In this study a single M. tuberculosis isolate (R190), with a S522L mutation, was detected that was resistant to RIF (MIC 32 μg/ml) but remained sensitive to RFB (MIC <0.8 μg/ml) (Table 3). Specific mutations in codon 522 have previously been shown to result in only low level RFB resistance from clinical isolates of MTB [7, 9]. The distribution of second round RFB mutations from R190 was strikingly different from that obtained from the parent strain (Table 2) and the frequency of resistant mutants increased > 10 fold when measured on two separate occasions. Additional second round mutations in the rpoB gene were identified in 24 of 26 (92%) of the R190 second round RFB resistant mutants (Table 2), only six of which (23%) were similar to those seen in the first round selection (either in codon 526 or 531). A single mutant with a change in codon 512 at the beginning of cluster I was also identified. Mutations in codon 512 have only been reported previously in association with additional mutations  as in this study (S512G + S522L). The remaining mutations were outside cluster 1, codons 144, 146, 148, and just before cluster I in codon 505 (Table 2).
This dramatic change in mutations identified indicates either, the range of viable SNPs resulting in high-level RFB resistance was significantly altered by the presence of the S522L mutation, or the range of spontaneous mutations occurring in this strain changed. Thus, it is possible, for example, that the mutations in codon 505 occurred in both cases but alone would not have resulted in resistance or in the absence of the S522L mutation may be lethal. Interestingly, these results could also be explained by a change in both the frequency and spectrum of spontaneous mutations after the first round of selection, ie. the rate of spontaneous mutations in codons 505, 512, 144, 146, and 148, has increased after the first round of selection. As 9 different codon changes were identified among the 26 isolates tested (Table 2), from two independent experiments (drug concentrations), a single spontaneous mutant in an early generation of this culture (a "Jackpot" mutation)  cannot explain this result.
The distribution of in vitro first round spontaneous mutations in M. tuberculosis has previously been reported  and 7 of the 9 codon changes identified were also seen in this study. The most striking difference between this data and the first round selection data presented here is the higher frequency of the C>T H526Y mutation in our study with both RIF and RFB. Some variation between random selections of mutants would be expected and methodological differences, notably the use of a different bacterial strain [5, 12], probably contributed to this effect.
Mutations in codons 526 and 531 predominate in most of the published studies but there is also some indication that certain strains may be prone to develop specific mutations [4, 12] and marked differences in the distribution of mutations have been observed in different geographical locations [4, 13]. Our data from the first round selection (Table 1) suggest that for the strain and conditions we used RFB may be more likely to select for C>T H526Y mutations than rifampicin.
All first round mutations identified in this study have been reported previously from clinical isolates [4, 5, 8–10, 14] except the one triple mutant identified (table 1). The second round mutations seen in codons 505, 512, 144, and 148 present in addition to the S522L mutation (Table 2), have to our knowledge not been reported previously from in vitro or clinical isolates. Although, it should be noted that these mutations lie outside the 81 bp hotspot region in a region of the rpoB gene that has been subjected to much less extensive investigation.
The possibility of an MTB strain with altered or raised mutation rate is important. Selection of a mutator phenotype is recognised as a consequence of antibiotic challenge in many bacterial species [11, 15]. The selection of strains with increased mutation rates will result in a greater chance of acquiring resistance to other drugs but may also impact on the pathogenicity of the strain. It has been reported that many MDR-MTB strains have, at least initially, reduced pathogenicity  and mutations in three putative mutator genes as well as evidence for a reversion back to a more pathogenic non-mutator state after the acquisition of drug resistance has been reported in W-Beijing strains . However, some rpoB mutations are associated with only a modest decrease in in vitro fitness . Interestingly, induction of the proposed MTB error prone DNA repair enzyme was associated with survival of the bacteria in vivo , so the effect of a mutator phenotype on pathogenicity is difficult to predict .
In conclusion, the presence of a different spectrum of secondary rifabutin mutations implies that either, the mutation rate of individual mutations has changed, due to a defect in DNA repair or replication, or that additional spontaneous mutations are viable (and lead to resistance) in the presence of the S522L mutation . A further consequence of this observation is that a proportion clinical isolates with mutations in cluster 1 of the rpoB gene associated with rifampicin resistance only may in fact be resistant to rifabutin as a consequence of addition mutations in other regions of this gene. We believe further study is warranted, of this and similar strains which should include generating mutants to other antimicrobials and measuring mutation rates [19, 20], allowing the contribution of each of these possible explanations to be explored. The details of how resistance mutations arise would be valuable when formulating standard treatment regimens with the aim of minimising the emergence of resistance in treated populations .