Validation of a 14-drug microtitre plate that includes delamanid and bedaquiline for susceptibility testing of M. tuberculosis

Background Universal access to drug susceptibility testing is key to ending TB. UKMYC5 is a 96-well microtitre plate designed by the Comprehensive Resistance Prediction for Tuberculosis: an International Consortium (CRyPTIC) which has potential to determine, at low cost, the MICs for 14 different anti-TB drugs, including several new and repurposed compounds. It is a dry-format plate and therefore easy to transport and store. Objectives Determine how long to incubate the plates before reading, and the optimal reading method. Establish the reproducibility of the UKMYC5 plate and compare it to established methods. Methods UKMYC5 plates were tested by seven laboratories on four continents using a panel of 19 external quality assessment (EQA) strains, including H37Rv. MICs were measured from each plate by two readers using three methods (mirrored-box, microscope and Vizion™ Digital viewing system) at four different timepoints. All EQA strains were whole-genome sequenced and phenotypically characterized by MGIT960, 7H10/7H11 agar and resazurin microtitre assay. Results The optimum duration to incubate a plate is 14 days. The within-and between-laboratory reproducibilities for the best performing methods (mirrored-box and Vizion) were both 95% and 92%, respectively. One site was identified as requiring re-training and one drug (para-aminosalicylic acid) produced inconsistent results. Conclusions MICs measured using the UKMYC5 microtitre plate (i) are reproducible, (ii) compare well with the results of several established methods and (iii) correlate with the presence or absence of genetic mutations that confer resistance. This study provides the evidence that this assay can be deployed by TB reference laboratories world-wide as a diagnostic and research tool.


INTRODUCTION
The proportion of tuberculosis (TB) cases that are multi-drug resistant (MDR) is increasing worldwide. Although set against a background of a falling global incidence of TB, the net effect is that the number of MDR-TB cases continues to grow. 1 Improving the treatment success rate for MDR-TB requires each patient to receive an individual antimicrobial regimen tailored to maximize efficacy whilst minimizing toxicity; this necessitates being able to measure minimum inhibitory concentrations (MIC), to direct both the choice of drug and dose. Universal access to prompt and comprehensive drug susceptibility testing (DST) is therefore a key component of the WHO's End TB Strategy. 2,3 Although molecular approaches have the potential to deliver universal DST methods, they require further development work and any resulting solutions are likely to be expensive.
Liquid and solid media assays that measure MICs for TB exist, [4][5][6][7][8][9] but are time consuming and, often, expensive. Microtitre plates offer a way of testing in parallel the effectiveness of a large number of drugs at a range of concentrations on small aliquots from a single clinical isolate. Broth microdilution methods, including several using colorimetric indicators, have previously been developed that assess the MICs for a panel of compounds using a single microtitre plate. 5,10,11 A dry-format, 96-well microtitre plate assay (the Sensititre™ MYCOTBI plate; Thermo Fisher Scientific Inc., USA) containing 12 drugs has been commercially available since 2010 and early validation studies have returned promising results. [12][13][14][15][16] No plate-based assays, however, have so far included both new and re-purposed drugs that will be key to the successful treatment of individual MDR-TB cases in the future.
In this paper, we validate a bespoke, dry-form 96-well microtitre plate (UKMYC5) that  (GenBank AL123456) 17 and was labelled as such. Eleven EQA strains were duplicated (one of which was also H37Rv) by the SRL prior to sending, along with one representative of eight additional EQA strains, bringing the total to 31 culture vials containing nineteen distinct strains. All vials, except the labelled H37Rv vial, were labelled CRY-1 to CRY-30, and therefore were blinded.
Preparation of replicates. Ten replicates were derived from the unblinded H37Rv ATCC27294 vial ( Figure 1B), each being sub-cultured on solid media as described below before being subbed onto a UKMYC5 plate. Two replicates were created from each of the remaining thirty vials and, again following sub-culture, were inoculated onto a UKMYC5 plate. Each participating centre therefore tested up to 70 UKMYC5 plates ( Figure 1C). The UKMYC5 plate was designed by the CRyPTIC consortium   and manufactured by Thermo Fisher Scientific Inc., UK. Fourteen anti-TB drugs   (rifampicin, rifabutin, isoniazid, ethambutol, levofloxacin, moxifloxacin, amikacin, kanamycin, ethionamide, clofazimine, para-aminosalicylic acid, linezolid, delamanid and bedaquiline) were included, at 5 to 8 doubling dilutions ( Figure S1, Table S1).

UKMYC5 design.
Janssen Pharmaceutica and Otsuka Pharmaceutical Co., Ltd provided, respectively, bedaquiline and delamanid pure substances. Although pyrazinamide-only plates containing lyophilised substance in different stocks were tested, poor performance due to suboptimal broth pH conditions resulted in pyrazinamide being excluded from UKMYC5.   Figure S3).

Measurement of MICs.
In each centre, two laboratory scientists independently read each microtitre plate using three different methods (Thermo Fisher Sensititre™ Vizion™ Digital MIC viewing system, a mirrored box and an inverted-light microscope) at 7, 10, 14 and 21 days post-inoculation ( Figure 1D). MIC results and additional data were recorded locally onto paper and into a shared web-enabled database (Table S2). An image of each plate was captured using the Vizion™ and was stored and subsequently analysed by software, the Automated Mycobacterial Growth Detection Algorithm (AMyGDA). 18 AMyGDA analysis was performed at the University of Oxford ( Figure S3).   Figure   1B) were phenotypically characterised by BACTEC™ MGIT960 (BD Lifesciences, New Jersey, USA), Middlebrook 7H10/7H11 agar dilution method (Table S3) and (with the exception of ethionamide and para-aminosalycilic acid) resazurin microtitre assay (REMA , Table S4) for drugs for which the WHO has endorsed critical concentrations (CCs). 19 All strains were also whole genome sequenced; genomic DNA was extracted from Löwenstein-Jensen cultures using either FastPrep24 for cell lysis and ethanol precipitation or the cetyl-trimethylammonium bromide method as described elsewhere. 20  According to the International Organization for Standardization (ISO-20776-2), a clinical antimicrobial susceptibility test is required to have a reproducibility of ≥ 95 %. 22 To assess the reproducibility within a site, the mode MIC for each drug was computed for that site, pooling the results across reading methods, days, replicates

The proportion of plates that can be read
The proportion of readable results (defined in the Materials and Methods) increased with elapsed time since inoculation, from 66.5% at day 7, to over 90% at day 14, to 96.8% at day 21 ( Figure 2A).
For Site F the proportion of readable results was anomalously high (> 94%), regardless of reading day ( Figure S4A), yet the reproducibility within the site was anomalously low, varying between 77.3%-80.2% depending on the reading day ( Figure S4C). Site F also had an anomalously low between site reproducibility of only 72.2%-75.3% ( Figure S4D). Logistic mixed-effect models confirmed that for Site F between reader OEA and within and between laboratory reproducibilities were all significantly lower than the other laboratories (p<0.001 for all comparisons; Table   S7). Data from Site F was consequently excluded from subsequent analyses.

Overall within and between laboratory reproducibility.
At least 92% of all MICs read by two scientists from the same laboratory were in overall essential agreement (OEA, within a doubling dilution), regardless of reading day or method ( Figure 2B

Selection of reading day and method.
After 7 days, only between 57.8%-66.1% of results were readable, depending on the method used, compared to over 88% at day 14, and between 95.9%-97.6% at day 21 ( Figure 2A). The corresponding logistic mixed-effects model demonstrated that significantly fewer results were readable at day 7 and 10 than at day 14, and that significantly more were readable at day 21 compared with day 14 (p<0.001 for all comparisons, Table S8). The proportion of results readable by inverted-light microscope or mirrored box was significantly lower than for Vizion™ (p<0.001 in both instances, Figure 2A, Table S8).
The OEA between two scientists examining the same plate within a laboratory increased with time ( Figure 2B). For Vizion™ and mirrored box, the OEA was >95% for all reading-days, while for the inverted-light microscope it increased from 92.7% at day 7 to 96.8% at day 21 ( Figure 2B). The corresponding model showed there was significantly lower OEA between readers when they used the inverted-light microscope compared to the Vizion™, or when they read the plates at day 7 compared to day 14 (p<0.001 for both comparisons, Table S8).
Overall, the greatest reproducibility (both within and between laboratories) was observed at day 14 ( Figure 2C & 2D), with both Vizion™ and mirrored box being 95% and 92% reproducible within and between sites, respectively. For the inverted-light microscope, the corresponding reproducibilities were 92.9% and 89.9%, respectively.
The logistic mixed-effects models confirmed that within and between site reproducibility for the inverted-light microscope was overall significantly lower with respect to Vizion™ (p<0.001 for all, Table S9). The Sensititre™ Vizion™ Digital MIC Viewing System was therefore selected for all subsequent analyses, although we note that the mirrored box also performed well. The individual MICs measured using Vizion™ were then analyzed to establish the optimal reading-day for each drug ( Figure S5). Para-aminosalicylic acid showed the lowest OEA between readers at each reading day and the lowest reproducibility both within and between sites at days 10, 14 and 21. The results of the corresponding logistic mixed-effects models (Table S10)  Reproducibility was assessed also on an individual drug basis by using rifampicin as a reference standard ( Figure S5 & Table S10). Although there was a statistically significant difference between many drugs, the greatest difference was seen for paraaminosalicylic acid where the reproducibilities within and between sites were significantly lower than rifampicin (p<0.001).

Results with AMyDGA software
MICs measured from photographs of the UKMYC5 plates using the AMyGDA software 18 were compared with those based on the Vizion™ reading method. The OEA between the two methods was above 87% for all days and increases with time since inoculation (from 87.9% at day 7, up to 93.8% at day 21, Figure 3). At day 14, the OEA between readings by AMyGDA and Vizion™ is above 90% for all drugs, except for moxifloxacin (89.3%) and para-aminosalicylic acid (73.8%), and therefore satisfies ISO-20776-2. 22 Using a logistic mixed-effect model, we found that the overall OEA at days 7, 10 and 21 is not significantly different from the overall OEA at day 14 (p=0.143, p=0.479 and p=0.525, respectively, Table S11). The OEA between Vizion and AMyGDA was not significantly different for all drugs at day 14 (Table   S11).   (Table S12). To infer 'sensitivity' or 'resistance' using the UKMYC5 plate, we assumed shared breakpoints with each comparator method (APM or MGIT). Discrepancies are shown in Table S13.

Comparison between genotype and UKMYC5 results
For all drugs, the presence of mutations in genes in which there is a high confidence that they confer resistance was associated with MICs greater than observed in wild type strains ( Figure 6). 25 Mutations whose role in the drug resistance was not well established in the literature were included in the analysis. Only bedaquiline, ethambutol, linezolid and the fluoroquinolones showed some overlap in MICs between wildtype strains and strains containing mutations in genes associated with resistance. For linezolid, no difference of MICs was observed between wild type and mutant strains, but this was expected since these mutations were shown to be susceptible by MGIT and APM (Table S6), although mutations in the rrL gene have been reported to shift MICs. 26 As expected, mutations associated with low-level resistance, such as inhA_C-15T, were associated with smaller MIC increases than mutations in katG that is associated with high-level resistance ( Figure S7). 27 Interestingly, the EQA panel included two isolates that were resistant to delamanid according to the APM. One of these had a mutation in the ddn gene, and the other, in Rv1173. The ddn mutation was associated with higher MICs than the Rv1173 mutation ( Figure S7).   This validation study moves the field forward: UKMYC5 is the first microtitre plate assay to incorporate both new (delamanid, bedaquiline) and repurposed drugs (linezolid and clofazimine) in a dry-well format that is convenient to transport and store. In this paper, we have demonstrated that the measurement of MICS within and between participating laboratories is reproducible for 13 of the 14 drugs on the plate.
Fourteen days after inoculation was shown to be the optimal time to read a plate and the Thermo Fisher Sensititre™ Vizion™ Digital MIC viewing system was selected as the optimal reading method. We note, however, that the mirrored box performed equally well and, since it is cheaper, is likely to be a better solution in low-income countries. The Vizion™, however, also allowed us to photograph each plate, thereby not only providing an audit trail but also enabling the use of the AMyGDA software.

TRANSPARENCY DECLARATIONS
None to declare

SUPPLEMENTARY DATA
Figures S1 to S7 and Tables S1 to S13.