The aim of this thesis is to describe patterns and mechanisms of bacterial adaptation to novel environments. The specific aims are;
• to determine the mutation rate among natural isolates of E. coli.
(paper I)
• to determine how mutation rate and population size influence the rate and extent of adaptation of S. typhimurium LT2 to mice. (paper II)
• to determine the mechanisms of fosfomycin resistance in E. coli. Also to estimate the biological cost of fosfomycin resistance in E. coli and evaluate its impact on clinical resistance development. (paper III)
• to identify mutations conferring resistance to actinonin in S.
typhimurium LT2 and to estimate their effect on bacterial fitness.
Further to perform compensatory evolution and identify the compensatory mutations. (paper IV)
• to reduce the genome size of S. typhimurium LT2 by means of spontaneous deletions. Also to identify the deletion junctions and evaluate the role of sequence homologies in deletion formation.
(paper V)
RESULTS AND DISCUSSION
The following presentation is a summary of the main findings in the papers included in this thesis. For a detailed description of the methods used and graphical presentation of the data, the reader is referred to the specific papers.
Weak Mutators are Enriched in Clinical Isolates(paper I)
In paper I we determined the frequency of strains with elevated mutation rates among natural isolates of E. coli. Strains were collected in three different European countries;
Spain, Denmark and Sweden. In Spain 100 strains were collected from positive urine cultures, 100 strains from blood cultures and 100 strains from stools of healthy volunteers.
170 strains were collected in Denmark from blood cultures and the Swedish collection consisted of 226 strains from urine cultures. In total 696 E. coli strains were included in the study. For all strains we determined the mutation frequency to rifampicin resistance, which is a common method to estimate mutation rates in bacteria. Based on their estimated mutation frequencies, the strains could be categorized into four different groups. Most strains (63.3 %) had mutation frequencies in a narrow interval between 8x10-9– 4x10-8, corresponding to the modal peak of the distribution. These strains were considered as normo-mutable and the mutation frequency to rifampicin resistance observed in this group corresponds well to the mutation frequency expected for a repair proficient E. coli strain (85). 13 % of the strains had a lower mutation frequency than 8x10-9 and were termed hypomutable. Among the E. coli isolates we observed an enrichment of strains with a moderate increase in mutation rate (4x10-8 – 4x10-7). This group constituted 23% of all isolates and was designated as weak mutators. Also, five strains with a strong mutator phenotype were identified, where the mutation frequency to rifampicin resistance was above 4x10-7. This high mutation frequency corresponds to that observed for MMR-deficient E. coli, however the genetic basis for the high mutation rate was not investigated in these strains (105, 166, 171, 172).
A clear difference in enrichment of weak mutators could be observed depending on the source of isolation. In the collection of Spanish isolates, weak mutators were more common among pathogenic E. coli isolated from blood (38%) and urine cultures (25%) than among commensal E. coli isolated from stool samples (11%). There was also a difference between pathogenic E. coli depending on the site of isolation. Enrichment of mutator strains among clinical isolates have been observed in several studies (65, 98, 135, 141, 157, 188, 218). Among Pseudomonas aeruginosa isolated from cystic fibrosis patients as many as 20% of the strains displayed a strong mutator phenotype (188). The life-style of
pathogenic bacteria seems to be especially prone to enrich for elevated mutation rates and pathogenic bacteria encounter a highly dynamic and heterogeneous environment in its host (194). Disease progression, with tissue destruction and activation of the immune system, contribute to a rapidly changing environment. Such environmental fluctuations are known to enrich for mutator strains due to their increased rate of formation of beneficial mutations (64, 228). Population structure is also important in the enrichment for mutators. Strains with a high mutation rate are more rapidly enriched in small populations and in populations passing through severe bottle-necks (64, 228). Both of these conditions may very well exist in the host environment. Strong mutators have been the focus of interest in most studies and little attention has been paid to weak mutators. In our study we identified weak mutators at far higher frequencies than strong mutators. There are several possible explanations for this finding. First, weak mutators might simply form at a higher rate than strong mutators due to a larger mutational target. Due to their abundance, weak mutators can therefore experience a higher probability than strong mutators of acquiring beneficial mutations with which they hitch-hike to higher frequencies in a population. Second, strong mutators could be efficiently counter-selected because of their increased rate of fixation of deleterious mutations. Due to their lower rate of mutation accumulation, the evolutionary success of a weak mutator could therefore be higher over time than for a strong mutator.
A high mutation rate has been suggested as a risk factor for antibiotic resistance development and the enrichment of mutator strains among clinical isolates could potentially speed up the rate of resistance development (50, 91, 153). If so, mutator strains should be more common among antibiotic resistant isolates than among fully susceptible strains. We determined the resistance level to ciprofloxacin resistance in a sub-set of E. coli isolates and compared the resistance level with mutation rate. We could not detect any correlation between mutation rate and ciprofloxacin resistance. These results are in concordance with data from a French collection of E. coli strains (65). Increased frequencies of mutator strains have however been identified among ciprofloxacin resistant E. coli isolates in other strain collections (135). The reasons for the discrepancies observed between different strain collections are not clear. The trivial explanation would be that an increased mutation rate is not a risk factor for resistance development and the correlation observed in some studies is purely incidental. Alternatively, a correlation between resistance and mutation rate could be obscured by other factors, for example mutation rate modifiers arising after resistance development.
Mutation Supply Rate is Limiting for Adaptation of
Salmonella
to Mice (paper II)In paper II we investigated the influence of mutation rate and population size on the rate and extent of adaptation of S. typhimurium to mice. The laboratory strain S. typhimurium LT2 cause a typhoid fever-like infection in BALB/c mice (156, 186). In the animal host, the bacteria replicate in macrophages and, a few days after infection, bacteria can be harvested from liver and spleen in large numbers. We evolved 18 different lineages of S. typhimurium in mice by serial passage for approximately 130 generations. Four different combinations of population size and mutation rate were used to vary mutation supply rate and fitness of the evolving populations was estimated at the end-point (132 generations) and mid-point (66 generations) of the experiment by performing competition experiments in mice. At the end point all populations had increased fitness. However, the fitness increase (s=0.1) was small for the populations evolved with the lowest mutation supply rate (i.e. small population size-low mutation rate) whereas the other evolved populations had reached the same high fitness level in mice (s=0.4). In populations evolved with a high mutation rate, the time of ascent of beneficial mutations seemed to be limiting for the rate of adaptation. Increasing mutation supply rate in these populations, by increasing population size, had little effect on the rate of adaptation. In the populations evolved with a low mutation rate however, mutation supply rate became limiting. The populations evolved with a small population size increased in fitness only marginally and increasing the mutation supply rate, by increasing the population size, significantly accelerated the rate of adaptation. By mathematical modeling we could estimate an apparent mutation rate for beneficial mutations in these lineages to be larger than 10-6/cell/generation. The experimental data is consistent with the fixation of two successive beneficial mutations, each with a selection coefficient of 0.2, and a target for beneficial mutations as large as 104 base pairs. The large target for beneficial mutation suggests that adaptation to mice can be conferred via a large number of single nucleotide substitutions or loss-of-function mutations in single genes.
Mutator Strains Evolved in Mice Display Fitness Trade-Offs due to Mutation Accumulation (paper II)
To determine if adaptation to mice resulted in decreased fitness in secondary environments we investigated the metabolic potential of the evolved populations. First, we screened the evolved populations for auxotrophic mutants that had lost the ability to synthesize specific amino acids. In the populations evolved with a high mutation rate, 0.7-39% of the bacteria was auxotrophs, whereas the frequency of auxotrophic mutants in the populations evolved with a low mutation rate was below 0.08%. We also investigated the capacity to metabolize
46 different carbohydrates for one random clone from each evolved population. All clones from populations evolved with a low mutation rate had the same metabolic potential as the unevolved ancestral strain, whereas clones from the evolved mutator populations had defects on at least one of the substrates tested. In conclusion, populations evolved with a high mutation rate displayed fitness trade-offs in secondary environments whereas populations evolved with a low mutation rate did not. Fitness trade-offs in secondary environments are thought to occur either via antagonistic pleiotrophy effects, where a beneficial mutation by itself causes decreased fitness in a secondary environment, or via mutation accumulation. In this case, antagonistic pleiotrophy is an unlikely explanation, since populations evolved with a low mutation rate displayed no fitness trade-offs and the fitness trade-offs observed in the mutator lineages were most likely caused by a higher rate of mutation accumulation.
Partially Different Spectra of Mutations Confer Resistance to Fosfomycin in Resistant Strains Isolated in the Laboratory and in Clinical Fosfomycin Resistant Isolates (paper III)
Fosfomycin is a cell-wall inhibitor mainly used in treatment of UTIs (119, 168). Under laboratory conditions fosfomycin resistance develops rapidly with a mutation frequency to resistance between 10-7-10-8. However, despite many years of usage the frequency of fosfomycin resistant clinical isolates is low, indicating that factors other than the rate with which resistant variants form are important for clinical resistance development (122). In paper III we isolated a set of fosfomycin resistant mutants of E. coli NU14 under laboratory conditions, identified the resistance mutations and estimated their effect on bacterial fitness. The results from the in vitro isolated strains were then compared with a set of fosfomycin resistant clinical isolates. Fosfomycin resistance in E. coli is mainly conferred via inhibited uptake of the antibiotic into the cell through the GlpT and UhpT transport systems. The expression of the two transporters is positively regulated by the cAMP levels in the cell and full expression of the UhpT system also requires the transcriptional regulators UhpABC and an external inducer in the form of glucose-6-phosphate (G6P) (117, 164, 187, 221). Fosfomycin resistance is conferred by loss-of-function mutations in the genes encoding either the transporters themselves, or in regulatory genes like cyaA and ptsI that control the cAMP levels in the cell (118, 119, 168). In the in vitro isolated fosfomycin resistant strains, resistance mutations were identified both in genes encoding the transporters as well as in cyaA and ptsI genes whereas in the clinical isolates only mutations in the transporter genes could be identified. To exclude the possibility that the clinical isolates carried unidentified mutations that interfere with the cAMP levels in the cell
we tested the ability of the fosfomycin resistant strains to grow on minimal medium supplemented with five different carbohydrates. Mutations in genes regulating cAMP levels in the cell confer many pleiotrophic effects and mutant strains become defective for uptake of several different carbohydrates (199). All fosfomycin resistant mutants isolated in vitro with mutations in cyaA or ptsI were unable to growth on at least one of the carbohydrates tested whereas no growth defects could be identified in the clinical fosfomycin resistant isolates. The results show that fosfomycin resistance is conferred by partially different spectra of mutations in the in vitro isolated strains as compared to the clinical fosfomycin resistant isolates. The absence of mutations in genes regulating cAMP expression in the clinical isolates suggests that this class of mutations is counter-selected under in vivo conditions. Reduced cAMP levels are known to decrease adhesion to epithelial cells, since pilus biosynthesis is reduced (22). Also, defects in uptake of carbohydrates could reduce growth rates both in the bladder and in the intestine.
Fosfomycin Resistance Generally Confers a High Fitness Cost that Contributes to the Low Prevalence of Fosfomycin Resistance among Clinical Isolates(paper III)
To estimate the fitness costs associated with fosfomycin resistance we measured exponential growth rates in LB and urine. Also, a mathematical model was developed to estimate the effect of fitness costs on the probability of resistance development in the bladder during antibiotic therapy. The model takes into account known parameters of bladder dynamics, e.g. flow rate of urine into the bladder and maximal and minimal volume of urine in the bladder before and after emptying (96). Microbial parameters, such as growth rate of resistant and susceptible strains and the mutation rate to resistance, were estimated in the laboratory. The mathematical modeling showed that fosfomycin resistance would develop with a high rate during antibiotic therapy, if the resistance mutations did not have any effect on bacterial fitness. However, the probability of resistance development was very sensitive to changes in growth rate and even a moderate decrease in fitness would drastically reduce the probability of resistance development. By measuring exponential growth rates we showed that fosfomycin resistance generally confers a fitness cost to the resistant bacteria. Growth rates of the in vitro isolated fosfomycin resistant mutants were between 10-25% slower than for the susceptible parental strain, both in LB and in urine.
For a resistant strain to establish an infection in the bladder during fosfomycin therapy growth rate in the presence of the antibiotic is also important. We therefore measured exponential growth rates in the presence of a range of fosfomycin concentrations. The addition of the antibiotic to the growth medium reduced growth rates of the resistant
strains even further compared to when fosfomycin was absent. For the severe fitness costs observed, the mathematical model showed that the probability of resistance development in the bladder during fosfomycin therapy is improbable. For the clinical fosfomycin resistant isolates we lacked access to an isogenic susceptible strain for comparisons. Instead, we compared growth rates between a set of fosfomycin resistant and fosfomycin susceptible clinical isolates. No significant differences in growth rate could be detected between the two groups. The results indicated that either the fitness cost of fosfomycin resistance in the clinical isolates had been ameliorated via compensatory mutations, or the clinical isolates carry resistance mutations that do not confer a high fitness cost.
Exponential growth rate is only one parameter affecting fitness of uropathogenic E. coli.
The ability to adhere to uroepithelial cells is also an important virulence factor and other studies have shown that fosfomycin resistant E. coli display reduced adherence to epithelial cells (47, 52, 63, 93, 130). However, rapid growth in urine is an important virulence factor and reduced growth rates in urine will, therefore, have significant effects on the virulence of fosfomycin resistant E. coli. Expression profiling of E. coli growing in the bladder of infected mice show that genes encoding ribosomal proteins are highly expressed, suggesting that the bacteria are growing rapidly (230). Also, uropathogenic E. coli have been demonstrated to grow significantly faster in urine than E. coli isolated from feces, indicating that rapid growth in urine is a specific trait for these strains (96). The low prevalence of fosfomycin resistance observed clinically has been suggested to be dependent on several parameters. For many antibiotics, resistance is rapidly enriched in the normal flora after treatment and can persist for many years. In these cases, the normal flora can function as a reservoir of resistance determinants that can then be transferred to pathogenic strains.
Fosfomycin therapy does not seem to enrich for resistant strains in the intestinal flora and resistance is mainly conferred by chromosomal mutations that do not spread horizontally (14, 101). We suggest that in addition to these factors, the biological cost of resistance, in terms of reduced growth rates, is an important contributor to the low prevalence of fosfomycin resistance observed clinically.
Actinonin Resistance Mutations are Associated with High Fitness Costs (paper IV)
For the success of any new antibiotic a low rate of clinical resistance development is essential. To evaluate the risk of resistance development clinically, the biological costs of resistance, and the ability to compensate for such costs, are important factors to estimate.
Deformylase inhibitors are a new class of antibiotics targeting a unique feature of bacterial translation (88, 154, 155). In paper IV we estimated the biological cost of resistance to the classic deformylase inhibitor actinonin, and the capacity to compensate for any such fitness costs. We also determined the genetic basis for resistance and compensation. Initiation of translation with a formylated methionine is an evolutionary conserved feature of bacterial translation. After translation is completed, the formyl group is removed by the enzyme peptide deformylase (PDF), a step that is essential for subsequent protein maturation (13, 24, 88, 183). Actinonin binds to and interferes with the activity of PDF, resulting in accumulation of formylated polypeptides in the cell and subsequent cell death. We isolated a set of actinonin resistant mutants in the laboratory from S. typhimurium LT2. Resistance mutations were identified by sequencing the fmt and folD genes. The fmt gene encodes formyl-methyl transferase (FMT), catalyzing the addition of the formyl group to the methionyl initiator-tRNA complex (13). The resistance mutations identified in this gene were mainly loss-of function mutations that disrupt FMT function. As a consequence, the ribosome is forced to initiate translation with an unformylated methionine and the need for a functional PDF is thereby by-passed. The folD gene encodes a bi-functional enzyme important for the production of the precursor molecule 10-formyl-H4folate, the donor of the formyl group (159). Similar to fmt mutations, loss of folD function is thought to reduce formylation of the methionyl initiator-tRNA complex. To determine the effect of the resistance mutations on bacterial fitness we measured exponential growth rates in LB. The majority of all actinonin resistant mutants displayed severely reduced growth rates and the effect was especially pronounced in strains carrying mutations in the fmt and folD genes. For a subset of the actinonin resistant mutations, fitness was also estimated in a mouse model and the results showed that for all but one strain, virulence in mice was severely perturbed.
The one strain where no cost was identified in the mouse model also had similar fitness as the parental susceptible strain in LB, indicating that this strain might carry a resistance mutation that does not confer any detectable fitness cost. The severe fitness costs observed can be attributed to a decreased translation rate, since initiation of translation with an unformylated methionine is a highly inefficient process (97). Taken together, the data show that actinonin resistance generally confers a heavy fitness burden on the resistant bacteria.
Over-expression of the Initiator tRNA via Gene Amplifications can Compensate for the Cost of Actinonin Resistance(paper IV)
The severe fitness costs associated with actinonin resistance suggests that the rate and extent of compensatory evolution will be important parameters to determine when evaluating the risk for clinical resistance development. We evolved in total 39 lineages from five different ancestral, actinonin resistant strains by serial passage in LB. The ancestral strains carried different resistance mutations in either fmt or folD, conferring severe fitness costs to the resistant strains (growth rates in LB were reduced between 42% - 76%). The first growth-compensated mutant to reach fixation in each lineage was isolated and used for further analysis and we could isolate growth-compensated mutants in all evolved lineages.
Some mutants increased fitness to levels similar to the susceptible wild-type strain, but the majority of growth-compensated mutants still displayed a significant fitness cost after acquisition of one compensatory mutation. The result indicates that restoring fitness to the same level as for the susceptible wild-type strain generally requires a sequential acquisition of several compensatory mutations.
Intragenic compensatory mutations were identified in 12 strains by sequencing the gene carrying the resistance mutation. The intragenic compensatory mutations identified were mainly in the same codon as the resistance mutations, reverting a stop codon to a sense codon. In 24 of the growth compensated strains, we could not identify any intragenic compensatory mutations and here we used transposon mutagenesis to identify the unknown extragenic mutations. In eight strains the transposon linked to the compensatory mutation was inserted close to the metZW genes that encode for the initiator-tRNA (tRNAi). Southern hybridization revealed gene amplifications spanning the metZW genes in all eight strains with a 5-fold to 40-fold increase in copy number. We could also show, by Northern hybridizations, that the increase in copy number was accompanied by a similar increase in the level of the tRNAi itself. An overproduction of the tRNAi can potentially compensate for the decreased translation rate of the resistant strains via either of two different mechanisms, or by a combination of both. First, if the mutated FMT enzyme retains some activity, over production of the tRNAi might increase the rate of formylation by increasing the concentration of one of the substrates (tRNAi) in the enzymatic reaction.
Second, an over production of the tRNAi can potentially increase the rate of translation initiation with an unformylated methionine by simply increasing the proportion of tRNAi relative to the elongator tRNAs in the cell. Presently, we can not distinguish between the two mechanisms, but the results will have important implications for resistance
development. A compensatory mechanism that requires restoration of formylation capacity will also concomitantly decrease the resistance level to actinonin.
We further wanted to determine the size of the amplified arrays and the structure of the amplification junctions in the growth-compensated strains. The formation of the first duplication is the rate limiting step in generating gene amplifications (231). Once the first duplication is formed, RecA can act on the large stretches of sequence identity and catalyze subsequent amplification via homologous recombination. The metZ and metW genes are identical and situated in close proximity to one another on the Salmonella chromosome, only separated by a spacer region of 31 bp. This tandem localization of the two genes constitutes an already preformed duplication that could potentially have been used to create the amplifications identified in the growth compensated strains. To investigate if this was the case, we performed PCRs with primers located directly upstream and downstream of the two genes. If the homologies in the two identical genes were used as substrates for the amplifications, the primers would be expected to produce a PCR product of the same size as the amplified array.
Figure 5. Model for the evolution of new gene functions by means of gene amplification.
However, the PCR products obtained were all of the same size as for the wild type S.
typhimurium strain and the metZW genes could therefore not have been used as substrates in the amplification process. In one of the eight compensated strains carrying gene amplifications, we could identify the duplication junction to 460 bp upstream and 1440 bp downstream of the start of the metZ gene, giving rise to an amplified array of 1908 bp. Here, we could not detect any sequence homologies at the junction. Gene duplications have been suggested to constitute a first step in evolution of new gene functions (128, 240). At the same time as one gene copy performs the original function, the second copy can evolve new functions through mutagenesis. There are two main problems with this theory. First, how can elevated copy numbers of a gene be maintained in a population long enough for genetic and phenotypic divergence to take place. Long stretches of amplified DNA are intrinsically unstable and rapidly segregate via homologous recombination, which rarely allows for enough time for diversification. Second, since most mutations will inactivate gene function the probability of evolving a new function is very low compared to the probability of gene inactivation if there is no selection for functionality of all gene copies. Here, we describe a system where gene amplifications can compensate for the cost of a deleterious mutation. In this system, both gene number and gene functionality is maintained by selection (Figure 5).
Bacterial Genome is a Rapid Process under Laboratory Conditions (paper V)
Evolution towards endosymbiosis is associated with extensive gene loss and concomitant reduction in genome size (129, 174, 252). In paper V we set up a simple experimental system to study the rate and mechanisms of DNA loss in real time (Figure 6). Different lineages of S. typhimurium were serially passaged on hematin agar plates, and with regular intervals samples from the evolving lineages were analyzed for DNA loss by performing PFGE and DNA microarrays. The evolving lineages were serially passaged by each day picking one random colony and streaking it on a fresh agar plate. By introducing a bottle-neck of one cell in each passage the effect of purifying selection was effectively reduced.
The majority of mutations, including deletions, are deleterious to bacterial fitness and would therefore be lost if purifying selection were not eliminated in the evolving lineages.
Reduction in population size and passage through severe bottle-necks are also characteristic features of evolution towards reduced genome size (129, 174, 252). The intracellular growth environment of endosymbionts provides many of the metabolic products needed to sustain bacterial growth and genes encoding metabolic functions are almost absent from reduced
Figure 6. Schematic presentation of the experimental system used for studying genome reduction under laboratory conditions.
genomes as they have been made redundant during evolution toward endosymbiosis. In our experimental set-up we used hematin agar plates to mimic the nutrient rich growth environment provided by the eukaryotic cell. The growth medium provides many of the amino acids, vitamins, co-factors and carbohydrates needed for bacterial growth and thereby selection for many of the metabolic functions encoded for on the Salmonella chromosome was thereby relaxed.
We evolved in total 72 lineages of S. typhimurium using this experimental set-up. Twelve lineages were evolved with a low mutation rate and 60 lineages with a high mutation rate.
The MMR system has a stabilizing effect on maintaining genome structure and the MutS -mutator strain used in this experiment displayed a 50-fold higher deletion rate than the wild-type strain in the mob-gal deletion assay system (143). After 60 cycles of serial passage (corresponding to approximately 1500 generations), deletions were identified in four of the mutator lineages, whereas no deletions could be detected in the wild type lineages after 300 cycles of serial passage (~7500 generations). The deletions identified varied in size from approximately 173 kbp to 1.2 kbp. From the data we calculated an average rate of DNA loss of 2.5 bp/chromosome/generation for the mutator strain and 0.05 bp/chromosome/generation for the wild-type strain. These rates most likely reflect maximal DNA loss rates, since successive deletions will decrease the target size for future deletions as the fraction of non-essential genes decreases. The data however, indicate that
genome size can be drastically reduced in a, from an evolutionary perspective, very short time period. If a generation time of one day is assumed, the above rates suggests that a genome can decrease in size by 1 Mbp in 50,000 years. The above calculations are obviously associated with a large degree of uncertainty since many factors like generation time, the extent of purifying selection and genetic drift, and the decline in deletion rate are unknown.
Reduction of genome size in endosymbionts is however known initially to have been a rapid process with large deletions occurring early in evolution, followed by a gradual reduction in deletion rate (236).
To evaluate the dependence of deletion formation on RecA-dependent homologous recombination and sequence homologies we identified the deletion end-points in two of the spontaneous deletions isolated in the evolution experiment and in four independent deletions isolated using the mob-gal deletion assay system. The homologies identified were 40 bp, 12 bp, 4 bp, 2 bp, 1 bp and 0 bp respectively. Since the RecA system requires sequence homologies of at least 25 bp, the results indicate that 5/6 deletions were probably formed by RecA-independent processes (222). Reduced genomes typically lack the recA gene and they are devoid of repeat sequences (12, 224). The lack of homologous recombination functions have been suggested as one explanation for the extraordinary genome stability observed in different strains of Buchnera (129, 252). Our data show that spontaneous deletions form at high frequencies via other pathways than classical RecA-dependent homologous recombination. Therefore, loss of repeat sequences and RecA function might not be enough to explain the genome stability observed in endosymbionts.
Reduced genomes have lost many genes involved in DNA repair and recombination, apart from RecA, and might therefore carry more general defects in recombination that affects several pathways of deletion formation (226). Alternatively, genome size and organization can be maintained effectively by selection (106). The evolutionary stasis observed for many reduced genomes may therefore be a reflection of selection acting to preserve genome organization.