Evaluation of the broth microdilution plate methodology for susceptibility testing of Mycobacterium tuberculosis in Peru

The study was carried out by the National Reference Laboratory for Mycobacteria (LRNM [Laboratorio de Referencia Nacional de Micobacterias]) of the National Institute of Health (INS [Instituto Nacional de Salud]) in collaboration with investigators at Universidad Peruana Cayetano Heredia (UPCH) and the London School of Hygiene and Tropical Medicine (LSHTM). Drug-resistant MTB strains (according to routine APM results), representative from all over Peru were selected. Strains were randomly selected by stratified sampling according to the local burden of MDR-TB in each one of the 24 departments (in addition to the constitutional province of Callao) of Peru, reported in the 2015–2018 period [4]. Each of the selected strains were previously isolated from samples of patients with pulmonary and/or extrapulmonary TB during the mentioned period. All strains had preliminary information on resistance profiles obtained by the routine APM and were obtained from the LRNM culture bank.

The susceptibility tests were carried out under programmatic conditions by the LRNM in the period 2015–2018 using the APM. The procedures established by the Clinical and Laboratory Standards Institute (CLSI) [9] were followed and the phenotypic susceptibility was determined for the drugs rifampicin, isoniazid, ethambutol, ethionamide, kanamycin and levofloxacin, according to the critical concentrations (CC) recommended by the World Health Organization (WHO) at the date of testing (Table 1). Briefly, four quadrant Petri dishes containing Middlebrook 7H10 medium (Becton–Dickinson, Sparks, Md., USA) were used. MTB cultures on Lowenstein Jensen were transported to the LRNM from regional laboratories. The strains were sub-cultured in Middlebrook 7H9 broth (Becton Dickinson, Sparks, USA) and incubated for 7 days at 37 °C. Subsequently, fresh broth cultures were standardized to a McFarland 0.5 turbidity scale. The standardized suspensions were diluted 10^–2 to allow the growth of countable colonies for interpretation. For each culture, the drug-containing quadrants, as well as the drug-free control quadrant, were inoculated with 100 µL of the diluted suspension. The plates were sealed in plastic bags and incubated at 37 °C for 21 days. The cultures were classified as resistant when the number of colonies developed in the drug quadrant was more than 1% of the number of colonies observed in the control quadrant, otherwise they were classified as susceptible.

Drug susceptibility testing was performed using the UKMYC6 96-well microdilution plate format, designed by the CRyPTIC project (Thermo Fisher Inc., UK). The UKMYC6 plate allowed for the determination of susceptibility against 13 antituberculosis drugs composed of agents used in drug-susceptible TB treatment (rifampicin, rifabutin, isoniazid) as well as longer MDR-TB treatment corresponding to groups A (levofloxacin, moxifloxacin, bedaquiline, linezolid), B (clofazimine), C (ethambutol, delamanid, amikacin, ethionamide) and kanamycin (Table 1). Testing for each drug entailed between 5 and 10 concentrations obtained by serial doubling dilution (Additional file 1: Fig. S1). Previously cryopreserved strains were reactivated in 7H9 liquid medium, supplemented with “oleic acid albumin dextrose catalase” (OADC) (Thermo Fisher, Scientific Inc., USA), for 7 days at 37 °C. These were then subcultured in Middlebrook 7H10 media for 25–30 days at 37 °C. From the solid cultures in 7H10 medium, 0.5 McFarland scale suspensions were prepared in Tween saline with glass beads (Thermo Fisher, Scientific Inc., USA). Then, 100 µL of the suspension was diluted in a 7H9 broth tube supplemented with OADC to give an approximate inoculum of 1 × 10⁵ CFU/mL. Subsequently, using the automated Sensititre Autoinoculator^®/AIM^® equipment (Thermo Fisher, Scientific Inc., USA), 100 µL of inoculum was dispensed into each well of the UKMYC6 plate. The plates were sealed using clear plastic and incubated aerobically at 35–37 °C. The H37Rv ATCC 27294 strain was used to perform periodic quality control tests of the analyzed drugs, as well as to quality control for contamination and adequate growth in two positive control (drug-free) wells of each plate used. All laboratory work related to the culture of live bacteria was carried out in the biosafety level 3 facilities of the INS and UPCH.

Plates were read using the semi-automated Vizion™ instrument. The results of the plates were considered valid only when the positive control wells showed acceptable growth and free of contamination. The plates were read 14 days after inoculation, as established by the CRyPTIC project [8]. Additionally, in accordance with the Standard Operating Procedure (SOP), if the growth of the positive control was weak or insufficient, a second reading was carried out at 21 days. The Vizion™ system captured and stored an image of the recorded MICs of each plate. The MIC of a drug was considered as the lowest concentration capable of inhibiting the visible growth of MTB in a given well. Based on the MICs, strains were categorized as susceptible, intermediate or resistant (ternary categorization) taking as reference the Epidemiologic Cut-off Values (ECOFF/ECV), as well as borderline concentrations (concentrations at which genetic mutations present different effects giving rise to mixed categories of susceptibility or resistance), established by the CRyPTIC project [11]. A result was determined as susceptible if the MIC was less than or equal to the established ECOFF/ECV; otherwise, it was defined as resistant. Isoniazid, ethambutol and ethionamide were categorized taking into account borderline concentrations (Additional file 1: Table S1).

WGS was performed for all strains for the primary objective of the parent trial. For this study, genomic results were only used for discrepant analysis to resolve discordant results between the APM and BMD tests. DNA extraction and BMD test were performed starting from the same solid culture (7H10 culture) to minimize phenotypic and genotypic variation. Genomic DNA was isolated using the phenol–chloroform method (Additional file 1: Supplementary Method). Sequencing libraries were generated using the Nextera XT Library Preparation Kit, following the manufacturer's recommendations. WGS was performed at the University of Oxford (UK) using the Illumina HiSeq platform (Illumina Inc., San Diego, CA, USA). Paired end 150 bp sequencing reads were generated and stored in fastq.gz files.

The quality of the sequencing reads was evaluated using the FastQC v0.11.9 [12] program. The paired reads were filtered with the Trimmomatic v0.38 [13] program using default values and a minimum Phred score of 20. The filtered reads were mapped against the H37Rv reference genome (NC_000962.3) using BWA v0.7.17 [14]. The elimination of duplicate readings was carried out with the program Picardtools v2.18.25 [15]. The variant call was made using the GATK v4 [16] program. Resistance-associated genes were evaluated for genotypic resistance to rifampin (rpoB), isoniazid (inhA, katG, ahpC, fabG1, kasA), fluoroquinolones (gyrA, gyrB), and second-line injectables (rrs, eis, tlyA). The genetic variants found were validated using the list of mutations published by the WHO in the “Catalogue of mutations in Mycobacterium tuberculosis complex and their association with drug resistance” [17]. Likewise, the resistance profiles were determined using the programs TBProfiler v3.0.4 [18] (database of mutations v. a2a234b) and Mykrobe v0.10 [19]. For each drug, the strains were classified as susceptible or resistant according to the absence or presence of mutations detected in the genes associated with resistance, respectively.

Descriptive analyses of the MICs obtained and development times were performed on the UKMYC6 plates. For the six drugs evaluated by both APM and BMD method, diagnostic performance indices (sensitivity, specificity, positive predictive value, negative predictive value, categorical agreement and Cohen's Kappa coefficient) were determined by taking as reference values ​​the results of the APM and comparing them to the susceptible/resistant categorization from the BMD method (results with borderline MICs were excluded from analysis). All the calculations were performed in the program R v4.0.5 [20] using the packages epiR v2.0.19 (https://​cran.​r-project.​org/​web/​packages/​epiR) and vcd v1.4 (https://​cran.​r-project.​org/​web/​packages/​vcd). To determine the strength of statistical agreement, the kappa (k) value was used and was graded as insignificant (0 ≤ k ≤ 0.2), medium (0.2 < k ≤ 0.4), moderate (0.4 < k ≤ 0.6), good (0.6 < k ≤ 0.8) or almost perfect (0.8 < k ≤ 1) according to the previously proposed classification [21].

The study analysed 496 MTB drug-resistant strains of which 70% (347/496) were from Lima and Callao, while the rest came from the remaining 23 departments of Peru with a range of 1 to 18 strains per department (Additional file 1: Table S2).

The percentages of strains with APM-defined phenotypic resistance included in the study for each one of the drugs were: 86% (427/496) resistant to rifampicin, 94% (464/496) to isoniazid, 45% (218/489) to ethambutol, 31% (155/493) to ethionamide, 11% (56/495) to kanamycin and 11% (54/486) to levofloxacin. In addition, according to the classification of drug resistance profile of the year 2020 [2], 5% (24/496) presented rifampicin mono-resistant TB (RR-TB), 12% (61/496) isoniazid mono-resistant (HR-TB), 76% (376/496) MDR-TB, with a further 5% (27/496) XDR-TB, and 2% (8/496) had different patterns of drug resistance.

The microbiological evaluation by the BMD system using the UKMYC6 plate determined that 80% (397/496) presented final growth readings at 14 days and 20% (99/496) at 21 days. Overall, 99% of the readings obtained for each of the drugs showed valid results (MICs inside and outside evaluated ranges). Of these, overall 12% of MIC readings were above the dilution range of the UKMYC6 plate, most frequently rifampicin (68%, 326/480) and rifabutin (46%, 229/494) (Additional file 1: Table S3).

Comparative analysis was performed between the APM phenotypic test results and UKMYC6 MICs for 6 drugs: rifampicin, isoniazid, ethambutol, ethionamide, kanamycin and levofloxacin. The phenotypically susceptible or resistant strains by APM were graphed separately by means of histograms showing the MICs values obtained from the UKMYC6 plates (Fig. 1A). The drugs rifampicin, isoniazid, kanamycin, and levofloxacin showed visibly different MIC distributions for APM-susceptible and resistant strains. However, there was considerable overlap of the MIC distribution for strains defined as resistant and susceptible to ethambutol and ethionamide by APM (Fig. 1A; Additional file 1: Fig. S2). For the remaining drugs (rifabutin, amikacin, moxifloxacin, bedaquiline, delamanid, clofazimine, and linezolid) no reference drug susceptibility test (susceptible/resistant) assignment result was available so the MIC distributions for all strains were plotted together (Fig. 1B). The drugs rifabutin, amikacin and moxifloxacin showed the presence of strains with high and low MICs values for which the use of ECOFF/ECVs established the presence of 68% (336/494), 6% (29/491) and 9% (43/495) prevalence of resistant strains, respectively. However, none of the strains presented MICs higher than the ECOFF/ECVs values for the new or repurposed drugs, so they were defined as phenotypically susceptible strains by the BMD plate methodology (Fig. 1B).

For isoniazid, ethambutol, and ethionamide borderline MICs were demonstrated in 16% (77/496), 25% (121/492), and 6% (32/495) of strains, respectively. In the case of isoniazid, all borderline MICs were determined in strains categorized as resistant by APM; while in the cases of ethambutol and ethionamide, borderline MICs were determined in both susceptible and resistant strains (Table 2).

Using the UKMYC6 ECOFF/ECVs values to categorise strains as susceptible or resistant (excluding results with borderline MICs) and comparing these data with APM results available for the six drugs mentioned above resulted in an average categorical agreement of 0.93. The highest categorical concordances were obtained for rifampicin, isoniazid, kanamycin and levofloxacin. Using APM category as the reference, the sensitivity of BMD UKMYC6 plate for detection of drug resistance was highest for isoniazid (0.98) and rifampicin (0.93) and lowest for ethionamide (0.66) and kanamycin (0.68). Specificity was high for all drugs, from 0.91 for isoniazid to 1.0 for kanamycin (Table 3).

Initially, 266 discordant results (present in 191 different strains) were identified for all six drugs tested by the BMD plate method and APM, which included 30 results for rifampicin, 9 for isoniazid, 117 for ethambutol, 83 for ethionamide, 17 for kanamycin, and 10 for levofloxacin (Additional file 1: Table S4). More than 75% (200/266) of the discordant cases were due to results for ethambutol and ethionamide. Subsequent analyses excluding borderline MIC results considerably reduced the number of discordant results. For ethambutol, 68 ‘BMD-susceptible/APM-resistant’ results were excluded, of which 97% (66/68) presented mutations associated with resistance. In contrast, for ethionamide 19 ‘BMD-susceptible/APM-resistant’ discordant results were excluded, of which 47% (9/19) presented mutations associated with resistance. No strain was excluded for the isoniazid results.

Finally, 179 discordant results were obtained (present in 122 different strains). There were no ‘BMD-resistant/APM-susceptible’ discrepancies for kanamycin but for all other agents there were discordant results in both directions. Ethambutol and ethionamide had the highest degrees of disagreement with 13% (49/367) and 14% (64/460), respectively. For most agents the results of genotypic susceptibility testing by WGS gave similar levels support to the results obtained by APM (52%, 93/179) and those obtained by the BMD system (48%, 86/179), with the exception of rifampicin, in which a greater support towards APM results was observed (Table 4).