In this study we carried out an in depth investigation of molecular resistance mechanisms by correlating particular genomic variants with phenotypic resistance in clinical isolates from a high-incidence setting in West Africa. For INH and RIF there is a close correlation between data from molecular and phenotypic resistance testing for resistance determination in the strains analyzed. Sensitivity and specificity of sequencing of katG for detection of drug resistance were 86.7% and 100% and for sequencing of rpoB 100% and 93.8%, respectively.
Overall, the correlation between molecular and phenotypic resistance testing for the determination of SM, EMB and PZA resistance was lower. Although specificities of sequencing of rpsL, embB and pncA were high (96-100%), sensitivities were lower (48-73%) due to so far unknown resistance mechanisms.
However, while our results in principle support molecular resistance testing, the finding that especially in rpoB and also in pncA particular mutations are not linked to high-level resistance is challenging and demonstrates that careful interpretation of molecular resistance assays is mandatory. Therefore, studies targeting new resistance mechanisms should include valid phenotypic resistance data and, to our opinion, a comprehensive database on genetic variations in resistance genes and the correlation with phenotypic resistance is necessary. Furthermore, the level of resistance mediated by particular mutations and the clinical consequences need to be thoroughly investigated. In addition, especially variations in gidB appear to be phylogenetically restricted rather than being involved in drug resistance development.
In our study the most frequent mutation among INH resistant strains has been detected in katG at codon 315. This SNP has been observed in numerous prior studies [24, 25] and has clearly been correlated with INH resistance by loss of catalase activity. In two strains, in addition to variations at katG315, mutations at codon 291 and 471 were detected. However neither mutation has been described in the literature before and the katG315 mutation therefore represents the likely mechanism for INH resistance in these strains. The mutation at codon 300 observed in one strain in our study has been previously reported by Richardson and co-workers , where loss of this mutation has resulted in reversion of INH resistance in a previously drug resistant strain. The mutation at codon 302 as well as the insertion at codon 329 has not been described previously. Since they are restricted to INH resistant strains in our highly diverse MTBC collection, they represent potential new INH resistance mechanisms. Experimental evidence is required to validate this hypothesis.
Of the four INH resistant strains that displayed a wild-type katG sequence, three had mutations in the promoter regions of inhA and ahpC. The inhA mutation has previously been described in the literature  as being the most common variation in the inhA promoter region related to INH resistance. Mutations in ahpC have been found before, however to our knowledge not at this position. In one of the resistant strains no mutation was found in neither the complete katG gene nor in inhA or in ahpC. This result suggests a so far unknown resistance mechanism as being responsible for INH resistance of this strain.
Mutations in rpoB at codons 526 and 531 occur most frequently in the RIF resistant strains analyzed. Those SNPs are located in the RRDR and are well known for mediating resistance [27, 28]. The mutation at codon 481, which only occurs in one RIF resistant isolate, has to our knowledge not been described previously.
The mutations at codon 511 (Leu → Pro), 516 (Asp → Tyr) and 533 (Leu → Pro) conferred low-level resistance in agreement with previous studies [29, 30]. It has been shown that various substitutions in the same codon lead to different levels of resistance. For example mutations at codon 516 can confer either low- or high-level resistance depending on the amino acid change . Furthermore, the phenomenon of RIF low-level resistance has only recently been described in a work by Van Deun and colleagues , where mutations at codon 511, 516 and 533 have been found in strains tested susceptible by the radiometric Bactec 460 TB and Bactec 960 MGIT methods. Our data confirm the existence of low-level RIF resistance mediated by specific mutations in rpoB that is not detected by standard drug susceptibility testing methods. However, MIC values, especially for the mutations at codon 516 and 533, are even lower (0.5-1.0 μg/ml) than have been described in the literature. This fact may be due to the presence of further mutations in the operon or in other regions of the genome.
In a recent study  the therapeutic challenge of low-level RIF resistance has been addressed and may, according to the authors, be overcome by the application of higher RIF doses (20 mg/kg) in treatment regimens. However, the clinical relevance and interpretation of these data is still not fully understood and needs further investigation in animal treatment models or clinical trials.
Despite these discordant findings, we found a good correlation between the results from molecular and phenotypic testing for INH and RIF, as has been observed in another study . In fact, the strains analyzed in this study predominantly harbour well described mutations which allows for the application of standard sequencing protocols or commercial line probe assays.
The analysis of SM resistance mechanisms revealed an interesting observation. None of the SM resistant strains carried a mutation in the rrs gene, although those mutations have been described as main resistance mechanisms that confer high-level SM resistance . Instead, the SM resistant strains in our study population carry mutations in rpsL at codon 43 or 88 or at various codons in the gidB gene. The two mutations in rpsL have been described previously to confer high-level SM resistance [28, 34]. Polymorphisms in gidB were reported to confer a lower level of SM resistance . However, due to a number of phylogenetic polymorphisms in gidB, cautious interpretation of sequencing data is mandatory. Leu16Arg (ctt/cgt) has been described previously as phylogenetic marker for the LAM genotype , which could be confirmed in this study. Additionally, a synonymous SNP at codon Ala205Ala (gca/gcg) was identified as being specific for the WA1, WA2 and Beijing genotypes, as well as a combination of Ala205Ala (gca/gcg) and Val110Val (gtg/gtt) was determined as phylogenetically specific for strains belonging to the EAI genotype. These mutations in gidB occurred both in SM susceptible and resistant strains, affirming their role as phylogentic SNPs rather than markers for SM resistance. Polymorphisms in gidB probably playing a role in SM resistance, as they occur exclusively in SM resistant strains and do not coincide with mutations in rpsL, were detected throughout the complete gene (codons 34, 65, 71, 88, 91, 100, 138, 200). However, the actual importance of these SNPs for SM resistance needs to be investigated in further studies.
Reasons for the absence of rrs mutations in the strains analyzed and the shift to mutations in rpsL and gidB are mainly unclear, but are in line with previous studies reporting a disequilibrium in the distribution of resistance conferring mutations in different geographical areas or among strains of different genotypes [36–38]. Our findings confirm that the performance of molecular assays that only target particular mutations can be influenced by the differential prevalence of particular mutations in a given geographical area. Therefore, strain diversity needs to be considered and investigated before the new implementation of molecular assays in a study region.
Among EMB resistant isolates, the most frequent mutation affected codon 306 (Met/Ile) of the embB gene. This mutation has been described in various studies as the main mutation mediating resistance to EMB [14, 39]. The mutation at codon 497 has also been previously described in clinical isolates . Moreover, both mutations have been shown to confer resistance by transfer in a wild type genetic background using allelic exchange experiments . However, the authors conclude that single mutations only modestly increase resistance to EMB and additional so far unknown mutations are necessary to cause high-level resistance.
The mutations at codon 332 and 1002 determined here have not been described before. The impact of these changes has to be investigated in further studies. In four resistant strains no mutations were detected in the embB region analyzed. Additional analysis of the embC and embA genes revealed a mutation in embC [Val981Leu (gtg/ctg)] in one strain, belonging to the Haarlem genotype. However, this mutation has been described earlier as being specific for the Haarlem genotype and is not associated with resistance to EMB . As mentioned above, other so far unknown resistance mediating mechanisms are probably responsible for the resistance phenotype in these four strains.
Mutations or insertions in the pncA gene are known to mediate PZA resistance [42, 43], as observed in our study. No hotspot region has been determined, since polymorphisms occur throughout the complete gene. However, according to our data some specific mutations do obviously not mediate resistance that is detectable by applying standard critical concentrations. In the panel of strains analyzed, two susceptible strains carry a SNP at codon 47 and one displays a mutation at codon 96. PZA-MIC determination for these strains revealed slightly elevated values for the strains carrying the mutation at codon 47 (25.0 μg/ml) compared to the H37Rv control. In a recent study it has been shown that the site of the mutation is leading to varying efficiencies of the mutated pyrazinamidase mediating a wide range of resistance levels from low to high . As the mutation at codon 47 has previously been described by Juréen and co-workers  in PZA resistant strains, further investigations are necessary to determine if additional mutations in other parts of the genome might be responsible for the observed low-level resistance in the strains analyzed in this study. Out of all PZA resistant strains three carried the pncA wild type sequence. This indicates that further mutations in as yet unidentified genes are also important for mediating PZA resistance.