Abstract
Background
The World Health Organization-endorsed Xpert MTB/XDR assay provides a rapid method to detect resistance to isoniazid, fluoroquinolones, injectables aminoglycosides and ethionamide, yet evaluation of its performance, particularly in endemic settings, remains limited.
Methods
We conducted a prospective multicentre study (June 2017 to March 2021) in nine sub-Saharan African countries, enrolling adults with pulmonary tuberculosis confirmed by Xpert MTB/RIF or Ultra. Xpert MTB/XDR results were compared to a World Health Organization-endorsed targeted next-generation sequencing reference used on the same sputum, with discordance resolved using whole genome sequencing and phenotypic drug-susceptibility testing when available. Diagnostic accuracy for each drug was calculated, also accounting for genotypic heteroresistance detection.
Results
Among 1238 included patients, Xpert MTB/XDR demonstrated high specificity (≥98%) across all drugs yet showed variable sensitivity, detecting 606 out of 637 isoniazid-resistant (95%, 95% CI 94–97%), 22 out of 33 fluoroquinolones-resistant (67%, 95% CI 48–81%) and 159 out of 279 ethionamide-resistant (57%, 95% CI 51–63%) samples. The assay reliably detected most common resistance-conferring mutations, such as katG_S315T, fabG1_C-15T, and gyrA_A90V and D94G, yet failed to detect low-frequency heteroresistance (≤10–35%) and off-target mutations, mostly for ethionamide. Amikacin resistance was rare (0.2%). Sensitivity for fluoroquinolones was higher (78%) among rifampicin-resistant samples, highlighting its utility as a reflex test in rifampicin-resistant patients.
Conclusions
Xpert MTB/XDR offers rapid diagnosis of resistance with high specificity. While limitations in detecting low-frequency and off-target variants affect its sensitivity, most frequent, fixed in-target mutations are readily detected. Future studies should evaluate strategies to integrate Xpert MTB/XDR with other diagnostic approaches in national tuberculosis programmes.
Shareable abstract
Xpert MTB/XDR rapidly detects drug resistance in samples from patients with confirmed tuberculosis, at high specificity. It may miss low-frequency mutants, yet targeted use for patients with rifampicin-resistant tuberculosis can improve treatment guidance. https://bit.ly/41rLI0F
Introduction
Treating drug resistant (DR) tuberculosis (TB) requires complex regimens combining multiple second-line drugs. Owing to their potent bactericidal properties, fluoroquinolones (FQ) are recommended for inclusion in regimens for DR-TB regimens, whether short (all-oral 9-month or all-oral 6-month) or longer treatment regimens, as well as in the 6-month regimen for rifampicin-susceptible (RS) yet isoniazid (INH)-resistant TB treatment [1]. Of note, the World Health Organization (WHO)-recommended BPaL regimen (bedaquiline (B), pretomanid (Pa), linezolid (L)) for rifampicin-resistant (RR)-TB, which does not include FQ, has a relapse rate of around 4%, after which few options remain [2]. Therefore, prompt and accurate drug-susceptibility testing (DST) for low-level and high-level FQ resistance before treatment initiation is crucial to select effective regimens [1]. Rapid testing for high-level INH resistance is likewise critical before initiating longer all-oral DR-TB regimens [1].
Few WHO-recommended rapid molecular diagnostic tests are available for detecting resistance to second-line drugs [3]. The GenoType MTBDRsl (Bruker, Germany) line probe assay requires molecular expertise and infrastructure, limiting its use to reference laboratories [4]. The WHO-endorsed Xpert MTB/XDR cartridge (Cepheid, USA) stands out as a low-complexity assay for detecting resistance to INH, FQ, amikacin (AMK) and ethionamide (ETH) in 90 min. It can be implemented in the widely rolled-out GeneXpert platform, offering a scalable solution, particularly in low-resource settings. The assay distinguishes between high and low resistance levels to INH and FQ. WHO recommends it as a reflex test directly in sputum specimens positive for Mycobacterium tuberculosis complex (MTBc) with any rapid molecular diagnostic test [3, 5]. However, this recommendation is conditional because only very low (ETH), low (AMK) and moderate (INH, FQ) certainty of evidence is available on its diagnostic accuracy, highlighting the need for further field-based data.
Here, we present the results of a large multicentre evaluation of Xpert MTB/XDR, conducted in programmatic conditions across nine sub-Saharan African countries in the DIAMA project (DIAgnostic for Multidrug resistant tuberculosis in Africa). Preliminary results on 522 patients contributed to the WHO endorsement of the assay [3]. In the current analysis, diagnostic accuracy was determined for 1238 patients against a composite reference standard based on the recently WHO-endorsed targeted next-generation sequencing assay Deeplex Myc-TB (GenoScreen, France) applied on the same sputum [3, 6], supplemented by available culture-based whole genome sequencing (WGS) and phenotypic DST (pDST) results for discrepancy analyses.
Methods
Study design and participants
This prospective, multicentre diagnostic accuracy study was conducted from June 2017 to March 2021 in Benin, Cameroon, the Democratic Republic of Congo, Ethiopia, Guinea, Mali, Nigeria, Rwanda and Senegal (ClinicalTrials.gov, NCT03303963).
Participants were recruited at the (sub)national level and samples were tested in the national reference laboratories at country level, before shipment of samples and isolates to (candidate) supranational reference laboratories (SRLs) in Cotonou (Benin), Kigali (Rwanda) or Antwerp (Belgium). Eligible patients were aged 15 years or older who had tested positive for TB by GeneXpert MTB/RIF or Ultra with or without RR detected. The study protocol (supplement 2, p12) was approved by relevant ethics committees prior to its start. All participants provided written inform consent before enrolment.
Xpert MTB/XDR
Sputum samples from patients enrolled before cartridge public release (July 2020) were preserved at −20°C. From 2020 onwards, testing was performed on fresh specimens across the participating laboratories, except in three sites due to shipment logistic challenges (supplement 2, p1–2). Samples from these sites were shipped in batches to the SRL Cotonou for Xpert MTB/XDR testing.
Deeplex Myc-TB
DNA extracted using the Maxwell 16 low elution volume DNA Purification Kits (Promega, USA) was tested as per manufacturer's recommendation and sequences analysed with the extended catalogue mode of the pipeline [6]. Detected resistance-associated sequence variants with frequencies between 1% and 80% were considered heteroresistant [6]. For samples with multiple variants co-detected in a gene target, the proportion of the most abundant variant was considered representative for that sample. Deeplex Myc-TB off-target mutations covered by Xpert MTB/XDR (solo inhA_G-154A or fabG1_Leu203Leu; n=2) were excluded for INH specificity calculation.
Reclassification of mutations
The WHO catalogue of MTBc mutations [7] was used to reclassify some mutations identified by Deeplex Myc-TB, with group 1 and 2 mutations considered as resistance variants.
pDST and WGS
WGS was performed on positive cultures, complemented by pDST on Löwenstein-Jensen or Middlebrook 7H11 agar at the specified critical concentrations [8]. Genomic DNA extracted at SRL Antwerp [9] was sent to MicrobeNG (Birmingham, UK) for sequencing with minimum 30× coverage. Variant calling was done using an in-house pipeline, combining MTBSeq and TB Profiler v4.4.2, with a 10% frequency threshold for minority variants [9].
Sample size, data management and statistical analyses
Based on published data [10], we assumed 1) INH, FQ and ETH resistance prevalence at 44%, 3% and 20%, respectively; 2) Xpert MTB/XDR expected sensitivity of 94% (INH), 90% (FQ) and 55% (ETH), with specificity of 98%, 95% and 95%, respectively; and 3) 10% missing data. A minimum of 1120 specimens was required to estimate accuracy with a precision of 2%, 11% and 8% for INH, FQ and ETH, respectively. Owing to its rarity, evaluating AMK resistance sensitivity was unachievable. Valid results (sensitive or resistant) for both index and reference tests were all included. Data were anonymised and collected using standardised forms on REDCap, hosted at WHO Geneva [11]. Concordance rates between Xpert MTB/XDR and Deeplex Myc-TB were estimated per drug and overall, with binomial exact 95% confidence intervals (CIs) and CIs adjusted for patient-level clustering. Associations were tested using the adjusted Wald test. Accuracy was calculated based on the final valid result, excluding repeats from invalid tests. Analyses were performed using R 4.4.1 (www.r-project.org) and Stata 18 (StataCorp, TX, USA).
Results
Of the 2269 participants tested positive by Xpert MTB/RIF (Ultra) enrolled across nine study sites, 1031 were excluded according to criteria described in figure 1 and in supplement 2: p2–3. Demographic characteristics of excluded patients were similar to those of included patients (supplement 2: tables 2 and 3). The median age of the 1238 included participants was 34 years, 25% were female and 20% were HIV co-infected (supplement 2: table 3). Most (71%) had a history of TB treatment and 63% were RR by Xpert MTB/RIF (Ultra). Susceptibility and resistance results by both tests were available for 1115 (INH), 1112 (FQ), 1146 (AMK) and 887 (ETH) participants (figure 2). The lower number for ETH principally reflects the fact that 402 samples were detected by Deeplex as carrying uncharacterised variants in ethA. Overall concordance between Xpert MTB/XDR and Deeplex Myc-TB was 93%, with similar rates for frozen and fresh samples. For concordant resistant calls in in-target regions, the same gene(s) were identified by both Xpert MTB/XDR and Deeplex in over 99% of cases (786 out of 788). Specificity was high for all the drugs evaluated but sensitivity varied from 95% (95% CI 94–97%) for INH to 67% (95% CI 48–81%) for FQ and 57% for ETH (95% CI 51–63%). Indeterminate Xpert MTB/XDR result rates were 2.8% for FQ, 1.1% for INH and 0.6% for AMK, and did not vary by HIV status or TB treatment history.
FIGURE 1.
Patient enrolment and exclusion criteria. AMK: amikacin; ETH: ethionamide; FQ: fluoroquinolones; INH: isoniazid; MTB: Mycobacterium tuberculosis; MTBc: M. tuberculosis complex; NTM: non-tuberculous mycobacteria. #: Deeplex Myc-TB tests not done owing to insufficient samples/storage issues; ¶: all 208 patients from site 7 excluded (81% concordance with the reference) and 22 patients from remaining sites excluded (15 with discordant lineages/single nucleotide polymorphism (Deeplex Myc-TB and whole genome sequencing) and seven with discordant rifampicin results (Deeplex Myc-TB and GeneXpert MTB/RIF or Ultra).
FIGURE 2.
Sensitivity and specificity of the Xpert MTB/XDR assay for detection of drug resistance compared to Deeplex Myc-TB. AMK: amikacin; ETH: ethionamide; FN: false negative; FP: false positive; FQ: fluoroquinolones; INH: isoniazid; NA: not available; TN: true negative; TP: true positive.
Among 606 INH-resistant (INH-R) samples carrying canonical resistance mutations detected by both tests, katG mutations were most frequent, occurring alone (n=423, 69.8%) and in combination with fabG1 (n=97, 16.0%), ahpC (n=17, 2.8%) and inhA (n=1, 0.2%). fabG1 mutations alone (n=62, 10.2%) or combined with inhA (n=6, 1.0%) were less frequent (table 1). Among the detected INH-R samples, 12 were heteroresistant (20–75% frequency, figure 3a and supplement 2: figures 2 and 3). Of 31 samples discordantly identified as INH-susceptible (INH-S) by Xpert MTB/XDR, 14 were due to off-target mutations (grey font, table 1). The remaining 17 out of 31 discordant Xpert MTB/XDR INH-S samples exhibited mutations within, or encompassing segments of, the katG gene or ahpC and fabG1 promoter regions covered by Xpert MTB/XDR. Eight were detected as minority single nucleotide polymorphisms (SNPs) at 3–25% within-sample frequencies by Deeplex Myc-TB (figure 3a and supplement 2: figures 2 and 3), indicating a limit of detection for Xpert MTB/XDR around this frequency for these mutations. Two other samples exhibited selective absence of coverage of the complete katG target by Deeplex Myc-TB, indicating a gene deletion beyond the region amplified by Xpert MTB/XDR. Finally, seven out of 31 INH-S discordant samples carried fixed S315T or S315I mutations (alone or combined with ahpC/fabG1 mutations), otherwise readily detected by Xpert MTB/XDR (see above). All seven samples also tested positive for RR by both Deeplex Myc-TB and Xpert MTB/RIF (with three also RR by available pDST/WGS results), supporting the likelihood of the reference results given the strong association of resistance between rifampicin and INH. Among 64 samples carrying solo fabG1 mutations, Xpert MTB/XDR detected variants in 62 (table 1), correctly classifying 60 as low-level INH–R (97%). In the two not classified as such, fabG1 mutations were fixed by Deeplex Myc-TB.
TABLE 1.
General overview of samples identified as isoniazid-resistant by Deeplex Myc-TB (n=637), stratified by the targeted genes/mutations, their classification according to the WHO catalogue v2 and Xpert MTB/XDR results
| Drug/gene/mutation | WHO Catalogue v2 classification | |||
|---|---|---|---|---|
| Marker of resistance | Uncertain significance | Total (n) | ||
| XDR-R (n) | XDR-S (n) | XDR-S (n) | ||
| ahpC alone | - | - | 3 | 3 |
| C-57T# | - | - | 1 | 1 |
| G-48A | - | - | 2§ | 2 |
| fabG1 alone | 62 | 2 | - | 64 |
| C-15T | 59 | 2§ | - | 61 |
| C-15T|T-8A | 1 | - | - | 1 |
| G-17T | 2 | - | - | 2 |
| fabG1/inhA combination | 6 | - | - | 6 |
| C-15T|T-8C/I94T | 1 | - | - | 1 |
| C-15T/S94A | 4 | - | - | 4 |
| T-8C/S94A | 1 | - | - | 1 |
| katG alone | 423 | 21 | 2 | 446 |
| W90R¶ | - | - | 1 § | 1 |
| N138S | - | 1 | - | 1 |
| W191R¶ | - | - | 1 § | 1 |
| G297V | - | 1 | - | 1 |
| W300G | - | 2 | - | 2 |
| S315I | 4 | 1 | - | 5 |
| S315N | 19 | 1§ | - | 20 |
| S315R | 5 | - | - | 5 |
| S315T | 394 | 4+2§ | - | 400 |
| S315T|A110V | 1 | - | - | 1 |
| FS | - | 5+2 § | - | 7 |
| GENE_DEL | - | 2 | - | 2 |
| katG/ahpC combination | 17 | 2 | - | 19 |
| FS/C-57T | 1 | - | - | 1 |
| FS/G-48A | 2 | - | - | 2 |
| GENE_DEL/G-48A|C-57T | 1 | - | - | 1 |
| S315T/G-48A | 13 | 1 | - | 14 |
| W300G/G-48A | - | 1§ | - | 1 |
| katG/fabG1 combination | 97 | 1 | - | 96 |
| N138S/C-15T | 1 | - | - | 1 |
| S315N|S315T/C-15T | 1 | - | - | 1 |
| S315T/C-15T | 41 | - | - | 41 |
| S315T/C-15T|T-8C | 1 | - | - | 1 |
| S315T/T-8A | 16 | - | - | 16 |
| S315T/T-8C | 36 | 1 | - | 35 |
| S315T/T-8C|T-8A | 1 | - | - | 1 |
| katG/inhA combination | 1 | - | - | 1 |
| S315T/S94A+ | 1 | - | - | 1 |
Grey/underlined font indicates mutations that fall outside the Xpert MTB/XDR target region(s) or the frameshift/STOP codon and do not impact the nucleotides within the Xpert MTB/XDR target regions. FS: frameshift; R: resistant (Xpert MTB/XDR); S: susceptible (Xpert MTB/XDR); WHO: World Health Organization. #: evidence for classifying resistant by Deeplex from [12, 13]; ¶: evidence for classifying resistant by Deeplex from [14]; +: mutations in inhA were interpreted by the Deeplex algorithm based on the WHO v1 catalogue, resulting in the classification of five extra samples with inhA mutation as of uncertain significance and therefore exclusion from the isoniazid-resistant samples; §: heteroresistance with <25% frequency of variant allele.
FIGURE 3.
a) Isoniazid (INH) resistance variant detection and variant frequencies by Deeplex Myc-TB for katG (n=446) versus detection results by Xpert MTB/XDR. b) Fluoroquinolone (FQ) resistance variant detection and variant frequencies by Deeplex Myc-TB for gyrA and gyrB (n=33) versus detection results by Xpert MTB/XDR. Each data point represents a sample, either found resistant or susceptible by Xpert MTB/XDR. The shape corresponds to the classification in the World Health Organization (WHO) Catalogue v2: dots indicate Category 1 and 2, marker of resistance; triangles indicate Category 3, uncertain significance; squares indicate mutations not classified. Black font indicates mutations within the Xpert MTB/XDR target region. Grey font indicates mutations located outside the target region.
Of 279 samples with ETH genotypic resistance by Deeplex Myc-TB, 159 (57.0%) were concordantly ETH-resistant (ETH-R) by Xpert MTB/XDR (table 2). Concordant samples had fabG1 mutations, alone (n=150) or combined with ethA (n=3) or inhA (n=6) (table 2). Among the 120 discordant ETH-susceptible (ETH-S) samples, 111 (92.5%) exhibited ETH-resistance mutations in the ethA gene and/or in an inhA gene region uncovered by Xpert MTB/XDR. The remaining nine discordant samples harboured in-target mutations in fabG1 (C-15T: n=5; T-8C: n=4). Three (two C-15T samples and one T-8C) had variant frequencies <10% (table 2 and supplement 2: figure 4), while among the remaining six (with unavailable pDST/WGS results), one had false susceptibility for INH, indicating consistent errors for both drugs.
TABLE 2.
General overview of samples identified as resistant to ethionamide by Deeplex Myc-TB (n=279), stratified by the targeted genes/mutations, their classification according to the WHO catalogue v2 and Xpert MTB/XDR results
| Drug/gene/mutation | WHO Catalogue v2 classification | Total (n) | ||
|---|---|---|---|---|
| Marker of resistance | Uncertain significance | |||
| XDR-R (n) | XDR-S(n) | XDR-S (n) | ||
| ethA or inhA | - | 111 | - | 111 |
| ethA alone | - | 103 | - | 103 |
| inhA alone# | - | 6 | - | 6 |
| ethA/inhA combination | - | 2 | - | 2 |
| fabG1 alone | 150 | 9 | - | 159 |
| C-15T | 98 | 3+2¶ | - | 103 |
| C-15T|T-8A | 1 | - | - | 1 |
| C-15T|T-8C | 1 | - | - | 1 |
| G-17T | 2 | - | - | 2 |
| T-8A | 16 | - | - | 16 |
| T-8C | 31 | 3+1¶ | - | 35 |
| T-8C|T-8A | 1 | - | - | 1 |
| fabG1/ethA combination | 3 | - | - | 3 |
| C-15T/R292STOP | 1 | - | - | 1 |
| T-8C/FS | 2 | - | - | 2 |
| fabG1/inhA combination | 6 | - | - | 6 |
| C-15T|T-8C/I194T | 1 | - | - | 1 |
| C-15T/S94A | 4 | - | - | 4 |
| T-8C/S94A | 1 | - | - | 1 |
| Total | 159 | 120 | 0 | 279 |
Grey/underlined font indicates mutations that fall outside the Xpert MTB/XDR target region(s) or the frameshift/STOP codon and do not impact the nucleotides within the Xpert MTB/XDR target regions. FS: frameshift; R: resistant (Xpert MTB/XDR); S: susceptible (Xpert MTB/XDR); WHO: World Health Organization. #: mutations in inhA were interpreted by the Deeplex algorithm based on the WHO v1 catalogue, resulting in the classification of five extra samples with inhA mutation as of uncertain significance and therefore exclusion from the isoniazid-resistant samples; ¶: heteroresistance with <25% frequency of variant allele.
For FQ, both tests concordantly identified 22 out of 33 (66.7%) samples as FQ-resistant (FQ-R), including 20 with a solo gyrA mutation and two with a gyrA/gyrB combination mutation (table 3). These included three samples with solo minority variants at 13–61% frequencies. Of the 11 samples discordantly identified as FQ-susceptible (FQ-S) by Xpert MTB/XDR, only one carried an off-target mutation, namely gyrB_E459K, unclassified in the WHO v2 mutation catalogue but categorised FQ-R by Deeplex Myc-TB (supplement 2: p7) [15, 16]. Nine other discordant Xpert MTB/XDR FQ-S samples harboured minority variants <25% in gyrA (n=4, 3.7–22.0%) and in gyrB (n=5, 1.0–4.5%), suggesting an overall limit of FQ heteroresistance detection of 15–25% (figure 3b). The last discordant sample (one out of 11) corresponded to a FQ-R gyrA_D94G mutation (94.0%, combined with A90V at 4.0%) and was confirmed as FQ-R by WGS and pDST. Of note, Xpert MTB/XDR correctly identified 21 out of 27 FQ-R samples (77.8%) in RR patients (supplement 2: figure 1). Among seven Xpert MTB/XDR FQ-R samples with gyrA_A90V (n=6, 61.5–99.9% frequency) or S91P (n=1, 100% frequency), six were concordantly classified as low-level FQ-R. The only low-level FQ-R case not classified as such had a 13.4% minority gyrA_A90V variant by Deeplex Myc-TB [17].
TABLE 3.
General overview of samples identified as resistant to fluoroquinolones by Deeplex Myc-TB (n=33), stratified by the targeted genes/mutations, their classification according to the WHO catalogue v2 and Xpert MTB/XDR results
| Drug/gene/mutation | WHO Catalogue v2 classification | Total (n) | ||
|---|---|---|---|---|
| Marker of resistance | Uncertain significance | |||
| XDR-R (n) | XDR-S (n) | XDR-S (n) | ||
| gyrA alone | 20 | 5 | - | 25 |
| A90V | 5 | 2# | - | 7 |
| A90V|D94A | 1 | - | - | 1 |
| A90V|D94G | 1 | 1 | - | 2 |
| A90V|S91P|D94A | 1 | - | - | 1 |
| D94G | 10 | 1# | - | 11 |
| D94H | - | 1# | - | 1 |
| G88C|D94G | 1 | - | - | 1 |
| S91P | 1 | - | - | 1 |
| gyrA/gyrB combination | 2 | - | - | 2 |
| A90V/A504T# | 1 | - | - | 1 |
| D94H/D461N | 1 | - | - | 1 |
| gyrB alone | - | 2 | 4 | 6 |
| E459K | - | - | 1 # | 1 |
| N499D | - | 2# | - | 2 |
| N499K | - | - | 1# | 1 |
| T500N | - | - | 1# | 1 |
| A504T | - | - | 1# | 1 |
| Total | 22 | 7 | 4 | 33 |
Grey font indicates mutations that fall outside the Xpert MTB/XDR target region(s) or the frameshift/STOP codon and do not impact the nucleotides within the Xpert MTB/XDR target regions. R: resistant (Xpert MTB/XDR); S: susceptible (Xpert MTB/XDR); WHO: World Health Organization. #: heteroresistance with <25% frequency of variant allele.
Only three AMK-resistant (AMK–R) samples were identified by Deeplex Myc-TB, indicating the very low prevalence of AMK resistance across our study settings, consistent with available surveillance data [18]. While they carried Xpert MTB/XDR in-target rrs_A1401G (n=1, frequency of 99.9%) or eis_C-14T (n=2, 2.6% and 1.3%) mutations, only the fixed rrs_A1401G was detected by Xpert MTB/XDR, as expected.
Finally, the specificity of Xpert MTB/XDR versus Deeplex Myc-TB ranged from 98% for INH to >99% for the three other drugs (figure 2). For 11 samples classified as INH-R by Xpert MTB/XDR, Deeplex Myc-TB classified them as susceptible, with pDST/WGS results available for four samples aligning with the Deeplex Myc-TB results. Only one out of 1146 for AMK, one out of 1112 for FQ and one out of 887 for ETH (also INH discordant) tested resistant by Xpert MTB/XDR but susceptible by Deeplex Myc-TB. None of these 13 discordantly resistant samples showed any mutation, whether resistance-associated (group 1/2) or not (group 4/5), in Xpert MTB/XDR-covered regions.
Discussion
Based on an exceptionally large dataset, our results show that Xpert MTB/XDR met minimum specificity criteria (≥98%), and was slightly (INH, 95%) or substantially (FQ, 67%; ETH, 57%) below minimum sensitivity criteria (≥98%) set by the WHO target product profile for next-generation DST, in comparison to genetic sequencing as a reference standard [19]. Two factors explain our findings: 1) Xpert MTB/XDR out-of-target resistance mutations, mostly for ETH, and 2) Xpert MTB/XDR in-target minority variants at 1–25% proportions, especially for FQ. Importantly, the reference was used on the same sputum, avoiding between-sample heterogeneity, potential culture biases and enabling thorough evaluation of heteroresistance detection through deep sequencing. Enrolment of patients and evaluation of the test under real-world, programmatic conditions across multiple sub-Saharan African countries supports the generalisability of the findings, at least in this large continental region.
The sensitivity of Xpert MTB/XDR for INH in our study is slightly above the pooled sensitivity of 94% relative to INH pDST among nine geographical settings [3, 5], initially considered by WHO for the assay's endorsement, while it is slightly inferior to the reported 96–98% range compared to culture-based WGS detection of INH resistance [10, 20, 21]. In contrast, the sensitivity of Xpert MTB/XDR for FQ and ETH reported in our study was significantly lower relative to pDST for FQ (93%) and sequencing of the inhA promoter (in isolates) for ETH (98%) in the WHO dataset [3]. Our values are also lower than the range of 90–97% sensitivity versus WGS for FQ and 62–98% for ETH reported in other studies [5, 10, 20, 21].
Predominantly missed ETH-resistance mutations in ethA correspond to 36.9% (103 out of 279) of the solo ETH-resistance mutations found in this dataset. More generally, ethA-resistance mutations represent an even larger fraction (59.4%, 1527 out of 2571) of the ETH-R phenotypes with solo resistance mutations in the global dataset of 52 000 genomes of the WHO catalogue, substantiating the large detection defect of ETH resistance by Xpert MTB/XDR, both in these African settings and globally. Of note, ethA mutations were not considered for determining the sensitivity for ETH in previous evaluations [10, 20]. Xpert MTB/XDR also missed the inhA_S94A mutation, which was much less common in this and the WHO datasets, as was inhA_I194T, categorised as uncertain significance in the WHO v2 catalogue. Finally, the test does not report ETH-R for fabG1 mutations (e.g. Leu203Leu), which confers cross-resistance to INH and ETH, while correctly reporting INH resistance, highlighting the need to update Xpert MTB/XDR's instructions for use in line with the WHO mutation catalogue. Although ETH is no longer included in the latest priority treatment regimens recommended by WHO for RR/MDR-TB [1], it may still be used for individualised treatments, e.g. in patients previously exposed to BPaL±moxifloxacin drugs [1]. Moreover, clinical trials are ongoing to evaluate ETH combined with the alpibectir booster, which allows the use of ETH at a lower, less toxic dose [22]. Although this combination can overcome resistance caused by ethA mutations, it fails to do so for ETH-resistance mutations in the inhA coding sequence, encountered as solo mutations or in combination in 14 patients with TB in our study.
The observed capture by Xpert MTB/XDR of unfixed FQ-resistance mutations with frequencies exceeding approximately 15% for gyrA_A90V and 35% for gyrA_D94G is consistent with the limit of detection found in experimental wild-type and mutant DNA mixtures [23]. Of note, tests such as GenoType MTBDRsl V2, still in use in high-burden countries, detected such variants at 5% in experimental mixtures, at least where this involved mutations targeted by a mutant probe [24]. Specimens with heteroresistant gyrA or gyrB variants below these frequencies, detected down to 1% by deep sequencing but generally missed by Xpert MTB/XDR, represented about a quarter (nine out of 33) of the total FQ-R samples in our cohort. Only two of them showed mixed phylogenetic SNP calls by Deeplex Myc-TB, indicating concurrent presence of FQ-S and FQ-R strains, suggesting de novo emergence for the others. Baseline genetic heteroresistance may be due to monotherapy for treating other presumed infections before TB diagnosis [25]. The global prevalence of heteroresistance to FQ has been estimated at 10% [26], further emphasising its importance. Especially in settings with an overall low prevalence of FQ resistance, as in this study, reduced sensitivity for low-frequency heteroresistance detection impacts the test's negative predictive value. However, the sensitivity for FQ was higher in the RR group (78%, supplement 2: p7) compared to the overall group, reflecting the expectedly higher proportion of fixed gyrA or gyrB mutations among RR specimens.
The 1% threshold for variant detection used by Deeplex Myc-TB aligns with the standard 1% critical proportion used in pDST to differentiate resistant and susceptible strains [8], with strong predictive value for clinical resistance and poor treatment outcome, at least for core drugs like FQ and rifampicin [27]. The extent to which these thresholds overlap for clinical interpretation remains to be fully established. However, a recent study in culture–specimen pairs provided evidence that, contrary to widely held assumptions, MTBc culture accurately mirrors the genetic diversity within diagnostic sputum, both qualitatively (SNP presence/absence) and quantitatively (variant frequency) [28]. Moreover, isolates showing the same gyrA resistance variants fixed or unfixed at various frequencies were shown to have similar minimum inhibitory concentration ranges [24], further supporting close alignment of phenotypic and genotypic heteroresistance thresholds.
Our study has some limitations. First, approximately half of the initially enrolled participants were excluded. However, the sociodemographic characteristics of the included patients remained comparable, minimising the risk of selection bias. Second, the small number of FQ-R samples, due to the low resistance prevalence settings and the notable impact of in-target low-frequency variants, resulted in wide confidence intervals and limited precision of sensitivity estimates. Also, the high proportion of smear-positive samples (90%, supplement 2: table 2) could have contributed to the observed performances, which might thus not be extrapolatable to patient populations with lower bacterial loads. ETH resistance was likely underestimated, given that neither index nor reference tests include fabG1_Leu203Leu mutations in their resistance call. Finally, although a pDST/WGS composite reference was used to resolve discrepancies, this is not currently WHO-recommended for the drugs assessed [7, 29].
Xpert MTB/XDR represents a pragmatic compromise, delivering rapid and reliable resistance results, although (hetero-)resistance may be missed. For treatment regimens in which excluding FQ resistance is required or desirable, additional tests for excluding heteroresistance may be warranted. Although the assay has limitations in detecting low-frequency variants, it specifically targets the mutations most frequently associated with resistance to INH and FQ, key anti-TB drugs. This makes it a valuable diagnostic tool, particularly in low-resource settings, and especially as a reflex test in RR cases, in which fixed (FQ-)resistance mutations are more likely to be detected.
Acknowledgments
We thank all National Tuberculosis Control Programs and tuberculosis patients who participated in this project. Artificial intelligence tools were used to improve language, correct grammar and enhance clarity. The generated content was further reviewed and approved by the authors to ensure scientific accuracy and integrity.
Footnotes
Provenance: Submitted article, peer reviewed.
Disclaimer: The views and opinions expressed by the authors in this publication do not necessarily reflect those of European and Developing Countries Clinical Trials Partnership. C.S. Merle, a staff member of the World Health Organization (WHO), is solely responsible for the views presented, which do not necessarily represent WHO's decisions, policies or positions.
Ethics statement: The protocol (supplement 2: p12) of this multicentre study was approved by all relevant ethics committees prior to its start. All participants provided written informed consent before enrolment. All data were anonymised before analysis to ensure confidentiality. This study complies with the ethical principles of the Declaration of Helsinki and the relevant national regulations on research ethics.
Conflict of interest: A. Ferré and N. Badalato are employees of GenoScreen, which developed the targeted next-generation sequencing assay used as a reference standard. A. Penn-Nicholson is employee of the Foundation for Innovative New Diagnostics (FIND), which provided the Xpert MTB/XDR cartridges free of charge, and S.G. Schumacher is a former employee of FIND. P. Supply is a scientific adviser for GenoScreen. The other authors declare no conflicts of interest.
Support statement: We acknowledge our funders: the European and Developing Countries Clinical Trials Partnership (EDCTP2, DRIA2014-326 – DIAMA RIA Diagnostics) supported by the European Union, and Cepheid for providing the Xpert MTB/XDR cartridges via the Foundation for Innovative Diagnostics. Funding information for this article has been deposited with the Open Funder Registry.
Supplementary material
Please note: supplementary material is not edited by the Editorial Office, and is uploaded as it has been supplied by the author.
Plain language summary
00427-2025.SUPPLEMENT
Supplementary material
00427-2025.SUPPLEMENT2
Data availability
Individual, de-identified data including data dictionary may be shared upon reasonable request. Deeplex Myc-TB FASTQ files have been deposited in the NCBI Sequence Read Archive under bioproject PRJNA1309160, and WGS FASTQ files have been deposited in the European Nucleotide Archive under project PRJEB95959.
References
- 1.World Health Organization (WHO) . WHO Consolidated Guidelines on Tuberculosis. Module 4: Treatment and Care. Geneva, WHO, 2025. [PubMed] [Google Scholar]
- 2.Nyang'wa BT, Berry C, Kazounis E, et al. Short oral regimens for pulmonary rifampicin-resistant tuberculosis (TB-PRACTECAL): an open-label, randomised, controlled, phase 2B-3, multi-arm, multicentre, non-inferiority trial. Lancet Respir Med 2024; 12: 117–128. doi: 10.1016/S2213-2600(23)00389-2 [DOI] [PubMed] [Google Scholar]
- 3.World Health Organization (WHO) . Module 3: Diagnosis WHO Consolidated Guidelines on Tuberculosis. Geneva, WHO, 2025. [PubMed] [Google Scholar]
- 4.Gilpin C, Korobitsyn A, Weyer K. Current tools available for the diagnosis of drug-resistant tuberculosis. Ther Adv Infect Dis 2016; 3: 145–151. doi: 10.1177/2049936116673553 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Pillay S, Steingart KR, Davies GR, et al. Xpert MTB/XDR for detection of pulmonary tuberculosis and resistance to isoniazid, fluoroquinolones, ethionamide, and amikacin. Cochrane Database Syst Rev 2022; 5: CD014841. doi: 10.1002/14651858.CD014841.pub2 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Jouet A, Gaudin C, Badalato N, et al. Deep amplicon sequencing for culture-free prediction of susceptibility or resistance to 13 anti-tuberculous drugs. Eur Resp J 2021; 57: 2002338. doi: 10.1183/13993003.02338-2020 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.World Health Organization (WHO) . Catalogue of Mutations in Mycobacterium tuberculosis Complex and Their Association With Drug Resistance, 2nd ed. Geneva, WHO, 2023. [Google Scholar]
- 8.World Health Organization (WHO) . Technical Report on Critical Concentrations for Drug Susceptibility Testing of Medicines Used in the Treatment of Drug-Resistant Tuberculosis. Geneva, WHO, 2018. [Google Scholar]
- 9.Lempens P, Meehan CJ, Vandelannoote K, et al. Isoniazid resistance levels of Mycobacterium tuberculosis can largely be predicted by high-confidence resistance-conferring mutations. Sci Rep 2018; 8: 3246. doi: 10.1038/s41598-018-21378-x [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Penn-Nicholson A, Georghiou SB, Ciobanu N, et al. Detection of isoniazid, fluoroquinolone, ethionamide, amikacin, kanamycin, and capreomycin resistance by the Xpert MTB/XDR assay: a cross-sectional multicentre diagnostic accuracy study. Lancet Infect Dis 2022; 22: 242–249. doi: 10.1016/S1473-3099(21)00452-7 [DOI] [PubMed] [Google Scholar]
- 11.Harris PA, Taylor R, Thielke R, et al. Research electronic data capture (REDCap): a metadata-driven methodology and workflow process for providing translational research informatics support. J Biomed Inform 2009; 42: 377–381. doi: 10.1016/j.jbi.2008.08.010 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Sherman DR, Mdluli K, Hickey MJ, et al. Compensatory ahpC gene expression in isoniazid-resistant mycobacterium tuberculosis. Science 1996; 272: 1641–1643. doi: 10.1126/science.272.5268.1641 [DOI] [PubMed] [Google Scholar]
- 13.Sherman DR, Mdluli K, Hickey MJ, et al. AhpC, oxidative stress and drug resistance in Mycobacterium tuberculosis. BioFactors 1999; 10: 211–217. doi: 10.1002/biof.5520100219 [DOI] [PubMed] [Google Scholar]
- 14.Napier G, Campino S, Phelan JE, et al. Large-scale genomic analysis of Mycobacterium tuberculosis reveals extent of target and compensatory mutations linked to multi-drug-resistant tuberculosis. Sci Rep 2023; 13: 623. doi: 10.1038/s41598-023-27516-4 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Miotto P, Cirillo DM, Migliori GB. Drug resistance in Mycobacterium tuberculosis: molecular mechanisms challenging fluoroquinolones and pyrazinamide effectiveness. Chest 2015; 147: 1135–1143. doi: 10.1378/chest.14-1286 [DOI] [PubMed] [Google Scholar]
- 16.Pantel A, Petrella S, Veziris N, et al. Extending the definition of the GyrB quinolone resistance-determining region in Mycobacterium tuberculosis DNA gyrase for assessing fluoroquinolone resistance in M. tuberculosis. Antimicrob Agents Chemother 2012; 56: 1990–1996. doi: 10.1128/AAC.06272-11 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Cao Y, Parmar H, Gaur LR, et al. Xpert MTB/XDR: a 10-color reflex assay suitable for point-of-care settings to detect isoniazid, fluoroquinolone, and second-line injectable-drug resistance directly from Mycobacterium tuberculosis-positive sputum. J Clin Microbiol 2021; 59: 1–18. doi:0.1128/JCM.02314-20 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Osei Sekyere J, Reta MA, Maningi NE, et al. Antibiotic resistance of Mycobacterium tuberculosis complex in Africa: a systematic review of current reports of molecular epidemiology, mechanisms and diagnostics. J Infect 2019; 79: 550–571. doi: 10.1016/j.jinf.2019.10.006 [DOI] [PubMed] [Google Scholar]
- 19.World Health Organization (WHO) . Target Product Profile for Next-Generation Drug-susceptibility Testing at Peripheral Centres. Geneva, WHO, 2021. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Omar SV, Louw G, Ismail F, et al. Performance evaluation of the Xpert MTB/XDR test for the detection of drug resistance to Mycobacterium tuberculosis among people diagnosed with tuberculosis in South Africa. J Clin Microbiol 2024; 62: e0022924. doi: 10.1128/jcm.00229-24 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Chen X, Li R, Ge S, et al. Rapid detection of extensive drug resistance by Xpert MTB/XDR optimizes therapeutic decision-making in rifampin-resistant tuberculosis patients. J Clin Microbiol 2023; 61: e0183222. doi: 10.1128/jcm.01832-22 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Gries R, Chhen J, van Gumpel E, et al. Discovery of dual-active ethionamide boosters inhibiting the Mycobacterium tuberculosis ESX-1 secretion system. Cell Chem Biol 2024; 31: 699–711. doi: 10.1016/j.chembiol.2023.12.007 [DOI] [PubMed] [Google Scholar]
- 23.Dippenaar A, Diels M, Keysers J, et al. Accuracy in detecting fluoroquinolone-heteroresistant M. tuberculosis mixtures by Xpert MTB/XDR. Int J Tuberc Lung Dis 2024; 28: S1–S814. [Google Scholar]
- 24.Rigouts L, Miotto P, Schats M, et al. Fluoroquinolone heteroresistance in Mycobacterium tuberculosis: detection by genotypic and phenotypic assays in experimentally mixed populations. Sci Rep 2019; 9: 11760. doi: 10.1038/s41598-019-48289-9 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25.Zhang D, Gomez JE, Chien JY, et al. Genomic analysis of the evolution of fluoroquinolone resistance in Mycobacterium tuberculosis prior to tuberculosis diagnosis. Antimicrob Agents Chemother 2016; 60: 6600–6608. doi: 10.1128/AAC.00664-16 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26.Ye M, Yuan W, Molaeipour L, et al. Antibiotic heteroresistance in Mycobacterium tuberculosis isolates: a systematic review and meta-analysis. Ann Clin Microbiol Antimicrob 2021; 20: 73. doi: 10.1186/s12941-021-00478-z [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27.Van Deun A, Decroo T, Piubello A, et al. Principles for constructing a tuberculosis treatment regimen: the role and definition of core and companion drugs. Int J Tuberc Lung Dis 2018; 22: 239–245. doi: 10.5588/ijtld.17.0660 [DOI] [PubMed] [Google Scholar]
- 28.Mariner-Llicer C, Goig GA, Torres-Puente M, et al. Genetic diversity within diagnostic sputum samples is mirrored in the culture of Mycobacterium tuberculosis across different settings. Nat Commun 2024; 15: 7114. doi: 10.1038/s41467-024-51266-0 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29.Köser CU, Robledo J, Shubladze N, et al. Guidance is needed to mitigate the consequences of analytic errors during antimicrobial susceptibility testing for TB. Int J Tuberc Lung Dis 2021; 10: 791–794. doi: 10.5588/ijtld.21.0428 [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Please note: supplementary material is not edited by the Editorial Office, and is uploaded as it has been supplied by the author.
Plain language summary
00427-2025.SUPPLEMENT
Supplementary material
00427-2025.SUPPLEMENT2
Data Availability Statement
Individual, de-identified data including data dictionary may be shared upon reasonable request. Deeplex Myc-TB FASTQ files have been deposited in the NCBI Sequence Read Archive under bioproject PRJNA1309160, and WGS FASTQ files have been deposited in the European Nucleotide Archive under project PRJEB95959.