Mycobacterium tuberculosis Population Structure Determines the Outcome of Genetics-Based Second-Line Drug Resistance Testing

Abstract

The global emergence of multidrug-resistant tuberculosis has highlighted the need for the development of rapid tests to identify resistance to second-line antituberculosis drugs. Resistance to fluoroquinolones and aminoglycosides develops through nonsynonymous single nucleotide polymorphisms in the gyrA and gyrB genes and the rrs gene, respectively. Using DNA sequencing as the gold standard for the detection of mutations conferring resistance, in conjunction with spoligotyping, we demonstrated heteroresistance in 25% and 16.3% of Mycobacterium tuberculosis isolates resistant to ofloxacin and amikacin, respectively. Characterization of follow-up isolates from the same patients showed that the population structure of clones may change during treatment, suggesting different phases in the emergence of resistance. The presence of underlying mutant clones was identified in isolates which failed to show a correlation between phenotypic resistance and mutation in the gyrA or rrs gene. These clones harbored previously described mutations in either the gyrA or rrs gene, suggesting that rare mutations conferring resistance to ofloxacin or amikacin may not be as important as was previously thought. We concluded that the absence of a correlation between genotypic and phenotypic resistance implies an early phase in the emergence of resistance within the patient. Thus, the diagnostic utility of genetics-based drug susceptibility tests will depend on the proportion of patients whose bacilli are in the process of acquiring resistance in the study setting. These data have implications for the interpretation of molecular and microbiological diagnostic tests for patients with drug-susceptible and drug-resistant tuberculosis who fail to respond to treatment and for those with discordant results.

INTRODUCTION

Estimates by the World Health Organization (WHO) suggest that approximately 440,000 multidrug-resistant tuberculosis (MDR-TB) cases (resistant to at least isoniazid and rifampin) were diagnosed in 2008 (42). Of these, 5.4% are estimated to be extensively drug-resistant TB (XDR-TB), i.e., resistant to isoniazid and rifampin and to at least one representative of each class of the most effective second-line drugs, i.e., fluoroquinolones and the injectable drugs kanamycin, amikacin, and capreomycin (44). By January 2010, XDR-TB cases had been reported in 58 countries (42). In response to the high mortality and rapid nosocomial spread which can be associated with XDR-TB (11, 14), the WHO/Stop TB partnership issued an eight-point response plan (41). This plan emphasized the need for the strengthening of diagnostic services due to their absence in many parts of the developing world. In addition, this plan highlighted the need for the development of rapid diagnostics (12). Currently, the gold standard drug susceptibility testing (DST) method measures phenotypic resistance using the culture-based indirect proportion method. This method relies on the detection of growth of >1% of the inoculum (24) on culture medium containing a critical concentration of an anti-TB drug (45). However, culture-based methods are slow due to the inherent growth rate of Mycobacterium tuberculosis, which in turn delays the initiation of appropriate treatment, which can lead to ongoing transmission and the risk of amplification of resistance (6, 26).

Drug resistance in M. tuberculosis develops spontaneously, and the resulting drug-resistant mutants are selected during periods of poor adherence, monotherapy, or the administration of inappropriate treatment. The relationship between a defined mutation and resistance to a specific anti-TB drug has become a dogma (23) and now forms the basis for the development of molecular biology-based drug resistance diagnostic assays (1, 4, 33). In 2008, the WHO approved the Hain GenoTypeMTBDRplus test for the rapid identification of M. tuberculosis complex strains in clinical isolates and the diagnosis of isoniazid and rifampin resistance by the detection of mutations in the inhA promoter and katG gene and in the rpoB gene, respectively (43). A recent meta-analysis demonstrated that the application of this test can reduce diagnostic delay (21), with a high sensitivity and specificity for identifying rifampin resistance, which is an important marker of MDR- and XDR-TB. Similarly, the XpertMTB/RIF test enables the genetic detection of rifampin resistance, with the added advantage that the result can be obtained in approximately 2 h for most smear-positive sputum samples and a significant number of smear-negative sputum samples (4, 35). Rollout of genetics-based DST methods promises to decrease the time to diagnosis of MDR-TB and to increase the number of cases screened.

However, the application of these methods presents new challenges to diagnostic algorithms, since the identification of resistance to rifampin needs to be complemented with additional tests to provide a more comprehensive resistance profile in order to tailor treatment (7). Since these additional tests are currently culture based, it is essential that rapid genetics-based methods be developed to detect mutations conferring resistance to the most important second-line anti-TB drugs (fluoroquinolones and aminoglycosides) due to the high burden of additional resistance associated with MDR-TB (8). Resistance to fluoroquinolones develops primarily through mutations in the gyrA and gyrB genes (29), while resistance to aminoglycosides develops mostly through mutations in the rrs gene (31). The newly developed GenoType MTBDRsl test detects the high-frequency mutations occurring in the gyrA, rrs, and embB (marker for ethambutol resistance) genes. Evaluation of the performance of this test has suggested that the sensitivity for detecting fluoroquinolone resistance is between 75.6% and 90.6%, and the sensitivity for detecting kanamycin resistance is between 77% and 100% (15, 20). In contrast, the sensitivity for detecting ethambutol resistance is only 64.2% (20).

The sensitivity of genetics-based DST methods depends on the inclusion of most, if not all, mutations conferring resistance and knowledge of the phenotypic resistance levels associated with particular mutations (5). A major limitation associated with genetics-based DST is the absence of data on the complete repertoire of mutations conferring resistance to specific anti-TB drugs. Currently, it is assumed that phenotypic drug resistance in the absence of known mutations is indicative of unknown mechanisms conferring resistance (25). An alternative explanation is that of heteroresistance (mixed population of genetically homogeneous drug-resistant and drug-susceptible strains in the same clinical isolate [28]), where the underlying resistant population is below the detection limit of the genetics-based assay.

We investigated the genetic basis of resistance to ofloxacin and amikacin in the first available M. tuberculosis isolate as well as follow-up isolates from patients with pre-XDR-TB and XDR-TB by comparing their gyrA and rrs gene sequences with routine culture-based DST results. Our analysis demonstrates the complexity of the bacterial population structure in these sputum isolates. Both mixed-strain infections (mixed populations of drug-resistant and drug-susceptible clones with different genotypes [36]) and heteroresistance were identified. We discuss our findings in terms of the diagnostic utility of genotypic drug resistance tests in settings where drug resistance is either acquired or transmitted.

MATERIALS AND METHODS

Study population.

In accordance with the National TB Control Programme of South Africa, sputum specimens were collected from high-risk patients (those previously treated for TB, failing first-line therapy, or in contact with a patient with drug-resistant TB) and submitted to the National Health Laboratory Service (NHLS) for DST with isoniazid and rifampin, using either the MGIT Bactec 960 indirect proportion method or the GenoType MTBDRplus line probe assay. Starting in November 2006, sputum isolates resistant to isoniazid and rifampin were subjected to a second round of DST in MGIT Bactec 960 medium containing 2 μg/ml, 4 μg/ml, and 5 μg/ml ofloxacin, amikacin, and ethionamide, respectively. All resistant isolates were submitted to Stellenbosch University for genotyping and sequencing of genes known to confer resistance. This study was approved by the ethics review committees of Stellenbosch University (REF N09/11/296) and the University of Cape Town (REF 038/2008).

Clinical files were accessible only for MDR-TB patients from Cape Town. Files were reviewed to establish whether the patients were receiving or had received treatment with either ofloxacin or kanamycin at the time that the first available isolate was collected.

Genotyping.

The first available M. tuberculosis isolate from each patient which showed resistance to either ofloxacin or amikacin by routine DST, in addition to follow-up isolates (if available), was analyzed to determine the relationship between phenotype and genotype. Patients whose first available isolate was contaminated, lost viability, or failed to show resistance on subsequent testing were excluded.

First available and follow-up isolates were genotyped by spoligotyping (18) according to the internationally standardized method. Strain genotypes were identified as previously described (34). In order to identify mixed infections with strains with identical spoligotypes, the entire pncA gene was sequenced. Mutations conferring ofloxacin resistance were determined by DNA sequencing of the quinolone resistance-determining region (QRDR) of the gyrA gene and flanking sequences (amplification product codons 18 to 132), while mutations conferring amikacin resistance were determined by DNA sequencing of the region encompassing nucleotide 1401 of the rrs gene (amplification product nucleotides 1339 to 1528). Briefly, a 200-μl aliquot of the MGIT culture was heat inactivated by incubation at 100°C for 30 min to generate a crude DNA lysate (40). PCR amplification was done in a reaction mixture containing 2 μl crude DNA template, 5 μl Q buffer, 2.5 μl 10× buffer, 2 μl 25 mM MgCl₂, 4 μl of deoxynucleoside triphosphates (dNTPs) (10 mM concentration of each), 1 μl of the QRDR primer set (50 pmol/μl) (forward, 5′-TGACATCGAGCAGGAGATGC-3′; and reverse, 5′-GGGCTTCGGTGTACCTCATC-3′), the rrs primer set (50 pmol/μl) (forward, 5′-GTAATCGCAGATCAGCAAC-3′; and reverse, 5′-GTGATCCAGCCGCACCTT-3′), or the pncA primer set (50 pmol/μl) (forward, 5′-AGTCGCCCGAACGTATGGTG-3′; and reverse, 5′-CAACAGTTCATCCCGGTTCG-3′), and 0.125 μl HotStarTaq DNA polymerase (Qiagen, Germany) and made up to 25 μl with distilled water (dH₂O). Amplification was initiated by incubation at 95°C for 15 min, followed by 35 to 45 cycles of 94°C for 45 s, 62°C for 45 s, and 72°C for 45 s. After the last cycle, the samples were incubated at 72°C for 10 min. To minimize laboratory cross-contamination, the preparation of the PCR mixes, the addition of DNA, and PCR amplification were conducted in physically separated rooms. Negative controls (water) were included to detect reagent contamination. Amplification was confirmed by electrophoretic fractionation in 1% agarose containing Tris-borate-EDTA (TBE), pH 8.3. Amplification products were sequenced using an ABI 3130XL genetic analyzer, and the resulting chromatograms were analyzed using Chromas software. The presence of more than one nucleotide at a defined position in the DNA sequence was assigned if the peak height of the underlying nucleotide was ≥2 times the height of the highest background peak.

Isolates which demonstrated phenotypic resistance in the absence of known mutations were subcultured onto Middlebrook 7H10 medium containing either 2 μg/ml ofloxacin or 5 μg/ml amikacin for 3 to 4 weeks at 37°C. Thereafter, single CFU were picked, suspended in 1,000 μl of enriched 7H9 medium (containing albumin, dextrose, and catalase), and incubated at 37°C for 4 days. Thereafter, a 500-μl aliquot was stored at −80°C, while the remaining aliquot was heat inactivated at 100°C to generate a crude DNA lysate. This DNA lysate was genotyped by spoligotyping and DNA sequencing of the QRDR, the rrs gene, or the pncA gene as described above.

Heteroresistance was defined for sputum isolates which demonstrated the presence of more than one QRDR or rrs sequence while showing only a single spoligotype and pncA sequence (27). Mixed infections were defined by the presence of different strains (according to spoligotype and pncA sequence) in the same sputum isolate (40).

RESULTS

Patients with ofloxacin-resistant MDR-TB.

During the period from November 2006 to December 2008, ofloxacin-resistant MDR-TB isolates cultured from 241 patients were submitted to Stellenbosch University for genotyping. A total of 181/241 (75%) isolates were available for analysis. Of these, 171 could be subcultured for subsequent spoligotyping and DNA sequencing of the gyrA gene encompassing the QRDR (5 were contaminated or lost viability, and 5 were found to be susceptible to ofloxacin as a result of either a transcription error, incorrect scoring of resistance, or overgrowth of the drug-susceptible clone due to an in vitro fitness advantage). DNA sequencing identified mutations in the QRDR or flanking sequence in the first available ofloxacin-resistant isolate for 162/171 (94.7%) patients. The DNA sequence chromatogram showed the presence of a single QRDR mutation in 126 (78%) of these 162 isolates, while mutations in the region flanking the QRDR sequence were identified in 2 (1.2%) of these 162 isolates (Table 1).

Table 1.

QRDR mutations identified in fluoroquinolone-resistant M. tuberculosis isolates

QRDR mutation in first available isolate	No. of patients with mutation
A90V	34
D94A	14
D94G	52
D94N	15
D94Y	2
G88C	1
S91P	8
H70R	1
H52Q	1

First available isolates from the remaining 34/162 patients (21%) showed multiple sequence peaks at one or more nucleotide positions in the QRDR. Clinical records were available for 22/34 patients, among which 16 indicated that the patient had received ofloxacin prior to or was receiving ofloxacin at the time of sputum collection (Table 2). Twelve of the 34 isolates showed the presence of “wild-type” as well as single mutant QRDR sequences, and 22 isolates showed the presence of multiple mutations in the QRDR (Table 2). Thirty-three of the 34 isolates showed spoligotypes with no underlying patterns, while the pncA sequence chromatograms confirmed clonality for 29/33 isolates. Together, these data suggest the absence of mixed infection with different strains, thereby implying heteroresistance. This was confirmed for 5 randomly selected isolates by spoligotyping and sequencing of the pncA gene from single CFU (8 to 10 colonies per isolate). The remaining isolate showed the presence of two distinct genotypes by spoligotyping, implying mixed infection. This was confirmed by spoligotyping of single CFU, which identified strains with either a LAM or Beijing genotype.

Table 2.

QRDR mutations identified for patients whose isolates showed heteroresistance^a

Patient	QRDR sequence variant(s) in first available isolate	Ofloxacin treatment for >30 days	Patient on ofloxacin treatment when sample was taken (for >30 days)	QRDR sequence variant(s) in follow-up isolates	No. of days between follow-up isolates
1	D94G/D94A	NA		D94A	400
2	D94G/D94A	NA		D94G	52
3	D94G/D94A	No	No	D94G	84
4	D94N/D94Y	Yes	No	WT/D94G/D94N/D94S	63
5	WT/A90P/D94G	NA		NA
6b	WT/A90V	Yes	Yes	A90V	9
7	WT/A90V	No	No	WT	578
8	WT/A90V	No	No	WT/A90V	902
9	WT/A90V	Yes	No	WT/A90V/D94G	185
10	WT/A90V	NA		NA
11	WT/A90V/D94G	No	No	A90V	34
12	WT/A90V/D94G	NA		A90V	82
13	WT/A90V/D94G	NA		D94G	372
14	WT/A90V/D94G	NA		WT/A90V/D94G
15	WT/A90V/D94G	Yes	Yes	NA
16	WT/A90V/D94G	Yes	Yes	NA
17	WT/A90V/D94G	No	No	NA
18	WT/A90V/D94G/D94Y/D94C	Yes	Yes	WT/D94G/D94Y/D94C/D94N/D94S	568
19	WT/D89G	Yes	No	D89G	22
20	WT/D94A	Yes	Yes	NA
21	WT/D94A	Yes	No	NA
22	WT/D94G	Yes	No	D94G	147
23	WT/D94G	NA		NA
24	WT/D94G	NA		NA
25	WT/D94G/A90V/S91P	Yes	No	NA
26	WT/D94G/D94N/D94S	NA		D94N	41
27	WT/D94G/D94N/D94S	Yes	No	WT/D94G/D94N/D94S	63
28	WT/D94G/D94N/D94S	NA		NA
29	WT/D94G/D94N/D94S	Yes	No	NA
30	WT/D94G/D94N/D94Y/D94S/D94C	NA		NA
31	WT/D94G/D94Y	Yes	No	NA
32	WT/D94Y/D94G/D94C	Yes	No	NA
33	WT/G88C/D94N	Yes	Yes	WT/G88C/D94G/D94N/D94S	84
34	WT/S91P	No	No	WT/D94G	181
35	WT	No	No	D94G	397
36	WT	Yes	Yes	WT	2
37	WT	Yes	Yes	NA
38	WT	NA		NA
39	WT	No	No	NA
40	WT	NA		NA
41	WT	Yes	No	NA
42	WT	Yes	Yes	NA
43	WT	NA		NA

^aNA, not available; WT, wild type.

^bMixed infection identified by spoligotyping.

Eighteen of the 33 patients with heteroresistance had follow-up isolates (Table 2). DNA sequence analysis of the QRDR in follow-up isolates from 9 of these patients showed the selection of a single QRDR mutant during subsequent treatment, while the spoligotypes remained identical to those of the first available isolates. Clonality was confirmed for 7/9 patients whose first available and follow-up isolates showed identical pncA mutations. This is exemplified in Fig. 1, which shows the DNA sequence chromatograms of follow-up isolates from three different patients. Follow-up isolates from 5/33 patients showed the emergence of additional QRDR mutations, while the initial QRDR mutational profile remained constant during the study period for three patients (data not shown). In all instances, the spoligotype was identical to that of the first available isolate. Furthermore, the mutant pncA sequence remained the same for both the first available and follow-up isolates for 6/8 patients. One patient's follow-up isolate reverted to a “wild-type” QRDR sequence after 19 months of treatment without ofloxacin (ofloxacin was removed from the treatment regimen once phenotypic resistance was noted). The spoligotype pattern and pncA sequence remained the same during the period of follow-up (family X1), suggesting overgrowth of an underlying ofloxacin-susceptible clone.

Fig 1.

DNA sequence chromatograms of the QRDR for follow-up isolates from three patients in whom fluoroquinolone resistance emerged. The sequence chromatograms for patient 1 show the emergence of two QRDR variants (D94G and D94A), followed by the selection of the D94G variant. The sequence chromatograms for patient 2 show the presence of at least 3 QRDR variants (D94G, D94S, and D94N), including the “wild-type” (WT) sequence, followed by the selection of the D94N variant. The sequence chromatograms for patient 3 show the persistence of three QRDR variants (WT, A90V, and D94G) for at least 16 months, followed by the selection of the D94G variant.

QRDR mutations were absent in the first available isolate for 9/171 (5.3%) patients, despite the isolates being classified as phenotypically resistant to ofloxacin by routine DST. Clinical records from 6 patients were available and showed that 4 had received or were receiving ofloxacin treatment at the time of sputum collection (Table 2). DNA sequencing of single CFU selected in the presence of ofloxacin (2 μg/ml) confirmed the existence of underlying clones with QRDR mutations in 7/9 (77.8%) patient isolates, while 2/9 (22.2%) patient isolates showed clones with only the “wild-type” QRDR sequence. Spoligotyping and sequencing of the pncA gene of single CFU (with and without QRDR mutations) selected from the first available isolate for 7 patients showed identical genotypes, thereby demonstrating heteroresistance. Two of the seven patients had follow-up isolates; in one patient, the underlying mutant clone was selected to become the dominant clone, while in the other patient no selection was observed and the dominant population remained “wild type” for the QRDR. In all instances, the spoligotypes and pncA sequences were identical to those of the first available isolate. Follow-up isolates were not available for the two patients whose first available isolates showed only “wild-type” QRDR sequences.

Patients with amikacin-resistant MDR-TB.

During the study period, amikacin-resistant MDR-TB isolates from 177 patients were received from the NHLS. Isolates from 160 (90.4%) patients were available for further analysis. Of these, 14 lost viability or were contaminated, and 6 were found to be susceptible to amikacin (as a result of either a transcription error, incorrect scoring of resistance, or overgrowth of the drug-susceptible clone due to an in vitro fitness advantage) and were excluded from further analysis. First available isolates from 140 patients were subjected to genotypic analysis by spoligotyping and DNA sequencing of the pncA and rrs genes. Analysis of the DNA sequencing chromatograms confirmed the presence of an A1401G mutation in the rrs gene in 117/140 (83.6%) patient isolates (not shown).

DNA sequence chromatograms of the first available isolates from 12 patients (8.6%) showed the presence of both the “wild-type” (A) and mutant (G) nucleotides at position 1401 of the rrs gene (Table 3). For all isolates, no underlying spoligotype patterns were identified, and for 10/12 isolates, no multiple peaks in the pncA sequence chromatograms were identified, suggesting heteroresistance. Clinical records were available for 3/12 patients, and 1 of these patients had received kanamycin at the time the sputum was collected, suggesting the emergence of resistant clones during treatment (Table 3). Follow-up isolates were available from 3/12 patients, of which 2/3 confirmed that the resistant clones had become the dominant population in the follow-up isolates (Table 3). In both instances, the spoligotypes and pncA mutant sequences were identical to those of the first available isolate.

Table 3.

rrs nucleotide 1401 mutations identified in amikacin-resistant M. tuberculosis isolates^a

Patient	rrs sequence variant(s) in first available isolate	Kanamycin treatment for >30 days	Patient on kanamycin treatment when sample was taken (for >30 days)	rrs mutation in follow-up isolates	No. of days between follow-up isolates
1	WT/A1401G	NA		A1401G	12
2	WT/A1401G	No	No	A1401G	52
3	WT/A1401G	Yes	Yes	WT	200
4	WT/A1401G	NA		NA
5	WT/A1401G	NA		NA
6	WT/A1401G	NA		NA
7	WT/A1401G	NA		NA
8	WT/A1401G	No	No	NA
9	WT/A1401G	NA		NA
10	WT/A1401G	NA		NA
11	WT/A1401G	NA		NA
12	WT/A1401G	NA		NA
13	WT	NA		WT	593
14	WT	NA		WT	609
15	WT	NA		WT	804
16	WT	NA		WT	664
17	WT	Yes	Yes	A1401G	325
18b	WT	No	No	A1401G	9
19	WT	NA		A1401G	57
20	WT	Yes	No	NA
21	WT	Yes	Yes	NA
22	WT	NA		NA
23	WT	NA		NA

^aNA, not available; WT, wild type.

^bMixed infection identified by spoligotyping.

rrs mutations were absent in the first available isolate for 11/140 (7.9%) patients. Available clinical records from 4 patients showed that 3 were receiving or had received kanamycin at the time the isolates were collected (Table 3). DNA sequencing of single CFU, selected in the presence of amikacin (5 μg/ml), confirmed the existence of underlying clones harboring the A1401G mutation in the rrs gene in all 11 patient isolates. The spoligotypes were identical for individual colonies with or without the rrs mutation for 10 of the patient isolates, suggesting heteroresistance. Clonality was confirmed for single colonies from 5 patient isolates by the presence of distinct pncA mutations. The isolate from the remaining patient showed a mixed infection (same isolate as the ofloxacin-resistant mixed infection described above). DNA sequencing of follow-up isolates from 7 of 11 patients confirmed the emergence of the rrs A1401G mutation in 3 patient isolates. In all instances, the spoligotypes and mutant pncA sequences were identical to those of the first available isolate. Follow-up isolates from the remaining 4 patients showed that the dominant population retained the “wild-type” rrs sequence.

DISCUSSION

Recent molecular epidemiological studies have demonstrated that the population structure of M. tuberculosis may not be homogeneous in sputum isolates (30, 40) and in different disease foci (13, 19). The clinical importance of heterogeneous M. tuberculosis populations is particularly relevant for patients where both drug-susceptible and drug-resistant strains are present in the same sputum (2, 36). Thus, heteroresistance poses a significant challenge to genetics-based drug resistance diagnostics, as the sensitivity of detecting genomic mutations conferring drug resistance will be dependent on the complexity of the bacterial population structure within the specimen being analyzed, including the number of different mutations conferring resistance to a single anti-TB drug as well as the ratio between resistant and susceptible clones (9, 16). In most instances, primary resistance would be associated with a homogeneous population of bacilli harboring the same single mutation. In contrast, drug resistance by acquisition, mixed infection, or reinfection would initially be associated with a heterogeneous population, possibly progressing to a homogeneous population after antibiotic selection has been sustained for an extended period (22). Thus, the relative proportion of patients whose bacilli are in the process of acquiring drug resistance during treatment or who have been reinfected with a drug-resistant strain, leading to mixed infection with a drug-susceptible strain, will influence the utility of genetics-based DST methods.

Using DNA sequencing, this study demonstrated a high level of complexity in the population structure of M. tuberculosis bacilli in sputum isolates which were phenotypically resistant to either a fluoroquinolone or an aminoglycoside. Our analysis showed that 79% of the first available isolates resistant to ofloxacin and 83.6% of the first available isolates resistant to amikacin displayed single mutations in the gyrA (QRDR region) and rrs (nucleotide 1401 region) genes, respectively (29). Twenty-one percent of the ofloxacin-resistant isolates and 8.6% of the amikacin-resistant isolates showed multiple QRDR sequences and rrs (nucleotide 1401 region) sequences, respectively. We propose that the relative proportions of resistant and susceptible clones or the number of different resistant clones present in a sputum isolate reflects different stages in the progression toward clonal resistance. This conclusion differs from a previous report which suggested that multiple mutations in the QRDR lead to higher levels of resistance (32). An alternative explanation could be that of laboratory cross-contamination. However, this would have been detected as mixed infection due to the high strain diversity in the study setting. The absence of mixed infections in this study population may be explained by the fact that many of the patients had previously been treated or were receiving treatment for MDR-TB, which would have selected for a defined drug-resistant M. tuberculosis population in each sputum specimen.

DNA sequencing of amplicons generated by amplification of DNAs from the entire M. tuberculosis population from each isolate failed to identify mutations conferring resistance in 5.3% of the ofloxacin-resistant isolates and 7.9% of the aminoglycoside-resistant isolates. In a subset of the isolates tested, underlying resistant clones with known gyrA or rrs mutations were identified after culture on the corresponding anti-TB drug. This confirms the insensitivity of the DNA sequencing method and thereby demonstrates that heteroresistance has the potential to influence the sensitivity of genetics-based tests. The sensitivity for detecting underlying populations of bacilli has not been reported extensively, although a recent study using the XpertMTB/RIF test suggested that a mutant population could be detected by this method only if it constituted >65% of the population (3). This is significantly less sensitive than the 1% proportion method. Thus, current genetics-based DST methods cannot be used to “rule out” the presence of resistant strains when testing the first available isolate. Repeating the test during the course of therapy may be essential in order to “rule in” drug resistance. However, due to the cost of genetics-based DST, this may not be possible in resource-limited settings. Therefore, there is a need to reduce the cost as well as improve the sensitivity of future tests.

Our observation that drug-resistant bacilli with known gyrA or rrs mutations could be selected from the first available isolate in cases where the total population showed phenotypic resistance in the absence of genotypic resistance contrasts with previous assumptions which have suggested that resistance in these isolates is due to mutations outside the gyrA/gyrB or rrs gene (29). Based on this finding, we suggest that additional mechanisms conferring drug resistance may not be as important as was previously assumed. However, this hypothesis needs to be tested in more settings. We suggest that future studies aimed at identifying novel mechanisms leading to drug resistance should follow a selection strategy to exclude the possibility that subpopulations (with known mutations) polarize the phenotypic DST classification as resistant by the 1% proportion method.

Our finding of heteroresistance in isolates resistant to second-line anti-TB drugs differs from what has been reported previously for resistance to first-line anti-TB drugs in the same setting (1). In that study, >95% of rifampin-resistant isolates demonstrated known mutations in the rifampin resistance-determining region (RRDR) of the rpoB gene (the remaining 5% of isolates were thought to have mutations outside the RRDR) (1). These contrasting scenarios suggest different epidemiological reasons for the emergence of first- and second-line resistance. Numerous molecular epidemiological studies have suggested that MDR-TB is transmitted primarily in the study setting, as measured by the high level of clustering (17, 37–39). Our observation of heteroresistance suggests that a proportion of resistance to the second-line drugs is acquired in this setting. This finding differs from the Tugela Ferry study in KwaZulu-Natal, South Africa, which suggested that primarily XDR-TB was transmitted (11, 14). Risk factors associated with the emergence of resistance to second-line anti-TB drugs remain largely unknown. Two recent studies have demonstrated the emergence of fluoroquinolone resistance despite excellent adherence (6, 10). It is therefore essential that further studies, including studies of the treatment regimens used and the quality of drugs, as well as pharmacokinetic studies of both HIV-negative and HIV-positive patients, be conducted to identify these risk factors.

The clinical implications of our findings are that heteroresistance may explain false-negative genotypic susceptibility results depending on the proportion of the dominant strain or clone. Thus, patients with drug-susceptible TB, isoniazid-monoresistant TB, or MDR-TB who fail to respond to appropriate treatment regimens in the absence of nonadherence may in fact have higher-grade resistance. Thus, in clinical practice, genotypic DST results must be interpreted in the clinical context and appropriate retesting performed when indicated. The caveat we highlight must be borne in mind when interpreting genotypic results, particularly now that molecular diagnostic tools such as the XpertMTB/RIF test are being rolled out in high-burden settings (35).

We acknowledge that this study has certain limitations. The limited access to clinical records may have influenced our interpretation that heteroresistance implied acquisition of drug resistance. It is possible, however, that some patients had not received anti-TB treatment and thus heteroresistance implied reinfection with a genetically closely related drug-resistant strain. Furthermore, the use of two genetic markers may be insufficient to differentiate heteroresistance from reinfection with a genetically closely related drug-resistant M. tuberculosis strain (strain with a common drug-resistant progenitor). Lastly, the absence of follow-up samples means that we were unable to conclusively determine the influence of anti-TB treatment on the population structure of clones in patient sputum isolates over time.

Despite these limitations, we concluded that the utility of a genetics-based test will be dependent on the proportion of patients who acquire drug resistance during treatment or who have recently been reinfected with a drug-resistant strain.