Distribution of Common and Rare Genetic Markers of Second-Line-Injectable-Drug Resistance in Mycobacterium tuberculosis Revealed by a Genome-Wide Association Study

ABSTRACT

Point mutations in the rrs gene and the eis promoter are known to confer resistance to the second-line injectable drugs (SLIDs) amikacin (AMK), capreomycin (CAP), and kanamycin (KAN). While mutations in these canonical genes confer the majority of SLID resistance, alternative mechanisms of resistance are not uncommon and threaten effective treatment decisions when using conventional molecular diagnostics. In total, 1,184 clinical Mycobacterium tuberculosis isolates from 7 countries were studied for genomic markers associated with phenotypic resistance. The markers rrs:A1401G and rrs:G1484T were associated with resistance to all three SLIDs, and three known markers in the eis promoter (eis:G-10A, eis:C-12T, and eis:C-14T) were similarly associated with kanamycin resistance (KAN-R). Among 325, 324, and 270 AMK-R, CAP-R, and KAN-R isolates, 274 (84.3%), 250 (77.2%), and 249 (92.3%) harbored canonical mutations, respectively. Thirteen isolates harbored more than one canonical mutation. Canonical mutations did not account for 103 of the phenotypically resistant isolates. A genome-wide association study identified three genes and promoters with mutations that, on aggregate, were associated with unexplained resistance to at least one SLID. Our analysis associated whiB7 5′-untranslated-region mutations with KAN resistance, supporting clinical relevance for this previously demonstrated mechanism of KAN resistance. We also provide evidence for the novel association of CAP resistance with the promoter of the Rv2680-Rv2681 operon, which encodes an exoribonuclease that may influence the binding of CAP to the ribosome. Aggregating mutations by gene can provide additional insight and therefore is recommended for identifying rare mechanisms of resistance when individual mutations carry insufficient statistical power.

INTRODUCTION

Tuberculosis (TB) remains a constant global public health threat due to rising cases of drug resistance among various strains of Mycobacterium tuberculosis. Half a million estimated TB cases were rifampicin resistant in 2020, including 3 to 4% of new TB cases and 18 to 21% of previously treated cases (1). This trend is exacerbated in countries of the former Soviet Union, where over half of the isolates from previously treated TB patients were rifampicin resistant (1). In 2018, 78% of rifampicin-resistant cases were also resistant to isoniazid, making them multidrug-resistant tuberculosis (MDR-TB) cases (2). In 2018, an estimated 6.2% of MDR-TB cases were extensively drug resistant (XDR), then defined as MDR-TB strains that were additionally resistant to at least a fluoroquinolone and a second-line injectable drug (SLID) (2).

Successful tuberculosis treatment relies on early identification and effective regimens, which can be ensured by rapid and accurate drug susceptibility testing (DST) to identify potential MDR-TB cases. MDR-TB should be treated with combinations of drugs shown in vitro to be effective (3). Phenotypic DST takes weeks, during which time patients may face ineffective treatment regimens with often debilitating side effects. Molecular diagnostics, in contrast, can be performed rapidly. As these rely on genetic markers of resistance, to improve their accuracy, we must understand the mechanisms behind resistance and comprehensively identify all their markers.

The SLIDs amikacin (AMK) and kanamycin (KAN) kill M. tuberculosis cells by binding to the 16S ribosomal subunit and disrupting translation (4). The bactericidal mechanism of the SLID capreomycin (CAP) is less understood. It is hypothesized to bind to the 70S ribosomal subunit and limit mRNA-tRNA translocation (5). In 2018, the WHO recommended against the use of CAP and KAN due to their side effects such as ototoxicity and nephrotoxicity (6) and their significant association with treatment failure (7). Even though AMK is now being phased out in favor of bedaquiline (8), it is still widely used. Although CAP and KAN are currently not recommended in treatment regimens, understanding the mechanisms of resistance to these drugs can inform and corroborate our understanding of AMK resistance (AMK-R), as the three drugs have similar mechanisms of action.

Genetic markers can rapidly identify SLID resistance (SLID-R) in clinical isolates (9). Across multiple studies, the most frequently observed SLID-R marker within the M. tuberculosis complex is rrs:A1401G (10–13). This single nucleotide polymorphism (SNP) in the 16S rRNA gene rrs typically causes cross-resistance to all three SLIDs (11, 13). Other mutations in rrs are also associated with SLID-R but are far rarer (14).

While rrs:A1401G typically causes resistance to all three SLIDs, not all isolates resistant to one or more SLIDs have this marker (11). Many KAN-resistant (KAN-R) isolates, for example, harbor variants in the promoter of eis (11, 15, 16). The eis gene encodes an N-acetyltransferase, which can inactivate KAN when overexpressed (15). Spectrophotometric assays have further shown that eis acetylates KAN (17). The eis promoter mutation eis:C-14T, a C to T transition located 14 nucleotides upstream of eis, has also been shown to cause resistance to amikacin (18). The amikacin MIC, when increased by eis:C-14T, is just below the critical concentration, and resistance is thus inconsistently detected by culture-based DST (18). These eis promoter markers are common in M. tuberculosis strains from the former Soviet Union, where the use of KAN has been high (16). Together with rrs:A1401G and rrs:G1484T, these known markers (Table 1) explain most SLID-R M. tuberculosis isolates (18).

TABLE 1.

Known SLID-R markers in M. tuberculosis^a

Antibiotic	Known markers
AMK	rrs:A1401G, rrs:G1484T, eis:C-14T
CAP	rrs:A1401G, rrs:G1484T
KAN	rrs:A1401G, rrs:G1484T, eis:G-10A, eis:C-12T, eis:C-14T

^aThis set of mutations was derived from the WHO-endorsed catalogue of resistance-associated mutations (18) and used to determine the expected resistance of clinical isolates to AMK, CAP, and KAN.

Previous studies have found SLID-R isolates with no known markers (11, 13). Mutations in several genes have been suggested to cause resistance in these isolates, including tlyA (19), whiB7 (20), vapC21 (21), and bfrB, also known as ferritin (22, 23). Knockout variants in tlyA induce CAP resistance (CAP-R) in vitro (5). Loss-of-function mutations in tlyA (5, 19, 24) are proposed to prevent CAP from binding to the 16S subunit (19) by preventing the methylation of 16S (25).

To find alternative mechanisms of SLID-R, we performed a genome-wide association study (GWAS) on 1,184 clinical M. tuberculosis isolates, including 111 SLID-R isolates with no known SLID-R markers. Our methods corroborated the association of whiB7 with kanamycin resistance and identified several putative amikacin resistance markers in ppe51, a transport mediator previously implicated in resistance to the drug candidate 3,3-bis-di(methylsulfonyl)propionamide (26).

RESULTS

This study surveyed 1,184 clinical isolates for injectable-drug resistance markers. AMK and CAP phenotypic DST data were available for 1,163 and 1,159 isolates, respectively (Table 2). Only 496 isolates had phenotypic DST data available for KAN (Table 2). The isolates were categorized based on phenotypic DST and the presence of known resistance markers. No isolate tested for all three SLIDs was monoresistant to AMK (see Fig. S1 at https://doi.org/10.5281/zenodo.6149149).

TABLE 2.

Phenotypic drug susceptibility testing results^a

Drug	No. of resistant isolates/no. of isolates tested	% resistant isolates
AMK	325/1,163	27.9
CAP	324/1,159	27.9
KAN	270/496	54.4

^aThe number of total resistant M. tuberculosis isolates out of all isolates with available phenotypic drug susceptibility testing (DST) results for each drug and the percentage of isolates with DST data for a given drug that show resistance are shown.

Mutations in rrs and the eis promoter are associated with SLID resistance and specific lineages.

Known SLID-R markers in the rrs gene and the eis promoter were associated strongly with resistance (Table 3). They predicted SLID-R with a sensitivity similar to that in a 2018 GWAS (10), although the sensitivity was lower than those in previous studies (5, 24). The lower sensitivity is likely due to the deliberate selection of clinical isolates with discordant genotypes and phenotypes. Variant rrs:A1401G was the most frequent marker (n = 260) (Table 4), while rrs:G1484T was the least frequent (n = 4) (Table 4). However, no isolate with rrs:G1484T was susceptible to any SLID that it was tested for (Table 4). The known marker rrs:A1401G was more frequent (181/700) within lineage 2 (East Asian) than within any other lineage (Fig. 1 and Table 5). Despite rrs:A1401G typically conferring full cross-resistance, rrs:A1401G was carried by 22 CAP-susceptible (CAP-S) isolates (Table 4), all of which were AMK-R.

TABLE 3.

Sensitivity and specificity of SLID-R prediction with known markers^a

Drug	No. of isolates				Sensitivity (%) (95% CI)	Specificity (%) (95% CI)
	Explained resistant	Explained susceptible	Unexplained resistant	Unexplained susceptible
AMK	276	829	51	27	84.3 (82.1–86.3)	96.8 (95.6–97.7)
CAP	250	813	74	22	77.2 (74.6–79.5)	97.4 (96.2–98.2)
KAN	249	223	21	3	92.3 (89.6–94.4)	98.7 (97.1–99.4)

^aA 95% confidence interval for each estimate was calculated using the score method with continuity correction (64). Isolates with genotypic-phenotypic concordance were classified as “explained.” Otherwise, they are “unexplained.”

TABLE 4.

Number of M. tuberculosis isolates with each known marker^a

Variant	No. of isolates
	AMK-R	AMK-S	CAP-R	CAP-S	KAN-R	KAN-S
rrs:A1401G	260	9	246	22	188	1
rrs:G1484T	4	0	4	0	3	0
eis:G-10A	5	49	3	51	10	0
eis:C-12T	10	95	19	84	42	2
eis:C-14T	11	18	3	26	10	0

^aIsolates are classified based on their resistance or susceptibility to each SLID. R, resistant; S, susceptible. Note that some isolates carried multiple markers.

FIG 1.

Frequency of known resistance markers within each lineage among 1,184 clinical M. tuberculosis isolates. The y axis represents the percentage of clinical M. tuberculosis isolates from each lineage that possess the mutations represented on the x axis. The lineages of 12 isolates could not be determined, including 1 isolate with eis:C-14T and 1 isolate with rrs:A1401G. One other isolate belonged to the M. africanum West African 1 lineage.

TABLE 5.

Frequency of known resistance markers within each lineage^a

Variant	Indo-Oceanic	East Asian	East African-Indian	Euro-American
rrs:A1401G	7	181	6	66
rrs:G1484T	1	1	2	0
eis:C-12T	0	54	0	53
eis:C-14T	0	25	0	3
eis:G-10A	3	47	3	2
Total	68	700	53	350

^aThe number of clinical isolates in each lineage carrying each known SLID-R marker, regardless of phenotypic DST, is shown. The lineages of 12 isolates could not be determined, including 1 isolate with eis:C-14T and 1 isolate with rrs:A1401G. One other isolate belonged to the Mycobacterium africanum West African 1 lineage.

The eis promoter mutation C-14T was carried by 11 AMK-R and 18 AMK-S isolates (Table 4). Ten of the AMK-R isolates with eis:C-14T carried no other SLID-R marker in rrs (Table 4). Three unexplained AMK-R isolates had evidence of heteroresistance, with rrs:A1401G being supported by 10 to 20% of the mapped reads (see Table S1 at https://doi.org/10.5281/zenodo.6149149). No unexplained AMK-R isolates had reads supporting rrs:G1484T.

Mutations in the eis promoter were especially common in isolates from Moldova (Table 6). Among the 270 KAN-R isolates, isolates with known KAN-R markers in the eis promoter were 27.8 times more likely to be from Moldova than isolates without known markers in the eis promoter (95% confidence interval [CI], 12.8 to 64.8; P < 2.2e−16 by Fisher’s exact test) (Table 6). The 89 isolates from Moldova belonged exclusively to lineage 4 (Euro-American, n = 51; and East Asian, n = 38). Similarly, among all 1,184 isolates, the only isolates with variant eis:C-12T were East Asian (54/700) or Euro-American (53/350) (Fig. 1). Ten isolates with known markers in the eis promoter also carried the marker rrs:A1401G (see Table S3 at the URL mentioned above). Of the three KAN-S isolates with known KAN-S markers, one had both eis:C-12T and a nonsynonymous eis mutation, eis:V163I. Promoter mutations in eis do not confer resistance if eis has a loss-of-function mutation (27); however, eis:V163I is not a frameshift or early stop codon and is thus less likely to be a loss-of-function mutation.

TABLE 6.

Geographic specificity of KAN resistance markers in the eis promoter^a

Specificity	No. of isolates
	Moldovan	Not Moldovan	Total
Kanamycin resistant
Carries known marker in the eis promoter	48 (2)	13 (1)	61 (3)
No known marker in the eis promoter	24 (21)	185 (170)	209 (191)
Total	72	198	270
Kanamycin susceptible
Carries known marker in the eis promoter	2	0	2
No known marker in the eis promoter	13	211	224
Total	15	211	226

^aThis contingency table reports the association between the possession of a known KAN-R marker in the eis promoter and collection from Moldova, stratified by KAN DST. Two Moldovan isolates had low sequencing coverage in the eis promoter and were thus excluded from this contingency table and all KAN analyses. The numbers of isolates with either of the SLID-R markers rrs:A1401G and rrs:G1484T are included in parentheses.

High cooccurrence of rrs:A1401G with the streptomycin resistance marker rrs:A514C.

Variants rrs:C492T, rrs:C517T, and rrs:A514C were the most frequent noncanonical rrs mutations among the clinical isolates (Fig. 2). The three mutations were not associated with resistance to any SLID. Both rrs:A514C and rrs:C517T are known streptomycin resistance markers (28). Variant rrs:A514C was frequently carried by isolates that also carried the SLID-R marker rrs:A1401G. Among all clinical isolates, rrs:A514C was 6.0 times more likely to be carried in isolates with rrs:A1401G than in isolates without rrs:A1401G (P = 2.424e−11 by Fisher’s exact test).

FIG 2.

Mutations in the eis promoter and rrs. Shown are numbers of clinical M. tuberculosis isolates with mutations in the rrs gene or the eis promoter, with resistance or susceptibility to AMK, CAP, and KAN. Each column reports the numbers of resistant (R) and susceptible (S) isolates carrying each mutation. For each drug, only isolates with phenotypic DST results for that drug were counted. A vertical line separates the mutations found in rrs from those in the eis promoter. Note that column “*57 mutations” represents a set of 57 rrs variants called in a single isolate, which were combined for brevity.

Insufficient statistical power to identify individual alternative resistance markers.

As there were only 21 unexplained KAN-R isolates in this study (Table 3), there was insufficient statistical power to identify potential rare KAN-R mechanisms. In this isolate set, aside from known markers, no significant association was found between KAN-R and any other mutations in the eis promoter. While previous studies found the mutation eis:G-37T in isolates with at least low-level resistance to KAN (15, 16), only one of our isolates carried it (Fig. 2). This isolate was phenotypically KAN-S, although it is possible that the MIC would reveal low-level resistance to KAN. Similarly, the mutation rrs:C1402T was carried by only three isolates, including 1 AMK-R, 1 AMK-S, 1 AMK-untested, 2 CAP-R, and 1 CAP-S isolates (Fig. 2).

Mutations in the RNA methylase gene tlyA confer CAP-R in vitro (5) and have previously been observed in clinical isolates (29). However, in this isolate set, no significant association was found between CAP-R and any mutation in tlyA. Only six nonsynonymous tlyA mutations were carried by any CAP-R isolates that did not also carry the known marker rrs:A1401G (see Table S4 at https://doi.org/10.5281/zenodo.6149149). None of these six mutations were carried by more than two CAP-R isolates (see Table S4 at the URL mentioned above).

Excluding isolates with known SLID-R markers, no individual mutation in the genome predicted resistance to any SLID with >21.8% sensitivity in this isolate set. No mutation with at least a 50% positive predictive value had more than 5.03% sensitivity, and no mutation exclusive to resistant isolates had more than 1.34% sensitivity. No AMK-R-exclusive mutation was carried by more than two unexplained AMK-R isolates. Similarly, no CAP-R-exclusive mutation was carried by more than two unexplained CAP-R isolates, and no KAN-R-exclusive mutation was carried by more than four unexplained KAN-R isolates. Due to their rarity, there was insufficient statistical power to determine whether any of these mutations were alternative resistance markers.

Our set of clinical isolates included 333 isolates sequenced on single-molecule real-time (SMRT) sequencers, which can resolve many known blind spots in the M. tuberculosis genome, such as in the proline-glutamic acid (PE) and proline-proline-glutamic acid (PPE) gene families (30). However, no mutations in PE or PPE genes were associated with resistance, even for a subset of only the SMRT-sequenced isolates.

Several genes have previously been suggested to affect SLID-R. A support vector machine approach identified variants in three genes as potential determinants of an XDR phenotype: vapC21, Rv3471c, and Rv3848 (21). Comparative proteomics suggested that 12 genes may be involved in resistance to AMK or KAN: atpA, tig, lpdC, tuf, moxR1, Rv2005c, 35kd_ag, prcA, Rv0148, bfrA, bfrB, and hspX (22). The efflux pump Rv1258c (31), the transcriptional regulator whiB7 (32), and the virulence gene whiB6 (33) have also been associated with SLID resistance. However, no individual variant within these 18 genes was present in more than one AMK-R or CAP-R isolate after removing variants also present in susceptible isolates. Variants present in one or more susceptible isolates also were not significantly associated with resistance in this isolate set.

whiB7 variants in aggregate are associated with kanamycin resistance.

Different variants in the same gene can cause the same change in phenotype (34). If multiple rare variants in the same gene cause resistance, they may be missed by a genome-wide association study on individual variants. To account for this, we measured the association between SLID-R and the variants in each M. tuberculosis gene in aggregate (Table 7).

TABLE 7.

Top genes from the genome-wide association study, aggregated by gene^a

Gene	SLID	No. of unexplained resistant isolates with resistance-exclusive mutations	Total no. of unexplained resistant isolates with mutations in the gene	Proportion	No. of unique resistance-exclusive mutations in unexplained resistant isolates	No. of isolates with resistance-exclusive mutations carried by unexplained resistant isolates
plcB	AMK	3	5	0.60	5	3
Rv0584	AMK	3	5	0.60	3	7
recB	AMK	3	6	0.50	3	3
smc	AMK	3	7	0.43	3	3
Rv2680_prom	CAP	7	7	1	15	8
cysG_prom	CAP	5	7	0.71	26	17
thrB_prom	CAP	5	8	0.62	30	29
plcB	CAP	4	7	0.57	10	4
cut1	CAP	4	7	0.57	4	12
Rv1726	CAP	4	7	0.57	5	4
whiB7_prom	KAN	9	9	1	9	16
cyp142	KAN	5	5	1	5	6
accE5	KAN	5	5	1	13	21
prfB	KAN	5	5	1	4	5
folC	KAN	5	5	1	4	37

^aThe number of unexplained resistant isolates with resistance-exclusive mutations for each SLID and each gene is the number of isolates that have unexplained resistance to that SLID and carry a mutation in that gene that is not carried by any isolate susceptible to that SLID. The total number of unexplained resistant isolates with mutations in the gene for each gene and each SLID is the number of isolates that have unexplained resistance to that SLID and carry any mutation in that gene. Genes below the mean for this count were removed (5 for AMK, 7 for CAP, and 5 for KAN). “Proportion” is the first count divided by the second. Genes associated with first-line-drug resistance were excluded.

For KAN, the whiB7 5′ untranslated region (UTR) had the strongest signal, with nine unexplained KAN-R isolates carrying resistance-exclusive mutations (Table 7). Each of the nine isolates carried a different whiB7 5′-UTR mutation. WhiB7 regulates eis (35), which is in turn associated with KAN-R. Two of these whiB7 5′-UTR mutations, whiB7:T-116C and whiB7:C-56T, were previously reported in KAN-R spontaneous mutants (32). Two of the nine isolates carried additional mutations in the coding DNA sequence (CDS) of whiB7, the missense mutation whiB7:E69A and an in-frame deletion of codons 84 through 87. While these are not nonsense mutations, if they still cause a loss of function, then these two KAN-R isolates remain unexplained despite their whiB7 5′-UTR mutations (27). None of the nine isolates carried mutations in the CDS of eis.

Meanwhile, seven unexplained CAP-R isolates carried CAP-R-exclusive mutations in the upstream region of Rv2680 (Table 7). These seven mutations were all within 82 bp of both the Rv2680 start codon and the Rv2680 transcription start site reported previously by Cortes et al. (36). While the function of Rv2680 is unknown, it is likely in an operon with Rv2681 (37), a homologue of the exoribonuclease RNase D (38). Rv2681 is on the same strand as Rv2680, and the start codon of Rv2681 is 2 bp after the stop codon of Rv2680.

DISCUSSION

While known markers in the eis promoter and rrs were associated strongly with SLID-R, 111 SLID-R isolates lacked known markers in this study of 1,184 clinical M. tuberculosis isolates. Some of the discordant isolates may be due to errors in phenotypic DST. Phenotypic/genotypic DST results sometimes disagree (39) and have shown an error rate as high as 2.2% for AMK (40). These discordant isolates also suggest the existence of rare, alternative mechanisms of resistance. However, identifying rare mechanisms is difficult as no single variant is carried by enough isolates for a strong association with resistance. To overcome this difficulty, we searched for rare mechanisms by aggregating the mutations around each gene.

The known marker rrs:A1401G was ubiquitous and strongly associated with cross-resistance to all three SLIDs (see Table S2 at https://doi.org/10.5281/zenodo.6149149). However, while the variant was carried by 246 CAP-R isolates, it was also carried by 22 CAP-S isolates (Table 4). The discordant isolates may still have low-level resistance to CAP, as a previous study reported a wide range of CAP MIC values (8 to 40 μg/mL) among clinical isolates carrying rrs:A1401G (41). However, the cause of this variable MIC is still unknown. The same study reported that mutagenesis of reference strains consistently resulted in a 40-μg/mL CAP MIC, suggesting that the inconsistency was due to the genetic background of the clinical isolates rather than the mechanism of rrs:A1401G itself (41). However, we observed no mutations common to the genetic background of our 22 discordant isolates.

The known marker rrs:G1484T was rare, carried by only four clinical isolates (Table 4), all cross-resistant. The low prevalence of rrs:G1484T has been reported previously despite a high MIC (14, 42). In a mutagenesis study on Mycobacterium smegmatis, rrs:G1484T mutants grew slower than rrs:A1401G mutants, suggesting that this rarity is due to a fitness cost of rrs:G1484T (43). However, a later study found that rrs:G1484T conferred no growth disadvantage to the M. tuberculosis reference strain H37Rv, although it did confer a disadvantage to a strain of the Beijing F2 sublineage (41). The full extent of the fitness cost of rrs:G1484T across clinical isolates is unknown.

KAN-R was most often explained by rrs:A1401G (Table 4), except among isolates from Moldova, which were enriched for known markers in the eis promoter (Table 6). Moldova is representative of countries in the region of the former Soviet Union, where this geographic trend has been reported previously (16). The prevalence of eis promoter mutations in this region is thought to be the result of the extensive use of KAN (16).

While known markers in the eis promoter and rrs were associated strongly with resistance, no other mutations within these genes were significantly associated with SLID-R in this isolate set (Fig. 2). The mutation rrs:C1402T was infrequent and carried by both resistant and susceptible isolates (Fig. 2). Mutagenesis has previously shown that rrs:C1402T reduces susceptibility to a level near the critical concentration (41). Both AMK-R and AMK-S isolates carried rrs:C1402T in a recent WHO study (18), and the marker is considered resistant by the GenoType MTBDRsl platform (Hain Lifescience).

The known eis promoter and rrs mutations (Table 1) were not carried by 21 of the 270 KAN-R isolates (Table 3), leaving the genetic basis of their resistance unexplained. Of these, seven isolates carried whiB7 5′-UTR mutations (Table 7), although these mutations were unique in each isolate. Thus, while no single whiB7 5′-UTR mutation was associated strongly with KAN-R, their aggregate signal suggests that whiB7 5′-UTR mutations are an alternative mechanism of KAN-R. This finding in clinical isolates is supported by previous mutagenesis experiments. Increased expression of whiB7 causes low-level streptomycin resistance and KAN-R in H37Rv mutants (32), while deletion of whiB7 in Mycobacterium abscessus lowers the MICs of erythromycin, tetracycline, streptomycin, and AMK (44). The whiB7 gene encodes a transcriptional activator that regulates ribosome protection and efflux pump genes (44). Moreover, whiB7 regulates eis, providing a plausible mechanism for KAN-R (32).

The promoter of the Rv2680-Rv2681 operon was associated with unexplained CAP-R, with a signal comparable to that of the whiB7 5′ UTR and KAN-R (Table 7). While the function of Rv2680 is unknown, Rv2681 is a homologue of the exoribonuclease RNase D (38). The sequence homology between Rv2681 and RNase D (38) is further supported by the similarity of its predicted protein structure (45). Mutations in the promoter of the Rv2680-Rv2681 operon may affect the transcription of Rv2681, in turn altering the structure of rRNA processed by the exoribonuclease encoded by Rv2681. CAP disrupts protein synthesis by binding to the ribosome (25). By altering rRNA, these mutations may prevent CAP from binding to the ribosome. This potential mechanism would explain seven CAP-R isolates that lacked known CAP-R markers but carried mutations in the promoter of the Rv2680-Rv2681 operon that were absent from all CAP-S isolates.

The known SLID-R markers are accurate predictors of resistance. However, they still do not explain all SLID-R cases. Rare, alternative mechanisms, such as whiB7 5′-UTR mutations, are likely responsible for these unexplained SLID-R isolates. For molecular diagnostics to fully replace phenotypic diagnostics, these rare mechanisms must also be understood. Finding these rare mechanisms will require sequences from larger sets of unexplained resistant isolates and more sensitive methods of association, such as the machine learning approaches employed previously (21) or aggregating mutations by gene as done here. This method independently corroborated the association between whiB7 5′-UTR mutations and KAN-R (44) and identified a new association between CAP-R and mutations in the promoter of the Rv2680-Rv2681 operon, which encodes an exoribonuclease (38) and may impact the binding between CAP and the ribosome.

MATERIALS AND METHODS

Sample collection.

As part of a previous study, 323 clinical M. tuberculosis isolates were collected for long-read PacBio sequencing (24). There were 89 isolates that originated from Hinduja National Hospital (PDHNH) in Mumbai, India; 89 that came from the Phthisiopneumology Institute (PPI) in Chisinau, Moldova; 48 that were from the Tropical Disease Foundation (TDF) in Manila, Philippines; and 97 that were from the National Health Laboratory Service of South Africa (NHLS) in Johannesburg, South Africa. All raw sequences were uploaded to the NCBI Sequence Read Archive (SRA) database under the BioProject accession number PRJNA353873. An additional 10 M. tuberculosis clinical isolates were collected from the Supranational Reference Laboratories in Stockholm, Sweden, and Antwerp, Belgium. These isolates were originally genotyped with a Hain Lifescience GenoType MTBDRsl line probe assay (9) and were chosen for sequencing due to discordance between their genotype and phenotypic DST results for any SLID. Another 851 whole-genome sequences were downloaded from the NCBI SRA database using the SRA Toolkit’s fastqdump (46). These 851 downloaded raw reads were previously sequenced on Illumina short-read platforms (47–50).

Phenotypic drug susceptibility testing.

DST for the PacBio-sequenced isolates was performed on the Bactec mycobacterial growth indicator tube (MGIT) 960 platform (BD Diagnostic Systems, Franklin Lakes, NJ, USA) using the 2008 WHO-recommended critical concentrations of 1.0 mg/L (AMK) and 2.5 mg/L (CAP/KAN) as described in previous studies (24, 51, 52). DST for Illumina-sequenced isolates was also performed on the MGIT 960 platform using contemporary WHO-recommended critical concentrations, as described previously (47–50). As of 2018, the recommended critical concentrations for AMK, CAP, and KAN remain at 1.0 mg/L, 2.5 mg/L, and 2.5 mg/L, respectively (53). Bacterial isolates were excluded from analysis if DST data were not available for at least one SLID.

DNA extraction and sequencing.

The DNAs of all 333 isolates collected for long-read PacBio sequencing, including those from the WHO Supranational Reference Laboratories in Stockholm and Antwerp and the NCBI SRA database, were extracted as described in a previous study (54). The SMRT sequencing protocol was described previously (55, 56). Sixty-four isolates were later resequenced due to low coverage. DNA extraction for the 851 downloaded public genomes was previously described (47–50). The downloaded genomes were sequenced on the Illumina Genome Analyzer, MiSeq, or HiSeq platform.

Genome assembly, alignment, and variant calling.

Genome assembly, alignment, and variant calling methods are described in the supplemental material at https://doi.org/10.5281/zenodo.6149149. Briefly, PBHoover (57) aligned 64 SMRT-sequenced isolates to H37Rv and called variants. Later, 269 SMRT-sequenced isolates were de novo assembled with canu (58) or HGAP2 (Pacific Biosciences), and their assembled genomes were then aligned to reference strain H37Rv using dnadiff (v1.3) (59) for variant calling, with the output converted to VCF format by a custom script, mummer-snps2vcf (https://gitlab.com/LPCDRP/mummer-extras/-/blob/master/src/mummer-snps2vcf). Reads from Illumina-sequenced isolates were aligned to H37Rv using bowtie2 (v2.2.4) (60), and variants were then called with VarScan2 (v2.3) (61).

Lineage identification.

Mycobacterial interspersed repetitive-unit–variable-number tandem-repeat (MIRU-VNTR) analysis and spoligotyping were previously performed on the initial 323 SMRT-sequenced isolates collected (51). The 10 isolates sent from Stockholm and Antwerp and the 851 downloaded Illumina-sequenced isolates underwent MIRU-VNTR analysis and spoligotyping with MiruHero, a custom Python script (https://gitlab.com/LPCDRP/miru-hero). MiruHero used the rule-based criteria from TB-Insight (62) to classify lineages.

Identifying known resistance-conferring mutations.

After variant calling, known SLID-R markers were searched for in the VCF file of each clinical isolate. The eis promoter mutations C-14T, C-12T, and G-10A are known KAN-R markers, and the rrs mutations G1484T and A1401G are known markers of resistance to all three SLIDs (24). The genomic positions and orientation of rrs and the eis promoter are noted in Table S5 at https://doi.org/10.5281/zenodo.6149149 (63). Known markers and phenotypic DST were used to estimate the sensitivity and specificity of predicting resistance to each SLID using known markers. A 95% confidence interval was calculated for sensitivity and specificity estimates using the score method with continuity correction (64).

Genome-wide association.

For each of the three drugs, separate genome-wide association studies were performed using a custom Python script (https://gitlab.com/LPCDRP/gwa) to identify novel markers for alternative mechanisms of resistance. To remove the overriding signal of known resistance markers, our GWAS excluded isolates that had their resistance explained by known markers (Table 1). This exclusion was necessary to avoid confounding associations with potential alternative mechanisms of resistance. Isolates with additional eis promoter resistance markers (Table 1) were excluded from our KAN-R analysis for similar reasons. We calculated the sensitivity and specificity of each variant’s prediction of resistance to each SLID (the proportion of unexplained resistant isolates with the variant and the proportion of susceptible isolates without the variant, respectively). For each SLID, we identified variants absent from susceptible isolates and ranked them by the number of unexplained resistant isolates that carried them.

In the gene-based association, for each SLID and each gene, we counted the number of unexplained resistant isolates with at least one variant in that gene that was absent from isolates susceptible to that SLID. Genes below the mean for this count were removed. This count was then divided by the total number of unexplained resistant isolates with any variant in that gene. Genes with known markers of resistance to first-line drugs were excluded, as most SLID-R isolates are also resistant to first-line drugs due to prior treatment with first-line drug regimens.