Effectiveness of GenoType MTBDRsl in excluding TB drug resistance in a clinical trial

Summary

Objectives

To assess the performance of the Geno-Type MTBDRsl v1, a line-probe assay (LPA), to exclude baseline resistance to fluoroquinolones (FQs) and second-line injectables (SLIs) in the Standard Treatment Regimen of Anti-tuberculosis Drugs for Patients With MDR-TB 1 (STREAM 1) trial.

Methods

Direct sputum MTBDRsl results in the site laboratories were compared to indirect phenotypic drug susceptibility testing (pDST) results in the central laboratory, with DNA sequencing as a reference standard.

Results

Of 413 multidrug-resistant TB (MDR-TB) patients tested using MTBDRsl and pDST, 389 (94.2%) were FQ-susceptible and 7 (1.7%) FQ-resistant, while 17 (4.1%) had an inconclusive MTBDRsl result. For SLI, 372 (90.1%) were susceptible, 5 (1.2%) resistant and 36 (8.7%) inconclusive. There were 9 (2.3%) FQ discordant pDST/MTBDRsl results, of which 3 revealed a mutation and 5 (1.3%) SLI discordant pDST/ MTBDRsl results, none of which were mutants on sequencing. Among the 17 FQ- and SLI MTBDRsl-inconclusive samples, sequencing showed 1 FQ- and zero SLI-resistant results, similar to frequencies among the conclusive MTBDRsl. The majority of inconclusive MTBDRsl results were associated with low bacillary load samples (acid-fast bacilli smear-negative or scantily positive) compared to conclusive results (P < 0.001).

Conclusion

MTBDRsl can facilitate the rapid exclusion of FQ and SLI resistances for enrolment in clinical trials.

Despite the availability of curative anti-TB therapy for nearly half a century, the emergence and spread of MDR-TB strains is a major public health concern and threatens global control of the disease.^1–4 In 2019, according to the WHO,⁵ an estimated 465,000 people developed rifampicin-resistant TB (RR-TB).

The evaluation of new drugs and new regimens requires randomised clinical trials.^6–8 Standard Treatment Regimen of Anti-tuberculosis Drugs for Patients With MDR-TB 1 (STREAM 1) was a Phase III, multicentre, open-label, randomised controlled trial that enrolled patients between 2012 to 2015 to evaluate the safety, efficacy and cost-effectiveness of a standardised shorter regimen for MDR-TB, similar to a regimen described by Van Deun et al. in 2010,⁷ compared to a long regimen recommended in 2011 by the WHO.⁹ Enrolment criteria and trial procedures have been previously reported elsewhere (STREAM 1, ISRCTN78372190 and clinicaltrials.gov NCT02409290).^8,10 The trial enrolled adults with RR/MDR-TB and no evidence of resistance to fluoroquinolones (FQs) or second-line injectables (SLIs) on line-probe assay (LPA) (GenoType MTBDRsl, Hain Lifesciences, Nehren, Germany) in the 4 weeks prior to randomisation.^8,11

Second-line drug LPAs are the fastest and most commonly used genotypic drug susceptibility testing (gDST) method available, although these still require an adequate level of molecular technical expertise.^12,13 The MTBDRsl LPA v1 detects the most frequent mutations that confer resistance to FQs in the gyrA and gyrB genes, and to SLIs in the 16S rRNA (rrs) and eis promoter genes.¹⁴ A systematic review from the Cochrane Database showed that MTBDRsl had sensitivities and specificities of 85.1% and 98.2% and 76.9% and 99.5% for the detection of FQ and SLI resistance from clinical samples, respectively.¹⁴ At the time STREAM 1 was planned, it was not clear whether this would lead to sufficiently high negative predictive values (NPVs) to reliably exclude FQ- and/or SLI-resistant cases from enrolment.¹⁵ Moreover, indeterminate results have been reported in 7.1% of FQ bands and 13.5% of SLI on MTBDRsl.^15,16

Using data and Mycobacterium tuberculosis isolates from the STREAM 1 trial, we assess MTBDRsl effectiveness and investigate the most appropriate interpretation of inconclusive MTBDRsl results directly from sputum, including their association with sputum bacillary load.

Materials And Methods

Study design and population

This study was a retrospective comparison of MTBDRsl results obtained from baseline clinical samples (defined as participant’s sputum specimens collected before initiation of MDR-TB treatment) from seven STREAM 1 sites located in Ethiopia, Mongolia, South Africa and Vietnam from patients randomised between July 2012 and June 2015 against phenotypic DST (pDST) results on baseline isolates.^8,11

Microscopy

Sputum smear microscopy was conducted at each STREAM trial site following WHO standard protocol.¹⁷ The site TB laboratories currently participated in external quality assessment (EQA) programmes ensuring the quality of acid-fast bacilli (AFB) smear microscopy results. Additionally, the reference laboratory at the Institute of Tropical Medicine (ITM), Antwerp, Belgium, warranted the quality of the site results through periodical monitoring visits.

Drug susceptibility testing

Phenotypic drug susceptibility testing (pDST) of M. tuberculosis isolates received from three countries (Ethiopia, South Africa and Vietnam) was performed at ITM using the proportion method on Middlebrook 7H11 agar at the WHO-recommended critical concentrations,¹⁸ whereas pDST of M. tuberculosis isolates from the Mongolian site was conducted at the National Tuberculosis Reference Laboratory (NTRL), Ulaanbaatar, Mongolia, following the standard WHO protocol, using the indirect proportion method on Löwenstein-Jensen for first and second-line drugs.¹⁸

GenoType MTBDRsl line-probe assay

MTBDRsl v1.0 testing was performed and interpreted according to the manufacturer instructions for clinical samples.¹⁹ Results were deemed inconclusive based on the presence of overall weak bands or absence of wild-type (WT) and mutated (MUT) bands, together with the absence of the locus control bands for one or more of the genes tested. The DNA extracted from the reference M. tuberculosis H37Rv strain and molecular-grade water were used as positive and negative controls for each run.

DNA sequencing

To resolve discordant results of FQ MTBDRsl and pDST, extracted DNA was amplified using gyrA/B specific primers and sequenced using the Sanger method, as previously described.²⁰ Sequences were compared to that of the H37Rv reference strain (National Center for Biotechnology Information, GenBank accession number NC_000962) using CLC Sequence Viewer software (Qiagen, Hilden, Germany). For SLIs, next-generation sequencing (NGS) was used to resolve discordance between pDST and MTBDRsl results. Illumina (Illumina, San Diego, CA, USA) NGS short sequencing was performed using the TGen/C-Path platform, in which the whole genome is analysed for variants that are known to SLI resistance-conferring genes.²¹ TB-profiler v2.6.0 (London School of Hygiene & Tropical Medicine, London, UK) was used to analyse raw fastq data.

Statistical analysis

Statistical analysis was performed using Stata v15.0 (Stata, College Station, TX, USA). MTBDRsl results discordant with those of pDST were resolved by sequencing as the reference standard. As evidence of FQ or SLI resistance excluded patients from the trial, only a small fraction of the strains obtained at the screening visit that had MTBDRsl-resistant results also had pDST done at the reference laboratories; therefore, only the predictive value of a susceptible, but not that of a resistant result in the population screened could be determined. Logistic regression was used for the analysis of the association between inconclusive results and a low bacillary load. The χ² test was used to compare proportions.

Ethical statement

This study used microbiological data and isolates from the STREAM 1 clinical trial. The isolates tested to resolve discordant MTBDRsl/pDST results were identified using the central laboratory accession number, without any patient identifiers. The Institutional Review Board of ITM, Antwerp, Belgium, was notified of these analyses. Full protocol review was not requested, in line with the low-risk nature of microbiological analyses.

Results

A total of 689 patients were screened at the seven sites, of whom 579 patients with both MTBDRsl and smear microscopy results were used to investigate the association between inconclusive MTBDRsl results and bacillary load. Only 413 patients with both MTBDRsl and pDST results available were also included for the assessment of MTBDRsl performance (Figure).

Figure.

Flow diagram displaying patients and samples included in the study. MDR-TB = multidrug-resistant TB; FQ-S = fluoroquinolone-susceptible; FQ-R = FQ-resistant; SLI-S = second-line injectable-susceptible; SLI-R = SLI-resistant; pDST = phenotypic drug susceptibility testing.

MTBDRsI line-probe assay results

Of 413 patients, 396 (95.9%) patients were FQ-conclusive on MTBDRsl, and 17 (4.1%) had inconclusive results (Table 1). MTBDRsl also identified as 377 (91.3%) SLI-conclusive, and 36 (8.7%) with inconclusive results (Table 1). Among all 579 patients, 503 (86.9%) were reported as FQ-susceptible (FQ-S), 26 (4.5%) as FQ-resistant (FQ-R) and 50 (8.6%) as having inconclusive results; 478 (82.6%) were SLI-susceptible (SLI-S), 25 (4.3%) SLI-resistant (SLI-R) and 76 (13.1%) patients had inconclusive results. Inconclusive MTBDRsl results were significantly more common in South Africa than at other trial sites (P < 0.001 and P < 0.001, respectively) (Supplementary Table S1).

Table 1. Cross-tabulation of MTBDRsl vs. pDST results.

FQ MTBDRsl	pDST			SLI MTBDRsl	pDST
	FQ-S (n = 403) n (%)	FQ-R (n = 10) n (%)	Total (n = 413) n		SLI-S (n = 406) n (%)	SLI-R (n = 7) n (%)	Total (n = 413) n
Conclusive	388 (98.0)	8 (2.0)	396	Conclusive	373 (98.9)	4 (1.1)	377
Susceptible	384 (98.7)	5 (1.3)	389	Susceptible	370 (99.5)	2 (0.5)	372
Resistant	4 (57.1)	3 (42.9)	7	Resistant	3 (60.0)	2 (40.0)	5
Inconclusive	15 (88.2)	2 (11.8)	17	Inconclusive	33 (91.7)	3 (8.3)	36

pDST = phenotypic drug susceptibility testing; FQ-S = fluoroquinolone-susceptible; FQ-R = FQ-resistant; SLI-S = second-line injectable-susceptible; SLI-R = SLI-resistant.

pDST vs. MTBDRsl results and sequencing

Of the total 413 pDST results (Figure and Table 1), respectively 403 (97.6%) and 406 (98.3%) M. tuberculosis isolates were FQ-S and SLI-S. Among 396 patients with conclusive MTBDRsl and pDST results for FQ, 9 patients showed discordant results: 5 were LPA-susceptible (LPA-S) but pDST-resistant (pDST-R), while 4 were LPA-resistant (LPA-R) but pDST-susceptible (pDST-S) (Table 1). For 377 patients with conclusive results for SLI, 5 had discordant results: 2 were LPA-S/pDST-R and 3 were LPA-R/pDST-S (Table 1). gyrA/B sequencing of the discordant results showed WT gyrA/B for all 4 LPA-R, whereas 2 WTs and 3 resistance-conferring mutations (2 Ala90Val, 1 Asp94Gly) were observed for the LPA-S. Together with the three concordant resistant cases, this brings the total FQ-R results to 6 (1.5%) (Supplementary Table S2). At this very low prevalence, the NPV of MTBDRsl for FQ resistance was very high, 99.2% (95% confidence interval [CI] 0.98–1.00; Table 2). In case of strains with SLI discordance between MTBDRsl and pDST, sequencing did not show SLI-R-conferring mutations. Hence, only two strains were finally classified as SLI-R (0.5%; Supplementary Table S2). The NPV for the exclusion of SLI-R using MTBDRsl was thus 100% (95% CI 0.99–1.00; Table 2). As patients identified as FQ-R or SLI-R on MTBDRsl screening were excluded from the trial, no pDST is available from strains for most of these patients. We were therefore not able to evaluate false resistance in the screened population. Sensitivity, specificity and positive predictive value (PPV) of the FQ-R or SLI-R results could thus not be determined in this study.

Table 2. MTBDRsl effectiveness in excluding resistance.

MTBDRsl conclusive results		DNA sequencing*			Negative predictive value % (95% CI)
		Resistant n	Susceptible n	Total n
Fluoroquinolone	Resistant	3	4	7
	Susceptible	3	386	389	99.2 (0.98–1.00)
Second-line injectable	Resistant	2	3	5
	Susceptible	0	372	372	100 (0.99–1.00)

^*DNA sequencing used as a reference standard to resolve discordance between pDST and MTBDRsl results. CI = confidence interval; pDST = phenotypic drug susceptibility testing.

Resistance missed by inconclusive MTBDRsl LPA results

pDST identified 15 FQ-S and 2 FQ-R samples among the 17 with FQ-inconclusive MTBDRsl results (Table 1). Sequence analysis could confirm only one gyrA Ala90Val resistance-conferring mutation among the two phenotypic FQ-R, all other samples being WT (Supplement Table 2). Thus, among the 17 inconclusive MTBDRsl results, one FQ-R sample had not been identified by LPA. Given the 26 (4.9%) FQ-R cases identified at the initial screening of 529 patients with conclusive results, the 1/17 (5.9%) confirmed by the reference standard from FQ-inconclusive MTBDRsl results is very similar.

pDST identified 3 (8.3%) SLI-R patients among 36 with SLI-inconclusive MTBDRsl results (Table 1). Only half (17) of these — and among them, only one pDST SLI-R — were available for sequencing. No SLI resistance-conferring mutations were detected for these 17 (Supplementary Table S2). Due to insufficient confirmatory testing, it was not possible to compare the proportions of resistance among the inconclusive vs. conclusive SLI MTBDRsl results.

Association of inconclusive MTBDRsl LPA results with bacillary load

We also analysed the association between the 579 MTBDRsl results and sputum bacillary load as determined using smear microscopy (Table 3). Among 140 (24.2%) isolates with low bacillary loads, 94 (16.3%) were smear-negative and 46 (7.9%) had scanty-positive smear microscopy results, while the 439 (75.8%) with high bacillary loads comprised 146 (25.2%) 1+, 118 (20.4%) 2+ and 175 (30.2%) 3+ smear-positive results. A higher proportion of patients with inconclusive FQ MTBDRsl and SLI MTBDRsl results had lower bacillary loads than those with conclusive results (P < 0.001; Table 3).

Table 3. Sputum bacillary load stratified by FQ or SLI MTBDRsl result.

MTBDRsl	Total n	Direct smear microscopy		P value‡	(95% CI)
		Low bacillary load* n (%)	High bacillary load† n (%)
FQ
Inconclusive	50	37 (74.0)	13 (26.0)	<0.001§	(0.04–0.16)§
Conclusive	529	103 (19.5)	426 (80.5)
Total	579	140 (24.2)	439 (75.8)
SLI
Inconclusive	76	46 (60.5)	30 (39.5)	<0.001§	(0.09–0.25)§
Conclusive	503	94 (18.7)	409 (81.3)
Total	579	140 (24.2)	439 (75.8)

^*Low bacillary load included negative and scanty smear microscopy results.

^†High bacillary load included 1+, 2+, and 3+ smear microscopy results.

^‡Univariable logistic regression analysis (MTBDRsl result vs. smear microscopy result).

^§Statistically significant.

FQ = fluoroquinolone; SLI = second-line injectable; CI = confidence interval.

Discussion

Rapid and accurate diagnostic tools to exclude pre-extensively drug-resistant TB (pre-XDR-TB) and XDR-TB patients are essential to decide on the eligibility of patients for enrolment in some MDR-TB clinical trials, and also for timely management of MDR-TB. In this study, we evaluate the performance of MTBDRsl in such a context, the STREAM 1 trial. The agreement between MTBDRsl and pDST results in detecting true susceptibility was excellent for both FQ and SLI. Although DNA sequencing showed that most of the discordant results for MTBDRsl FQ and/ or SLI resistance were false, MTBDRsl nevertheless performed very well as a clinical trial screening tool. Similar to results achieved in other settings,^14,22 MTBDRsl performance in the STREAM trial in excluding resistance was very good, with a NPV of over 99% for both FQ and SLI. In addition to minimising testing delay, MTBDRsl were almost 100% effective in identifying FQ-R and/or SLI-R TB patients in trial settings with a low prevalence of resistance, and thus a low pre-test probability such as Ethiopia and Vietnam (Supplementary Table S1). Others have also evaluated MTBDRsl v2.0 and have reported the specificity of the MTBDRsl to be close to 99%.²³ Therefore, provided that DNA sequencing is used to resolve discordances with pDST, the NPV should still be sufficiently good for a trial setting in high-prevalence settings; only five FQ-R cases would be missed when screening 1,000 patients at 50% prevalence. Although the proportion of patients with SLI resistance missed among the inconclusive results could not be determined in this study, these test failures may constitute a serious problem where SLI resistance prevalence is high, since these were not rare, and not as strongly associated with a low bacillary load.

In addition to overall prevalence, the distribution of specific FQ-R mutations might vary geographically,^24–26 partly explaining variations in MTBDRsl performance. In the study by Pantel et al.,²⁷ compensatory mutations that restore FQ susceptibility in M. tuberculosis strains have also been described during MDR-TB screening.

The assay’s good performance at all STREAM 1 sites in exclusing resistance to FQ and SLI is similar to that reported in a feasibility study describing the molecular assay to screen patients in TB clinical trials,²⁸ except for some false MTBDRsl results, particularly in South Africa. The WHO Expert Group also noted that given the high assay NPV for detecting resistance to FQ and SLI, MTBDRsl results could be used for screening, pending pDST results,²² by avoiding placing patients who have resistance to FQ and SLI on the regimen and start eligible patients on treatment.^11,29

Our analysis showed an overall rate of inconclusive LPA results of 8.6% for FQ and 13.1% for SLI, which is within the reported range from other studies.^16,30 Although not fully understood, poor test execution or suboptimal amounts of DNA in low bacillary burden samples are obvious possible reasons for this.¹⁶ However, in principle, inconclusive results could also be associated with mutations or deletions in the locus control region, as well as the complete or partial deletion of a target gene.^30,31 A significantly higher proportion of inconclusive MTBDRsl results was seen in patients from South Africa, where the majority had no or only scanty bacilli on sputum microscopy, possibly due to a higher proportion of enrolled patients with HIV co-infection. Previous studies have also documented more inconclusive results in smear-negative patients, those with lower bacillary load specimens^32,33 and in the HIV-positive.^13,32,34 The WHO recommends that the direct use of sputum for the MTBDRsl test is not suitable for smear-negative clinical specimens.^13,34 Ongoing DR-TB trials increasingly enrol larger proportions of patients with low bacillary burdens, e.g., those who have a low Xpert score and are AFB-negative. As shown in our study, such samples are more likely to yield inconclusive LPA results, lowering LPA utility for the rapid exclusion of baseline resistance.

In our study, the proportion of FQ-R organisms among the inconclusive MTBDRsl results did not appear to differ from that observed among conclusive results, but we could not arrive at any conclusion in this regard for SLI due to insufficient confirmatory testing. At least for FQ, enrolment of patients with an inconclusive MTBDRsl result can be fully justified at low FQ-resistance prevalence, and it would seem reasonable to believe this to be true for SLI.

This study had some limitations. As we could not review the strips for each MTBDRsl performed at the sites, we were not able to investigate banding profiles suggestive of heteroresistance or reading errors at these laboratories as possible causes of the false results observed. Exclusion of FQ-R and SLI-R MTBDRsl samples from the trial further limited our ability to estimate the level of false resistance in the screened population, which made it impossible to determine the sensitivity and PPV of resistant results. Furthermore, we could only apply the reference standard—sequencing—to half the samples with inconclusive results for SLI, leaving some uncertainty regarding their importance and the most suitable interpretation for clinical trial patient screening.

Conclusion

In this study, MTBDRsl proved a good screening tool for the rapid exclusion of FQ and SLI resistance using a composite reference standard of pDST plus DNA sequencing to resolve discordant results. Although the prevalence of both FQ and SLI resistance was low to moderate in our study populations, the high specificity suggests that MTBDRsl can facilitate screening for FQ and SLI resistance in high-prevalence settings as well. Relatively frequent inconclusive results, particularly for SLI, may constitute a larger problem, especially when enrolling patients with AFB smear-negative disease, such as those diagnosed on Xpert. Inconclusive results may conceal a proportion of resistance proportional to the prevalence of FQ or SLI resistance in the test setting.