The mutation rate does not limit the emergence of novel traits without immediate benefit
To study how an increased supply of mutations might affect the evolution of viability in new environments, we used both a wild-type strain of E. coli and a ‘mutator’ strain with a twenty-two-fold higher mutation rate (Methods)18. Prior to experimental evolution we determined the phenotypic differences between the wild-type and mutator strains. Such differences may be generated by the mutator strain’s intrinsically higher mutation rate, which may cause more mutations and their ensuing phenotypic effects even during initial strain cultivation18. To identify phenotypic differences between the two strains, we used a set of ten Biolog Phenotyping microarrays (PM11-20, Biolog Inc., USA). These microarrays comprise 236 different environments that inhibit microbial growth. Each environment harbours a different antimicrobial agent chosen from a wide range of categories, including antibiotics, detergents, surfactants, ion chelators, oxidising agents, and pyridine analogues. To determine a strain’s viability in each of these phenotyping environments, we randomly chose two clones of the strain from an LB (Luria Bertani medium) agar plate incubated overnight, and required that at least one of the clones was able to survive and grow in the environment (OD600 after 48 h of growth >0.3, Methods).
By this criterion, our wild-type strain was inviable in 95 of the 236 environments, as reported previously5. The mutator strain was inviable in 58 of the 236 environments (Fig. 1A). The mutator strain was thus viable in more (178 = 236−58) environments than the wild-type strain (141 = 236−95). Both strains were inviable in the same 52 environments. In other words, the wild-type strain was inviable in 43 (=95–52) environments where the mutator strain was viable, whereas the mutator strain was inviable in only 6 (=58–52) environments where the wild-type strain was inviable (Fig. 1A). The environments in which the wild-type or the mutator strains were viable harbour antimicrobials with diverse mechanisms of action. They include iron chelators like 2,2-dipyridyl19 and lawsone20, the oxidising agent diamide21,22,23, as well as several antibiotics like ciprofloxacin, polymyxin B, and vancomycin (Table S1). We sequenced the genomes of both mutator clones using Illumina HiSeq (Illumina, CA, USA) to at least 30-fold genomic coverage per clone to identify candidate mutations that may have increased the mutator’s viability, and found five such mutations (Supplementary note S1, Table S2 for the details).
Starting from the ancestral mutator strain, we next conducted five separate evolution experiments identical to those we previously described for the wild-type strain5. Specifically, we performed each of these experiments in five different environments, where each environment contained one of five different single antibiotics. We refer to these environments as simple evolution environments. The antibiotics are ampicillin (amp), azithromycin (azi), nalidixic acid (nal), streptomycin (strep), and trimethoprim (tri). We choose these antibiotics because they have distinct cellular targets and different modes of action24,25.
Analogous to our previously described experiments with the wild-type strain, we evolved eight replicate populations of the mutator strain for ~100–200 generations in each of the five simple environments (Methods, Table S3), until all the populations could grow at the IC90 of the antibiotic they evolved in. The IC90 is the concentration of an antibiotic that kills 90% of all cells in the wild-type ancestral strain1. At the end of the evolution experiment, we identified two representative evolved clones from each antibiotic environment for further analyses. We chose these clones to represent the central tendency of the growth rates of the evolved populations (Methods, Table S4 and Fig. S2). We then sequenced all evolved clones using Illumina HiSeq to at least 30-fold coverage (Illumina, CA, USA, Methods), which also confirmed that all evolved mutator clones retained the 103 bp insertion upstream of the mutL gene that endows them with their higher mutation rate.
As expected, we observed significantly more genomic mutations in the evolved mutator clones than in the evolved wild-type clones (Fig. 1B, Two-way ANOVA, F = 115.77, df = 1, p = 8 × 10-7, Methods). This difference ranged between a threefold greater number of mutations for the mutator in the streptomycin environment, and a twelve-fold greater number in the trimethoprim environment. Thus, the evolved mutator clones did not only experience more mutations as a result of their higher mutation rate, they also retained more mutations after experimental evolution.
We next asked whether our evolved strains had become viable in any of the 236 phenotyping environments of the Biolog phenotyping microarrays. We called viability novel in a given environment if both clones that had evolved on the same antibiotic were able to survive and grow in this environment, even though neither ancestral strains were able to. For instance, both the wild-type and mutator ancestral clones could not grow on the antibiotic spectinomycin which inhibits protein synthesis26, but both mutator clones evolved in ampicillin could. This novel ability was without any immediate benefit, because spectinomycin was not present in the ampicillin environment in which the clones evolved. In four out of five simple environments, except the evolution environment with streptomycin, wild-type clones had evolved more novel traits than mutator clones. However, this difference was not statistically significant (Two-sided Wilcoxon rank sum test, W = 14.5, n = 5 and 5, p = 0.78). The antimicrobials on which these clones had evolved viability inhibit growth through diverse mechanisms but these mechanisms differed from that of the antibiotic in the respective evolution environment in many phenotyping environments (Supplementary table 10.)
More importantly, however, evolved mutator clones did not show more novel traits (Fig. 1C), even though they experienced more mutations and retained more of the resulting genetic variation during evolution (Fig. 1B). Thus, the supply of mutations does not limit the evolution of latent novel traits at mutation rates that exceed those of the wild-type.
Complex antibiotic environments facilitate the emergence of latent novel traits
We next investigated the role of the selection environment in the evolution of novel traits without immediate benefit. So far we had studied simple environments that harboured only a single antibiotic. We next evolved our strains in three complex environments which contained more than one antibiotic (Fig. 2A). Two out of the three environments contained three antibiotics each. Specifically, environment 3A1 (for ‘three antibiotics’) harboured ampicillin, nalidixic acid, and trimethoprim. Environment 3A2 harboured azithromycin, nalidixic acid, and streptomycin. The remaining complex environment ‘5A’ contained all five antibiotics that we had used in the evolution experiment with simple environments (Fig. 1C). We performed six evolution experiments in the complex environments, three for the wild-type strain and three for the mutator strain. In each experiment we evolved eight replicate populations of wild-type or mutator E. coli until the populations could grow at the IC90 of each of the antibiotics present in the environment. We used a procedure which ensured that the populations evolved for a similar amount of time to acquire this ability (∼600 generations or ∼147 days, Methods, Supplementary note S2). After experimental evolution, we chose two representative evolved wild-type and mutator clones from each of the three environments for phenotyping and genome sequencing (2 clones × 2 strains × 3 environments = 12 clones in total, Methods, Table S5 and Fig. S4).
The greater the number of antibiotics present in the evolution environment was, the greater was the percentage of phenotypic environments in which our strains gained viability during experimental evolution. This holds both for the wild-type strain (Fig. 2B, Spearman’s correlation, n = 8, R = 0.87, p = 0.004, Methods) and for the mutator strain (Fig. 2C, Spearman’s correlation, n = 8, R = 0.82, p = 0.012, Methods).
In sum, increasing the complexity of the environment in which our strains evolved increases the number of novel traits without immediate benefit, irrespective of the mutation supply. Thus, the selection environment plays the predominant role in the evolution of novel traits without immediate benefits.
The nature of genetic variation determines the extent of novel trait evolution
An important confounding factor in our analysis is the time that our populations spent evolving. Specifically, evolution in complex environments lasted for almost ~600 generations, approximately three times longer than evolution in simple environments (~100–200 generations). As a result, wild-type and mutator populations that evolved in complex environments have experienced more mutations than their counterparts that evolved in simple environments. In consequence, the higher incidence of novel trait evolution in complex environments (Fig. 2B, C) could be caused by this higher number of mutations, as a result of the longer time our populations spent evolving in complex environments. In fact, not only the supply of mutations but also the number of mutations retained after experimental evolution is significantly higher in complex environments for both the wild-type and mutator strain (Fig. S5).
To control for this confounding factor, we first quantified the partial correlation between environmental complexity and the number of evolved novel traits, while controlling for the number of generations. In this analysis, high environmental complexity remained significantly associated with a high number of evolved novel traits, both for the wild-type (partial Spearman’s R = 0.95, n = 8, p = 0.0007) and the mutator strain (partial Spearman’s R = 0.95, n = 8, p = 0.0009).
Second, we compared the extent of novel trait evolution for clones evolved in the two kinds of complex environments (3A and 5A), because our populations had evolved for an identical amount of time in these environments. This analysis suggests a predominant role of environmental complexity, but not for the number of mutations, in driving novel trait evolution (Fig. S6). However, it also lacks statistical power, because we evolved populations only in a single environment containing five antibiotics.
Third, we compared mutator clones evolved in simple antibiotic environments to wild-type clones evolved in complex antibiotic environments. A simple calculation shows that these two kinds of clones experienced a similar number of mutations during experimental evolution (Supplementary note S3). In other words, the increased time spent by wild-type clones in complex antibiotic evolution environment compensated partly for the higher mutation rate of mutator clones. In addition, our genomic analysis showed that the number of genetic variants retained after evolution did not differ significantly between the mutator clones evolved in the simple environments and wild-type clones evolved in the complex environments (Fig. 3A, Two-sided Wilcoxon rank sum test, n = 10 and 6, W = 28, p = 0.86). Based on these observations, we reasoned that it may be appropriate to compare novel trait evolution between these two types of clones. This comparison also supports our previous observations. That is, wild-type clones from complex environments evolved a significantly higher number of novel traits without immediate benefit than mutator clones evolved in simple environments (Fig. 3B, Two-sided Wilcoxon rank sum test, n = 5 and 3, W = 0, p = 0.035). Once again, selection exerted by the evolution environment, and not the mutation supply or the amount of genetic variation retained during evolution, is the key force behind the evolution of novel traits without immediate benefits.
Complex environments preferentially select for pleiotropic mutations that can increase viability in multiple environments
We next asked what kinds of genetic variation may be responsible for the higher prevalence of novel traits in complex environments. To answer this question, we compared again the mutator clones evolved in simple environments and the wild-type clones evolved in complex environments, because they received a similar mutation supply, and harboured similar amounts of genetic variation.
Specifically, we first focused on genetic variants in genes that encode cellular targets of antibiotics in the evolution environment, or proteins that directly interact with such targets. Mutations in such genes do not only cause resistance to these antibiotics, but have pleiotropic effects that can bring forth novel traits1,5,27,28. We found pertinent variants in all evolved clones, except for one mutator clone evolved in azithromycin (Fig. 3C, Table S6). However, clones evolved in complex environments harboured many more mutations in antibiotic target genes than clones evolved in simple environments. Specifically, six evolved clones from complex environments harboured 24 variants in genes encoding antibiotic targets, whereas the greater number (ten) of mutator clones evolved in simple environments harboured merely 16 variants in such genes. Many of these variants have known pleiotropic effects. For example, the gyrA gene was mutated at least once in all the wild-type clones evolved in complex environments, but only in two of the mutator clones evolved in a simple antibiotic environment, namely that harbouring nalidixic acid. Mutations in gyrA can confer not just resistance against nalidixic acid, but also against β-lactams and aminoglycosides, likely by modifying the supercoiling of DNA and global gene expression with it28,29. Similarly, one of the wild-type clones evolved in the 3A2 environment, which contained streptomycin, harboured a mutation in the gene infB (Fig. 3C), whereas none of the two mutator clones evolved on streptomycin harboured a mutation in infB. The gene encodes the translation initiation factor IF-2, which interacts closely with the 30S ribosomal subunit, the cellular target of streptomycin30,31. Mutations in infB can also confer resistance against macrolide antibiotics that target the 50S ribosomal subunit in protein synthesis32. In a similar vein, one of the wild-type clones evolved in the 3A1 environment, but none of the mutator clones evolved on a single antibiotic, harboured a mutation in the gene ampC. Mutations in this gene can confer resistance to carbapenems and the combination drugs ceftolozane-tazobactam and ceftazidime-avibactam33,34. Likewise the gene ftsI was only mutated in one of the wild-type clones from 3A1 environment (Fig. 3C). The genes codes for peptidoglycan D,D-transpeptidase35, and mutations in it can also increase resistance to mecillinam, cephalexin, and sefsulodin36. Taken together, these observations suggest that complex environments select for the spreading of pleiotropic mutations that can affect viability in multiple environments.
Further support for this hypothesis comes from a closer examination of genes that encode proteins involved in multi-drug efflux, which are well-known to have pleiotropic effects in multiple environments37,38,39,40. Overall, we found twenty-nine mutations across ten different genes implicated in multi-drug resistance. Twenty-five of these mutations occurred in the wild-type clones evolved in complex environments, while only three occurred in mutator clones evolved in the simple environments (Fig. 3D, Table S6). Affected genes included those encoding global transcription regulators, such as emrR (mprA)37,41, efflux pumps, such as mdt38 and yojI40,42, as well sensory and regulatory proteins that respond to environmental stress, such as envZ39,43 and phoQ28. Mutations in emrR can increase resistance to many antibiotics, such as nalidixic acid, nitrofurantoin and erythromycin, while mutations in the gene phoQ can also confer increased resistance against antibiotics with diverse cellular targets, such as ampicillin, ciprofloxacin, nalidixic acid, kanamycin and tobramycin28. Similarly, mutations in the gene envZ can increase resistance to ampicillin, cefoxitin, ciprofloxacin, chloramphenicol, erythromycin, and tetracycline28. Most notably, all wild-type clones evolved in complex environments harboured at least one mutation in the genes coding for the AcrAB-TolC efflux pump, while none of the mutator clones evolved in simple antibiotic environments did. AcrAB-TolC is an important multi-drug efflux system in E. coli that exports antibiotics with diverse cellular targets, as well as non-antibiotic toxins44,45.
In sum, these observations help explain why complex antibiotic environments promote the evolution of latent traits without immediate benefit. They promote the spreading of pleiotropic mutations that can help bacteria become resistant against multiple antibiotics they encounter during evolution. As a by-product, these mutations also convey viability in other environments that the bacteria have not encountered.