A new oligonucleotide array for the detection of multidrug and extensively drug-resistance tuberculosis

Drug-resistant tuberculosis (TB) is a global crisis and a threat to health security. Since conventional drug susceptibility testing (DST) takes several weeks, we herein described a molecular assay to rapidly identify multidrug-resistant (MDR) and extensively drug-resistant (XDR) and reveal transmission associated-mutations of Mycobacterium tuberculosis complex (MTBC) isolates in 6 to 7 hours. An array was designed with 12 pairs of primers and 60 single nucleotide polymorphisms of 9 genes: rpoB, katG, inhA, ahpC, embB, rpsL, gyrA, rrs and eis. We assessed the performance of the array using 176 clinical MTBC isolates. The results of culture-based DST were used as the gold standard, the GenoType MTBDRplus and MTBDRsl tests were used for parallel comparison, and gene sequencing was performed to resolve the discordance. The sensitivities and specificities of the array are comparable to those of the MTBDRplus test for resistance to isoniazid (INH) (100.0%, 96.7%) and rifampicin (RIF) (99.4%, 96.7%) and of the MTBDRsl test for resistance to fluoroquinolones (FQs) (100%, 100%) and second-line injectable drugs (SLIDs) (98.3%, 100%). The sensitivities of the array for detecting resistance to ethambutol and streptomycin were 79.3% and 64.9%, respectively. The array has potential as a powerful tool for clinical diagnosis and epidemiological investigations.

www.nature.com/scientificreports www.nature.com/scientificreports/ Detection of drug resistance with the array. As an example of the interpretation of the final results obtained from the array, Fig. 2 shows the detection of H37Rv DNA and 2 isolates with drug resistance using the array. Excluding one of the embB probe signals, the detection limit of the array was 300 pg per test. Compared with conventional DST, the sensitivities of the array for detecting resistance to the 8 drugs were over 90%, except for those of EMB and SM, which were 79.3% and 64.9%, respectively. The specificities of the array ranged from 80.0% to 100% (Table 5).
Analysis of divergent results between the molecular assays and phenotypic Dst. For detecting RIF resistance, we obtained divergent results with 1.8% (3/176) of the isolates. One of these three was a phenotypic RIF-susceptible isolate that lacked WT bands 2, 3 and 4 on the RIF-determining region of the GenoType MTBDRplus strip and the WT signal of the rpoB codon 511 when using the array. rpoB gene sequencing revealed the presence of the rpoB D516A mutation, which was not included in the array (Fig. 1). Of the 2 phenotypic RIF-resistant isolates, one isolate harbored two mutations, rpoB L511P and D516G, as revealed using rpoB gene sequencing; conversely, the array showed the WT signal. The partial overlap between the sequences of the probes  Total   RRRR  9  10  2  3  2  23  7  3  5  13  77   RRRS  2  1  15  6  2  4  9  39   RRSR  4  2  1  2  8  17   RRSS  2  1  1  4  3  2  1  13  27   SSSS  16  16   Total  17  13  3  3  3  42  17  9 10 59 176 www.nature.com/scientificreports www.nature.com/scientificreports/ for detecting codons 511 and 513 may cause a false negative. The other isolate had the rpoB R529K mutation, which was not included in the array but was correctly identified by the GenoType MTBDRplus test. In addition, a fully susceptible isolate with the rpoB gene L533P mutation was detected by the array, the GenoType MTBDRplus test and gene sequencing ( Ins TTC at codon 514   Table 2). The results from sequencing katG, the inhA locus and the oxyR-ahpC intergenic region revealed that among these isolates, only 2 harbored mutations in the oxyR-ahpC intergenic region, G-8A and C-17T, which were not included in the array or the GenoType MTBDRplus test. One isolate harbored the C-10T mutation in the oxyR-ahpC intergenic region according to the array and sequencing, but this  Table 3. Profiles of ethambutol and streptomycin resistance identified by phenotypic and molecular assays. R, resistant; S, susceptible; WT, wild-type; ∆, lack of wild-type signal; *discordant results between each assay.
www.nature.com/scientificreports www.nature.com/scientificreports/ mutation was missed by the GenoType MTBDRplus test. The other isolate showed C-15T mutations in the inhA locus and the C-10T mutation in the oxyR-ahpC intergenic region (Table 2).
For detecting SLID resistance, we found that 14.2% (25/176) isolates afforded divergent results. Of the 75 phenotypic SLID-resistant isolates, 2 (2.6%) CAP-resistant isolates were identified as susceptible by the array, the GenoType MTBDRsl test and rrs gene sequencing. Furthermore, 21 (28%) isolates with the rrs A1401G mutation detected by all three molecular methods were deemed susceptible by phenotypic DST. Two isolates that harbored the eis C-12T and G-37T mutations respectively, as detected by the 3 molecular methods, were identified as phenotypic KM susceptible. In addition, one fully SLID-susceptible isolate revealed that deletion at eis -8 was not identified by the array; however, this isolate was correctly identified by the GenoType MTBDRsl test as lacking eis WT2.

Discussion
Because of the urgent need to rapidly and accurately diagnose drug-resistant TB, the WHO has recommended the GenoType MTBDRplus and GenoType MTBDRsl tests for detecting resistance to first-and second-line drugs 1 . We developed an oligonucleotide array consisting of 60 mutant probes for detecting resistance to 8 anti-TB drugs in 6 to 7 hours. The sensitivity and specificity of the array were comparable to those of the WHO 6,[26][27][28][29] . Moreover, our array contains multiple mutation probes targeting the same alleles as the other aforementioned tests to achieve better sensitivity and accuracy.
This array can identify exact nucleotide substitutions in the mutated codon(s) to provide a clearer interpretation than those LPAs with fewer mutation probes. In addition, our system can simultaneously determine first-and second-line drug resistance in a single array, whereas the above commercial LPAs detect resistance to first-and second-line drug in separate strips, which may delay the turnaround time for determining the genotypic susceptibility to second-line drugs when cases are identified as RIF-resistant or MDR-TB. Furthermore, the expenditure for multi-step diagnosis is a huge burden for low-to middle-income countries.
The high concordance for detecting RIF resistance was expected because over 95% of RIF-resistant strains harbor mutations in the rifampicin-resistance determining (RRDR) region 30,31 . The one discordant profile obtained from phenotypic and molecular assays had the rpoB L533P mutation, which confers low-level resistance to RIF 32,33 . Low-level drug resistance is a major challenge in clinical diagnosis, especially for those laboratories that use the Mycobacteria growth indicator tube (MGIT) as the phenotypic assay 34 .
Concerning INH resistance, eleven INH-resistant strains were missed by the array. Thus, we considered potentially including other mutations of resistance-related gene in our bacteria set, such as the inhA structural gene specifically, the S94A and I194T mutations, which have been reported to be associated with high-level resistance to both INH and ethionamide in MDR M. tuberculosis strains identified in Latin American-Mediterranean families in Lisbon 35 . The sensitivity and specificity of the array for detecting INH resistance were comparable to those of GenoType MTBDRplus, which were 100% and 96.7%, respectively. We also designed two specific probes for detecting mutations in the oxyR-ahpC intergenic region, i.e., C-10T and C-12T, which are absent from the INH-resistance determining region of the GenoType MTBDRplus test. Accordingly, the array can correctly identify resistant isolates in agreement with the sequencing results. However, the G-8A and C-17T substitutions in the oxyR-ahpC intergenic region, which were found by gene sequencing, were not included in our probe panel; thus, we perhaps should consider these mutations as alternatives for detecting INH resistance.  www.nature.com/scientificreports www.nature.com/scientificreports/ We observed low sensitivity for detecting EMB resistance, which might be due to the presence of mutations in the embA and embC genes 36 . In addition, some rare resistance-related mutations that are identified by gene sequencing, such as the embB codon Y319S mutation, were also not included in the array. Three mutations corresponding to EMB resistance were missed by array: the embB codon M306I, G406A and Q497R mutations. We re-checked the raw data of the array for these three isolates, which revealed that the signal of the mutation probes of the aforementioned three locations, were too weak to identify, causing us to misjudge the output at first. The main reason for weak signals include low yield of embB PCR products, low melting temperature of the probe, probe length, probe self-binding and the formation of secondary structures within the PCR products. All probes designed in this study were checked for internal repeat, self-binding, secondary structure, and GC content by using software Vector NTI (Invitrogen Corporation, Carlsbad, Cal. USA). However, the software may not be able to reveal all drawbacks inherently present in the probes and therefore probe efficacy cannot be guaranteed. In this study, normally 5-10 probes were designed, checked by software, pretested by array hybridization, and finally the one with the best hybridization signal was used in the current array. Longer probes can have a stronger hybridization signal, but at the same time, longer probes may decrease the specificity of the probes. Therefore, a probe used in the array is a compromise of sensitivity and specificity. Conversely, six isolates showed the same genetic profile with triple mutations of embB L355L, E378A and A505V; we reviewed the registry of all six cases and found that it was an outbreak within a family. The selection bias was due to the selection of drug-resistant strains according to only phenotypic DST without initially confirming the clinical data.
Concerning SM, 28 of 33 SM-resistant isolates were identified as sensitive by the array and sequencing; the remaining 5 isolates carrying the rpsL K88R mutation as revealed by gene sequencing were missed by the array ( Table 5). The 5 missing results were mainly due to the same reason mentioned above for EMB. The discordant phenotypic/genotypic profile harbored the rpsL K43R, rrs A513C and rrs C516T mutations, which were found in 5 SM-susceptible isolates. In various studies, rpsL and rrs mutations were found in 36.6% to over 90% of SM-resistant isolates [37][38][39][40] . Additionally, recent studies indicated that the gidB gene encoding a 7-methylguanosine (m7G) methyltransferase specific to the 16 S rRNA was associated with low-level SM resistance in M. tuberculosis 41,42 .
To improve the array performance in detecting SM, we might need to consider including some specific primers and probes for detecting the gidB gene.
With regard to FQs, the array results were 100% consistent with those from the GenoType MTBDRsl test. However, 4 FQ-resistant isolates were identified as FQ susceptible by the array, GenoType sl test and sequencing. Furthermore, we found one phenotypic FQ-susceptible isolate harboring the gyrA S91P mutation, as identified by the array, the GenoType MTBDRsl test and sequencing. The gyrA S91P mutation confers moderate-to low-level FQ resistance 43 .
Regarding SLIDs, two CAP-resistant isolates were identified as susceptible by the array, the GenoType MTBDRsl test and sequencing. The discordant results may be due to the involvement of other resistance-related genes, such as tlyA 44 . Nevertheless, 21 phenotypic susceptible strains containing the rrs A1401G mutation were correctly identified by the array, the GenoType MTBDRsl test and sequencing. On the one hand, some studies found that the rrs A1401G mutation confers high-level resistance to KM and AM but only low-level CAP resistance 45,46 , which suggests that CAP is still an alternative option for combination chemotherapy 47,48 . On the other hand, some discordant results were found in 3 isolates with eis mutations, i.e., deletion at -8, C-12T and G-37T. The substitution of eis-8 was not detected by our detection probes but was correctly identified by the GenoType MTBDRsl test, as WT band deletion in the GenoType strip should be considered indicative of resistance. The mutations within the eis promoter region conferring KM resistance still need to be confirmed. Several studies indicated that the -10 or -12 substitution has very little or no association with resistance to KM 49,50 .
Nucleic acid amplification tests (NAATs) are widely used in clinical diagnosis, such as real-time PCR assays (e.g., the GeneXpert MTB/RIF test), LPAs (e.g., the GenoType MTBDRplus test), and reverse hybridization microarrays (e.g., the BluePoint MtbDR array). DNA concentration is the principal limitation of the multiplex PCR assay; when the DNA concentration is inadequate for performing amplification, false negatives and low sensitivity may result. Furthermore, DNA quantification could play an important role in TB clinical management for characterizing the disease, transmission, and response to therapy 51 . In our study, the signals from the probes for embB and rpsL were too weak for some isolates, preventing us from making a correct identification. The misjudgments were probably due to low DNA concentration. In addition, the probes or PCR products of interest may form a secondary structure and hence decrease the hybridization efficiency 52 . Thus, preventing the formation of secondary structure of the probes and PCR products, which significantly decreases the detection signal, may become a challenging problem for multiplex PCR in a single tube.
Real-time PCR assays such as the GeneXpert MTB/RIF test, the Abbott RealTime MTB INH/RIF Assays (Abbott, USA) and the Anyplex plus MTB/NTM MDR-TB products (Seegene, South Korea) are extensively used to diagnose TB with high accuracy and a fast turnaround time 53,54 . Although array hybridization takes multiple steps, other assays require an expensive instrument for detecting fluorescent signals and/or high-cost cartridges. In contrast, the cost of the array is approximately 10 to 15 US dollars, almost half the price of one GeneXpert MTB/RIF test 55 . Other than cost, no expensive equipment is required and the logistics are much simpler than either the Xpert test (imported with in vitro diagnostics approval by Taiwan Food and Drug Administration) or the 2 LPAs (imported without in vitro diagnostics approval by Taiwan Food and Drug Administration) assays. The only instrument used in this study is a small incubator with a shaking function for performing the hybridization process, rendering the array an affordable way to routinely conduct molecular surveillance in a high burden area. Compared to the Xpert test, the array could detect 8 drugs in a single assay. Compared to the LPA, when testing for XDR-TB, one will need to perform 2 LPAs, which is time-consuming and costly. Besides, our array could detect the exact mutations conferring drug resistance and avoid the detection of false RIF resistance, such as L511P by the Xpert test and the GenoTypeDRplus test. Furthermore, we could incorporate other probes to other resistance alleles into the array to enhance the performance, or we can remove some probes (incidental www.nature.com/scientificreports www.nature.com/scientificreports/ polymorphisms unrelated to resistance) with much lower detection rates than the others to improve the hybridization signal.
In conclusion, the results of an oligonucleotide array strongly agreed with those obtained using 2 WHO-recommended GenoType tests. The array is suitable for detecting MDR or XDR-TB and is a useful tool for epidemiological investigations with a short turnaround time. The array can have a significant impact on treating patients and on preventing the spread of drug-resistant TB.

Methods
Mycobacterium tuberculosis complex isolates. To  Drug susceptibility testing. MTBC isolates were subjected to DST using the proportion method with 7H10 medium (Becton, Dickinson and Company, Spark, MD, USA). Resistance was defined as 1% of the colonies growing in the presence of the following critical concentrations of first-line drugs: INH, 0.2 μg/ml; RIF, 1 μg/ml; SM, 2 μg/ml; and EMB, 5 μg/ml; and second-line drugs: AM, 6 μg/ml; KM, 6 μg/ml; CAP, 10 μg/ml; and ofloxacin (OFX), 2 μg/ml 25 . Growth on the control medium was compared to the growth on the drug-containing medium to determine susceptibility. The DST results were categorized as resistant or susceptible. The tests were validated by determining the susceptibility of M. tuberculosis H37Rv. MDR-TB is defined as an M. tuberculosis isolate that is resistant to at least INH and RIF. XDR-TB is defined as an MDR MTBC isolate that is resistant to at least one FQ and one SLID, whereas pre-XDR-TB is defined as an MDR-TB isolate that is resistant to either FQs (pre-XDR-FQ) or at least one of the injectable drugs (pre-XDR-INJ). Molecular assays. DNA was prepared according to the protocol provided by the manufacturer of the GenoType MTBDRplus v2.0 and MTBDRsl v2.0 kits and used for the array and DNA sequencing. The LPA assays were performed and results were read according to the manufacturer's instructions.
The layout of the probes in our array is illustrated in Fig. 1. The array can be used in a biosafety level 1 laboratory. Our array hybridization procedures consisted of the amplification of the regions containing 9 genes using a multiplex PCR. The digoxigenin-labeled amplicons were hybridized with probes immobilized on the array. Most reagents, except buffers, were included in the DIG Nucleic Acid Detection kit (Roche, Cat. No. 1175041). The protocol used for array hybridization was as follows: 20 ml of 0.5 × SSC [(1 × SSC is 0.15 M NaCl, 0.015 M sodium citrate) plus 0.1% sodium dodecyl sulfate (SDS)] wash buffer was added to a 9-cm Petri dish containing the arrays for 10 min at room temperature, and the wash buffer was subsequently discarded. Each individual array was placed into a 24-well cell culture plate, and 300 μl of hybridization buffer [5× SSC, 1% (w/v) blocking reagent, 0.1% N-laurylsarcosine and 0.02% SDS] was added to each well. PCR products obtained by multiplex were denatured at 99 °C for 5 min and put on ice immediately. Hybridization was performed by adding 20 μl of digoxigenin-labeled PCR products into each well and incubating at 50 °C and 150 rpm for 90 min. After hybridization, all the following procedures were performed at room temperature. The arrays were washed twice with 0.5 × SSC plus 0.1% SDS for 5 min and were incubated with 20 ml of 1 × blocking solution [1% (w/v) blocking reagent dissolved in maleic acid buffer (0.1 M maleic acid, 0.15 M NaCl, pH 7.5)] containing 8 μl diluted 1:2500 alkaline phosphatase-conjugated sheep anti-digoxigenin antibodies (Roche, Cat. No. 11093274910) for 1 hour at 150 rpm. The antibody solution was removed, and then the arrays were washed with 20 ml of maleic acid buffer (0.1 M maleic acid, 0.15 M NaCl, pH 7.5) and 20 ml of detection buffer (0.1 M Tris-HCl, 0.15 M NaCl, pH 9.5). Color development of each array was achieved by incubation with 75 μl of alkaline phosphatase substrate (stock solution of nitroblue tetrazolium chloride/5-bromo-4-chloro-3-indolylphosphate diluted 1:50 in detection buffer) for 15 min away from light. After color development, the arrays were washed twice with distilled water for 5 min each time and dried out in an oven. Drug resistance was established when the hybridization signal of the mutation probe was stronger than that of the corresponding wild-type (WT) probes.
Determination of detection limit of the array. Genomic DNA of M. tuberculosis H37Rv was used to determine the detection limit of the array. The DNA was serially diluted 10-fold (1 ng/μl to 100 fg/μl) with a carrier DNA (polyadenylic acid, Sigma-Aldrich, Saint Louis, MO, USA; 1 μg/ml in phosphate-buffered saline). The diluted DNA (2.5 μl) was used for PCR and then hybridized to the array. The detection limit of the array was defined as the highest dilution with signal that could be identified. Discordance analysis. DNA sequencing was performed by the Sanger method to resolve the divergent results between the array, the GenoType MTBDRplus v2.0 and MTBDRsl v2.0 tests, and phenotypic DST. DNA sequencing targeted mutations in rpoB, the inhA locus, katG, the oxyR-ahpC intergenic region, embB, rrs, rpsL, the eis promoter region, and gyrA. These regions were amplified by PCR and then further sequenced as previously described to detect mutations 25 . Data analysis. The sensitivity, specificity and agreement of the molecular assays were calculated using phenotypic DST as the gold standard. The degree of agreement between the array and the GenoType tests was assessed using the kappa statistic. Values of the kappa coefficient over 0.75 indicated excellent agreement, 0.40