Insights into the processes that drive the evolution of drug resistance in Mycobacterium tuberculosis

Abstract At present, the successful transmission of drug‐resistant Mycobacterium tuberculosis, including multidrug‐resistant (MDR) and extensively drug‐resistant (XDR) strains, in human populations, threatens tuberculosis control worldwide. Differently from many other bacteria, M. tuberculosis drug resistance is acquired mainly through mutations in specific drug resistance‐associated genes. The panel of mutations is highly diverse, but depends on the affected gene and M. tuberculosis genetic background. The variety of genetic profiles observed in drug‐resistant clinical isolates underlines different evolutionary trajectories towards multiple drug resistance, although some mutation patterns are prominent. This review discusses the intrinsic processes that may influence drug resistance evolution in M. tuberculosis, such as mutation rate, drug resistance‐associated mutations, fitness cost, compensatory mutations and epistasis. This knowledge should help to better predict the risk of emergence of highly resistant M. tuberculosis strains and to develop new tools and strategies to limit the development and spread of MDR and XDR strains.

As example of intrinsic mechanisms, epistasis (Box 1), which plays an important role in the evolution of organisms in general, is also known to drive the evolution of antibiotic resistance (Lehner, 2011). Epistasis can occur between mutations in the same gene or in different genes and can lead to negative or positive effects (Box 1) (Lehner, 2011;Wong, 2017). This mechanism may generate the combination of a set of alleles from different loci, also called linkage disequilibrium (Box 1). The spread of these sets of co-adapted alleles in the population is then favoured by the clonal reproductive mode of M. tuberculosis.
Regarding the drug resistance, many studies underline that epistatic interactions can occur between different drug resistance mutations, between drug resistance mutations and compensatory mutations and/or the genetic background of the organism Lehner, 2011;Trindade et al., 2009;Wong, 2017).
In this review, we focus on the intrinsic factors influencing the drug resistance evolution in M. tuberculosis, particularly the mutation rate, drug resistance-associated mutations, fitness cost of resistance mutations, compensatory mutations and epistasis (Box 1). An understanding of the role of these intrinsic factors is essential to get insights into the evolutionary trajectories of drug resistance in M. tuberculosis and to help identifying the best strategies to control the emergence and spread of highly drugresistant strains.

| MUTATI ON R ATE AND DRUG RE S IS TAN CE ACQUIS ITION
Mycobacterium tuberculosis is characterized by a low mutation rate (about 2 × 10 −10 mutations/bp/generation) (Ford et al., 2011), with an estimated evolutionary rate of 0.4-0.5 single nucleotide polymorphisms (SNPs)/genome/year and a divergence rarely higher than five SNPs in 3 years Walker et al., 2013).
Despite this low mutation rate, the number of drug resistant, especially MDR and XDR TB cases, due to the acquisition of mutations, is progressively increasing worldwide.
Besides innate drug resistance mechanisms (for instance, the specific characteristics of the cell envelope of M. tuberculosis and the active drug efflux mechanism) (Sarathy, Dartois, & Lee, 2012), chromosomal mutations are the major mechanism of drug resistance acquisition in M. tuberculosis (Table 1) (Sandgren et al., 2009;Zhang & Yew, 2015). The rate for evolution of drug resistance to major firstand second-line drugs ranges from 10 −6 to 10 −10 mutations/bacterial cell/generation (McGrath, Gey van Pittius, van Helden, Warren, & Warner, 2014). This rate might also be affected by the drug concentration in the medium, the drug resistance profile of the strain and its genetic background (Ford et al., 2013;McGrath et al., 2014).
As the genes responsible for resistance to the various anti-TB drugs are generally not functionally related, the risk of emergence of spontaneous double, triple and quadruple drug-resistant mutants is theoretically extremely low, ranging from about 10 −14 mutants (for isoniazid and rifampicin) to 10 −25 mutants per population (for isoniazid, rifampicin, ethambutol and pyrazinamide). Furthermore, clinical data estimated that the population size in active pulmonary disease ranges between 10 7 and 10 10 bacilli (Nachega & Chaisson, 2003), thus the risk of spontaneous drug resistance-associated mutations should be very low. In addition, drug resistance-associated mutations may impose a fitness cost because they target essential cell biological functions (Melnyk, Wong, & Kassen, 2015). Therefore, in theory the chance of drug resistance acquisition should be negligible when the four effective first-line drugs are used in combination. However, by mathematical modelling, Colijn, Cohen, Ganesh, & Murray (2011) estimated that the probability of acquisition of resistance to both isoniazid and rifampicin is as high as 10 −4 to 10 −5 mutants/bacterial population. Similar to that, in a clinical study, Gao et al. (2016) found that 3.7% (62/1671) of pan-susceptible clinical isolates (susceptible to isoniazid, rifampicin, streptomycin and ethambutol) acquired different resistance patterns during the standard short-course chemotherapy according to the Directly Observed Treatment (DOT) guidelines. Among the 62 strains with acquired drug resistance, approximately 10% were resistant to four drugs, 22.6% to three drugs, 21% to two drugs and the remaining 46.8% were mono-drug resistant. These data underline that multiple drug resistance acquisition emerges at higher rate under strong drug selection pressure than theoretically predicted.
Regarding the genetic background, Ford et al. (2013) demonstrated that overall, the mutation rates for drug resistance acquisition are higher in M. tuberculosis lineage 2 (East Asia, mainly Beijing strains) than in lineage 4 (Euro-American). In addition, the authors also demonstrated that the risk of de novo MDR acquisition before treatment is higher (approximately 22-fold) in macaques infected with M. tuberculosis strains of lineage 2 than in animals infected with lineage 4 strains. These data are consistent with the high drug resistance potential of lineage 2 observed in many epidemiological studies (Casali et al., 2014;Merker et al., 2015). Besides the effect of genetic background, Ford et al. (2013) also showed that differences in target size (defined as the number of resistance-associated mutations) contribute to the two-to thirty-five-fold differences in rifampicin resistance rates that they have measured in their samples.
It is worth noting that extrinsic factors such as the economic and social situation of individuals or populations, the major political events (e.g., the fall of the Soviet Union) and the quality of TB control programmes also strongly influence the speed of drug resistance spread (Eldholm et al., 2016;Klopper et al., 2013;Müller, Chihota, et al., 2013). In vivo, the drug resistance acquisition also greatly varies depending on the location of bacterial populations in the body and the characteristics of the drugs (Kempker et al., 2015;Warner et al., 2015). Indeed, even under an optimal treatment, the bacteria can be exposed to suboptimal drug concentrations due to variable degrees of tissue penetration linked to the tissue and/or drug. In particular, the poor penetration ability of drugs in cavitary lesions is an important risk factor for the emergence of drug resistance in TB patients under long-term treatment course (Kempker et al., 2015).
Furthermore, using mathematical modelling, Moreno-Gamez et al. (2015) demonstrated that the imperfect drug penetrance leads to spatial mono-therapy and thus to a rapid evolution towards MDR.
In addition, variations in drug absorption in patients (pharmacokinetic variability) can be also a factor of emergence of MDR TB (Pasipanodya & Gumbo, 2011).

| INTR A-HOS T G ENE TI C VARIAB ILIT Y OF DRUG -RE S IS TANT P OPUL ATIONS
The intra-host evolution of bacterial resistance patterns is one of the key aspects of drug resistance emergence and spread (Eldholm et al., 2014;Meacci et al., 2005;Merker et al., 2013). The existence of genetically variable drug-resistant bacterial populations within a single patient is now acknowledged and could affect drug resistance evolution (Black et al., 2015;Eldholm et al., 2014;Müller, Borrell, et al., 2013;Shamputa et al., 2004). Meacci et al. investigated M. tu-berculosis population evolution in a noncompliant patient during more than 12 years of active disease. They identified the emergence of a MDR M. tuberculosis population from one single parental strain that was composed of discrete subpopulations with different drug resistance-associated gene variants (Meacci et al., 2005). This suggests that the intra-host bacterial population evolved over time by acquiring and accumulating different gene mutations associated with resistance to isoniazid, rifampicin and streptomycin. This led to the emergence, in one single patient, of different coexisting populations that harbour different drug susceptibility profiles. In another

BOX 1 Glossary
Acquired resistance: the ability of a bacterial population to resist the activity of a particular drug to which it was previously susceptible.
Biological fitness: the capability of an individual with a certain genotype to reproduce and survive in a competitive environment.
Clonal interference: competition between lineages ("clones") arising from different beneficial mutations to reach the fixation in asexual organisms.
Compensatory mutation: a second-site mutation acquired that arises after the acquisition of resistance mutation and that lessens or alleviates the fitness cost associated with the acquisition of the resistance-associated mutation.
Cross-resistance: the acquisition by a microbe of resistance to one drug through direct exposure and the gain, in parallel, of resistance to one or more other drugs to which it has not been exposed.
Epistasis: a form of interaction between genes or mutations that influences a phenotype. Epistasis occurs when the combined fitness effect of multiple alleles from same locus or different loci is different from the sum of the individual allele effects.
Extensively drug resistance (XDR): MDR (see below) Mycobacterium tuberculosis also resistant to at least one of the three-second-line injectable drugs (kanamycin, amikacin and capreomycin) and one fluoroquinolone (ofloxacin, levofloxacin, moxifloxacin or gatifloxacin).
Fitness cost of resistance mutation: a decrease in the relative fitness of the drug-resistant mutants in comparison with their drug-susceptible counterparts.
Genetic drift: random changes in the frequency of alleles over time usually in small populations.
Innate resistance: the innate ability of a bacterial species to resist activity of a particular drug.
Linkage disequilibrium: the nonrandom association of alleles at different loci.
Multidrug resistance (MDR): Mycobacterium tuberculosis resistant at least to isoniazid and rifampicin, the two more potent fist-line drugs.
Mutation rate: the number of mutations per nucleotide site (bp) per generation. It is worth noting that in the case of antibiotic resistance, the mutation rate is frequently defined as the rate of resistance acquisition (see below).
Negative epistasis: if the cost of double mutant in the absence of antimicrobial use is higher than the total cost of each resistance determinant on its own. Parallel evolution: evolution of a similar trait in closely related, independently evolving lineages.
Positive epistasis: if the cost of double mutant in the absence of antimicrobial use is smaller than the total cost of each resistance determinant on its own.
Purifying selection: selection reducing the frequency of deleterious alleles in a population. showed that drug resistance-associated mutations were acquired multiple times by individual clones, but only one expanded and replaced the other clones. In an ultimate manner, adaptive mutants are fixed and become dominant while others are lost by competition, referred as clonal interference (Box 1, Figure 1a) (Gerrish & Lenski, 1998). In addition, recent studies also demonstrated that M. tuberculosis populations can evolve measurably in response to selection pressures imposed by the environment within hosts (Lieberman et al., 2016;O'Neill, Mortimer, & Pepperell, 2015). This process can lead to the spatial structuring of the bacterial population within host (lungs) into related subpopulations that will evolve independently (parallel evolution, Box 1) as demonstrated previously (Gygli, Borrell, Trauner, & Gagneux, 2017). All these studies underline the constant genome evolution due to the acquisition of multiple independent mutations in the bacterial population despite the evolutionary bottleneck imposed by purifying selection (Box 1) due to drug selective pressure and clonal interference (Figure 1a).

| CHAR AC TERIS TI C S AND D IVER S IT Y OF DRUG RE S IS TAN CE-A SSOCIATED M UTATI O N S
The mutation frequency and type vary in function of different parameters, such as the geographic region, the drug resistance pattern and genetic background (Fenner et al., 2012 (Chang et al., 2015;Hughes & Andersson, 2015)). The figure represents the evolution of bacteria from wild type to drug-resistant mutants with fitness advantage and illustrates several different mechanisms of fitness increase. (a) Wild type can acquire different drug resistance-associated mutations in a same gene with high or low biological cost. The bacteria with low biological cost mutations will be selected under drug pressure by clonal interference and will propagate. (b) Under drug pressure, positive epistasis may favour the acquisition of compensatory mutations to alleviate the fitness cost exerted by certain drug resistance-associated mutations. (c) Driven by positive epistasis, drug-resistant mutants are likely to be more prone to accumulate drug resistance-associated mutations at higher frequencies (Nguyen, Nguyen, et al., 2017;Trindade et al., 2009)  hundreds of rpoB mutations have been described (not all were associated with rifampicin resistance), but more than 80% of rifampicin-resistant isolates display mutations in three codons rpoB531, 526 and 516 (Campbell et al., 2011;Lipin et al., 2007;Pozzi et al., 1999;Telenti et al., 1993). Similar to that, among the approximately 300 mutations found in the katG gene, the prevalence of katG S315T mutation can vary between 32% and 95% in isoniazid-resistant clinical isolates depending on the geographic regions and drug resistance patterns (Hazbon et al., 2006;Lipin et al., 2007;Mokrousov et al., 2002;Vilcheze & Jacobs, 2014).
impart a biological cost that leads to reduced fitness of the resistant strains in comparison with the sensitive ones, in the absence of antibiotics (Melnyk et al., 2015). According to that, several studies showed that drug-resistant M. tuberculosis mutants are characterized by reduced fitness (Billington, McHugh, & Gillespie, 1999;Gagneux, Long, et al., 2006;Mariam, Mengistu, Hoffner, & Andersson, 2004).
However, the extent of the biological cost depends on the mutation and the strain genetic background (Billington et al., 1999;Bottger, Springer, Pletschette, & Sander, 1998;Gagneux, Long, et al., 2006;Pym, Saint-Joanis, & Cole, 2002). Furthermore, in the absence of genetic drift (Box 1), drug-resistant mutations with low or no biological cost are more likely to be selected and maintained in the populations (Farhat et al., 2013;Osorio et al., 2013). For instance, the predominant mutations associated with high level of drug resistance and a low or no biological cost, such as katG S315T, rpoB S531L, rpsL K43R and gyrA D94G (conferring resistance to isoniazid, rifampicin, streptomycin and fluoroquinolones respectively), are more frequently found in clinical drug-resistant isolates (Billington et al., 1999;Bottger et al., 1998;Campbell et al., 2011;Casali et al., 2014;Gagneux, Burgos, et al., 2006;Gagneux, Long, et al., 2006;Mariam et al., 2004;Pym et al., 2002). Indeed, some of these resistance mutations do not reduce bacterial fitness in the absence of treatment Several works investigated the link between M. tuberculosis genetic background and the cost of drug resistance mutations. Gagneux, Long, et al. (2006) found differences in biological cost for the rifampicin resistance-associated rpoB H526D mutation between the lineages 2 and 4, while the rpoB S531L mutants showed similar costs in both lineages. The inhA-15 and katG S315T mutations are strongly associated with lineage 1 and modern lineages, respectively (Casali et al., 2014;Fenner et al., 2012;Gagneux, Burgos, et al., 2006). On the contrary, katG mutations other than katG S315T that likely abrogate enzyme activity result in high biological cost and seem to be more associated with lineage 2 (Gagneux, Burgos, et al., 2006). Thus, lineage 2 could be better adapted to compensate for the loss or reduced activity of this catalase-peroxidase enzyme in the context of isoniazid resistance. This hypothesis could also explain why the Beijing strains are generally strongly associated with resistance to isoniazid, regardless of the type of katG resistantassociated mutations and country (Fenner et al., 2012;Gagneux, Burgos, et al., 2006;Mokrousov et al., 2002;Ribeiro et al., 2014;van Soolingen et al., 2000).
Almost all laboratory-generated mutants with a rifampicin resistance-associated mutation in the RRDR of rpoB show a significant fitness deficit compared with their drug-susceptible ancestors when grown in the absence of this drug. Therefore, it was hypothesized that the fitness cost linked to rifampicin resistance could be reduced by compensatory mutations in clinical isolates (Billington et al., 1999;Comas et al., 2012;Mariam et al., 2004;de Vos et al., 2013).
Nonsynonymous mutations in the rpoA and rpoC genes that encode the α and β' subunits of RNA polymerase, respectively, could play the role of fitness-compensatory mutations in rifampicin-resistant rpoB mutants (Comas et al., 2012;de Vos et al., 2013). Indeed, it was reported that part of rifampicin-resistant isolates with a rpoB mutation also carry a nonsynonymous mutation in the rpoA or rpoC gene  (Comas et al., 2012;Li et al., 2016;Song et al., 2014).
In addition, clinical isolates that carry mutations in the RRDR of rpoB and also in rpoA/rpoC display higher competitive fitness in vitro and in vivo compared with laboratory-generated rifampicin-resistant mutants that carry only the same rpoB RRDR mutation and that belong to the same phylogenetic lineage (Brandis & Hughes, 2013;Comas et al., 2012;Song et al., 2014). These data suggest that mutations in the rpoA/rpoC genes are fitness-compensatory mutations that alleviate the costs of rpoB mutations. Furthermore, genetic reconstructions in a Salmonella model demonstrated that mutations not only in rpoA and rpoC, but also in rpoB are associated with higher growth rate (Brandis & Hughes, 2013;Brandis et al., 2012). In fact, many previous studies showed that rifampicin-resistant M. tuberculosis clinical isolates carry multiple (double, triple and quadruple) mutations in the rpoB gene (Bahrmand, Titov, Tasbiti, Yari, & Graviss, 2009;Casali et al., 2014;Nguyen, Nguyen, et al., 2017;Song et al., 2014). This could be the result of compensatory mechanisms to alleviate the fitness cost exerted by specific mutations (Brandis & Hughes, 2013;Brandis et al., 2012). It is worth noting that compensatory mutations are more commonly identified in the dominant MDR, pre-XDR and XDR clones in high MDR TB burden countries, suggesting that high drug-resistant mutants harbouring these mutations can be successfully transmitted in human populations (Casali et al., 2014;Cohen et al., 2015;Comas et al., 2012;Klopper et al., 2013;Li et al., 2016;de Vos et al., 2013).
These studies also showed that the rpoC mutation is significantly associated with the rpoB S531L mutation, suggesting an interaction between a fitness-compensatory mutation and a specific drug resistance-associated mutation (Casali et al., 2014;Li et al., 2016;de Vos et al., 2013). This may explain why rpoB S531L is the most common mutation observed in rifampicin-resistant clinical isolates and displays a low biological cost. The presence of compensatory mutations seems to be associated with Beijing strains, especially those harbouring the rpoB S531L variant (Casali et al., 2014;Li et al., 2016). Nevertheless, the frequency of compensatory mutations differs according to the Beijing genotype subclades, suggesting epistatic interactions (Box 1, see below) between drug resistance mutations, compensatory mutations and genetic background (Casali et al., 2014).

| Accumulation of drug resistance-associated mutations
The high diversity of mutations in M. tuberculosis suggests different evolutionary trajectories towards highly resistant genotypes, in response to various selection pressures. Nevertheless, for almost all drug resistance-associated genes, the predominance of some specific mutations, generally known to be associated with high level of resistance and low biological cost, has been described (see Table 2).
As a result, combinations of at least two specific mutations, such as rpoB531, katG315, rpsL43, embB306 and gyrA94, are favoured  Although the quality of the treatment undoubtedly plays a role in the emergence of particular drug resistance mutations, the strains with drug resistance-associated mutations seem to have higher propensity to accumulate other drug resistance mutations in the same gene or in different genes (Bahrmand et al., 2009;Jagielski et al., 2014;Nguyen, Nguyen, et al., 2017;Shen et al., 2007). For instance, the katG315, embB306 or pncA mutations are more frequently observed in MDR than in non-MDR isolates (Hazbon et al., 2006;Nguyen, Contamin, et al., 2017;Salvatore et al., 2016;Shen et al., 2007). All these data suggest a cumulative effect of mutations that are specifically associated with drug resistance and the occurrence of epistasis ( Figure 1c). Besides epistatic interactions, Chang et al. (2015) in their review of the causes of the excess of MDR infections suggest that the associated linkage selection can also be at the origin of the proliferation of multiple drug-resistant bacteria. This is especially the case for M. tuberculosis which follows a basic clonal evolution model (see above). This model generates a strong linkage disequilibrium that may favour the coexistence of two or more particular drug resistance-associated alleles.

| Epistasis between drug-resistant mutations
Although little is known about epistasis between drug resistance mutations in M. tuberculosis, a finding suggests that it could play an important role in the emergence and evolution of MDR and XDR M. tuberculosis strains   (Spies et al., 2013). This suggests that the interaction between these mutations may offer a fitness advantage to the double mutants. Indeed, these double mutations increase the fitness of drug-resistant E. coli and drive the evolution of MDR acquisition (Trindade et al., 2009 The authors demonstrated that 35% (6/17) of double mutants carrying specific rpoB and gyrA mutations associated with rifampicin and fluoroquinolone resistance have a significant higher fitness than the corresponding single drug-resistant mutants. In particular, the gyrA N94G mutation was associated with improved fitness in all double mutants, irrespectively of the rpoB mutation. In an interesting manner, the mutation combinations obtained in vitro in M. smegmatis correspond to the most common mutations detected among MDR and XDR clinical isolates in high MDR TB burden countries Casali et al., 2014;Comas et al., 2012). These authors also found some double mutants bearing higher biological cost, which can be a sign of negative epistasis. Furthermore, the acquisition of a secondary mutation (linked or not linked to drug resistance) in the same gene, for example rpoB, was associated with a reduction of biological cost (Brandis & Hughes, 2013;Song et al., 2014).

| Epistasis between drug resistance-associated mutations and compensatory mutations
The progressively increasing identification of drug resistant, including MDR and XDR isolates without reduction in fitness, suggests the presence of epistatic interactions between drug resistance muta- have not been investigated in M. tuberculosis. Nevertheless, the findings that many (27%-70%) clinical rifampicin-resistant mutants carry putative compensatory mutations in either rpoA or rpoC genes support the hypothesis that these two mutation types interact also in M. tuberculosis (Casali et al., 2014;Comas et al., 2012;Li et al., 2016;Song et al., 2014). As example, rifampicin-resistant M. tuberculosis strains carrying the rpoB S531L mutation are often associated with putative compensatory mutations in the rpoA or rpoC genes (Casali et al., 2014;Song et al., 2014;de Vos et al., 2013).
Concerning the evolution of MDR strains, clinical and molecular studies suggest that isoniazid resistance, due to the katG S315T mutation, has preceded the emergence of rpoB gene mutations leading to the acquisition of rifampicin resistance Gegia, Winters, Benedetti, van Soolingen, & Menzies, 2017;Manson et al., 2017;Salvatore et al., 2016). The combination of katG315 and rpoB531 mutations with a rifampicin-resistant fitness-compensatory mutation (e.g., rpoC mutation) is favoured in clinical MDR isolates, suggesting that these genotypes lead to primary MDR infections. In an important way, the emergence of XDR TB seems to be caused by the transmission of XDR strains directly from person to person rather than by inadequate MDR treatment (Shah et al., 2017). Thus, compensatory evolution and epistasis could play an important role in the emergence and spread of highly resistant strains in the community.

| Epistasis between resistance determinants and genetic background
For many bacteria, epistatic interactions have been also described between resistance determinants and their genetic background (Chang et al., 2015). In M. tuberculosis, epidemiological and molecular data have shown the emergence and the successful spread of MDR/ XDR clones belonging to Beijing or LAM families carrying specific mutations associated with high level of drug resistance and compensatory mutations (Casali et al., 2014;Cohen et al., 2015;Eldholm et al., 2015). Indeed, in Beijing family, it was demonstrated that the biological costs of resistance mutations are smaller than those in other families, or the acquisition of compensatory mutations appears easier, possibly explaining the association between Beijing genotype and MDR (Casali et al., 2014;Gagneux, Long, et al., 2006).
Altogether, the interactions between different drug resistance mutations, between drug resistance mutations and compensatory mutations and between drug resistance mutations and the genetic background underline the key role of epistasis in the evolution of multiple drug resistance in M. tuberculosis. In an important way, the Beijing lineage is rapidly spreading worldwide. This lineage is associated with MDR TB as well as with high level of drug resistance and fitness-compensatory mutations (Casali et al., 2014;Manson et al., 2017). This suggests a worrying scenario in which drug resistance evolves towards very fit and highly drug-resistant genotypes and the successful transmission of deadly drug-resistant mutants. This could seriously challenge the success of TB control programmes worldwide.

| CON CLUDING REMARK S
It is unfortunate that, some drug resistance mechanisms remain unclear and many mechanisms of fitness-compensatory evolution and epistasis have not been investigated in M. tuberculosis. More work is needed to increase our knowledge on all the forces that drive drug resistance in M. tuberculosis for better controlling the emergence and rapid spread of highly drug-resistant strains. As suggested by the levels of drug resistance reached globally, we are losing the arms race against bacteria including M. tuberculosis (Bañuls et al., 2018). M. tuberculosis, as many pathogens, has a complex ecology and evolution and is also evolving and fluctuating through time and space according to local contexts (Bañuls et al., 2015;Comas et al., 2013;Eldholm et al., 2016;Müller, Borrell, et al., 2013;O'Neill et al., 2015;Trauner et al., 2014). For instance, our review underlines that strains carrying multiple drug-resistant mutations reveal a high ability to acquire other resistances or compensatory mutations by epistatic interactions in reducing the biological cost imposed on the fitness of bacteria. These evolutionary processes suggest that, to limit the drug resistance escalation, molecules acting simultaneously on multiple bacterial targets are urgently needed to replace singletarget drugs that now require to be used in more and more complex combinations (the standard treatment of TB disease is composed by a minimum of four drugs). In addition, the detailed knowledge of evolutionary mechanisms will help develop accurate models to predict the evolution of drug resistance and thus to better control it as underlined by other authors (Lehtinen et al., 2017;Schenk & de Visser, 2013).

ACK N OWLED G EM ENTS
This review was written in the framework of PHC Lotus project "Application of DNA chip technology for the development of Andermacher for assistance in preparing and editing the manuscript.

CO N FLI C T O F I NTE R E S T
None declared.