**4. Discussion**

In this retrospective analysis of drug resistant isolates from Botswana, it is shown that there is a high proportion of rifampicin (R) and isoniazid (H) monoresistance posing an increased risk of the development of MDR-TB [21,22]. It was observed that 7% (4/57) of the isolates were pre-XDR-TB. These isolates had resistance to fluoroquinolone only while 5% (3/57) were XDR-TB. The majority, 88% (50/57) of the isolates did not have any resistance to second-line drugs. It is interesting to see that in this sample, set SLID resistance occurred after fluoroquinolone resistance and not the other way around. Fluoroquinolones are used to treat other bacterial infections other than TB which could play a role in the increasing levels of resistance to this class of drugs in *M.tb* (in both pre-XDR and XDR cases) [5]. The introduction of Pretomanid, recently approved by the Food and Drug Administration (FDA) for the treatment of R resistant TB, has greatly shortened treatment and has been seen to improve treatment success [23,24]. However, Pretomanid is not ye<sup>t</sup> available in Botswana. The presence of both pre-XDR-TB and XDR-TB cases are indicators that there are gaps in the control of TB. The genotyping methods used in this study are not sufficiently discriminatory to investigate transmission, and whole genome sequencing has not been done on these strains. However, the possibility exists that patient to patient transmission exists as there were 2 patients who were infected with a strain of the same spoligotype and second-line drug resistance pattern. Further analysis would however be needed to investigate TB transmission in this population.

The GenoType MTBDR*sl* assay can detect mutations in the quinolone resistance determining region (QRDR) of the genes *gyrA* and *gyrB*. The most common *gyrA* mutation detected by the MTBDR*sl* among the pre-XDR and XDR cases was A90V (57%). This mutation confers resistance to levofloxacin and is associated with low level resistance to moxifloxacin [8]. Some resistance mutations are characterized by the absence of hybridization at the respective wild type probes [25]. The absence of the wild type bands in the line probe assay can be used to infer that there could be resistance to flouroquinolones but it does not allow the determination of the genotypic changes and the resulting phenotypic resistance to specific drugs. The targeted sequencing is therefore required to identify the specific drug resistance mutations. For example, in our cohort, there were two isolates which had an undefined mutation shown by the absence of both the wild type (*gyrA* WT3) and the mutation band in the *gyrA* gene. In this case, resistance to fluoroquinolones, particularly levofloxacin, can only be inferred since the specific mutation is not known [25]. This information can nevertheless help select a treatment regimen that could be more beneficial to the patients (e.g., excluding fluoroquinolones). The future implementation of new drugs, such as Bedaquiline and Pretomanid, could change the genetic drug resistance patterns observed in this study. The pre-XDR and XDR-TB patients with drug resistance patterns described in this study could still be successfully treated with these drugs, thereby reducing the spread of these *M.tb* strains. The genotypic drug resistance patterns therefore need to be closely monitored to be able to adapt treatment guidelines if required.

This study found that all the XDR-TB patients had unsuccessful treatment outcomes. These strains being XDR probably resulted in (almost) none of the prescribed drugs being efficient in killing the bacteria. The pre-XDR and XDR-TB patients in this study were managed with regimens which contained levofloxacin. Previous studies have shown that in cases of levofloxacin resistance, moxifloxacin may be the preferred drug of choice since *gyrA* A90V mutation has a smaller effect on moxifloxacin activity [8,26]. Therefore, these patients could have benefited from a regimen containing Moxifloxacin if the specific resistance markers had been determined timely. Among the pre-XDR isolates and MDR-TB isolates, 75% and 73% of the patients had successful treatment outcomes respectively, however all isoniazid monoresistant and 55% of the rifampicin monoresistant patients had unsuccessful treatment outcomes. Previous studies have shown that isoniazid monoresistance is associated with poor treatment outcomes [3,27].

There were no mutations detected in the *gyrB* gene in any of the isolates in this study. The mutations in the *gyrB* gene are usually associated with low level resistance to fluoroquinolones and are not as common as those in the *gyrA* gene [2]. The *rrs* MUT1 A1401G mutation which leads to a high level second-line injectable drug resistance was detected among 5% of the MDR-TB isolates. This mutation causes high level resistance to KAN and cross resistance to AM and CAP [6]. The presence of these mutations shows that there is a need to routinely test for second-line drug resistance among MDR-TB cases in Botswana. In this study, gene mutations that are associated with low level drug resistance induced by mutations in the promoter area of the *eis* gene were not detected. There was no association between the drug resistance profile and HIV status. The data on HIV viral load and CD4 cell counts were not available for this study and their association with drug resistance could therefore not be analyzed. However, previous studies have shown that there is an association between drug resistance and HIV status [28–30]. Haar et al. and Fenner et al. have shown that patients with high viral loads are more likely to have multi-drug resistant TB than those who are virally suppressed [31,32].

Even though this study is informative and provides data on the genetic mutations that are associated with second-line drug resistance in Botswana, it had some limitations. The small sample size may not reflect the true burden of second-line drug resistance in the entire country and there is limited statistical power to detect other drug resistance mutations within the population. Due to the small sample size, there is insu fficient statistical power to fully address the association with various risk factors and treatment outcomes. This was a retrospective study therefore, there may be other unknown confounding factors. One of the limitations of the line probe assays (LPAs) is that there may be a false detection of resistance due to some synonymous mutations. Some studies have reported synonymous mutations which can result in false-positive results (false detection of resistance). However, in such instances, appropriate confirmatory testing should be done promptly [33]. The lack of hybridization of the wild type probes is not a reliable indication of phenotypic resistance, hence these kinds of hybridization patterns need to be verified with phenotypic DST [7]. There were some mutations that were undefined in our study therefore, other mechanisms of fluoroquinolone resistance need to be investigated further using techniques, such as whole genome sequencing or targeted gene sequencing. The identification of gyrase mutations can aid in predicting fluoroquinolone resistance as well as estimating the levels of resistance to various flouroquinolones. This may assist clinicians to determine the most e ffective dose of fluoroquinolones [34]. There is also a need to carry out this study in a larger population in order to determine the association of several risk factors with treatment outcomes as well as to determine the prevalence of second-line drug resistance in Botswana.
