Mutation rate and the emergence of drug resistance in Mycobacterium tuberculosis

The emergence and spread of multidrug-resistant strains of Mycobacterium tuberculosis remains a majorconcern of tuberculosis control programmes worldwide, as treatment depends on low-efﬁcacy, toxic compounds that often lead to poor outcomes. M. tuberculosis develops drug resistance exclusively through chromosomal mutations, in particular single-nucleotide polymorphisms. Moreover, in laboratoryassaysthe organism exhibits a spontaneous mutation rate that is at the lowerend of the bacterial spectrum. Despite this, whole-genome sequencing technology has identiﬁed unexpected genetic diversity among clinical M. tuberculosis populations. This suggests that the mycobacterial mutation rate may be modulated within the host and, in turn, implies a potential role for constitutive and/or transient mutator strains in adaptive evolution. It also raises the possibility that environmental factors might act as key mutagens during M. tuberculosis infection. Here we consider the elements that might inﬂuence the mycobacterial mutation rate in vivo and evaluate the potential roles of constitutive and tran-sientmutatorstatesinthegenerationofdrugresistancemutations.Inaddition,weidentifykeyresearchquestionsthatwillinﬂuencefutureeffortstodevelopnoveltherapeuticstrategiesforadiseasethatcontinuestoimposeasigniﬁcantglobalhealthburden. tuberculosis strains identiﬁcation immunefactorsthatcouldbemanipulated, aswellasinformthe design of new drug regimens for the treatment of M. tuberculosis


Introduction
Mycobacterium tuberculosis is the causative agent of tuberculosis (TB) in humans. TB remains a massive global health problem and the emergence of multidrug-resistant (MDR) (resistant to rifampicin and isoniazid) and extensively drug-resistant (XDR) (MDR with added resistance to a fluoroquinolone and an injectable secondline agent) strains, coupled with the lack of new effective drugs to combat the disease, is of major concern. 1 Recently, totally drug-resistant (TDR) cases have been described, but WHO has yet to define this form of drug-resistant TB. 2 Treatment of MDR-TB and XDR-TB requires prolonged chemotherapy with expensive and less efficient drugs that often lead to undesirable side effects. 3 TDR-TB might be untreatable with currently available drugs. For this reason, a greater understanding of how drug resistance arises and might be prevented is of critical importance.
In M. tuberculosis, genetically encoded drug resistance arises exclusively through chromosomal mutations, 4,5 the majority of which are single-nucleotide polymorphisms (SNPs). 6 In general, mutations occur spontaneously 7 as a consequence of errors that arise during DNA replication at a rate of one mutation per 10 000-100000 bp per round of replication. 8 Correction by proof-reading exonucleases reduces the rate 2 log-fold, 8 with post-replicative DNA mismatch repair ensuring a further 3 log-fold reduction. 9 The bacterial mutation rate is, in turn, defined as the probability of a mutation occurring per cell division, and so is determined per bacterium per generation. 10 The calculated rates for the in vitro evolution of resistance to various antibiotics are shown for M. tuberculosis and related mycobacteria in Table 1.
In this review, we discuss the different factors that might contribute to the emergence of drug resistance mutations in M. tuberculosis during host infection. In particular, we evaluate whether spontaneous errors during DNA replication and repair are sufficient to drive the evolution of drug-resistant M. tuberculosis strains within the host or if additional factors contribute to the emergence of these mutant strains. Understanding the mechanisms and pathways that influence the mutation rate might identify targets for novel agents designed to prevent the development of drug resistance 21 or to potentiate the activity of existing antituberculars. 22 In addition, it could aid the identification of environmental or host immune factors that could be manipulated, 23 as well as inform the design of new drug regimens for the treatment of M. tuberculosis.

M. tuberculosis mutates at a low rate in vitro
M. tuberculosis does not exhibit an elevated mutation rate relative to most other bacteria; 11 under in vitro conditions, it is estimated that the organism makes a mutational error at two bases in every 10 000 genomes copied. 16 This is despite the fact that the DNA repair complement of M. tuberculosis does not include homologues of known mismatch repair (MMR) enzymes. 24 Escherichia coli mutants lacking MMR components are considered strong heritable mutator strains; 25 however, this does not apply to M. tuberculosis. In this review, the term 'mutator' is applied both to a strain that exhibits a higher mutation rate compared with its progenitor and to a gene that, when mutated, confers a higher mutation rate on the organism. Instead, it has been proposed that MMR deficiency may, in fact, exert selective pressure on M. tuberculosis, resulting in a more stable genome over time, 26 as is suggested by the lower than expected degree of polymorphism amongst simple sequence repeats (SSRs) in the M. tuberculosis genome. SSRs (also called microsatellites) consist of repeats of 1 -6 bp units and are particularly prone to polymerase slippage, which can lead to frameshift mutations. 26 -28 Long tracts comprising more than seven repeats of the microsatellite unit are more prone to slippage than short repeats, and may contribute to genome instability. A lack of MMR would, therefore, be expected to result in an elevated number of polymorphic SSRs. 28 However, M. tuberculosis contains fewer long sequence repeats than might be predicted on the assumption that nucleotides are randomly distributed, 26 and in comparison with E. coli. 27 Consistent with the inference that the M. tuberculosis genome is under selective pressure to limit features that might contribute to inherent variability, Machowski et al. 29 showed that PGRS repeats-found among PE_PGRS members of the highly conserved PE protein family, which is characterized by the presence of proline -glutamate residues in the Units are per bacterium per generation. c Mutation frequency, the proportion of mutants within a bacterial population at a given timepoint. No data were available on mutation rate. Review N-terminal region-are not associated with a higher rate of mutation relative to base substitutions.
The absence of MMR in M. tuberculosis might nevertheless contribute to genetic variation. Amongst other functions, MMR is known to prevent recombination between non-identical sequences 30 during repair of double-stranded DNA breaks in other bacterial systems. In Mycobacterium smegmatis, abrogation of nucleotide excision repair (NER) function results in an increased number of gene conversion events arising from mismatch errors during homologous recombination, suggesting that NER may (partially) compensate for the loss of MMR in mycobacteria. 31 However, the fact that NER does not distinguish between the parent and daughter strands, whereas MMR does, raises the possibility that mutation fixation-rather than avoidance-might be a common outcome. 24 The lack of an MMR system might also promote exchange of divergent DNA sequences, thus driving genome evolution, 31 since evidence suggests that recombination can occur in M. smegmatis when divergent loci are flanked by sequences of perfect homology. 31 The mechanisms ensuring genomic stability in M. tuberculosis therefore represent an important area for future research.
Generation of drug resistance-conferring mutations in M. tuberculosis during infection As noted above, M. tuberculosis does not display an elevated in vitro mutation rate compared with other bacteria. For this reason, drug-resistant mutants of M. tuberculosis are expected to be rare, and are predicted to arise as a consequence of spontaneous errors in DNA replication that are subsequently selected under applied drug pressure. Multiple factors could influence the rate of selection of drug-resistant mutants in the host, including the relative fitness of individual mutants, 32,33 patient compliance with prescribed drug regimens, 34 pharmacokinetic variability amongst patients, 35 spatial heterogeneity in drug distribution 36,37 and the size of the infecting bacterial population. 38 The rate of selection of a particular mutation should, however, not be confused with the basal mutation rate. The fundamental principle remains; errors in DNA replication must occur in order to generate the base substitutions and other types of mutations that lead to drug resistance. It is not entirely clear, though, whether a relatively low mutation rate is sufficient to account for the elevated rates of acquired drug resistance observed clinically. Is the mutation rate within the host modulated in some way to bring about sufficient diversity from which drug resistance can be selected?
In a pioneering study, Ford et al. 16 applied whole-genome sequencing (WGS) technology to detect the mutational events that arose in M. tuberculosis during experimental infection of nonhuman primates. Using a flexible range of predicted mycobacterial replication rates, the authors estimated the rate at which the observed mutations must have arisen in vivo. Unexpectedly, this estimated in vivo mutation rate was very similar to that which was inferred from in vitro fluctuation analyses, and, moreover, did not differ for active versus latently infected animals. In other words, M. tuberculosis does not appear to enter into a hypermutable state within the host. Work by Saunders et al. 39 reinforced the idea that drug resistance mutations are rare. In this case, the authors sequenced the genomes of serial isolates obtained from a TB patient over 12 years, a period that correlated with the emergence of strains resistant to isoniazid and, subsequently, rifampicin. WGS analysis suggested that only two SNPs-in katG and rpoB-differentiated the drug-resistant isolate from the susceptible strain, indicating that the mutation rate in the host is very low. The reference genome employed in this study was assembled from pooled data from all the serial isolates. It is possible, therefore, that other SNPs present at low frequencies were not detected. In addition, the authors estimated the bacillary population to be in the order of 10 13 cells/lung, which far exceeds predicted values for M. tuberculosis infections. 40 In contrast, Sun et al. 41 utilized more sensitive WGS technology to track genomic changes in serial sputum samples obtained from three patients over the course of anti-TB treatment. In this case, the authors reported a high degree of diversity in the serial clinical specimens, an observation that is consistent with the idea that mutation rates in vivo might be higher than previously proposed; for example, between 8 and 41 SNPs arose during treatment in each sample, and as many as 34 SNPs were unique to a single sample. Interestingly, the majority of the SNPs were detected at frequencies below 20% in each sample, indicating that the sputum bacillary population was characterized by significant microheterogeneity at each timepoint. These observations are supported by other studies that employed Sanger sequencing to investigate the resistance-determining regions of specific target genes. For example, serial M. tuberculosis isolates obtained from patients over a period of 12 years revealed a diversity of mutations within rpsL. 42,43 Similarly, Mariam et al. 43 detected different rpsL and rrs mutations in serial isolates obtained from a single patient. These authors also reported transient co-existence of some mutations as well as successive clonal sweeps of other mutations, suggesting a dynamic interplay between the genetic variability of the infecting bacillus and the selective pressures associated with host infection.
The extent to which the sputum bacillary population can be considered representative of the total infecting mycobacterial population is not known. For this reason, the above studies might be limited by their dependence on organisms obtained from sputum. In a key study, bacilli isolated from discrete pulmonary lesions in patients with chronic TB were shown to possess different drug resistance alleles. 44 It seems likely, therefore, that the microenvironments represented by each lesion may contribute to the heterogeneity observed within the TB population in the host. However, WGS analyses in the non-human primate model revealed significant heterogeneity among M. tuberculosis bacilli isolated within single lesions. 16 The mycobacterial replication rate The levels of genetic diversity identified in the studies above imply that M. tuberculosis might have an elevated mutation rate within the host compared with that calculated in vitro. However, it is important to remember that the fixation of spontaneous mutations occurs during replication; the mutation rate is, therefore, dependent on the rate of replication. This raises the possibility that the replication rate may be higher in the host than previously thought. Until recently, the prevailing dogma held that the rate of acquisition of dual resistance (required for MDR-TB) was the product of the individual mutation rates for rifampicin and isoniazid; 45,46 that is, in the order of 10 216 . This requires that, for the evolution of MDR strains, a total population of at least 10 16 bacilli must Review be present within any infected individual prior to the initiation of treatment. A recently proposed mathematical model 38 suggests otherwise. In developing the model, Colijn et al. 38 allowed for the possibility that a single drug-resistant mutant may arise early following infection, and could replicate to a large enough population from which the probability of the emergence of a second drug-resistance mutation would not be so low. Critically, they also considered the possibility that a greater number of replication events might occur during the initial infection period than has previously been assumed from in vitro estimates, an idea that is supported by recent evidence from experimental infections of macrophages 47 and mice. 48 Applying this set of revised parameters, Colijn et al. 38 showed that the likelihood of emergence of drug resistance prior to initiation of anti-TB therapy is much higher than previously expected, even when basing the calculation on published in vitro mutation rates for specific TB drugs.
Human TB can be usefully divided into three broad phases: active, chronic and reactivated TB. 49 Based on data from animal models of infection, it is assumed that the total number of M. tuberculosis bacilli remains stable during chronic TB and that these are in a state of slow replication or non-replication. 50,51 Recent evidence suggests, however, that the apparently stable bacillary numbers conceal a population that is cycling continuously between active replication and death. 48 Utilizing a 'clock' plasmid that is lost from daughter cells during division, Gill et al. 48 demonstrated that, during chronic infection in the mouse model, a balance is established between bacillary replication and death. This implies that the replication rate may be higher than previously thought during chronic TB in humans and is consistent with an emerging appreciation of latent TB as a continuum of mycobacterial growth states. 52,53 To date, the clock plasmid has only been assessed in the mouse model, 48 which fails to recapitulate key features of human disease. 54,55 It will be interesting, therefore, to see whether an alternative model, such as the non-human primate, 16,56 yields similar results.
Is the mycobacterial mutation rate altered during host infection, and might this explain the observed rates of acquired drug resistance? From the above discussion, it is clear that additional research is required to address this question. There are many factors that might influence the mutation rate; these are considered below and summarized in Table 2, together with the corresponding knowledge gaps.
A role for constitutive mutator strains?
Individual cells within an apparently homogeneous bacterial population are unlikely to share an identical mutation rate.

JAC
Spontaneous mutations in genes coding for DNA metabolic proteins can lead to the emergence of mutator strains that carry a short-term selective advantage owing to their capacity to produce a higher number of adaptive mutations. 58 In some cases, these mutators are maintained owing to the linkage of the mutator allele with other beneficial mutations, but only as long as the fitness gain counterbalances (or exceeds) the cost inherent in the increased risk of generating deleterious mutations. For this reason, mutations associated with elevated mutation rates are expected to be lost from bacterial populations over time. The adaptive mutations may, however, become fixed if the mutator alleles revert to wild-type, or if suppressor mutations occur. 58 There has been considerable interest in the potential association of mutator alleles with the emergence of drug resistance in M. tuberculosis. 59 For example, mutations in mutT4 and mutT2 appeared to define certain sublineages of Beijing isolates. 60 Beijing strains have been linked with the emergence of drug resistance in some settings; 61 however, the apparent association of specific strains with drug resistance could reflect transmission, 62 a possibility that is consistent with the fact that no direct link has been demonstrated between the reported DNA repair mutations and the emergence of drug resistance. 63 -65 Instead, biochemical evidence suggests that the observed mutations might not impact DNA repair function at all. 66 Moreover, although mutT2 and mutT4 are annotated as DNA repair-type Nudix hydrolases, the substrate specificity of these enzymes in vitro suggests a role separate from DNA repair, 66,67 perhaps in regulating nucleotide availability. 66 Consistent with this idea, MutT2 66,68 and MutT4 67 do not function as antimutator proteins. It is possible, however, that the observed mutations confer other phenotypes that indirectly affect the acquisition of drug resistance, but not via a DNA repair pathway-e.g. the G58R mutation in mutT2 has been proposed to increase the replication rate in macrophages. 66 The evidence for the association of Beijing strains with an elevated mutation rate is mixed. 13,69,70 In one study, various Beijing strains as well as non-Beijing strains were subjected to fluctuation analyses to determine their respective in vitro mutation rates on the assumption that mutator strains would exhibit at least a 10-fold higher mutation rate than corresponding 'non-mutator' strains; however, the authors observed no difference between Beijing and non-Beijing isolates. 13 It has subsequently been suggested that the Beijing strains employed in this analysis were not adequately defined 67 specifically in terms of the presence of putative mutator alleles 71 and so were not true mutators. This limitation has not been addressed directly. It is also possible that Beijing strains are not characterized by elevated mutation rates in vitro, but are instead dependent on specific factors in vivosuch as increased oxidative stress and/or exposure to antitubercular drugs-to elicit an elevated mutation rate. In support of this idea, a recent study 69 demonstrated that Beijing strains were associated with significantly higher mutation frequencies compared with the East African Indian (EAI) strains when cultured in the presence of rifampicin. This effect was not observed with other antibiotics, implying that global mutation rates were not affected. The same strains were then applied in a mouse model of TB to determine the impact of treatment non-compliance on the acquisition of resistance in vivo. 70 Notably, drug-resistant mutants arose in mice infected with the Beijing strain following exposure to the most severe conditions of non-compliance, and did not emerge in mice infected with the EAI strain. However, an acknowledged limitation of this study was that it utilized Beijing strains that were likely to have been closely related, so the results might not be applicable to all Beijing strains. For this reason, it may be more useful to consider sublineages of Beijing strains. For example, studies from Japan and Taiwan reported that specific Beijing sublineages differed in their associations with drug-resistant TB. 62 Moreover, a comparative genomics analysis 57 of polymorphisms in 3R genes revealed that a specific Beijing strain had an accumulation of non-synonymous mutations in genes encoding DNA glycosylase family enzymes.
The potential association of Beijing strains with acquired drug resistance could be ascribed to factors other than mutation rate, such as better developed efflux systems, a higher rate of replication 66 or an increased ability to adapt to the fitness cost of resistance. 72 In the study by de Steenwinkel et al., 70 the resistant mutants selected under the model of non-compliance with isoniazid treatment carried no mutations in katG or inhA, the usual sites of isoniazid resistance. It seems, therefore, that Beijing strains might be prone to acquiring mutations in genes other than target genes, enabling them to better adapt to drug stress. 70 A compelling case for the phenomenon of strain-specific mutation rates was made recently by Ford et al., 73 who reported that M. tuberculosis strains from the East Asian lineage acquired drug resistance more rapidly than strains from the Euro-American lineage. Their results suggest that this is not due to an increased ability to adapt to antibiotic pressure or to an increased number of possible mutations. Instead, it appears that an elevated mutation rate in the absence of antibiotic pressure might contribute to the observed difference, although the mechanism(s) underlying the inferred difference in mutation rate remains to be determined.
There is also some evidence indicating that the presence of resistance-conferring mutations impacts the basal mutation rate. For example, a rifampicin-resistant mutant carrying an S522L mutation in rpoB was associated with a higher frequency of mutants than the progenitor, rifampicin-susceptible strain when selected on rifabutin. 74 In a follow-up study 75 that investigated the induction in rpoB mutants of recA and dnaE2-encoding the principal regulator and key mutagenic polymerase of the mycobacterial SOS response, respectively-expression of dnaE2 was moderately, but significantly, up-regulated in S522L, H526D and S531W rpoB mutants, with S522L showing the highest constitutive expression. This is an intriguing observation; however, given that DnaE2 depends on the activity of the imuA ′ -and imuB-encoded accessory factors-which, like dnaE2, are included in the mycobacterial SOS regulon 76 -the functional consequences of increased levels of DnaE2 alone remain unclear. In other work, the same authors demonstrated that an isoniazid-resistant Haarlem strain harbouring an S315T mutation in katG was also associated with an increased mutation rate. 18 Again, this is an intriguing observation, and suggests that other unidentified strain-dependent mutations could interact with drug resistance-conferring mutations to confer a mutator phenotype.

Transient mutagenesis
In contrast to constitutive mutator strains, a transient mutator phenotype results in a temporary increase in the mutation rate. There are a number of mechanisms by which this may occur: mistranslation of proteins, especially those involved in accurate Review replication and repair of DNA, 25 transcriptional mutagenesis 77 -82 and the up-regulation of error-prone DNA polymerases. 77,83 Consistent with the multiplicity of mechanisms that can generate this phenotype, it is likely that transient mutator strains are responsible for the majority of adaptive mutations in bacterial populations. 25,84 Transcriptional mutagenesis occurs when RNA polymerase bypasses DNA lesions, inserting incorrect nucleotides into the mRNA. 81 If one of the resulting mutant proteins confers an altered growth phenotype, it may in turn lead to DNA replication past the mutagenic lesion, which could lead to the fixation of the mutation. 81 Alternatively, the mRNA containing an error may encode a protein involved in DNA repair or replication, which could lead to a transient mutator phenotype. The transcriptionrepair coupling factor, Mfd, recruits the NER repair machinery to sites where RNA polymerase is stalled. 85 Mfd was shown to play a role in mutagenesis in Bacillus subtilis 78 and to mediate point mutations conferring fluoroquinolone resistance in Campylobacter jejuni. 79 Currently, nothing is known about transcriptional mutagenesis in M. tuberculosis, suggesting this as a potentially useful area of future research.
Mistranslation describes any error that affects the accurate translation of mRNA into protein. 86 A major mechanism by which this occurs is through mischarging of tRNAs, 87 which can happen when the tRNA gene is mutated 88 or when aminoacyl-tRNA synthetases are defective in their editing capabilities. 87 Mistranslation was associated with an elevated mutation rate under conditions of stress in Candida albicans, Acinetobacter baylyi 86 and E. coli. 25 To date, there are no published data on the impact of mistranslation on the mycobacterial mutation rate.
Perhaps the most intensively studied transient mutator system is the bacterial SOS response, which is induced by DNA damage. In most bacteria, the SOS response is regulated by the RecA/LexA system; during normal growth, SOS genes are negatively regulated by the LexA repressor protein. Following genotoxic damage, RecA protein binds to single-stranded DNA (ssDNA) at sites of DNA lesions and replicon arrest to form RecA -ssDNA filaments that can interact with LexA, stimulating its cleavage and de-repressing SOS regulon genes. Unlike bacteria such as E. coli, whose SOS regulons include multiple Y-family DNA polymerases capable of translesion synthesis ('TLS'), the DNA damage response in M. tuberculosis is limited to the dnaE1-and dnaE2-encoded catalytic (a) subunits of the C-family DNA polymerase III. 76,83,89,90 M. tuberculosis encodes two DinB-like Y-family polymerases; 15 however, neither is induced as part of the mycobacterial SOS response. Critically, deletion of dnaE2 has been shown to reduce the frequency of rifampicin resistance emergence in the mouse model, 83 thereby implicating the mutagenic cassette as a major mechanism driving adaptive evolution during chronic infection. It is tempting, therefore, to speculate that DnaE2-dependent transient mutagenesis functions in mycobacterial evolution; however, this requires further investigation.
Another low-fidelity repair system recently characterized in M. smegmatis, a model organism for M. tuberculosis, is that of non-homologous end joining (NHEJ). 91,92 Three proteins seem to be key to this process: Ku, LigD and LigC. 91 Even though NHEJ cannot strictly be considered a transient mutator system, it has the potential to elevate mutagenesis if up-regulated under certain conditions, 93,94 a possibility that requires further investigation.

Environmental mutagens
Antibiotics Antibiotics target essential functions and therefore impose a strong selective force for genetic resistance. At subinhibitory concentrations, antibiotics can also act as mutagens, driving the emergence of drug-resistant mutants, a phenomenon that has been demonstrated in vitro for different bacterial species and a variety of antibacterial classes. 95 -101 M. tuberculosis may encounter subinhibitory drug concentrations as a result of patient noncompliance, 102 poor absorption by the gastrointestinal tract, 103 poor penetration of-or activity in-certain parts of the lung 104 and host genetic factors that impact drug metabolism and clearance. 105 M. tuberculosis might also inadvertently be exposed to broad-spectrum antibiotics (e.g. fluoroquinolones) in patients undergoing treatment for other respiratory infections (e.g. community-acquired pneumonia). 106 Subinhibitory concentrations of ciprofloxacin, a fluoroquinolone whose mechanism of action results in the induction of the mycobacterial SOS response, 107 have been shown to elevate in vitro mutagenesis in related mycobacterial species. 19,108 Moreover, there is some evidence to suggest that fluoroquinolone use prior to diagnosis of TB may be associated with first-line drug resistance, 109 especially if multiple prescriptions have been given. 106 However, this topic is controversial owing to conflicting evidence. 110 It is likely, too, that antibiotics that do not act directly on DNA metabolic processes can lead to DNA damage. For example, multiple studies have reported that bactericidal antibiotics of very different classes-including b-lactams, fluoroquinolones and aminoglycosides-are united by a common mechanism of killing that results in the generation of hydroxyl radicals. 98,111,112 Although very recent studies have questioned the general applicability of the model, 113,114 there is some evidence implicating the antibiotic-dependent formation of reactive oxygen species (ROS) in an elevated mutation rate. 98 In M. tuberculosis, the production of ROS has been directly correlated with the ability to eliminate an antibiotic-tolerant bacillary population in vitro; 112 however, to our knowledge, the link between ROS and mutagenesis has not been explored. Cirz et al. 115 proposed an intriguing mechanism by which antibiotics other than fluoroquinolones could also lead to DNA damage; on the assumption that disruption of metabolism could alter ATP/ADP ratios, it was argued that depletion of ATP-an essential energy source for DNA gyrase function-may result in stalling of replication forks and, potentially, SOS induction. Again, this possibility requires further investigation.
Genome-wide expression studies demonstrated that fluoroquinolones induce up-regulation of SOS clusters in M. tuberculosis, 116 whereas inhibitors of translation do not. 116 Consistent with the prediction that isoniazid might potentially generate ROS, 117 genomewide studies have detected up-regulation of the SOS response 118 and recA 119 following isoniazid exposure. Similarly, the ligD gene, essential for mycobacterial NHEJ, 91 was also induced. 119 To date, however, the effect of subinhibitory concentrations of anti-TB antibiotics on the mutation rate of the organism has not been investigated in vitro, or in patients developing drug-resistant TB. Although challenging, an analysis of the drug regimens or drug combinations that could lead to an increased rate of emergence of drug resistance in M. tuberculosis-infected patients could, therefore, be of Review 297 JAC great value in estimating the risk of acquisition of drug resistance during chemotherapy.

Antiretroviral drugs
It was previously suggested that antiretroviral drugs may affect the mutation rate of M. tuberculosis. 120 However, the evidence for an association between HIV status and M. tuberculosis drug resistance on an individual level is mixed. 121 One class of antiretroviral comprises nucleoside reverse transcriptase inhibitors (NRTIs), which have been associated with mutagenic effects in animal models and humans. 122 NRTIs are nucleoside analogues and may be genotoxic in bacteria owing to their ability to be incorporated into bacterial DNA, causing chain termination, 123 which itself may lead to the induction of the SOS response. 83,124 Alteration of nucleotide pools and conformational changes in DNA polymerase might similarly also lead to reduced replication fidelity. However, while NRTIs are known to elevate the mutation rate of the HIV genome, 125 there are no published studies that have assessed their potential effects on M. tuberculosis.

TB pathogenesis and the host environment
The discrepancies observed between apparent mutation rates (and frequencies) in vivo and those measured in vitro suggest that the factors modulating the mutation rate during host infection might be very different from those operating under laboratory conditions. For example, a Beijing strain was associated with an elevated in vitro frequency of rifampicin resistance-but not isoniazid resistance-compared with EAI strains; 69 yet the converse was observed in a mouse model of infection. 70 Similarly, Bergval et al. 12 detected different types of isoniazid resistance-conferring mutations in vitro versus those generally observed clinically. Therefore, determining the mutagenic stimuli and DNA-damaging events encountered in vivo may be critical to the identification of those factors that drive adaptive evolution and the emergence of drug resistance during host infection.
M. tuberculosis encounters many DNA-damaging influences during the infectious lifecycle (reviewed by Gorna et al. 93 ). For example, during aerosol transportation, bacilli are likely to encounter UV radiation and desiccation. 93 UV radiation induces expression of lexA, dnaE2 and recA, 83 while M. smegmatis mutants defective in ku and ligD are susceptible to desiccation. 94 Expression of Ku was also higher in granulomas than in broth. 126 In addition, the mutT2 gene, which may be involved in hydrolysis of 8-oxo-dGTP, 67 was down-regulated in response to lung surfactant. 127 This could lead to an increase in 8-oxo-dGTP in nucleotide pools and elevated mutagenesis. 128 Oxidative stress is a major mutagenic influence encountered by M. tuberculosis in the host. Macrophages, as an antibacterial defence mechanism, produce ROS and reactive nitrogen intermediates (RNI). 129 These reactive species interact with nucleotides, resulting in chemical modifications that can lead to base mispairing and DNA damage. 128,130 -132 DNA damage can, in turn, lead to an up-regulation of error-prone DNA repair. 89 M. tuberculosis is well armed against oxidative stress; moreover, mechanisms to detoxify ROS and RNI are essential for the survival of this pathogen in the host. 129 The high levels of redundancy in these and specific DNA repair 21 pathways might, therefore, indicate a critical role in pathogenesis. 129 The high GC content of the genome also suggests that M. tuberculosis may be especially vulnerable to oxidative damage. 133,134 Similarly, observations from the non-human primate model 16 suggest a role for ROS in the spectrum of mutations observed.
In contrast, analyses by O'Sullivan et al. 135 of mutations in rpoB and pncA in clinical isolates suggest a pattern inconsistent with oxidative stress as the major driver of bacillary evolution. Their study 135 was based on the assumption that resistance to isoniazid most often occurs before resistance to rifampicin (conferred by mutations in rpoB) or pyrazinamide (conferred by mutations in pncA). Isoniazid resistance is conferred by mutations in katG, encoding an enzyme that functions in the oxidative stress response. 136 Mutations in katG, including the clinically dominant S315T mutation, have been shown to have a detrimental effect on the activity of this enzyme. 137 Therefore, katG mutants are expected to undergo mutagenesis driven by oxidative stress. However, O'Sullivan et al. 135 did not observe the expected overall increase in G to A or C to T mutations in the genes analysed.
It is possible, though, that fitness plays a greater role in defining the mutation spectra of resistance genes in clinical isolates. This interpretation is consistent with evidence from another study that investigated the effect of lowering the pH on mutational events during in vitro culture of M. tuberculosis. 138 Although the decrease in pH impacted the mutation frequency only slightly, there was a major effect on the mutation spectrum, in which rare, 'less-fit' mutations were observed. As discussed above, there is the possibility that less-fit mutants are associated with up-regulated expression of dnaE2, 75 which might provide the opportunity for the development of compensatory mutations, a possibility that requires further investigation.

Smoking and air pollution
An association between drug resistance and smoking or tobacco use has been observed in some cases. 139,140 Cigarette smoke contains mutagenic chemicals 141,142 and smoking and environmental pollutants could also alter the redox balance, in turn affecting the mutation rate. 143 A compound typically generated during combustion, 1,6-dinitropyrene ('1,6-DNP'), increased the incidence of drug resistance in Pseudomonas aeruginosa. 144 Whether this compound, and other similar compounds, are important mutagens in the development of drug resistance in M. tuberculosis remains to be determined.

Conclusions
M. tuberculosis has a low mutation rate in vitro, yet seems capable of generating surprisingly high levels of genomic diversity within the host. However, determining whether the mutation rate is modulated during host infection is complicated by the fact that it is difficult to distinguish between the rate of selection for drug resistance and the rate at which the genotypic diversity for selection is itself generated. Moreover, the extent to which the mycobacterial replication rate itself is modulated during host infection remains unknown. Therefore, it is not clear whether the mutation rate of M. tuberculosis is higher in the host than in vitro or whether there are factors specific to the in vivo environment that might drive mutagenesis. Knowledge of the contribution made by transient and constitutive mutator strains to the mutation rate and, therefore, the evolution of drug resistance in M. tuberculosis represents an Review important area for future research as part of continued efforts to inhibit-and prevent-the acquisition of drug resistance.

Funding
This work was funded by a grant from the National Research Foundation (to D. F. W.). M. M. has financial support from the National Research Foundation and the Ernst and Ethel Eriksen Trust.

Transparency declarations
None to declare.