Newly arisen mutations can have very different impacts on the fitness of the organism, ranging from deleterious through neutral to beneficial. However, they appear at very different rates. For example, for Escherichia coli K-12, the rate of deleterious mutations per genome per replication is, at least, 2–8 × 10−4 (Kibota and Lynch, 1996; Boe et al., 2000), while that of beneficial mutations is, at least, 2 × 10−9 (Imhof and Schlotterer, 2001). The rate of spontaneous mutations results from a balance between the effects of deleterious mutations and the metabolic costs of further reducing mutation rates (Drake et al., 1998). Any variant that has increased mutation rates is expected to have reduced fitness due to increased production of deleterious mutations. However, contrary to these intuitive predictions, strains having high mutation rates (mutators) are not rare in natural bacterial populations. They have been found in populations of E. coli (Matic et al., 1997), Salmonella enterica (LeClerc et al., 1996), Neisseria meningitidis (Richardson et al., 2002), Haemophilus influenzae (Watson et al., 2004), Staphylococcus aureus (Prunier et al. 2003), Helicobacter pylori (Bjorkholm et al., 2001), Streptococcus pneumoniae (del Campo et al., 2005) and Pseudomonas aeruginosa (Oliver et al., 2000), with frequencies ranging from 0.1% to above 60%. Experimental (Mao et al., 1997; LeClerc et al., 1998) and theoretical (Boe et al., 2000) studies indicate that the frequency of mutators observed among natural isolates is much higher than expected from mutation/selection equilibrium, which suggests that there are situations in nature where being a mutator confers a selective advantage. In this review, we will discuss the role of different factors involved in shaping mutation rates in natural bacterial populations, as well as their role in the evolution of pathogenicity and antibiotic resistance.
Evolutionary success of bacteria relies on the constant fine-tuning of their mutation rates, which optimizes their adaptability to constantly changing environmental conditions. When adaptation is limited by the mutation supply rate, under some conditions, natural selection favours increased mutation rates by acting on allelic variation of the genetic systems that control fidelity of DNA replication and repair. Mutator alleles are carried to high frequency through hitchhiking with the adaptive mutations they generate. However, when fitness gain no longer counterbalances the fitness loss due to continuous generation of deleterious mutations, natural selection favours reduction of mutation rates. Selection and counter-selection of high mutation rates depends on many factors: the number of mutations required for adaptation, the strength of mutator alleles, bacterial population size, competition with other strains, migration, and spatial and temporal environmental heterogeneity. Such modulations of mutation rates may also play a role in the evolution of antibiotic resistance.
Mismatch repair-deficient mutators
Mutator phenotypes in general result from alterations in genes coding for DNA repair enzymes and for proteins that assure accuracy of DNA replications. These mutant genes are called mutator alleles. The majority of strong mutators found in the laboratory [E. coli (Sniegowski et al., 1997); S. enterica serovar Typhimurium (LeClerc et al., 1998)] and in nature [E. coli (Matic et al., 1997); S. enterica (LeClerc et al. (1996); N. meningitidis (Richardson et al., 2002); P. aeruginosa (Oliver et al., 2002)] have a defective mismatch repair system due to the inactivation of mutS or mutL genes. This repair system controls the fidelity of DNA replication by eliminating biosynthetic errors (Schofield and Hsieh, 2003), and by participating in processing DNA lesions during transcription-coupled repair (Mellon and Champe, 1996). In addition, the mismatch repair system is involved in the maintenance of chromosomal structural integrity and in the control of horizontal gene transfer by preventing recombination between non-identical DNA sequences. The mismatch repair system involves several proteins, of which two, MutS and MutL, have been highly conserved during evolution. MutS protein recognizes seven of eight possible base pair mismatches, C-C mismatches (the least frequent replication error) being refractory. In addition, MutS protein binds up to four unpaired bases, allowing for repair of frameshift errors. The efficiency with which different mismatches are repaired is determined by the affinities of MutS protein for various mismatches. MutL plays the role of ‘molecular matchmaker’ between MutS–mismatch complexes and other proteins involved in the repair process. The inactivation of mutS or mutL genes results in a strong mutator phenotype, with a 102-fold increased rate of transition (G:C→A:T and A:T→G:C) and 103-fold increased rate of frameshift mutations. In addition, mutS or mutL knockout mutants have a strong hyper-recombination effect, resulting in a 101−103-fold increase in the rate of chromosomal rearrangements. Molecular characterization of E. coli and P. aeruginosa natural mutS and mutL mutants has revealed that these genes are inactivated by a variety of mechanisms: frameshifts, insertions, premature stop codons and deletions (Oliver et al., 2002; Li et al., 2003).
Any bacterial population is expected to harbour a subpopulation of mismatch repair mutants due to spontaneous mutations in mismatch repair genes. The frequency of mismatch repair-deficient mutators in cultures of E. coli K-12 that are not subjected to any selective pressure was estimated to be less than 3 × 10−5 (Mao et al., 1997; Boe et al., 2000), which corresponds to a mutation rate from non-mutators to mutators of 5 × 10−6 per cell per generation (Boe et al., 2000). For S. enterica serovar Typhimurium, the frequency of mutators in the unselected population is even lower, 1–4 × 10−6 (LeClerc et al., 1998). How can mismatch repair-deficient mutators outgrow non-mutator cells when they produce such vastly increased numbers of mutations that are predominantly deleterious? There are two possibilities: (i) mismatch repair mutators have higher rates of replication than non-mutators due to the absence of the metabolic load imposed by DNA repair enzymes, or (ii) cells with a high mutation rate have a selective advantage because they adapt faster than non-mutators. These two hypotheses have been experimentally and theoretically tested and the first one was rejected. If there is a selective advantage due to decreased metabolic load, then this advantage should be independent of the initial ratio of mutator to non-mutator cells. However, this is not the case. The mutator outgrows the non-mutator strain only when the ratio of mutator : non-mutator population size is above a certain threshold. This was observed for mutS (Fig. 1 and Giraud et al., 2001; Labat et al., 2005) and mutT (Chao and Cox, 1983) mutators. This threshold is determined by the ratio of the frequency of mutants carrying beneficial alleles in mutator versus non-mutator population. In each population, the frequency of these mutants depends on the mutation rate and the population size (Chat et al., 2006). These studies strongly suggest that mutators are selected because they produce more adaptive mutations. This selective advantage occurs despite the fact that mismatch repair-defective mutators start off with a small selective disadvantage (about 1%) relative to non-mutators (Tröbner and Piechocki, 1984a; Boe et al., 2000). The selective advantage of mutators over non-mutator strains is not restricted to mutS mutants, but is also observed for mutL mismatch repair-deficient mutants (Tröbner and Piechocki, 1981).
Selection of high mutation rates
Natural selection favours mutator cells by favouring beneficial mutations generated in these cells due to its mutator phenotype. However, it also indirectly favours the mutator allele due to its physical linkage with beneficial mutations. The linkage between beneficial mutations and mutator alleles is particularly strong in bacteria because the rate of gene exchange in these asexual organisms is, in general, very low. Mutators can increase in frequency even when their initial population size is very small (see above) when several beneficial mutations are required for adaptation (Mao et al., 1997; Giraud et al., 2002). For example, if the probability of generating each beneficial mutation is 102-fold higher in mutators than in non-mutators, then the probability that two beneficial mutations will be generated in mutators is 104-fold higher than in non-mutators. The hitchhiking of mutator alleles with beneficial mutations is possible only when fitness gain counterbalances the fitness loss due to generation of deleterious mutations (Taddei et al., 1997a).
Selection of mutator alleles also depends on many other phenomena. For example, the increase in frequency of mutators depends on total population size (Tenaillon et al., 1999), on mutator strength [i.e. increase of mutator mutation rate relative to non-mutator mutation rate (Taddei et al., 1997a)], and on the rate of gene exchange (Tenaillon et al., 2000). Furthermore, it depends on the stability of the environment. For example, mutator alleles are particularly advantageous upon a shift in environmental conditions (Travis and Travis, 2002; Tanaka et al., 2003). The dynamics of selection of mutators depends also on environmental spatial heterogeneity, which will allow or prevent the competition between the cells carrying different adaptive mutations. Theoretical modelling predicts that mutators will be particularly favoured in temporally and spatially heterogeneous environmental conditions (Travis and Travis, 2004).
Counter-selection of mutators
Experimental and theoretical studies showed that the frequency of mutator strains in a population could rapidly increase to almost 100%. However, the majority of natural isolates are not mutators. A major factor that diminishes the fitness of mutators is the continuous production of deleterious mutations, particularly in a stable environment, once adaptation is achieved (Giraud et al., 2001). A second factor is that neutral, beneficial and deleterious mutations can have very different impacts on fitness in different environments. Consequently, no single genotype is optimally adapted to all environments. For example, an adaptive mutation in one environment can be deleterious in another [a phenomenon called ‘antagonistic pleiotropy’ (Cooper and Lenski, 2000)]. Therefore, migration from one environment to another might contribute to the reduction of fitness of mutators in natural populations, as observed in in vivo laboratory experiments for E. coli mutS (Giraud et al., 2001) and S. enterica serovar Typhimurium mutS (Nilsson et al., 2004) mutators. Finally, continuous passages through strong bottlenecks result in the accumulation of deleterious mutations due to genetic drift. This phenomenon, called Muller’s ratchet, is particularly deleterious for strong mutator populations. For example, when wild-type and mutS defective cells were passaged through single-cell bottlenecks, only 3% of the wild-type lineages had phenotypically detectable mutations after 40 cycles. On the contrary, 4% of mutS lineages had died out, 55% had auxotrophic requirements, 70% had defects in at least one sugar or catabolic pathway, 33% had a defect in cell motility, and 26% were either temperature-sensitive or cold-sensitive lethals (Funchain et al., 2000). The loss of fitness of mutator strains should be positively correlated to the mutator strength (Taddei et al., 1997a).
Therefore, in the long run, fitness cost associated with high mutation rates is expected to cause the loss of adaptive mutations from bacterial populations, with consequential elimination of the mutator genome. However, some adaptive mutations generated in mutator backgrounds can be saved either by their horizontal transfer to a non-mutator background or by a reduction in the mutation rate of the adapted mutator strain before the load of deleterious mutations becomes too high. The reduction of mutation rate might be achieved by several mechanisms: reversion of the mutator mutation, acquisition of suppressor mutations [e.g. in populations of mutT mutators (Tröbner and Piechocki, 1984b)], or by reacquisition of a wild-type allele of anti-mutator gene from non-mutator bacteria via horizontal gene exchange (see below and Denamur et al., 2000).
Some bacterial species, like H. influenzae and N. meningitides (Bayliss et al., 2001; Meyers et al., 2003), possess mechanisms allowing them to permanently maintain high mutation rates at some loci, avoiding fitness costs associated with high genome-wide mutation rates. The hypermutability of these loci results from the mutational properties of repetitive DNA sequences located within the gene or within its controlling elements. These genes code for evasins, lipopolysaccharide biosynthesis enzymes, adhesins, iron acquisition proteins, and restriction–modification systems. Repetitive DNA sequences experience high rates of insertion and deletion mutations through replication slippage, which results in alternating loss-of-functions and reversions. Such mutagenesis can increase bacteria fitness by allowing evasion of the host’s immune system. However, the type of variation produced by localized mutator activity might not always be sufficient for adaptation, as suggested by the presence of strong generalized mutators in natural populations of N. meningitides and H. influenzae (Richardson et al., 2002; Watson et al., 2004).
Why are mismatch repair-deficient mutants so frequent?
Low spontaneous mutation rates are maintained by the activity of many molecular mechanisms that protect and repair DNA, as well as by the mechanisms that assure high-fidelity DNA replication. Inactivation of over 20 different E. coli genes can confer mutator phenotypes of different strengths (for review see Horst et al., 1999). So, why do nearly all strong mutators found in nature and in the laboratory have defective mismatch repair systems? One explanation for this phenomenon is that inactivation of the other genes involved in important aspects of DNA or RNA metabolism might have too high a cost to be compensated by advantageous mutations. For example, competition experiments in chemostats have shown that E. coli mutT mutators can also be selected for by beneficial mutations they generate (Chao and Cox, 1983). However, mutT mutators were never found in E. coli natural populations. This may be explained by the fact that inactivation of mutT gene, which codes for the protein that eliminates 8-oxo-G from the nucleotide pool (Horst et al., 1999), increases replication, but also transcriptional errors (Taddei et al., 1997b), which might considerably reduce fitness of the mutant cell.
The specific advantage of mismatch repair-deficient strains over other mutator alleles, which might also explain their abundance in nature, is their hyper-recombination phenotype. Recombination can also increase adaptability by increasing genetic variability. Consequently, genotypes with increased recombination rates might be selected for by virtue of the favourable genotypes they generate by allowing the association of beneficial mutations that have appeared in different individuals. Indeed, the frequency of mismatch repair-deficient mutants has been reported to increase rapidly in E. coli populations by hitchhiking with the recombination events they generate (Funchain et al., 2001). Irrespective of whether selection of mismatch repair mutators is due to their hyper-recombination or mutator phenotypes, the cost of high genetic instability remains the same. However, when selective pressure for increased genetic variability is no longer present, the hyper-recombination phenotype of mismatch repair-deficient strains might facilitate the reacquisition of the functional mismatch repair genes via horizontal gene transfer. This can ‘save’ the adapted mutator genome from being overburdened with deleterious mutations and allows survival. Such events seem to happen frequently during E. coli evolution (Fig. 2; Denamur et al., 2000).
Mutators, pathogenicity and antibiotic resistance
Although mutator strains have been found in commensal populations (Matic et al., 1997), most of them were isolated from populations of pathogenic bacteria. Furthermore, pathogens are exposed to changing stressful environments due to host defences and antibiotic treatments, i.e. conditions expected to favour mutators. Therefore, mutator and hyper-recombination phenotypes might accelerate the evolution of pathogenic strains by, e.g. increasing variation of surface antigens, as well as by facilitating acquisition of pathogenic determinants and antibiotic resistance (Taddei et al., 1997c). Indeed, the highest frequency of mutators within E. coli species was found among isolates from urinary tract infections (Denamur et al., 2002) and N. meningitidis strains displaying high phase variation rates (Richardson et al., 2002). Also, antibiotics resistance was significantly higher in mutator than in non-mutator P. aeruginosa pathogenic isolates (Oliver et al., 2000). In addition to these epidemiological studies, several experimental studies have confirmed the link between mutators, pathogenicity and antibiotic resistance.
For example, the measurements of the fitness of the uropathogenic E. coli CFT073 strain relative to its mismatch repair-defective derivative showed that the mutator was advantaged, in vitro, in urine (Fig. 1), and in vivo, in the late stages of urinary tract infection in a mouse model (Labat et al., 2005). Hence, increased mutagenesis could play an important role in the development of persistent reservoirs of uropathogenic E. coli observed in the mammalian bladder epithelial cells (Mulvey et al., 2000).
In vitro and in vivo studies have also provided clear evidence that antibiotic treatment can contribute to the selection of the mutators (Mao et al., 1997; Giraud et al., 2002). Mutators are favoured under such conditions because they generate antibiotic-resistance-conferring mutations at a higher rate than non-mutators (Fig. 3). This phenomenon is not specific to one antibiotic or to one family of antibiotics, suggesting that mutations can contribute to various modes of evolution of antibiotic resistance. Indeed, whenever mutations can confer or increase resistance to the antibiotics, or reduce the biological cost of resistance on bacterial fitness, it is more likely that they will appear in the populations of cells with higher mutation rates. Furthermore, when resistance is associated with the acquisition of several mutations, the advantage of being a mutator increases significantly (Tenaillon et al., 1999). For example, the sequential appearance of several mutations responsible for the evolution of extended spectrum β-lactamases is favoured by high mutation rates (Orencia et al., 2001). Therefore, the presence of mutator bacteria is a risk factor in the treatment of infectious diseases, as they might be responsible for therapeutic failures.
Fine tuning of mutation rates
A positive correlation between multiple-antibiotic resistance and high mutation rates is expected to be frequent in natural populations of bacterial pathogens (Blazquez, 2003). Such a correlation was found for P. aeruginosa isolated from lungs of patients with cystic fibrosis, bronchiectasis and chronic obstructive pulmonary disease (Oliver et al., 2000; Macia et al., 2005). This is probably due to extensive antibiotic therapy and chronicity of the infection as well as to strong compartmentalization and low migration rates between hosts that limit competition between strains. These variables would therefore increase the probability of persistence of strong mutators. Indeed, the prevalence of P. aeruginosa mutator strains increases with time in the lungs of cystic fibrosis patients (Ciofu et al., 2005). However, this correlation was not observed in P. aeruginosa isolated from acutely infected patients (Gutierrez et al., 2004). The explanation for this discrepancy is provided by studies of the relationship between antibiotic resistance and mutation rates in E. coli natural populations (Gustafsson et al., 2003; Lindgren et al., 2003; Baquero et al., 2005; Denamur et al., 2005).
The correlation between multiple-antibiotic resistance and high mutation rates was also absent in E. coli natural populations, despite the capacity of strong E. coli mutators to generate mutations conferring resistance to anti-biotics in vivo and in vitro (Fig. 3; Mao et al., 1997; Tanabe et al., 1999; Denamur et al., 2002; Giraud et al., 2002). Among E. coli natural isolates, the highest frequency of antibiotic resistance was found among weak mutators, while strains with high mutation rates had significantly lower antibiotic resistance (Denamur et al., 2005). The results of several other studies also suggested that even modest increase in mutation rate can significantly influence the evolution of antibiotic resistance (Gustafsson et al., 2003; Lindgren et al., 2003; Baquero et al., 2005).
The absence of the correlation between high mutation rates and antibiotic resistance can be explained by the dynamics of selection and counter-selection of mutator alleles. Selection acting on mutator alleles is modulated by many factors including changes in the environment and the opportunity for competition between strains. Mutator clones can rapidly specialize to one environment, which renders them less fit in other environments (Giraud et al., 2001). The initial gain and long-term loss of fitness is proportional to the strength of the mutator allele. Computer simulations predict that weak mutators will be selected less rapidly than strong mutators but, once selected, they will persist much longer than strong mutators (Taddei et al., 1997a). This may explain why there is around 1% of strong mutators and around 10–30% of weak mutators in E. coli natural populations (Matic et al., 1997; Baquero et al., 2004). Therefore, in bacterial species that cycle between different environments, e.g. E. coli, weak mutator strains might have more chance to accumulate multiple-antibiotic resistance than the strong mutators (Denamur et al., 2005). In conclusion, spatial and temporal environmental variations, together with the nature and the strength of selective pressures, constantly adjust mutation rates to levels that favour adaptation to each unique environment.
We are grateful to Richard Moxon and Olivier Tenaillon for discussing the manuscript. This study was partially supported by grants from the ‘Fondation pour la Recherche Médicale’ (E. D.). We apologize to all the contributors in the field for the numerous papers that we were not able to cite due to space limitation.