Pyrazinamide (PZA) is one of the most important drugs for tuberculosis (TB) treatment. It has a unique sterilizing effect on killing semi-dormant bacilli and is also very effective against multi-drug (MDR)-TB [1] [2] . Pyrozinoic acid (POA) is the active form of the drug and is metabolized by pyrazinamidase (PZase) enzyme which is encoded by the pncA gene [3] . Although number of studies suggest that PZA resistance is not caused by a single mechanism, mutations in the pncA gene are considered as the main mechanism of PZA resistance in M. tuberculosis [4] . The lack of reliable culture-based methods which require an acidic condition and high cost of the molecular techniques makes the testing of PZA resistance generally very difficult to perform [5] . It is very important to develop rapid and accurate drug susceptibility testing (DST) to prevent the spread of multi-drug resistant (MDR)-TB as well as extremely drug resistant (XDR)-TB.

Several molecular techniques have been described for the detection of anti-TB drug resistance-associated mutations, including the line probe assays GenoType MTBDRplus (Hain Lifescience GmbH, Nehren, Germany), GenoType MTBDRsl (Hain Lifescience GmbH, Nehren, Germany), INNO-LiPA Rif.TB (Innogenetics, Ghent, Belgium), and Xpert MTB/RIF (Cepheid, Sunnyvale, CA) [6] [7] . However, these are probe based assays and only suitable for mutations that are located in hotspot regions of specific genes. Mutations in the pncA are dispersed throughout the gene and its upstream promoter; this makes the development of probe-based methods challenging [8] . Direct sequencing of pncA amplicons remains the best genotypic strategy, however this is costly.

High-resolution melting (HRM) curve analysis is a simple technique and has been widely used for many applications. PCR amplicons are heated and fluorescence loss is monitored in real-time. Variations in sequences are detected by difference in melting point (Tm) compared to reference DNA; it does not require the use specific probes [9] [10] . In the TB field, HRM has been used for detecting rifampin (RIF), isoniazid (INH), streptomycin (STR), and fluoroquinolone resistant M. tuberculosis [11] . In this study, we describe an HRM technique to detect pncA mutations in M. tuberculosis clinical isolates and compare its results to the phenotypic PZA susceptibility testing (The Bactec MGIT 960). Isolates that had discordant results were sent for sequencing. This method uses the Roche LightCycler 480^® and involves simultaneous amplification of three overlapping fragments.

The study was conducted at the NHLS-TB Laboratory Tshwane Academic Division (Diagnostic division of the Department of Medical Microbiology, University of Pretoria from July 2013 to October 2013. Ethics approval was obtained (number 57/2012) prior to commencement of the study. The study isolates (95) were a mixture of MDR-TB and non-MDR-TB collected from the National Health Laboratory Service (NHLS) TB Laboratory. All isolates were presumptively identified as M. tuberculosis complex using Ziehl-Neelsen (ZN) stain and tested for susceptibility using the GenoType MTBDR plus line probe assay (HainLifescience GmbH, Germany). Isolates were then categorized as either MDRs or Non-MDRs (Mono-RIF resistance, Mono-INH resistance or fully sensitive) [7] .

Purity of all isolates by ZN staining was confirmed before susceptibility testing was performed. To exclude bacterial and fungal contamination, another amount of a well- mixed suspension was inoculated using sterile glass pipette on chocolate agar (produced in-house, NHLS Laboratory, Tshwane Academic Division) and incubated aerobically at 37˚C. All plates were inspected for bacterial growth every 24 hrs for two days.

Isolates were sub-cultured onto 7H10 agar medium (Becton Dickinson, Sparks, MD, USA) and incubated for three weeks aerobically at 37˚C until typical M. tuberculosis colonies appeared (dry, wrinkled, warty, with colourless rough surface). Colonies were picked and sub-cultured into MGIT tubes (Becton Dickinson, Sparks, MD, USA) according to the manufacturer’s procedure.

Each isolate was tested for PZA susceptibility when the Bactec MGIT 960 indicated positive culture growth. The day the Bactec MGIT 960 (Becton Dickinson, Sparks, MD, USA) gives a positive signal is considered as day zero. Once positive, tubes were removed from the Bactec MGIT 960 (Becton Dickinson, Sparks, MD, USA) they were transferred into an incubator 37˚C ± 1˚C until susceptibility testing was performed (within five days). Undiluted inoculum was used if the isolate was processed on days one or two. On days three, four or five, the growth culture was diluted 1:5 using normal saline (0.9%) (SABAX Pour Saline, Adcock Ingram, South Africa) prior to susceptibility testing. PZA susceptibility testing was performed according to the manufacturer’s procedure (Becton Dickinson, Sparks, MD, USA). Susceptibility results were available within four to 21 days. Once the test was completed, a report was generated and printed. Each isolate was categorized as “S” (susceptible) or “R” (resistant).

A total of 120 M. tuberculosis isolates were collected from the NHLS-TB Laboratory Tshwane Academic Division (Diagnostic division of the Department of Medical Microbiology, University of Pretoria. Seven isolates were excluded due to duplication. Additional 18 isolates had no HRM results and were excluded from analysis. Susceptibility of the 95 M. tuberculosis isolates included in the final analysis to the first-line anti-TB drugs are shown in Figure 1. Isolates were categorized as MDR-TB (Resistant to both RIF and INH) and Non-MDR-TB (susceptible to at least RIF or/and INH).

Although PZA forms an integral part of both first line and MDR TB management, routine susceptibility testing of this drug is not performed due to various challenges present in the currently available techniques. This study evaluated the rapid technique of HRM for the detection of PZA resistance against the phenotypic gold standard (Bactec

MGIT = Mycobacterial Growth Indicator Tube; HRM: High Resolution Melting; PZA: Pyrazinamide; PPV: Positive predictive value; NPV: Negative predictive value.

MGIT 960).

In this study, the Bactec MGIT 960 results showed a prevalence of PZA resistance of 57.5% (23/40) among the MDR-TB isolates tested, while the prevalence among non- MDR isolates was 10.9% (6/55). High rates of PZA resistance among MDR-TB have been observed in previous studies, ranging from 49% in Thailand [13] , to 50% in Central Africa [14] , 52% in South Africa [15] , 53% in Japan [16] and 55% in Taiwan [17] . Studies from Pakistan [18] , South Korea Kim et al., 2012 [19] and India [20] have shown higher rates of resistance, 77%, 85% and 85% respectively. These high rates of PZA resistance in TB endemic countries may be attributed to the widespread use of PZA in re-treatment regimens in these areas, particularly when used without relying on susceptibility testing results. The high occurrence of false resistance by phenotypic susceptibility testing could also cause the overestimation of PZA resistance among MDR- TB [21] .

In 2013, Nagai et al. developed and evaluated HRM curve analysis for mutation detection in four major anti-TB drugs, namely RIF, INH, EMB, and STR. Their HRM assay was successful in detecting mutations in genes associated with resistance (rpoB, katG, inhA, ambB, rpsL and rrs) and results were completely consistent with those of DNA sequencing. Sensitivity and specificity of HRM were 100% and 100% for RIF, 88.8% and 100% for INH, 100% and 100% for EMB, and 100% and 93.7% for STR, respectively [22] .

Comparable results were also achieved by another two recent studies when HRM was used for the detection of PZA resistance. Hong et al. reported a sensitivity and specificity of 85.5% and 98.5%, respectively. Whereas Pholwat et al. showed concordance of 84% between HRM and phenotypic PZA susceptibility testing [23] [24] . The agreement between the phenotypic Bactec MGIT 960 and HRM for PZA susceptibility in this study was only 68.4%. Sensitivity and specificity of HRM were 47.7% and 74.6%, respectively. These values were much lower than previously reported [24] [25] [26] . Discrepancies between the two methods detected on 19 PZA-resistant isolates that were reported wild-type by HRM, and on 10 PZA-susceptible isolates that showed mutations by HRM. Among these 19 isolates, sequencing detected mutations in 14 of them suggesting that HRM was not able to detect these mutants which were also phenotypically resistant. The majority variants were SNPs with five of the 19 occurring at codon 135.

Conversely, five of the ten susceptible isolates which were reported as variants by HRM showed no mutation on sequencing. Although this would suggest that these were false on HRM, it is also possible that the HRM result is true since HRM is PCR based and has a greater likelihood of detecting mixed populations which may not be detected by Sanger sequencing, and these sub-populations may not be large enough to show resistance phenotypically. We did not perform next generation sequencing which may have provided an answer to this uncertainty.

Recent studies have indicated that the new “gold standard” for PZA resistance determination should be sequencing. When considering the discordant isolates only and comparing HRM and MGIT 960 against this “new” gold standard, the accuracy for HRM was 47.9% while that of the Bactec MGIT 960 was slightly better at 66.7% (Table 7).

Phenotypic resistance to PZA has been correlated with mutations in the pncA gene in several previous studies. Some of these studies have reported this correlation to be inconsistent, ranging between 41% and 80% [25] - [31] , whereas other studies showed higher correlation ranging from 91% - 97% [15] [17] [32] [33] [34] . In the current study, based on sequencing results, a correlation of 0.67 (significant at p-value < 0.05) between phenotypic resistance to PZA and pncA mutation was observed, this is comparable to majority of the previous studies, more especially to the study that was conducted in South Africa by Bishop et al., in which a correlation of 67% was also reported [35] .

The mutations detected in our study were found to be scattered along the entire pncA gene with no major hot spots identified. However, mutations at codon 135 were found in five isolates, whereas mutations at codons 14 and 139 were shown by three isolates each. These observations of scattered mutations are supported by similar findings in previous studies [27] [36] . The high diversity of pncA mutations limits its inclusion in the current molecular techniques such as GenoType MTBDRplus (Hain Lifescience) [37] .

Absence of mutations in the pncA gene or its upper promoter in 26.7% of the PZA- resistant isolates we tested correlated with the study that was done by Sreevatsan et al., in which they reported no pncA mutations in 28% of PZA-resistant M. tuberculosis strains tested [15] . These cases could also be explained by another mechanism of resistance to PZA. Recently, rpsA gene has been shown to play a role in PZA resistance and has been recommended to be used as an additional target for the molecular detection of PZA susceptibility [9] [38] , however in another study, none of the PZA-resistant strains harboured mutations in the rpsA gene [39] . Based on these variations, the association between PZA resistance and rpsA mutations remains un-established and requires further investigations in the future.

The low sensitivity and specificity of HRM in this study could be due to the use of three sets of primers as compared to five or seven as used in previous studies [40] [41] . Unlike other studies, the current HRM was developed and evaluated using the Roche LightCycler^® 480. So far, only one study has utilized the Roche LightCycler 480^® for PZA susceptibility testing and used the Bactec MGIT 960 as a reference method [42] . They have shown a sensitivity and specificity of 85.5% and 98.5%, respectively and concluded that HRM is a rapid and accurate test for the detection of PZA resistance and can be used as a screening method [26] . However, recently, HRM results generated from the Rotor-gene system for the detection of PZA resistance have been found to be easier to interpret than other platforms [25] .

Another explanation for the low correlation between PZA resistance and pncA mutations in this study is the possibility of phenotypic false resistance. This observation has been proven elsewhere [25] . PZA phenotypic susceptibility testing by the Bactec MGIT 960 is normally carried out at pH of 5.9, resistance could be falsely reported as a result of the ammonia produced during mycobacterial growth; ammonia elevates the pH of the media thus inactivating the drug [26] . According to the Henderson-Hassel- balch equation, M. tuberculosis should be tested for PZA susceptibility at a concentration of at least 156 µg/ml [10] , rather than the current cut-off used in the Bactec MGIT 960 method (100 µg/ml). PZA resistance could also be caused by the action of efflux pumps which has been shown to play a role in mycobacterial resistance [43] . Detection of pncA mutations in the susceptible isolates suggests that these mutations might not be fully expressed to show phenotypic PZA resistance.

In this study, apart from simplicity and speed, the current HRM genotypic method to determine PZA susceptibility using the gene scanning software of the Roche LightCycler 480 was shown to be moderately specific with low sensitivity. Further modifications to improve its performance are required. Pyrazinamide resistance was common among MDR-TB and the current study results showed low association of the phenotypic PZA resistance with mutations in the pncA gene. This warrants further studies to determine potential alternate genetic mechanisms of resistance to PZA in our context and re- evaluation of the MGIT 960 for detection of phenotypic resistance with an alternate phenotypic method such as the Wayne’s test.

This study provides an important baseline for more extensive evaluation studies to improve and validate the use of HRM in determining PZA susceptibility. In the future, using more than three fragments (six or seven) should be considered as this could increase the sensitivity and specificity of HRM in detecting pncA mutations. However, this might also increase the cost of the test. Additionally, combining such data to results from the biochemical analysis of PZase enzyme (Wayne’s test) would add to its value and ease the interpretation of the final results.

The authors declare that they have no financial or non-financial competing interests.

This project was supported by grant from the National Health Laboratory Service of South Africa.

Osman, F., Ismail, F., Osman, A., Omar, S., Said, H. and Ismail, N. (2016) High Resolution Melting Curve Analysis for Rapid Detection of Pyrazinamide Resistance in Mycobacterium tuberculosis Clinical Isolates. Journal of Tuberculosis Research, 4, 155-172. http://dx.doi.org/10.4236/jtr.2016.44019