutant was recovered. That concentration defined a threshold for preventing the selection of resistance. The pattern of variants selected, which was also strongly influenced by antibiotic structure, readily explained the variants present in clinical isolates. Thus, resistance arises from selection of mutants whose identity depends on drug concentration and structure, both of which can be manipulated to restrict selection.
Antibiotic resistance has become a major problem in the management of infectious disease. For example, the resurgence of tuberculosis in New York City during the early 1990s was accompanied by a striking incidence of resistance in the causative organism, Mycobacterium tuberculosis: about one-third of the isolates exhibited resistance [1–3], some to as many as 8 distinct agents . A similar phenomenon, but with many more resistant cases, is now occurring in the Russian prison system [5, 6], and comparable scenarios are emerging with other pathogens . To halt selection of resistance, new dosing strategies are being developed , and structures of antibacterial agents, such as the fluoroquinolones, are being refined to restrict mutant selection [9–11]. Because resistance is not an absolute term, many levels can be attained through allelic diversity (for examples, see ). This diversity can be substantial, as illustrated by quinolone resistance in M. tuberculosis: clinical resistance has been associated with 6 distinct mutations [13–15] that confer at least a 4-fold range of susceptibility . Such diversity could be important in choosing antibiotic derivatives and concentrations best suited for preventing the selection of resistance; however, the factors determining the diversity of antibioticresistance alleles are too poorly defined to derive a general description of mutant selection .
Allelic diversity can be studied by selecting resistant mutants under a variety of conditions, isolating DNA from those mutants, and then determining nucleotide sequence changes associated with resistance. A central question is whether protein variants conferring the most resistance predominate or whether a broad, concentration-dependent spectrum arises. Also of interest is whether derivatives within the same class of antibiotic select different variants, and, if so, why. To address these issues, we have examined resistance to fluoroquinolones, because these compounds exhibit most of the activities generally seen with antibiotics: they are both bacteriostatic and bactericidal , they display target-based and nontarget-based resistance , and they are mildly mutagenic [19, 20]. Moreover, many derivatives are readily available for examining relationships between structure and resistance-allele selection. As test organisms, mycobacteria are good choices because they have only 1 fluoroquinolone target, DNA gyrase , which simplifies data interpretation. Furthermore, quinolone-resistant mutants are selected over a broad concentration range , and selection is sensitive to antibiotic structure: addition of a C-8-methoxy group to N-1-cyclopropyl fluoroquinolones dramatically shortens the concentration range over which mutants are recovered .
In the present study, we obtained fluoroquinolone-resistant mutants of M. smegmatis and M. tuberculosis by plating cells on agar containing one of several structurally related fluoroquinolones at a variety of concentrations. We then determined the nucleotide sequences of the quinolone-resistance—determining regions (QRDR) in the gyrase genes (gyrA and gyrB), the genes responsible for topoisomerase-mediated fluoroquinolone resistance in mycobacteria. A distinct pattern was seen with respect to antibiotic concentration. None of the resistant mutants recovered after challenge with low concentrations had alterations in the primary drug target, the GyrA protein, whereas at moderate concentrations many distinct changes were observed in this protein. The highest drug concentrations selected gyrase variants that conferred the most resistance. Drug structure also influenced allele selection. In a striking example, a shift in the position of a fluoroquinolone ethyl group changed the identity of the most resistant variant selected and suggested how fluoroquinolones might be oriented in gyrase-DNA complexes. These effects of fluoroquinolone concentration and structure provide a general description of allelic diversity arising from selection of antibiotic-resistant mutants.
Materials and Methods
Bacterial strains and growth conditions
Derivatives of M. smegmatis strain mc2155 (KD1163) and M. tuberculosis TN1626 were grown as liquid cultures in Middlebrook 7H9 medium containing 10% albumin dextrose complex (ADC) and 0.05% Tween 80 and as colonies on Middlebrook 7H10 agar plates . Agar plates containing fluoroquinolone were prepared by adding concentrated solutions to molten agar. Incubation was carried out at 37°C. Work with M. tuberculosis TN1 626, which is a fluoroquinolone-sensitive, multidrug-resistant (isoniazid, streptomycin, rifampicin, ethambutol, and kanamycin) clinical isolate , was performed in a biosafety level 3 containment facility. In this isolate, amino acid 95 is threonine , which makes its GyrA QRDR identical to that of M. smegmatis (Genbank accession numbers X94224 and L27512). The M. smegmatis isolate encoded a threonine at position 507 of the GyrB protein rather than the arginine reported elsewhere .
Ciprofloxacin was obtained from Miles Laboratories (Kankakee, IL), and fluoroquinolones PD160788 (C-8-H), PD160793 (C-8-H), PD161144 (C-8-methoxy) and PD161148 (C-8-methoxy) were provided by Parke-Davis (Ann Arbor, MI; for structures, see figure 1). Fluoroquinolones (10 mg) were first dissolved in 0.1 mL 1 N NaOH (0.1 of final volume), and then sterile water was added to give a final concentration of 10 mg/mL. This stock solution was divided into 50-µL aliquots and stored at −80°C. Dilution series were prepared with autoclaved water. Solutions were sometimes stored at −20°C for several weeks.
Selection of fluoroquinolone-resistant mutants
M. smegmatis was grown to stationary phase (Klett = 400), and the cells (6 × 108 to 2 × 109 cfu/mL) were poured directly on predried, fluoroquinolone-containing 7H10 agar. Incubation was at 37°C for 7–10 days. For M. tuberculosis, cells were grown to stationary phase in 7H9 medium (OD600 = 1.3), concentrated by centrifugation (3000 g for 10 min), resuspended in fresh 7H9 medium, and applied to agar plates in 1-mL aliquots at an estimated concentration of 5 × 108 to 5 × 109 cfu/mL. Colonies were counted after incubation at 37°C for 4 weeks. With both organisms, mutants were retested for growth on fluoroquinolone-containing agar.
MIC was measured to characterize mutant response to the fluoroquinolones. 7H10 agar plates containing various concentrations of fluoroquinolone were prepared as above. Stationary-phase cells were diluted by 10-fold increments, using 7H9 liquid medium as diluent, and 10-µL portions were spotted on plates. Colonies on each plate were counted after incubation at 37°C for 5–7 days (M. smegmatis) or 3–4 weeks (M. tuberculosis). MIC was defined as the fluoroquinolone concentration required to inhibit colony formation by 99%.
DNA isolation and nucleotide sequence determination
Cultures of stationary phase M. smegmatis (1.5 mL) were harvested by centrifugation and treated with lysozyme, as described elsewhere for M. tuberculosis . The cell suspension (0.5 mL) was then lysed by adding 70 µL 10% SDS and 4 µL 20 mg/mL proteinase K, followed by incubation at 55°C for 20 min with agitation. NaCl and hexadecyltrimethyl ammonium bromide (CTAB) were then added to 0.7 M and 10%, respectively. After incubation for an additional 10 min at 55°C and extraction with chloroform/isoamyl alcohol (24: 1), RNase A was added to 100 µg/mL, followed by incubation for 30 min at 37°C. The DNA-containing solution was extracted with phenol/chloroform, precipitated with isopropanol, washed with 70% ethanol, and dissolved in 200 µL TE (0.01 M Tris-HCl pH 8, 0.001 M EDTA). M. tuberculosis was treated in a similar way , although after centrifugation of 0.5 mL cell culture and resuspension in the same volume of TE, the cells were killed by treatment at 80°C for 30 min. Cells were next incubated at 60°C with shaking for 1 h after adding 100 µL proteinase K (10 mg/mL) and 80 µL 10% SDS. NaCl and CTAB were then added to 0.7 M and 10%, respectively. The resulting mixture was incubated at 60°C for 15 min, frozen, and then incubated again for 15 min at 60°C. DNA was extracted with chloroform/isoamyl alcohol, precipitated with isopropanol, washed with 80% ethanol, and dissolved in 50 µL TE.
After chromosomal DNA was extracted from M. smegmatis and M. tuberculosis, PCR was performed to amplify the QRDR of gyrA and gyrB. The nucleotide sequence of that region was determined by automated DNA sequencing using an ABI model 377 instrument. Oligonucleotides used for amplification were 5′-ACGA-ACTGTTCTCCA TCCTCATGGGCGAAG′ (forward primer), 5′-ACATCTCCATCGCCAACGGAGTG-3′ (reverse primer), and 5′-CAGCGCAGCTACATCGACTACGC-3′ (sequencing primer), for gyrA of M. smegmatis; 5′-CAGCTACATCGACTATGCGA-3′ (forward and sequencing primer) and 5′-GGGCTTCGGTGT-ACCTCAT-3′ (reverse primer), for gyrA of M. tuberculosis; 5′-CCGACTGCCGTTCGACGGAT-3′ (forward and sequencing primer) and 5′-CGGCCATCAACACGATCTTG-3′ (reverse primer), for gyrB of M. smegmatis; and 5′-CGCAAGTCCGA-ACTGTATGTCGTAG-3′ (forward and sequencing primer) and 5′-GTTGTGCCAAAAACACATGC-3′ (reverse primer), for gyrB of M. tuberculosis.
Effect of fluoroquinolone concentration and structure on recovery of resistant colonies
To provide a collection of mutants for analysis, M. smegmatis was applied to agar plates containing various concentrations of fluoroquinolone (for fluoroquinolone structures, see figure 1). Fluoroquinolone concentrations slightly above the MIC of wild-type M. smegmatis caused the fraction of colonies recovered to drop sharply until a plateau was reached; a second sharp drop occurred at higher concentrations (figure 2A). Even by this relatively coarse assay, differences among the compounds could be seen. For example, the commonly used compound ciprofloxacin allowed the initial drop to be much more gradual than did a similar compound that contained an ethyl group attached to its C-7 ring (figure 2A; compare ciprofloxacin [triangles] with PD160788 [circles]). Thus, many more mutants were selected at low concentrations with ciprofloxacin than with PD160788. A C-8-methoxy group added to the overall potency and shifted the curve to lower concentrations; it also shortened the plateau (figure 2A; compare PD161144 [squares] with PD160788 [circles]). Thus, 2 ways were revealed in which fluoroquinolone structure influences selection of resistant mutants.
A clinical isolate of M. tuberculosis, strain TN1 626, behaved in the same general way as did M. smegmatis, but the initial drop seen with ciprofloxacin was much sharper. Thus, low-level resistance differs between the 2 organisms. The C-8-methoxy fluoroquinolone (PD161148) used to obtain the data in figure 2B was chosen for studies with M. tuberculosis, because this compound had been tested previously with this organism [10, 26]. To compare the 2 bacteria directly, PD161148 was also tested against M. smegmatis. The latter was 2–3 times more susceptible than M. tuberculosis when the concentration at the second drop in mutant recovery was compared. It was surprising that the concentration at which the second sharp drop occurred with M. smegmatis was quite different for the 2 methoxy compounds (figure 2C; compare PD161148 [diamonds] and PD161144 [squares]). The 2 compounds were identical except that the more active one, PD161148, had its C-7-ring ethyl group attached to a carbon, whereas the ethyl in PD161144 was attached to the ring nitrogen (figure 1). This result strongly reinforced the idea that antibiotic structure can affect mutant selection.
Effect of fluoroquinolone concentration and structure on the spectrum of gyrA alleles recovered
To relate fluoroquinolone structure and concentration to particular alleles, we performed nucleotide sequence analysis on DNA extracted from colonies obtained on agar containing various concentrations of fluoroquinolone (figure 2). In many cases, all colonies on a plate were selected and examined; when hundreds of colonies were present, they were assigned numbers and chosen randomly for sequence analysis. The results are shown in figures 3 and 4, where each fluoroquinolone concentration is represented by a panel; the relative position of each sample in the mutant-selection curve is shown in the right portion of the figure. Casual inspection shows that, as drug concentration increased, the pattern of selected variants changed in a way that reflected fluoroquinolone structure.
With M. smegmatis, low concentrations of fluoroquinolone selected only non-GyrA mutants; at slightly higher concentrations, 10 distinct GyrA variants were obtained (figure 3). The large numbers of colonies selected at low concentrations by ciprofloxacin were largely nongyrase mutants. At higher concentrations, D95G GyrA mutants dominated. For the other compounds, each of which had a substituent on its C-7 ring, nongyrase mutants were also selected at low concentrations, but many distinct GyrA variants were obtained from agar plates containing intermediate concentrations.
At high concentrations the 2 C-8-H compounds (ciprofloxacin and PD160788) tended to select D95G variants. There was, however, a hint that at very high concentrations the G89C variant might be selected when the C-7 ring contained an Nethyl group (PD160788): the MIC for PD160788 was highest with a G89C mutant (table 1). With this compound, colony recovery dropped very sharply after the point labeled D in the right-hand panel of figure 3, and only 1 colony, a G89C mutant, was recovered beyond that point. Selection of the G89C mutant became obvious when a C-8-methoxy group and the C-7 ring ethyl were both present, as seen with PD161144 (figure 3D, PD161144). With this compound, the G89C mutant had by far the highest MIC (table 1). When the C-7-ring ethyl was attached to a carbon (PD161148), 3 variants had similar MICs (table 1), and all 3 were selected at high concentrations (figure 3). These data are consistent with many distinct resistance variants being present in the bacterial population and with subsets being selected on the basis of MIC. Susceptibility, in turn, depended on antibiotic structure.
With M. tuberculosis, low concentrations of fluoroquinolone also selected non-GyrA mutants (figure 4 and table 2), and these mutants became less prevalent as fluoroquinolone concentration increased. Unlike the situation in M. smegmatis, ciprofloxacin at moderate concentrations selected a wide variety of GyrA variants. Because a similar variety of resistant variants was seen when many clinical isolates were examined [4, 13], it is likely that ciprofloxacin concentrations in patients have been only moderate. The 2 compounds with an ethyl group on the C-7 ring both selected a narrower range of mutants than was observed with ciprofloxacin. At the highest fluoroquinolone concentration tested, the C-8-methoxy compound, PD161148, more often selected the D94N mutant than the D95G mutant observed with M. smegmatis. The basis for the differences between M. smegmatis and the multidrug-resistant M. tuberculosis isolate has not been determined.
Characteristics of non-GyrA mutants
With M. smegmatis, low concentrations of quinolone failed to select variants of the gyrA QRDR. Hence, we suspected that low-level resistance mutants might have mutations that confer resistance to other antibiotics . When we determined the MIC for ampicillin and chloramphenicol for several GyrA and non-GyrA mutants, we found that 3 of 4 non-GyrA mutants were resistant to ampicillin and chloramphenicol (table 1); however, none of the 3 GyrA mutants tested was resistant to either compound. Twenty-eight additional non-GyrA mutants were then plated on agar containing 10 µg/mL chloramphenicol or 50 µg/mL ampicillin. Fifteen strains were resistant to both antibiotics; the remaining 13 mutants were resistant to neither. Thus, almost 60% of the non-GyrA mutants had unidentified mutations that confer multidrug resistance.
Because gyrB alleles can also confer quinolone resistance [14, 27], we examined the QRDR of gyrB for representative, non-GyrA mutants recovered from agar plates containing low concentrations of fluoroquinolone. Of 89 M. smegmatis mutants obtained at low concentrations of the 4 compounds, none had a quinolone-resistant gyrB allele (not shown). We also failed to find gyrB-mediated resistance in 3 non-GyrA mutants selected by high levels of the C-8-methoxy compound (PD161144). In contrast, non-GyrA mutants of M. tuberculosis tended to map in gyrB, as shown using comparable conditions (similar frequency of mutant recovery; table 2). Indeed, we had to examine mutants obtained at very low concentrations of ciprofloxacin to observe nongyrase mutants (figure 4, ciprofloxacin, open portion of non-GyrA bar in panels A and B). Overall, >80% (51/62) of the non-GyrA mutants of M. tuberculosis had gyrB mutations (figure 4).
Bacterial populations often become resistant to antibiotics gradually. One pattern is illustrated by the clinical development of penicillin resistance in Streptococcuspneumoniae, which arose over a number of years through increases in the prevalence of strains with decreasing susceptibility (reviewed in ). In this case, selection occurred among many patients during relatively brief encounters with the pathogen. Another pattern occurs with tuberculosis. Patients are generally treated with several antibacterial agents for long periods of time, and multidrug resistance is often associated with treatment failure . The factors involved in selection of antibiotic resistance can be examined by mixing populations of cells having various levels of susceptibility and then challenging them with antibiotic under a variety of conditions. Such experiments, with β-lactams used against S. pneumoniae, revealed complex relationships among population structure, antibiotic structure, and antibiotic concentration . We used a simpler approach in which susceptible, wild-type mycobacteria were plated on agar containing fluoroquinolone, and the resulting mutants were examined by nucleotide sequence analysis. Plating was expected to select cells containing single mutations; and in 1100 isolates of M. tuberculosis, cells contained either a gyrA or a gyrB mutation but not both. Our results revealed that different alleles are prevalent at different challenge concentrations, that antibiotic structure affects the pattern of variants observed, and that an antibiotic concentration can be achieved at which no mutant is recovered even when 1010 cells were applied to plates. The latter concentration represents a threshold that, if exceeded, would severely restrict the selection of resistance.
The array of mutations recovered was surprisingly diverse. With M. smegmatis, we recovered 10 GyrA variants; with M. tuberculosis, we obtained 12 GyrA variants and 10 GyrB variants. Both species also yielded unidentified nongyrase mutations. The relative abundance of particular variants depended strongly on fluoroquinolone concentration, according to a simple pattern: challenge of M. smegmatis at low drug concentrations produced many nontarget (GyrA) mutants, more than half of which were multidrug resistant. As the concentrations increased, a large variety of target (GyrA) mutants was recovered. Further increases in concentration gradually reduced the variety until only the most resistant mutants were recovered. High-level mutants were not detected at low drug concentrations with sample sizes of 10–30; thus, under these conditions, low-level resistance mutants were much more abundant than high-level ones. Although selection of resistant pathogens is likely to be more complex in patients than in the laboratory , our data collectively suggest that the recovery of many types of M. tuberculosis mutants from patients [4, 13] reflects treatment with low-to-moderate concentrations of fluoroquinolone.
Why the nongyrase mutants were so abundant with M. smegmatis has not been established. These mutants grew at the same rate or more slowly than the GyrA mutants (data not shown); thus, their overrepresentation was not easily explained by growth rate differences. Many more ways may exist to acquire nongyrase- than gyrase-based resistance. For example, large numbers of nongyrase mutants might arise from the mild mutagenic action of the SOS response induced by quinolone treatment [19, 31, 32]. Gyrase alleles are less likely to arise in this way, because cells acquiring these recessive mutations would be killed before resistance was expressed. Thus, gyrase mutations were probably present prior to fluoroquinolone challenge. The influence of SOS-dependent mutations is being tested with Escherichia coli, using lexA mutations that block the SOS response.
Mutants exhibiting low-level resistance will be clinically important when fluctuations in antibiotic concentration occur, because, as concentrations drop toward the MIC, the number of resistant mutants selected will increase dramatically. Such mutants could constitute a population from which higher levels of resistance could be selected when antibiotic concentrations subsequently increase. These considerations, together with our observation that nongyrase mutants of M. smegmatis are often multidrug resistant, make low-dose fluoroquinolone therapy inadvisable.
The strong influence of fluoroquinolone concentration on the identity of resistance alleles selected also applies to other bacteria. For example, nontopoisomerase mutants of Streptococcus pneumoniae are obtained by challenge with low fluoroquinolone concentrations . Subsequent challenge of these first-step mutants with higher concentrations selects parC (topoisomerase IV) mutations. With Staphylococcus aureus, a case has been found in which fluoroquinolone concentration was high enough to select parC alleles first. Then a small incremental increase selected second-step mutants having nontopoisomerase mutations . Thus, the order in which topoisomerase and nontopoisomerase mutations are selected may not be important when only 1 drug target is considered. However, when 2 drug targets (e.g., topoisomerase IV and gyrase) are considered, target mutations follow a strict order: mutations in the more susceptible target are selected first, because resistance alleles in the less susceptible target confer no resistance by themselves. That is why parC alleles are usually selected before gyrA alleles in gram-positive bacteria; the reverse is seen with gram-negative bacteria.
Antibiotic structure also has a strong influence on allele selection. For example, placement of the C-7-ring ethyl at the N position of a C-8-methoxy fluoroquinolone (PD161144), rather than at the C-3′ position (PD161148), tended to select the G89C GyrA variant rather than the D95G variant as the most resistant (figure 3). As shown in table 1, the MIC of the G89C mutant was more than twice that of the D95G mutant when the N-ethyl compound (PD161144) was examined, whereas little difference between the alleles was seen with the C-3′-ethyl derivative (PD161148). Both amino acid substitutions are located in α-helix 4 of the breakage-reunion domain of gyrA , which is likely to be an important part of the fluoroquinolone binding site . One explanation for structure-dependent selection of the G89C allele is that Cys-89 interferes more with fluoroquinolone binding when an ethyl group is attached to the C-7 piperizinyl ring nitrogen than when it is bound to a ring carbon. This explanation orients fluoroquinolones on a-helix 4, placing the distal end of the C-7 ring near position 89 and the remainder of the quinolone between amino acids 91 and 95. Such an orientation would explain why a mutation in E. coli at the equivalent of M. smegmatis position 89 confers resistance to fluoroquinolones but not to nalidixic acid , a compound that lacks a C-7 ring.
Although the 2 mycobacterial species we studied responded in the same general way to fluoroquinolone challenge, significant differences were seen. For example, with M. smegmatis, ciprofloxacin selected few GyrA variants, and PD161148 selected many (figure 3), whereas the opposite occurred with M. tuberculosis (figure 4). Moreover, none of 89 non-GyrA mutants of M. smegmatis had mutations in the QRDR of gyrB. With M. tuberculosis, almost all of the non-GyrA mutants (80%) selected at low fluoroquinolone concentrations contained gyrB alleles (figure 4). Additional experiments are required to establish that these observations reflect species differences, because the M. tuberculosis isolate we examined was multidrug resistant, whereas this was not true for M. smegmatis.
The molecular basis for GyrB-mediated resistance is unknown. In the present study, 1 allele, N510Y, predominated (table 2). This mutation, plus the 9 less common ones (table 2), can now be combined with fluoroquinolone structure variants to study drug-gyrase interactions (see, for example, ).
In conclusion, the data presented in this report emphasize the importance of antibiotic concentration on the selection of resistance alleles. Concentrations that exceed the MIC of the most resistant mutant require a bacterial cell to acquire 2 concurrent mutations to exhibit resistance. Because that is a rare event, bacterial populations challenged with those concentrations will rarely become resistant [8, 37]. We are now examining other antibacterial agents to determine whether selection of resistance follows principles similar to those observed with fluoroquinolones.
We thank the following for critical comments on the manuscript: M. Gellert, M. Gennaro, S. Kayman, T. Palzkill, and B. Shopsin.