Detection of First- and Second-Line Drug Resistance in Mycobacterium tuberculosis Clinical Isolates by Pyrosequencing

Abstract

Conventional phenotypic drug susceptibility testing (DST) methods for Mycobacterium tuberculosis are laborious and very time-consuming. Early detection of drug-resistant tuberculosis (TB) is essential for prevention and control of TB transmission. We have developed a pyrosequencing method for simultaneous detection of mutations associated with resistance to rifampin, isoniazid, ethambutol, amikacin, kanamycin, capreomycin, and ofloxacin. Seven pyrosequencing assays were optimized for following loci: rpoB, katG, embB, rrs, gyrA, and the promoter regions of inhA and eis. The molecular method was evaluated on a panel of 290 clinical isolates of M. tuberculosis. In comparison to phenotypic DST, the pyrosequencing method demonstrated high specificity (100%) and sensitivity (94.6%) for detection of multidrug-resistant M. tuberculosis as well as high specificity (99.3%) and sensitivity (86.9%) for detection of extensively drug-resistant M. tuberculosis. The short turnaround time combined with multilocus sequencing of several isolates in parallel makes pyrosequencing an attractive method for drug resistance screening in M. tuberculosis.

INTRODUCTION

Tuberculosis (TB), caused by Mycobacterium tuberculosis, persists as a global public health problem. Drug-resistant bacteria, especially multidrug-resistant (MDR) M. tuberculosis and extensively drug-resistant (XDR) M. tuberculosis, pose challenges for the prevention and control of this deadly disease (64). MDR M. tuberculosis is defined as an isolate resistant to at least the two main first-line anti-TB drugs, rifampin (RIF) and isoniazid (INH), whereas XDR M. tuberculosis is defined as an MDR isolate that is also resistant to a fluoroquinolone (FQ) and at least one of the following second-line injectable agents: amikacin (AMK), kanamycin (KAN), and capreomycin (CAP) (8). TB control and prevention programs are based on early diagnosis followed by rapid identification of drug resistance (62). Generally, detection of drug-resistant M. tuberculosis is performed by culture and drug susceptibility testing (DST) in liquid or on solid media. However, these procedures are laborious and take several weeks to months to complete. The development of rapid molecular methods, which can be performed within 1 or 2 days, is important for the timely identification of appropriate TB treatment. In fact, the World Health Organization (WHO) endorses the use of genotypic methods that target specific molecular mutations associated with resistance to individual drugs (63). Two examples of commercially available methods are Xpert MTB/RIF and GenoType MTBDR (16–18), which are predominately based on a lack of probe hybridization to wild-type loci, indirectly indicating the presence of mutations. The Xpert MTB/RIF system is also limited in that it detects only RIF resistance.

Spontaneous chromosomal mutations are the genetic basis for drug resistance in M. tuberculosis (40, 68), and a limited number of mutations account for the majority of phenotypic resistance to anti-TB drugs. RIF interacts with the ß-subunit (encoded by rpoB) of the RNA polymerase and inhibits the early steps of transcription (5). Resistance to RIF is almost entirely coupled to mutations within an 81-bp region of the rpoB gene, called the RIF resistance-determining region (RRDR), comprising codons 507 to 533 (14, 40, 46). In accordance with standard practice, these codon numbers correspond to those found in the Escherichia coli rpoB sequence, as designated on the basis of alignment of the translated sequence, and do not represent the actual codon positions in M. tuberculosis rpoB (30, 32, 58). The mechanisms of resistance to isoniazid (INH), though extensively investigated, remain incompletely understood (60). INH is a prodrug that is activated by the catalase-peroxidase enzyme KatG, encoded by the katG gene (22). The primary target of INH in M. tuberculosis is believed to be the enoyl-acyl carrier protein reductase InhA, which is involved in mycolic acid synthesis (3, 39, 43). Mutations in katG, specifically at codon 315, lead to high-level INH resistance (40, 67). Low-level resistance to INH is attributed to mutations in the promoter region of the mabA-inhA operon (here denoted inhA) leading to overexpression of InhA (26, 60). Ethambutol (EMB) inhibits arabinosyl transferases, encoded by the embCAB operon, thereby interfering with biosynthesis of the cell wall component arabinogalactan (55, 59). Mutations associated with resistance to EMB are primarily found in the embB gene (40). Amikacin (AMK), kanamycin (KAN), and capreomycin (CAP) inhibit protein synthesis by binding to the ribosome (21, 31, 38). Resistance to all three drugs is associated with mutations in the 16S rRNA gene rrs, specifically at nucleotide positions 1401, 1402, and 1484 (1, 11, 28, 54, 57). A mutation at nucleotide 1401 or 1484 is associated with resistance to all these agents, whereas a mutation at nucleotide 1402 is associated with CAP resistance and low-level KAN resistance (11, 28). Overexpression of eis (encoding the aminoglycoside acetyltransferase Eis), caused by mutations in the promoter region, confers low-level resistance to KAN (66). Fluoroquinolones (FQs), e.g., ofloxacin (OFL), which bind to DNA gyrase, inhibit proper regulation of supercoiling and cause chromosomal double-strand breaks (65). DNA gyrase is a heterotetramer consisting of two A and two B subunits, encoded by the genes gyrA and gyrB (34, 61). FQ resistance-associated mutations are predominantly found in a short region, codons 88 to 94, of the gyrA gene (53, 56, 65).

Pyrosequencing is a semiautomated sequencing method based on real-time monitoring of DNA synthesis, optimized to analyze short DNA sequences. It is based on the quantitative detection of released pyrophosphate during DNA synthesis. In a cascade of enzymatic reactions, light is generated at intensities proportional to the numbers of incorporated nucleotides (33, 42). Pyrosequencing has previously been described as a method for detection of drug resistance in M. tuberculosis (2, 4, 23).

The objective of this study was to develop and evaluate a molecular method for detection of drug resistance-associated mutations in clinical isolates of M. tuberculosis by pyrosequencing technology. Apart from further optimization of the current rpoB assay (23), we have designed pyrosequencing assays for katG, embB, rrs, gyrA, and the promoter regions of inhA and eis. These assays facilitate rapid molecular detection of resistance to RIF, INH, EMB, AMK, KAN, CAP, and OFL in clinical isolates. Phenotypic and genotypic DST results were compared for determination of the sensitivity and specificity of the molecular assays for each drug.

MATERIALS AND METHODS

Bacterial strains and phenotypic drug susceptibility testing.

The pansusceptible reference strain M. tuberculosis H37Rv (ATCC 25618) and 290 clinical isolates of M. tuberculosis from Europe, Asia, Africa, and Latin America were selected from the culture collection at the WHO Supranational TB Reference Laboratory at the Swedish Institute for Communicable Disease Control in Solna, Sweden, and included in this study. The classification of strains as susceptible or resistant was based on DST performed earlier using WHO-recommended standardized and quality assured methods for M. tuberculosis and, in some cases, complemented by MIC determinations (11).

Gene amplification.

Genomic DNA was isolated as described elsewhere (23), or by the method of Sandegren and colleagues (45), but using only one chloroform extraction step. Parts of genes rpoB, katG, embB, rrs, and gyrA and of the promoter regions of inhA and eis were amplified using PCR. Primers and amplicon sizes are presented in Table S1 in the supplemental material. Amplifications were performed in a final volume of 50 μl with 1× PCR buffer, 2 mM MgCl₂ for all loci (except for rpoB, for which 1 mM MgCl₂ was used), 100 μM each dNTP (Applied Biosystems, Warrington, United Kingdom) (for rpoB, inhA, eis and gyrA) or 200 μM each dNTP (for embB, rrs and katG), 200 nM each primer, 1 U of AmpliTaq Gold (Applied Biosystems, Branchburg, NJ), and 10 ng of DNA. Thermocycling conditions are specified in Table S1 in the supplemental material.

Pyrosequencing.

Seven pyrosequencing assays were optimized for the following loci: rpoB, katG, embB, rrs, gyrA, and the promoter regions of inhA and eis. A previously reported rpoB assay was further optimized by using only two sequencing primers, F1 and F9 (see Table S1 in the supplemental material) and a directed dispensation order instead of cyclic dispensation (23), thus reducing the number of sequencing reactions needed per sample. The directed dispensation order for rpoB was designed to detect all mutations in the RRDR reported by Ramaswamy and Musser (40). All four nucleotides are repeatedly dispensed in cyclic dispensation, whereas the nucleotide dispensation order is preprogrammed according to a known sequence in directed dispensation. Cyclic dispensation order was used for katG, embB, gyrA, and the promoter region of inhA. Apart from the rpoB locus, directed dispensation order was also used for rrs and the promoter region of eis.

Preparation of templates and sequencing reactions were performed according to the instructions of the manufacturers (Biotage AB, Uppsala, Sweden, and Qiagen GmbH, Hilden, Germany). Briefly, each locus was amplified by PCR with a biotinylated reverse primer for immobilization onto streptavidin-coated Sepharose beads (GE Healthcare, Uppsala, Sweden). The PCR product was converted into a single-stranded DNA template and purified using a vacuum preparation tool. A sequencing primer (0.4 μM) was subsequently annealed to the single-stranded DNA (see Table S1 in the supplemental material). Pyrosequencing was carried out in a PSQ 96MA instrument (Pyrosequencing AB, Uppsala, Sweden) using a PSQ Gold 96 SQA reagent kit (Biotage AB, Uppsala, Sweden, and Qiagen GmbH, Hilden, Germany) and a nucleotide dispensation order as specified in Table S1 in the supplemental material. Pyrograms were interpreted and mutations were detected using both peak height and sequence output. Sequences were analyzed in BioEdit version 7.0.4.1 (13) and Geneious Pro version 4.8.4 (Biomatters Ltd., Auckland, New Zealand) by alignment to the M. tuberculosis H37Rv loci (GenBank accession no. NC_000962; NCBI bank) (7) and the sequence results of M. tuberculosis H37Rv (ATCC 25618).

Sensitivity and specificity.

Phenotypic DST was considered the gold standard for comparison with the results obtained by pyrosequencing. Sensitivity was calculated as the number of true-positive specimens (phenotypically resistant and pyrosequencing resistant) divided by the number of true-positive specimens plus the number of false-negative specimens (phenotypically resistant and pyrosequencing susceptible). Specificity was calculated as the number of true-negative specimens (phenotypically susceptible and pyrosequencing susceptible) divided by the number of true-negative specimens plus the number of false-positive specimens (phenotypically susceptible and pyrosequencing resistant).

RESULTS

We have developed a pyrosequencing method for simultaneous detection of mutations associated with resistance to RIF, INH, EMB, AMK, KAN, CAP, and OFL in M. tuberculosis clinical isolates. In comparison to phenotypic DST, the molecular method demonstrated an overall specificity of 97.2% and a sensitivity of 83.2%. For detection of MDR M. tuberculosis isolates, a specificity of 100% and a sensitivity of 94.6% were noted. The corresponding figures for detection of XDR M. tuberculosis isolates were 99.3% and 86.9%, respectively.

For RIF resistance detection, the sensitivity was 94.7% and the specificity was 100% (Table 1). Of 209 RIF-resistant isolates, 198 had a mutation in the RRDR, and the vast majority (189 of 198) had a mutation in codon 516, 526, or 531. The rpoB locus of all but five isolates could be sequenced using directed dispensation order and sequencing primers F1 and F9. Four of these had insertions or deletions not designed for the directed dispensation order; instead, these were sequenced using four primers (F1, F7, F9, and F13) (23) and cyclic dispensation order. One isolate could not be sequenced with primer F1, and as a control, the RRDR was sequenced by Sanger sequencing (24). This isolate harbored an L511P mutation but also a mutation in codon 505 (F→V), which is where the F1 primer anneals. The vast majority of isolates had a missense mutation in the RRDR (Table 2). In one case, the silent mutation G536G occurred together with the S531L mutation. Eleven of the RIF-resistant clinical isolates had a wild-type rpoB locus.

Table 1.

Performance of pyrosequencing as a molecular drug susceptibility test method

Pyrosequencing resultc	No. of isolates with the indicated result by phenotypic DST or % sensitivity or specificitya
	RIF		INH		EMB		AMK		KANb		CAP		OFL
	R	S	R	S	R	S	R	S	R	S	R	S	R	S
R	198	0	203	0	82	21	69	0	82	2	62	4	54	0
S	11	71	12	74	52	117	15	155	16	126	16	139	8	184
Sensitivityd (%)	94.7		94.4		61.2		82.1		83.7		79.5		87.1
Specificityd (%)	100.0		100.0		84.8		100.0		98.4		97.2		100.0

^aR, resistant; S, susceptible.

^bIsolates harboring eis promoter mutations at positions −10 and −12 were considered pyrosequencing susceptible.

^cR, isolates with a mutation in rpoB (RIF), katG or inhA (INH), embB (EMB), rrs (AMK and CAP), rrs and eis (KAN), or gyrA (OFL); S, isolates with a wild-type locus.

^dData were determined on the basis of phenotypic drug susceptibility testing as the gold standard.

Table 2.

Mutations identified within loci associated with resistance to the first-line drugs rifampin, isoniazid, and ethambutol in clinical isolates of M. tuberculosis^a

Drug	Locus(i)	Amino acid change(s)	Codon change(s)	Nucleotide change	No. of isolates with indicated phenotypic DST result
					Resistant	Susceptible
RIF	rpoB	del TS 508-509	del nt 1279–1284 (ACCAGC)		1
		del SQL 509-511	del nt 1282–1290 (AGCCAGCTG)		1b
		S509T + D516Y	AGC/ACC + GAC/TAC		1
		Q510H + D516Y	CAG/CAT + GAC/TAC		1
		L511P	CTG/CCG		1
		L511P + D516Y	CTG/CCG + GAC/TAC		2
		L511P + D516G	CTG/CCG + GAC/GGC		1
		L511R + D516Y	CTG/CGG + GAC/TAC		1
		Q513P	CAA/CCA		2
		D516G + L533P	GAC/GGC + CTG/CCG		2
		D516V	GAC/GTC		11
		D516Y	GAC/TAC		6
		D516Y + H526Y + P535L	GAC/TAC + CAC/TAC + CCC/CTC		1
		del N 519	del nt 1312–1314 (AAC)		1
		H526D	CAC/GAC		3
		H526L	CAC/GAC		3
		H526N	CAC/AAC		2
		H526R	CAC/CGC		1
		H526Y	CAC/TAC		11
		H526S + K527Q	ins nt 1331 (CGT) + del nt 1335–1337 (CAA)		1
		H526Y + P535R	CAC/TAC + CCC/CGC		1
		S531L	TCG/TTG		137
		S531L + G536G	TCG/TTG + GGC/GGT		1
		S531W	TCG/TGG		3
		L533P	CTG/CCG		3
INH	katGc	S315T	AGC/ACC		156
		S315T	AGC/ACG		1
		S315N	AGC/AAC		3
		S315R	AGC/AGA		1
	katG + inhA	S315T	AGC/ACC	T-8C	3
		S315T	AGC/ACC	T-8G	1
		S315T	AGC/ACC	C-15T	27
		S315N	AGC/AAC	C-15T	1
		S315T	AGC/ACC	G-17T	1
	inhA			C-15T	9
EMB	embB	M306V	ATG/GTG		58	5
		M306V	ATG/GTA		5
		M306I	ATG/ATA		8	6
		M306I	ATG/ATT		7	8
		M306I	ATG/ATC		1
		M306L	ATG/CTG		1	2
		A313V	GCC/GTC		1
		Y315C	TAC/TGC		1

^ains, insertion; del, deletion; nt, nucleotide.

^bHeteroresistant isolate, i.e., mixed wild-type and mutated sequences.

^cThe katG locus could not be amplified for one of the phenotypically resistant isolates.

The sensitivity and specificity for INH were 94.4% and 100%, respectively (Table 1). In total, 215 isolates were resistant to INH, and 203 of those had a mutation in katG (the majority had an S315T mutation), in the promoter region of inhA (the majority had a C-15T mutation), or in both loci (Table 2). Twelve clinical isolates were phenotypically determined to be resistant to INH but had wild-type katG and inhA loci.

The sensitivity for EMB was 61.2%, and the specificity was 84.8% (Table 1). Only 82 of the 134 phenotypically resistant isolates harbored a mutation in the embB locus. Also, among the EMB-susceptible isolates, 21 had an embB 306 mutation. Mutations in embB codons 313 and 315 were detected in EMB-resistant isolates (Table 2).

The sensitivity and specificity for AMK were 81.2% and 100%, respectively. For KAN, the sensitivity was 83.7% and the specificity was 98.4% (Table 1). The majority the AMK- and KAN-resistant isolates harbored a mutation in the rrs locus (Table 3). All but one isolate with −14 or −37 eis mutations were KAN resistant, and the majority of −10 and −12 mutations were found among susceptible isolates. A C-15G mutation was detected in a KAN-susceptible isolate. The sensitivity and specificity for CAP were 79.5% and 97.2%, respectively (Table 1). Most CAP-resistant isolates had an rrs mutation at position 1401; however, this mutation was also detected in four susceptible isolates. Few possible mutations in rrs and a directed dispensation order made it possible to achieve a read length of 95 bp, which covered nucleotide positions 1401, 1402, and 1484 in one sequence run.

Table 3.

Mutations identified within loci associated with resistance to the second-line drugs amikacin, kanamycin, capreomycin, and ofloxacin in clinical isolates of M. tuberculosis

Drug	Locus(i)	Amino acid change(s)	Codon change(s)	Nucleotide change(s)	No. of isolates with indicated phenotypic DST result
					Resistant	Susceptible
AMK	rrs			A1401G	69a
KAN	rrs			A1401G	65b
	rrs + eis			A1401G + G-6T	1
				A1401G + C-12T	1
				A1401G + G-37T	1c
	eis			G-10A	14	38
				C-12T	1	5
				C-14T	11	1
				C-15G		1
				G-37T	3
CAP	rrs			A1401G	62b	4b
OFL	gyrA	D89N	GAC/AAC		1
		A90V	GCG/GTG		23
		A90V + S91P	GCG/GTC + TCG/CCG		1
		A90V + D94N	GCG/GTG + GAC/AAC		1
		S91P	TCG/CCG		4b
		D94G	GAC/GGC		13
		D94G	GAC/GGT		1
		D94A	GAC/GCC		6
		D94N	GAC/AAC		3
		D94H	GAC/CAC		1
		D94Y	GAC/TAC		1

^aTwo isolates were heteroresistant, i.e., mixed wild-type and mutated sequences.

^bOne isolate was heteroresistant.

^cHeteroresistant isolate.

For OFL, the sensitivity and specificity were 87.1% and 100%, respectively (Table 1). Missense mutations in gyrA, specifically in codon 94, were detected in the vast majority of OFL-resistant isolates but were absent in all OFL-susceptible isolates.

By investigation of the pyrosequencing pyrograms, five heteroresistant isolates, i.e., a mixed template consisting of both wild-type and mutated sequences, were identified (Tables 2 and 3). The total turnaround time from inactivation of bacteria to interpreted sequencing results was approximately 8 h.

DISCUSSION

Bacteriological methods are highly sensitive and specific for detection of resistance to most anti-TB drugs; however, they are labor-intensive and very time-consuming. Consequently, molecular diagnostics can be used for rapid detection of drug-resistant M. tuberculosis. We have developed seven pyrosequencing assays for detection of mutations associated with resistance to first- and second-line anti-TB drugs and evaluated the method on a panel of 290 clinical isolates of M. tuberculosis.

Pyrosequencing is a robust technique that is easy to perform in a high-throughput manner. In the method developed here, 12 strains could simultaneously be sequenced in a 96-well plate. Some of the most frequent resistance-associated mutations in M. tuberculosis are located in very short regions of genes or promoter regions, whereas others are more dispersed. By using directed pyrosequencing dispensation order for rpoB, rrs, and the promoter region of eis, the entire resistance-associated locus could be covered in only one or two sequencing reactions. The successful sequencing of 95 bp of the rrs gene shows not only that it is possible to optimize longer reads for pyrosequencing but also that it is possible to do so using a GC-rich template (65%) (7). Cyclic dispensation was used for katG, the promoter region of inhA, embB, and gyrA, where mutations are confined to a short stretch of the locus. Cyclic dispensation increases tolerance for variations and eliminates the risk of failing to detect as-yet-unidentified mutations but typically renders shorter sequence reads. An advantage of sequencing compared to line-probe assays is the discriminatory power on the genotypic level. For instance, sequencing can distinguish between missense and silent mutations and discriminate against mutations not associated with drug resistance. This is in contrast to the commercially available GenoType MTBDR assays, which partly define resistance by the absence of wild-type sequence (17, 18). Pyrosequencing also requires less hands-on time than GenoType MTBDR and is more flexible, as the number of assays and the choice of included loci easily can be adjusted. Furthermore, the pyrosequencing method offers sensitivity and specificity comparable to those of Sanger sequencing (6) but is easier to handle and thus more suitable for investigation of a large number of samples. On the other hand, a drawback with pyrosequencing, as well as with other commercially available molecular methods (16–18), is that they cannot cover whole genes. Thus, pncA gene mutations, associated with resistance to the first-line drug pyrazinamide (48), cannot be sufficiently covered by these methods. Pyrogram analysis revealed that five phenotypically resistant isolates were heteroresistant. Heteroresistance is due either to segregation of a single strain into resistant and susceptible organisms or to superinfection with two different strains (one susceptible and one resistant) (19). The former scenario describes the evolution of resistance that arises from suboptimal treatment.

Introduction of directed dispensation order for the rpoB locus reduced the number of sequencing reactions needed for the RRDR to only two without compromising the detection of missense mutations. However, four isolates had in-frame deletions and insertions; thus, cyclic dispensation and four sequencing primers (F1, F7, F9, and F13) had to be employed in those cases. The rpoB assay is suboptimal in cases where missense mutations, insertions, or deletions have occurred specifically at the site where the sequencing primer anneals. This is, however, not a concern for the downstream sequences of F1, as the read length is generally sufficient to cover two sequencing primers, and mutations at the F1 primer site are rarely found among clinical isolates. In fact, sequencing primer F1 anneals upstream of the RRDR. Still, repeatedly failed sequencing reactions provide an indication that the sequencing primer cannot anneal to the specific annealing site, and we consider that the method failed to detect drug resistance in the case where an isolate had mutations in codons 505 and 511. Eleven phenotypically RIF-resistant isolates in this study did not harbor a mutation in the RRDR and were thus misclassified as susceptible by the molecular method. This may reflect the inherent limitations of phenotypic DST but could also be due to rpoB mutations outside the RRDR (14, 49) or elsewhere in the genome.

The INH assay (katG and inhA promoter region) was 100% specific and demonstrated high sensitivity. The link between INH and catalase peroxidase (KatG) originally came from the observation that some highly resistant clinical isolates of M. tuberculosis, with katG deleted, were catalase negative (67). We suspect that one of the INH-resistant isolates in this study had this genotype, as we were consistently unable to amplify the katG gene of this isolate, and this result was confirmed by the failure of another independent katG PCR assay. As the pyrosequencing assay is designed to detect mutations, we consider that the method failed to detect drug resistance in this case. It is possible that the 12 phenotypically resistant isolates lacking a mutation in either locus have a mutation elsewhere in the genome. Mutations in the open reading frame of inhA confer INH resistance (3); however, these mutations are rarely found in clinical isolates and usually occur together with mutations in katG and the promoter region of inhA (60). Consequently, we chose not to include the opening reading frame of inhA in the molecular method, as well as other genes, such as aphC and kasA, as there are limited data on their role in INH resistance (25, 60).

The embB pyrosequencing assay detected mutations in both phenotypically EMB-resistant and EMB-susceptible isolates (Table 2), suggesting that these mutations are a poor indicator of EMB resistance. The sensitivity (61.2%) and specificity (84.8%) of detection of EMB resistance were indeed lower than were seen with the other drugs; however, this does not necessarily mean that the assay has a low predictive value for clinically relevant EMB resistance. Conventional phenotypic EMB DST for M. tuberculosis is notoriously problematic (12, 27). Several studies, including allelic exchange experiments, have demonstrated a strong association between embB mutations, specifically at codon 306, and EMB resistance (35–37, 44, 51, 52). It has also been shown that there is a narrow range between the MICs of EMB-susceptible and EMB-resistant isolates of M. tuberculosis (15, 47). Some mutations, e.g., the M306I substitution, give rise to a modest increase in the MIC (37, 44, 52), which may explain why this particular mutation was found in so many EMB-susceptible isolates in this study. It is likely that the EMB MIC increases only moderately for the embB A313V substitution, as we detected it in a phenotypically resistant isolate, and others have found it in EMB-susceptible isolates (6). A possible clinical implication is that all embB 306 mutants should be treated as isolates with altered EMB susceptibility, even if they appear to be EMB susceptible by conventional DST. Mutations in codon 497 in the embB gene have been reported in EMB-resistant clinical isolates (20) and could possibly be included to increase the sensitivity of the method. Furthermore, mutations in embC and embA may also be involved (41).

Despite the relatively long distance between the mutations in rrs, the pyrosequencing assay for this locus could be optimized for only one sequencing reaction without compromising mutation detection. The specificities were high for AMK, KAN, and CAP (100%, 98.4%, and 97.2%, respectively), whereas the sensitivities were lower (82.1%, 83.7%, and 79.5%). The A1401G mutation in rrs was observed in isolates resistant to all the three agents. One isolate, which was found to be phenotypically resistant to AMK, KAN, and CAP, was identified as susceptible by the molecular method. It is possible that this isolate was heteroresistant but that the proportion of resistant organisms was not large enough to be detected by the pyrosequencing assay. Nevertheless, alternative drug resistance mechanisms cannot be ruled out. Four CAP-susceptible isolates were identified as resistant by the pyrosequencing assay (rrs 1401 mutation); however, the MIC (4 mg/liter) for these strains was close to the critical concentration for CAP (11). Overexpression of eis confers low-level KAN resistance (66). Mutations at certain positions, such as −14 and −37, give rise to a higher overexpression of eis than other eis promoter mutations (66) and can thus be considered more sensitive genetic markers for KAN resistance. All but one of the isolates with −14 or −37 mutations in this study were classified as phenotypically resistant. Mutations at position −10 and, in particular, at position −12 seem to give rise to a MIC close to the critical concentration, as isolates harboring these mutations have been reported as both KAN susceptible and KAN resistant (6, 11, 66). Because of their poor predictive value, we chose not to take mutations at these positions into account when calculating sensitivity and specificity for KAN. The previously unreported C-15T eis promoter mutation was detected in a KAN-susceptible isolate; however, further studies must be undertaken in order to clarify its role in eis expression. It should also be noted that all but one of the AMK-resistant clinical isolates lacking an rrs mutation did have a mutation (G-10A, C-14T, or G-37T) in the promoter region of eis. Overexpression of the Eis protein also leads to inactivation of AMK, but the inactivation is 3-fold less than that seen for KAN (66). As with EMB, this suggests that isolates with eis promoter mutations should be treated as isolates with altered KAN, and plausibly AMK, susceptibility, regardless of the phenotypic DST results. CAP resistance is also known to be conferred by mutations in tlyA (29); however, we chose not to include this gene in the molecular method, since tlyA mutations are rare (6, 11) and some have also been found also in CAP-susceptible isolates (11). Furthermore, in order to include tlyA as a genetic marker for CAP, the entire open reading frame must be analyzed, and this is not practically feasible by pyrosequencing.

The specificity of the OFL assay was 100%, although the sensitivity was lower. Eight of the 62 phenotypically OFL-resistant isolates lacked a mutation in the gyrA locus. In these cases, it is not certain that a mutation accounting for this phenotype would be found elsewhere in gyrA, or in gyrB, as other drug resistance mechanisms, such as efflux pumps, have been suggested to contribute to FQ resistance in M. tuberculosis (10). The gyrB locus was not included in this assay, as mutations in this gene occur at a much lower frequency and usually in association with gyrA mutations (9, 68). There were equal distributions of the gyrA S95T mutation among both susceptible and resistant clinical isolates; in fact, it was detected in 88% of the cases in both groups. This mutation is considered to be a polymorphism not associated with resistance to FQs (50).

In general, the noted sensitivity rates for detection of resistance (Table 1) reflect the incomplete understanding of resistance mechanisms associated with all drugs investigated in this study. Phenotypically resistant isolates lacking mutations in known resistance-associated loci present a problem for all molecular tests designed to detect mutations associated with resistance. This emphasizes the need for intensifying research that aims at identifying novel resistance mechanisms in M. tuberculosis.

We have developed a high-throughput molecular method for simultaneous detection of resistance to RIF, INH, EMB, AMK, KAN, CAP, and OFL in M. tuberculosis. A rapid and specific diagnosis is highly important not only for the individual patient but also from the general public perspective, since rapid modification to an effective drug therapy would reduce the spread of TB in the society. The short turnaround time in combination with a multilocus sequencing of several isolates in parallel makes pyrosequencing an attractive method for drug resistance screening in M. tuberculosis. As the method is PCR based, it has a great potential to be further developed for application directly to clinical samples, e.g., sputum samples.